ADVANCING INTEGRATED FIELD CROP DISEASE MANAGEMENT THROUGH EPIDEMIOLOGICAL INSIGHTS INTO PHYLLACHORA MAYDIS AND SCLEROTINIA SCLEROTIORUM By Jillian Clara Check A DISSERTATION Submitted to Michigan State University in partial fulfillment of the requirements for the degree of Plant Pathology – Doctor of Philosophy 2025 ABSTRACT Due to the humid continental climate of Michigan, field crops suffer from a plethora of fungal pathogens causing infections of roots, stems, foliage and reproductive structures. In 2022, corn, soybean, potatoes and dry beans accounted for 6.2 million acres in Michigan, approximately 65% of the state’s agricultural production land. Phyllachora maydis is a fungal pathogen of corn that is emerging in the region, causing tar spot of corn. Sclerotinia sclerotiorum is a long- established fungal pathogen across many economically important hosts produced in Michigan, including leguminous and vegetable crops. Both P. maydis and S. sclerotiorum are highly influenced by their environment, resulting in erratic and unpredictable year-to-year economic impacts due to the prevailing weather conditions regionally. Therefore, the focus of this dissertation was to improve our understanding of these pathogens’ interactions with their environment to create new disease management recommendations and tools. In chapter 1, I provide a review of the current literature on P. maydis and S. sclerotiorum biology, infection strategy and the management of their respective diseases. I also briefly discuss the use of spore traps in plant pathogen epidemiology research and disease management, as well as the history and current state of predictive modeling for white mold management. In chapter 2, I conducted a field trial to address key questions about cultural management of tar spot of corn. This study established that neither high nor low nitrogen application rate influences tar spot development and that low planting density increases tar spot severity. However, an economic analysis of the optimal planting density revealed that increasing planting density to reduce tar spot severity was not a viable management strategy. Therefore, other disease management strategies, such as planting partially resistant hybrids, should be employed rather than altering established agronomic practices. In chapter 3, I used a previously developed P. maydis qPCR assay, Burkard and rotating- arm spore traps to quantify spore capture in tar spot-infested fields. Using data collected from 6 different environments, logistic regression modeling showed that spore release was negatively related to maximum precipitation rate, minimum and mean temperature, and maximum relative humidity, and positively related to minimum wind speed. The best performing logistic regression model used mean temperature and maximum relative humidity to predict the capture of P. maydis spores, achieving a balanced accuracy of 85%. As the spore traps were used in this study, they were not able to detect spores prior to the onset of visible disease symptoms but were able to detect spores prior to tar spot incidence reaching 100%. With future improvement of spore trap deployment and qPCR sensitivity, Michigan farmers could benefit from an early tar spot warning system to aid in expediting disease management decision making. In chapter 4, I used on-site weather monitoring systems and supervised machine learning to predict S. sclerotiorum apothecia presence in irrigated environments. Apothecia monitoring was conducted in soybean, dry bean and potato fields representing 20 site-years to develop a multi-crop apothecia prediction model. Decision tree (DT) models performed best, with an accuracy of 77% when developed using gridded weather data and 89% after incorporating soil weather data collected on-site. Interpretability of models did not have to be sacrificed to achieve a high prediction accuracy, as the DT developed using untransformed features performed comparably to the DT developed using principal components. Through this study, a robust framework for developing multi-crop S. sclerotiorum apothecia prediction models was demonstrated, but future model validation will be required to assess the value of the prediction models for SSR disease control. Lastly, in chapter 5, I discuss the conclusions and impacts from the work described herein. Overall, these studies contribute to the current understanding of P. maydis and S. sclerotiorum epidemiology and the management of their associated diseases on Michigan field crops. Copyright by JILLIAN CLARA CHECK 2025 This dissertation is dedicated to my Aunt Jeanne and Uncle José, for their love, encouragement and support. v ACKNOWLEDGEMENTS To begin, I would like to thank my advisor, Dr. Martin Chilvers, for his guidance and support. Thank you for allowing me the freedom to learn in your lab while leading with a gentle hand. To Dr. Austin McCoy and Dr. Emily Roggenkamp, thank you for showing me the ropes and for always sharing your advice. Thank you to our lab and field management teams, Janette Jacob, Micalah Herendeen, Bill Widdicombe, Adam Byrne and John Boyse, as this research would not have been possible without you. Finally, thank you to my committee, Dr. Jaime Willbur, Dr. Younsuk Dong and Dr. Richard Wade Webster for sharing your knowledge, expertise and feedback on my work. For their support and partnership, I would like to acknowledge my external collaborators from the following institutions: North Dakota State University, Purdue University, University of Minnesota, University of Wisconsin-Madison and the USDA-ARS. Thank you also to my MSU collaborators, including those from the Forest Pathology lab, Small Fruit and Hop Pathology lab, Sugar Beet and Potato Pathology lab, Plant and Pest Diagnostics lab, Cropping Systems Agronomy lab, Soil Fertility and Nutrient Management lab and MSU Extension. Thank you to my funding sources for supporting this work, including the National Predictive Modeling Tool Initiative, Project GREEEN, Corn Marketing Program of Michigan, and National Corn Growers Association. Thank you to the facilitators of the National Science Foundation IMPACTS Research Traineeship Program for affording me the opportunity to specialize in computational plant pathology. Finally, thank you to my friends and family. Thank you first and foremost to my partner Erik Dams. I will forever be grateful for your love and support and am excited for our next chapter in life together. Thank you to my father, Grandma, Aunt Jeanne, Uncle José and the rest of the Checks for always reassuring me I am smart and capable. Thank you to my mother, Grammy, Aunt Wendy and the rest of the Westbergs for being my biggest supporters. Thank you to my best friend, Jordan Priebe, for your love, encouragement, and providing me much-needed solace from graduate school. Finally, thank you to Rebecca Harkness and Lexi Heger for the endless hours of scientific discussion, wonderful comedic relief, and friendship. You both have taught me so much, and I am a better scientist having known you. I am so excited to see your careers develop and will always be cheering for you. vi TABLE OF CONTENTS CHAPTER 1: LITERATURE REVIEW .................................................................................... 1 Phyllachora maydis background ................................................................................................. 1 Tar spot disease management ...................................................................................................... 3 Spore trapping for botanical epidemiology and disease management ......................................... 6 Sclerotinia sclerotiorum background ........................................................................................... 7 White mold disease management ................................................................................................ 9 Sclerotinia sclerotiorum predictive modeling ........................................................................... 12 LITERATURE CITED .............................................................................................................. 15 CHAPTER 2: EFFECTS OF NITROGEN APPLICATION RATE AND PLANT DENSITY ON SEVERITY OF TAR SPOT OF CORN ............................................................................. 28 Summary ................................................................................................................................... 28 Source ........................................................................................................................................ 28 CHAPTER 3: UNRAVELING THE ENVIRONMENTAL DRIVERS OF PHYLLACHORA MAYDIS SPORE RELEASE USING SPORE TRAPPING AND QPCR .............................. 29 Abstract ..................................................................................................................................... 29 Introduction ............................................................................................................................... 30 Materials and Methods .............................................................................................................. 34 Results ....................................................................................................................................... 39 Discussion ................................................................................................................................. 42 LITERATURE CITED ............................................................................................................. 58 APPENDIX ............................................................................................................................... 66 CHAPTER 4: MULTI-CROP SCLEROTINIA SCLEROTIORUM APOTHECIA PREDICTION MODELS FOR IRRIGATED ENVIRONMENTS ARE IMPROVED BY ON-SITE WEATHER MONITORING AND SUPERVISED MACHINE LEARNING ..... 68 Abstract ..................................................................................................................................... 68 Introduction ............................................................................................................................... 69 Materials and Methods .............................................................................................................. 73 Results ....................................................................................................................................... 79 Discussion ................................................................................................................................. 83 Conclusions and future directions ............................................................................................. 87 LITERATURE CITED ............................................................................................................. 98 APPENDIX ............................................................................................................................. 105 CHAPTER 5: CONCLUSIONS ............................................................................................... 111 Conclusions ............................................................................................................................. 111 List of published and submitted works from the duration of my PhD ..................................... 114 vii CHAPTER 1: LITERATURE REVIEW Phyllachora maydis background Phyllachora maydis Maubl. is an obligate biotrophic fungal plant pathogen responsible for tar spot disease in corn (Zea mays L.). Tar spot is an emerging disease with significant economic consequences for corn production in the U.S. and Canada. Following its initial identification in Illinois and Indiana in 2015 (Ruhl et al. 2016), P. maydis has been confirmed in 20 U.S. states and 2 Canadian provinces (https://corn.ipmpipe.org/tar-spot/). Between 2018 and 2023, annual yield losses from tar spot were estimated at 106 million bushels, with an economic impact of approximately 521 million USD (Crop Protection Network 2025a). P. maydis was first reported in Mexico in 1904 (Maublanc 1904) and has since been documented across Central and South America (Mottaleb et al. 2019). There, it is part of a disease complex involving interactions with Monographella maydis E. Müll. & Samuels (syn. Microdochium maydis) and the mycoparasite Coniothyrium phyllachorae Maubl. (Hock et al. 1992). However, in the U.S., P. maydis appears to be the sole pathogen responsible for tar spot and associated yield losses (Caldwell et al. 2024; Luis et al. 2023; McCoy et al. 2019). Multiple possible routes of introduction of P. maydis into the U.S. have been proposed, including transport of infected plant material across borders (Mottaleb et al. 2019) or transport of spores by weather events originating in Mexico or Central America (Ruhl et al. 2016). The emergence and spread of tar spot were aided by the lack of genetic resistance in commercially available hybrids (Rocco da Silva et al. 2021) and the similarity in climate between the native range of P. maydis and the North Central US (Mottaleb et al. 2019). Once in contact with susceptible host tissue, an ascospore forms an appressorium to penetrate the leaf surface (Parbery 1963). The hyphal growth within the epidermal cells facilitates colonization, resulting in the formation of the clypeus, a shield-like structure composed of both host epidermal cells and melanized hyphae, which protects the developing fungal hyphae and fruiting bodies (i.e., perithecia and pycnidia) within the forming stroma (Caldwell et al. 2024; Parbery 1963). Hyphal growth extends longitudinally and laterally into the leaf tissue, causing stroma expansion on the abaxial epidermis (Caldwell et al. 2024). Pycnidia develop at the center of the clypeus, surrounded by perithecia. Histological studies by Caldwell et al. (2024) suggested that asexual conidia released from pycnidia function as spermatia, providing the complementary mating-type idiomorph necessary for perithecia 1 formation and the production of ascospores, thus contributing to the characteristic ascomata development in host leaf tissue. This hypothesis is supported by MacCready et al. (2023), who showed that P. maydis is heterothallic, requiring compatible mating types for sexual reproduction. This was confirmed using sequence information from a mixed population of ascospores and conidia, where nearly equal proportions of mat1-1 and mat1-2 mating type idiomorphs were identified. Additionally, inoculation experiments revealed that using primarily conidia resulted in poor disease symptom development (less than 10% incidence or no symptoms), while inoculating with ascospores led to 100% incidence, further suggesting that ascospores are primarily responsible for disease development and conidia act as spermatia (Breunig et al. 2023). Ultimately, perithecia mature, swell, and release ascospores through the ostiole, completing the cycle of inoculum production and infection in the polycyclic disease process. The P. maydis genome assembly PM_02 utilized transcriptomic analysis to describe gene expression during the tar spot lesion formation (MacCready et al. 2023). As common in obligate fungal pathogens, significant gene loss due to genetic drift was found, especially in the degradation of inorganic nitrogen utilization pathways. However, transcriptomic analysis found the nitrogen metabolic pathway was among the most highly expressed during the early infection process, with glutamine synthetase and glutamine-fructose-6-phosphate transaminase among the highest expression genes within the nitrogen assimilation pathway. These results together suggest that P. maydis primarily uses ammonium, glutamate and glutamine as nitrogen sources rather than nitrate and nitrate, which requires energy-consuming assimilation into ammonium. Additionally, P. maydis prioritizes autophagy and secretion pathways, which have previously been implicated in pathogenesis, at the expense of DNA replication and cell division pathways during early infection (MacCready et al. 2023). Little is currently known about the molecular host-pathogen interactions of P. maydis due to the difficulty of working with an obligate pathogen. Subcellular effector protein localization targets multiple compartments including the host nucleus, nucleolus, plasma membrane, stroma of the chloroplasts and cytosol, but further work is needed to characterize the function of these candidate effector proteins (Helm et al. 2022; Rogers et al. 2024). One putative effector, PM02_g115, was found to have structural similarity to the PevD1 effector from Verticillium dahlia which is involved in disrupting host defenses through suppression of the GhPR5 protein 2 to facilitate disease development on cotton (Rogers et al. 2024). Using a heterologous expression system in Nicotiana benthamiana, three candidate effector proteins, PM02_g115, PM02_g7882 and PM02_g8240 were found to suppress chitin-mediated reactive oxygen species production, suggesting an important role in suppressing host immune responses (Rogers et al. 2024). Future work on molecular-host pathogen interactions in P. maydis and corn will be aided by the development of multiple methods for artificial inoculation in controlled environments (Breunig et al. 2023; Gongora-Canul et al. 2023; Solorzano et al. 2023). A recent investigation by Broders et al. (2022) explored the genetic variation within the Phyllachora genus by performing a phylogenetic analysis on samples gathered from diverse geographic regions and grass species using the internal transcribed spacer (ITS) gene region. The authors identified five distinct genetic groups, each corresponding to a unique ITS haplotype. Genetic cluster 1 was composed exclusively of contemporary P. maydis isolates from the U.S., while 2 and 3 included both modern and historical P. maydis isolates from North, Central, and South America. Interestingly, genetic cluster 3 grouped P. maydis isolates with other Phyllachora species, such as P. vulgata, P. sylvaticum, P. rottboellia, P. junci, P. heraclei, P. graminis, P. euphoribaceae, P. epicampis, P. diplocarpa, and P. chaetocloae, found across the world. This finding challenges the earlier assumption that Phyllachora species are strictly host- specific, given their obligate biotrophic nature (Mardones et al. 2017; Parbery 1967). The existence of multiple ITS haplotypes aligns with McCoy et al. (2019), who conducted a fungal community network analysis using the ITS1 region on corn leaf samples from a tar spot-affected field in Michigan. Their results indicated that 95% of the samples were coinfected with at least two distinct P. maydis operational taxonomic units (OTUs), underscoring the high level of genetic diversity and coinfection within the P. maydis population in that region. For a complete overview of P. maydis and tar spot history, biology, epidemiology, management and perspectives on future research directives, see da Silva, C. R., Check, J. C. et al. 2021. Tar spot disease management Effective tar spot management integrates genetic, chemical and cultural control methods. There is no complete resistance available in commercial hybrids, but planting moderately resistant hybrids has a significant impact on tar spot development and yield loss (Check et al. 2023; Pereyda-Hernandez et al. 2009; Ross et al. 2023). Screening exotic and native corn populations have identified sources of resistance to tar spot in inbred breeding lines that can be 3 used to map candidate resistance loci and improve resistance in commercial hybrids (Lipps et al. 2022; Singh et al. 2023). Using single-nucleotide polymorphism and genome-wide association mapping, quantitative trait loci (QTL) for tar spot resistance have been identified (Cao et al. 2021; Mahuku et al. 2016; Ren et al. 2022; Yan et al. 2022). A major QTL qRtsc8-1 accounted for 18-43% of the variation in observed tar spot resistance phenotypes, and a haplotype within the region occurring in 3.5% of the population increased resistance by 14% (Mahuku et al. 2016). This major QTL was confirmed by Yan et al. 2022 who identified an additional 4 minor QTLs explaining 10.18%, 12.72%, 10.14% and 13.38% of phenotypic variance in tar spot resistance. Fine mapping of the major QTL qRtsc8-1 found it to be flanked by two genes encoding a membrane protein-like and a leucine-rich repeat receptor-like protein kinase, both of which play a role in basal defense responses in the host (Ren et al. 2022) With the identification of tar spot resistance QTLs, marker-assisted and genomic selection can be further improved to advance tar spot resistance breeding efforts for the management of tar spot (Cao et al. 2021; Ren et al. 2022). With a lack of strong genetic resistance available in commercial hybrids, fungicides are critical tools in tar spot disease management. Quinone outside inhibitors (QoI), demethylation inhibitors (DMI) and succinate dehydrogenase inhibitors (SDHI) are the primary modes of action (MOA) used for foliar corn disease control in the United States and have all showed efficacy against tar spot (Ross et al. 2024; Telenko et al. 2022a, 2022b; Wise et al. 2019). Formulations containing two or three MOAs were found to reduce tar spot severity significantly more than single MOAs, and yield losses were significantly reduced compared to the non-treated control when using fungicides containing three MOAs (Telenko et al. 2022b). Fungicide application efficacy is also dependent on timing (Ross et al. 2024; Webster et al. 2023a). Single fungicide applications made prior to the VT/R1 growth stage do not provide adequate residual activity to protect against tar spot through harvest (Ross et al. 2024). Based on multistate trials from 2019 and 2021, single fungicide applications made between R2 and R3 have been shown to offer the most consistent results in yield protection and economic return (Ross et al. 2024). However, yield protection is largely dependent on disease conditions, with higher economic returns in environments with high disease pressure. Therefore, careful consideration of field history and scouting is warranted prior to fungicide application. The development of risk prediction models can aid in fungicide application decision making by advising farmers and crop consultants of 4 periods of high risk for tar spot development (Webster et al. 2023a). In the development of Tarspotter, the tar spot decision support system, a multi-model ensemble of two logistic regression models using 30-day moving averages of mean temperature, daily maximum relative humidity and 14-day moving averages of daily total nighttime hours with relative humidity over 80% most accurately predicted tar spot development among logistic regression and machine learning models (Webster et al. 2023a). Tarspotter and similar tools will provide important input in tar spot management decision to improve the efficacy and return on investment of fungicide applications. The demonstration of consistent, effective cultural control methods for tar spot management are still lacking. Crop rotation and tillage likely do not offer significant disease control as fields with no previous history of tar spot have been found to become heavily infested in a single growing season (da Silva et al. 2021; Ross et al. 2023). Observations of tar spot development under center pivot irrigation has found that frequent irrigation events can exacerbate tar spot development compared to non-irrigated dry corners (da Silva et al. 2021). Initial investigations into fertility management have found that nitrogen application rate has no effect on tar spot severity when applied in furrow at early middle vegetative growth stages (Check et al. 2023), but further research is needed to investigate other micro- and macro- nutrients and nitrogen application timings. Planting density has an inverse relationship with tar spot severity, but reducing planting rate is not an economically viable option for managing tar spot (Check et al. 2023). Previous descriptions of the casual agents of tar spot disease complex in Central and South America included the hyperparasite Coniothyrium phyllachorae (Hock et al. 1992). The existence of a hyperparasite of P. maydis suggests that biological control may be an effective tar spot disease management method. While the presence of C. phyllachorae in the U.S. has not been confirmed, multiple potential biocontrol agents for tar spot control have been proposed using culture-based and culture-independent molecular methods for describing the P. maydis stromata microbiome (Johnson et al. 2023; McCoy et al. 2019). From overwintered stromata, Johnson et al. 2023 isolated bacterial and fungal species and assessed their biological control activity against P. maydis, which they defined as the repression of stromata germination following application of a spore suspension or commercial formulation. The authors isolated 7 bacterial species and 2 fungal species with previously reported biological control properties, and 5 demonstrated biological control properties of 2 fungal species, A. alternata/A. arborescens and Cladosporium rectoides. Biological control of P. maydis using Gliocladium catenulatum formulated as the commercial biofungicide LALSTOP G46 WG (Lallemand, Montreal, CA) was also demonstrated. Through fungal community network analyses using ITS barcoding, McCoy et al. 2019 identified Fusarium sporotrichiodes and Paraphaeosphaeria neglecta as negatively associated with tar spot lesions, suggesting potential biological control activity. However, these fungal species were also positively associated with P. maydis in samples with “fish-eye” lesions, a secondary symptomology of tar spot where a necrotic halo surrounds the black stromata. Further research is warranted to observe interactions between P. maydis and these proposed biological control agents to identify the most promising species and describe their modes of action. Spore trapping for botanical epidemiology and disease management Spore traps have been fundamental tools for examining environmental factors influencing spore release for many years (Granke et al. 2014; Hernandez and KC 2024; King and Polley 1976; Pearson et al. 1980; Timmer et al. 1998). Using local weather data and site-specific management factors, models are built to identify conditions that promote pathogen sporulation and release. Among the various types, impaction traps, particularly the Hirst-type and rotating- arm models (Check et al. 2024a; Hirst 1952; West and Kimber 2015), are widely favored by plant disease epidemiologists. Spore traps offer crucial advantages for real-time monitoring of inoculum presence and its progression over the course of a growing season (Thiessen 2024). Paired with molecular detection methods, these tools can identify airborne spores before visible disease symptoms appear, enabling early scouting and prompt disease management decisions (Munir et al. 2020; Pizolotto et al. 2021). Furthermore, integrating inoculum data into disease forecasting models can enhance the precision of fungicide application timing, optimizing disease control and maximizing return on investment of fungicide applications (Newlands et al. 2018; Van der Heyden 2021). Spore trap networks are commonly found in regions where high-value crops are grown intensively, such as wine grapes in California, Oregon, and France, or cucurbits in the Eastern U.S. (Coastal Viticulture Consultants, https://www.coastalvit.com; Laurent et al. 2022; Rahman et al. 2021; Theissen et al. 2016). Such networks have also been suggested for managing diseases of commodity crops like soybean rust (Isard et al. 2011) and Fusarium head 6 blight in wheat (Wang et al. 2024a). Recently, the private sector has shown increased interest in inoculum surveillance, with companies offering advanced pathogen detection systems to growers using various trapping and quantification technologies (Pollensense™ (www.pollensense.com); Root Applied Sciences (www.rootappliedsciences.com); Scanit Technologies, Inc (www.scanittech.com); Spornado (www.spornadosampler.com)). However, supporting the infrastructure and personnel required to manage a spore trap network employing any of these technologies is challenging, and each trapping method will have its unique challenges and weaknesses. For further discussion on rotating-arm spore trap applications in plant pathology research and their limitations, see Check et al. 2024a and Check et al. 2024b. Sclerotinia sclerotiorum background Sclerotinia sclerotiorum (Lib.) de Bary is a necrotrophic plant pathogen with a global distribution (CABI 2005) and an extensive host range compromising of plant species in 64 families (Purdy 1979). On soybean alone, S. sclerotiorum is responsible for an estimated yearly loss of US$4.12 per acre from 2013 to 2023 (Crop Protection Network 2025b) and consistently ranks in the top 10 most destructive diseases in the Northern United States (Bradley et al. 2019). Due to its large host range, the diseases caused by S. sclerotiorum are known under a variety of names, but is known to cause white mold, also known as sclerotinia stem rot, on soybeans, dry beans and potatoes in Michigan. The white mold disease cycle is initiated by the germination of sclerotia, the pathogen’s overwintering survival structure made of tightly-packed melanized hyphae (Grau and Hartman 2015). Carpogenic germination of sclerotia produces apothecia to facilitate ascosporic infection, while myceliogenic germination produces hyphae to facilitate basal stem infection (Willbur et al. 2019a). However, it is believed that basal stem infection less epidemiologically relevant as it occurs rarely under field conditions (Abawi and Grogan 1975; Newton and Sequeira 1972; Willets and Wong 1980). Sclerotia buried in the top 2-3 cm of soil are most likely to successfully produce stipes long enough to reach the soil surface, as stipes longer than 3 cm are rarely produced in nature (Abawi and Grogan 1979). Conditioning of sclerotia requires 8 weeks of cool temperatures (8-16ºC) and 1 to 2 weeks of soil water potentials of -100 kPa and temperatures between 10 and 25ºC for 1 to 2 weeks to form apothecia (Clarkson et al. 2004; Dillard et al. 1995). Apothecia development requires proper light quality (between 276 and 319 nm) which is aided by the canopy development (Thaning and Nilsson 2000; Fall et al. 2018). 7 Under high light intensity (120-130 mol m-2 s-1), larger and more apothecia are developed than at low light intensity (80-90 mol m-2 s-1) (Sun and Yang 2000). Ascospores borne on the apothecia are forcibly ejected from their ascus in response to changes in relative humidity to cooperatively generate air flow to increase their dispersal distance before interception (Hartill and Underhill 1976; Roper et al. 2010). Ascospores land on flowering or other senescent tissue of nearby plants to initiate infection, which is favored by cool temperatures (15-25ºC) and 2 to 4 hours of free water on the plant surface (Abawi and Grogan 1979; Young et al. 2004). Following infection by ascospores, stem tissue becomes colonized by S. sclerotiorum to produce white, fluffy mycelia, water-soaked lesions, wilting and sclerotia. During harvest, sclerotia that were developed in the infected host’s stem tissue are deposited in the soil to serve as the future growing seasons’ inoculum source. S. sclerotiorum has a large arsenal of cell wall degrading enzymes (CDWEs) used to infect and colonize its host. The S. sclerotiorum genome is estimated to contain 106 carbohydrate active enzymes and affiliated proteins (Amselem et al. 2011). A large repertoire of CDWEs is primary characteristic of plant pathogens classified as having necrotrophic lifestyles. However, the infection strategy of S. sclerotiorum is described in two distinct phases, first suppressing host defenses to establish infection by secreting oxalic acid (OA) and effector proteins (Williams et al. 2011; Zhu et al. 2013), then using CDWEs to acquire nutrition from its host, which more closely resembles the lifestyle of hemibiotrophic plant pathogens (Kabbage et al. 2015). OA is a primary virulence factor of S. sclerotiorum, and a loss of pathogenicity is observed in OA knockout mutants (Kabbage et al. 2013). OA plays multiple roles throughout the infection process, including acidifying the infection court to increase the activity of CWDEs, decreasing the host’s production of reactive oxygen species (ROS), and as a calcium chelating agent. In contrast to the role of OA in host immunity suppression during infection establishment, OA also promotes host cell apoptosis to enhance nutrient acquisition later in the infection process by increasing host ROS production (Uloth et al. 2015; Williams et al. 2011). Regional populations of S. sclerotiorum are believed to be largely clonal due to the low dispersal potential of ascospores and homothallic life cycle (Anderson and Kohn 1995; Cubeta et al. 1997). Diversity within populations can be introduced through the dissemination of sclerotia by the movement of soil between fields by contaminated equipment, clothing, animal feed or irrigation runoff (Saharan and Mehta 2008). Early S. sclerotiorum population genetic 8 studies utilized mycelial compatibility groups (MCGs) and restriction fragment length polymorphism (RFLP) DNA fingerprinting to describe genetic diversity (Cubeta et al. 1997; Hambleton et al. 2002; Kohli et al. 1992, 1995). However, grouping by MCG and RFLP fingerprints may not be sufficient as MCG grouping does not measure the relative genetic relationship to isolates from other MCGs and RFLP is limited in the number of unique fingerprints that can be resolved accurately (Attanayake et al. 2019). Instead, microsatellite markers are more capable of describing genetic diversity with specific primers designed for multiple polymorphic loci. Recent studies using microsatellite markers detected random association of alleles, hinting at possible outcrossing to explain genetic diversity observed within S. sclerotiorum populations (Atallah et al. 2004; Attanayake et al. 2012; Attanayake et al. 2014; Chitrampalam et al. 2015). However, the mechanism for genetic recombination remains unclear. As there is no current evidence of fertilization through microconidia, as seen in the S. trifoliorum (Uhm and Fujii 1983), one suggested mechanism of outcrossing is heterokaryon formation from distinct genotypes co-infecting the same host during sclerotia formation (Ford et al. 1995; Sexton et al. 2006). Outcrossing and stable heterokaryon formation under controlled conditions has been demonstrated to successfully produce apothecia and ascospores that segregate to both parental genotypes in equal proportion (Ford et al. 1995). Further investigations into the population structure and sexual recombination mechanisms in S. sclerotiorum populations will improve our understanding of its behavior in natural systems and its evolutionary potential. White mold disease management As white mold is a monocyclic disease, management targets the reduction of inoculum production or quantity reaching susceptible host tissues during a single critical time point (host crop flowering). In addition to preserving yield, proper management of white mold is critical for reducing sclerotia populations in the soil to curb future epidemics in years with favorable weather conditions (Peltier et al. 2012). Across soybean, dry bean and potato, white mold management integrates genetic, chemical and cultural control. Cultivar selection is an important factor in white mold control, but only partial resistance has been identified in soybean germplasm (Roth et al. 2023; Willbur et al. 2019a). Genetic resistance to white mold is controlled by many quantitative trait loci with small, additive effects, and these QTLs may contribute to either physical disease escape phenotypes (klendusity) or 9 physiological resistance mechanisms (McCaghey et al. 2017). These two forms of resistance are selected for differently based testing environment and procedures which leads to conflicting results between greenhouse and field screenings of potential genotypes. Soybean phenotypes that might favor white mold escape include increased plant height, resistance to lodging and later date of maturity (Kim and Diers 2000). In soybean, few physiological sources of resistance to white mold are well described, but studies suggest that genes involved in the phenylpropanoid pathway may be involved in antifungal compound accumulation to interrupt ergosterol biosynthesis in S. sclerotiorum in resistant genotypes (Ranjan et al. 2019; Xiao et al. 2023). An additional difficulty in breeding for white mold resistance is specific genotype by isolate interactions that result in differing disease outcomes (Willbur et al. 2017). To account for these interactions, Willbur et al. 2017 prepared a panel of nine S. sclerotiorum isolates to capture diversity populations in the soybean-growing regions of the United States to be used to challenge breeding lines. Future breeding efforts will also benefit from the use of this isolate screening panel, as well as soybean check lines, which have been selected based on reproducible results in different environments and against diverse isolates for consistent assessment of white mold resistance in breeding populations (Webster et al. 2021). For chemical control of white mold, commonly used fungicides include those from the DMI, SDHI, QoI, methyl benzimidazole carbamates (MBC) and 2,6-dinitro-aninlines classes (Willbur et al. 2019b). A meta-analysis leveraging 25 site years of soybean fungicide efficacy trial data from across the Midwestern and Eastern United States found that fungicide programs using boscalid or picoxystrobin most consistently reduced disease severity and protected yield among the top 10 most popular active ingredients (Willbur et al. 2019b). The same analysis revealed that while two-spray programs applied during flowering offered the best disease control, more economical single-spray fungicide programs applied during the R1 or R2 growth stage provided greater disease protection than applications made at the R3 growth stage (Willbur et al. 2019b). S. sclerotiorum is considered at ‘low risk’ for developing fungicide resistance development based on previous reports of resistance (FRAC 2019). A recent screening of S. sclerotiorum isolates from dry bean and soybean fields in the United States found no practical fungicide resistance in S. sclerotiorum populations to tetraconazole (DMI), boscalid (SDHI), picoxystrobin (QoI) or thiophanate-methyl (MBC), although small shifts in sensitivity to boscalid and picoxystrobin were found (Nieto-Lopez et al. 2023). In addition to fungicides, 10 chemical control through application of the herbicide lactofen is achieved by modifying canopy growth, delaying flowering and inducing systemic acquired resistance through the accumulation of phytoalexins (Dann et al. 1999; Willbur et al. 2019b). Cultural control of white mold involves modifying the microclimate within the canopy to hinder the germination of sclerotia and the progression of disease. This can be achieved through various agronomic choices. For instance, reducing plant density by increasing in-row spacing, widening between-row spacing and lowering seeding rates can all reduce the risk of white mold development (Vieira et al. 2010; Webster et al. 2023b). However, these adjustments may also lead to a reduction in overall yield. Excessive nitrogen fertilization should be avoided as it can exacerbate white mold infection by promoting vegetative growth within the host canopy (Schmidt et al. 2001; Webster et al. 2023b). Limiting irrigation prior to and during flowering helps reduce moisture and prevents cool canopy temperatures, both of which can inhibit disease development (Blad et al. 1978; Weiss et al. 1980). Planting cultivars with open canopies, upright growth habits and resistance to lodging can significantly reduce disease development (Blad et al. 1978; Kader et al. 2018; Weiss et al. 1980). Crop rotation is generally ineffective for white mold management as sclerotia can survive in the soil for up to 5 years to initiate new infections and epidemics (Adams and Ayers 1979; Mueller et al. 2002). The impact of tillage on white mold incidence is inconsistent as tillage can both bury and exhume sclerotia when overturning soil (Mueller et al. 2002; Peltier et al. 2012). Work performed by Pethybridge et al. (2020) demonstrated that planting soybeans and dry beans into a rolled-crimped rye cover crop significantly reduced competition by weeds and white mold development by encouraging apothecia germination which exhausts the infection potential of the sclerotia prior to soybean flowering. Additionally, they found that the increased shade provided by the cereal rye led to malformed apothecia which are likely less able to produce viable ascospores. However, the overall effect on yield was unclear as some yield components were found to be reduced in cover crop-planted soybeans (plant density), while some were greatly increased (pod weight). Irrigation regime also influences apothecia development, with light, daily irrigation events resulting in greater apothecia development than heavy, daily and weekly irrigation in clay loam soils, suggesting that oversaturation in heavy soils can hinder apothecia development (Twengstrom et al. 1998a). Additionally, under light, frequent irrigation, greater apothecia development was seen in loamy sand soils than loam soils, demonstrating an interaction between soil type and irrigation 11 events in determining apothecia production. Multiple fungal and bacterial biocontrol agents are labeled for white mold control (Zeng et al. 2012). Biocontrol of S. sclerotiorum relies primarily on the colonization and degradation of sclerotia in the soil to reduce local inoculum reservoirs. In pot assays, commercial biofungicide strains of Bacillus subtilis, Coniothyrium minitans, Streptomyces lydicus and Trichoderma harzianum were found to significantly reduce sclerotia by between 29.6 and 50% (Zeng et al. 2012). However, the success of in vitro laboratory or greenhouse assays are difficult to translate to economical field-level disease control (Conrad and Telenko 2023; Wang et al. 2024b). Over the past 30 years, mycoviruses from 17 different families have been identified in the Sclerotinia genus, making them potential biocontrol agents for white mold (Khan et al. 2023), but very little data is available demonstrating their efficacy in controlling white mold in comparison to fungal and bacterial biocontrol agents (Wang et al. 2024b). One exception is S. sclerotiorum hypovirulence-associated DNA virus 1 (SsHADV-1), where hypovirulence was observed in SsHADV-1-carrying strains compared to original strains on Phaseolus vulgaris (Fu et al. 2024). Biopriming of dry bean, pea and sunflower seeds with hyphal fragment suspensions of SsHADV-1 infected S. sclerotiorum strains also effectively reduced S. sclerotiorum infection by eliciting the expression of defense-related genes in the host (Fu et al. 2024). Sclerotinia sclerotiorum predictive modeling S. sclerotiorum is a prime candidate for disease forecasting due to its monocyclic disease cycle and high environmental influence. Many disease forecasts have been produced for different S. sclerotiorum host species, such as canola (Koch et al. 2007; Twengstrom et al. 1998b), carrot (Foster et al. 2011), common bean (Jones et al. 2011), lettuce (Clarkson et al. 2014) and soybean (Willbur et al. 2018a). However, there is often a lack of infrastructure to maintain and deliver disease forecasting tools (Saharan and Mehta 2008). One exception is Sporecaster, which was developed for white mold management in soybean in the North Central United States for rainfed and irrigated environments (Willbur et al. 2018a). Sporecaster is widely available to farmers, crop consultants, industry representatives and university extension personnel and free to download as a smartphone application (https://ipcm.wisc.edu/apps/sporecaster/). Sporecaster aims to inform fungicide timing based on the percent risk of apothecia presence from 30-day moving averages of weather conditions and soybean row spacing. Due to grain prices and expense of fungicide applications, multi-spray programs are often not economically feasible in soybean. 12 Therefore, fungicide timing can be optimized by targeting application during periods of high risk of infection by ascospores. During model validation, Sporecaster demonstrated to be highly accurate in predicting apothecia presence in research and commercial trials, achieving 82% accuracy (Willbur et al. 2018b). Recently, Webster et al. 2023c aimed to improve white mold risk prediction by adjusting fungicide application action thresholds based on genetic resistance level of soybean genotypes. The authors found a low action threshold (5% for irrigated and 20% for rainfed environments) resulted in the most accurate predictions of white mold development for susceptible soybean varieties and increasing the action threshold (20% for irrigated and 75% for rainfed environments) was most accurate for resistant varieties. Despite the large body of work describing risk factors of S. sclerotiorum development and disease forecasting tools, the difficulty of managing of white mold persists. Reich and Chatterton (2023) performed a scoping literature review to summarize quantitative relationships between environmental variables and life stages of S. sclerotiorum. The authors found that despite the important role of soil moisture and temperature established in laboratory studies, few studies have established these relationships using field data. Additionally, few studies have established relationships between S. sclerotiorum apothecia development and precipitation, despite a positive relationship between irrigation quantity and frequency and apothecia presence. The authors also found that relationships with environmental variables are often moderated by features typically not included in model development, such as the irrigation, soil type, row spacing and host cultivar. This scoping review offers insights into future directions for S. sclerotiorum disease forecasting through thoughtful selection of environmental and agronomic variables to better represent the underlying biological processes in the S. sclerotiorum life cycle. Mechanistic models operate based on the description and quantification of underlying biological processes in various stages of the pathogen’s life cycle which can require extensive experimentation (Gauriau et al. 2024). Pathogen and disease development is modeled in response to influencing variables and can be expressed as fixed parameters or functions (Gonzalez-Dominguez et al. 2023). For example, a mechanistic model for S. sclerotiorum development on soybean, white bean and carrot used information on environmental conditions required for different life stages extracted from previous literature and field experiments to predict disease progress (Salotti and Rossi 2023). However, mechanistic modeling of S. sclerotiorum life cycles is confounded by contradictory findings of laboratory and field experiments quantifying 13 the relationships between the environment and development of S. sclerotiorum (Reich and Chatterton 2023). Empirical models have the advantage of predicting disease outcomes without explicit description of cause-and-effect relationships but require large datasets to represent yearly and geographic variability (Gonzalez-Dominguez et al. 2023). Logistic regression modeling has been long favored by disease forecast developers (Johnson et al. 1998; Shah et al. 2013; Webster et al. 2023a; Willbur et al. 2018a). Recently, both categorical and regression supervised machine learning algorithms have been explored for predicting crop disease or pathogen development, and common algorithms include support vector machines, decision trees, random forests, artificial neural networks and anomaly detection algorithms for predicting rare events (Carisse and Fall 2020; Gauriau et al. 2024; Shah et al. 2023; Shahoveisi et al. 2022; Skelsey 2022; Webster et al. 2023a). A study by Shahoveisi et al. 2022 used data from laboratory experiments to model the relationship between S. sclerotiorum apothecia development, temperature and leaf wetness duration on canola and dry bean flowers under controlled incubation conditions. The authors found that machine learning algorithms provided greater accuracy in apothecia development than logistic regression, improving prediction accuracy from 78% to 89% and 88% to 92% for canola and dry bean infections, respectively. Machine learning algorithms have also proved advantageous over logistic regression modeling in other plant-pathogen systems, such as in predicting Fusarium head blight of wheat epidemics (Shah et al. 2014) and predicting Monilinia vaccinii-corymbosi ascospore release (Harteveld et al. 2017) but do not always improve predictions, as seen in the prediction of airborne ascospores of S. sclerotiorum (Reich et al. 2024) and tar spot of corn development (Webster et al. 2023a). Therefore, exploring both approaches is worthwhile during the development of disease forecasting tools to find a balance between model complexity and accuracy. Additionally, machine learning algorithms are not as easily interpretable and deployable as the less computationally costly logistic regression models, which may reduce their accessibility to users. 14 LITERATURE CITED Abawi, G. S. and Grogan, R. G. 1975. Source of primary inoculum and effects of temperatute and moisture on infection of beans by Whetzelinia sclerotiorum. Phytopathology 65:300- 309. doi.org/10.1094/Phyto-65-300 Abawi, G. S. and Grogan, R. G. 1979. Epidemiology of diseases caused by Sclerotinia species. Phytopathology 69:899-904. doi.org/10.1094/Phyto-69-899 Adams, P. B. and Ayers, W. A. 1979. Ecology of Sclerotinia species. Phytopathology 69:896- 899. doi.org/10.1094/Phyto-69-896. Amselem, J., Cuomo, C. A., van Kan, J. A. L., Viaud, M., Benito, E. P., Couloux, A., Coutinho, P. M., de Vries, R. P., Dyer, P. S., Fillinger, S., Fournier, E., Gout, L., Hahn, M., Kohn, L., Lapalu, N., Plummer, K. M., Pradier, J.-M., Quévillon, E., Sharon, A., Simon, A., ten Have, A., Tudzynski, B., Tudzynski, P., Wincker, P., Andrew, M., Anthouard, V., Beever, R. E., Beffa, R., Benoit, I., Bouzid, O., Brault, B., Chen, Z., Choquer, M., Collémare, J., Cotton, P., Danchin, E. G., Da Silva, C., Gautier, A., Giraud, C., Giraud, T., Gonzalez, C., Grossetete, S., Güldener, U., Henrissat, B., Howlett, B., Kodira, C., Kretschmer, M., Lappartient, A., Leroch, M., Levis, C., Mauceli, E., Neuvéglise, C., Oeser, B., Pearson, M., Poulain, J., Poussereau, N., Quesneville, H., Rascle, C., Schumacher, J., Ségurens, B., Sexton, A., Silva, E., Sirven, C., Soanes, D. M., Talbot, N. J., Templeton, M., Yandava, C., Yarden, O., Zeng, Q., Rollins, J. A., Lebrun, M.-H., and Dickman, M. 2011. Genomic analysis of the necrotrophic fungal pathogens Sclerotinia sclerotiorum and Botrytis cinerea. PLoS Genetics 7:e1002230. doi.org/10.1371/journal.pgen.1002230 Anderson, J. B. and Kohn, L. M. 1995. Clonality in soilborne, plant-pathogenic fungi. Annu. Rev. Phytopathol. 33:369-391. doi.org/10.1146/annurev.py.33.090195.002101 Attalah, Z. K., Larget, B., Chen, X. and Johnson, D. A. 2004. High genetic diversity, phenotypic uniformity, and evidence of outcrossing in Sclerotinia sclerotiorum in the Columbia Basin of Washington state. Phytopathology 94:737-742. doi.org/10.1094/PHYTO.2004.94.7.737 Attanayake, R. N., Porter, L., Johnson, D. A. and Chen, W. 2012. Genetic and phenotypic diversity and random association of DNA markers of isolates of the fungal plant pathogen Sclerotinia sclerotiorum from soil on a fine geographic scale. Soil Biol. Biochem. 55:28-36. doi.org/10.1016/j.soilbio.2012.06.002 Attanayake, R. N., Tennekoon, V., Johnson, D. A., Porter, L. D., del Rio-Mendoza, L., Jiang, D. and Chen, W. 2014. Inferring outcrossing in the homothallic fungus Sclerotinia sclerotiorum using linkage disequilibrium decay. Heredity 113:353-363. Doi.org/10.1038/hdy.2014.37 Attanayake, R. N., Xu, L. and Chen, W. 2019. Sclerotinia sclerotiorum populations: clonal or recombining? Trop. Plant Pathol. 44:23-31. doi.org/10.1007/s40858-018-0248-7 Blad, B. L., Steadman, J. R. and Weiss, A. 1978. Canopy structure and irrigation influence white mold disease and microclimate of dry edible beans. Phytopathology 68:1431-1437. doi.org/10.1094/Phyto-68-1431 15 Bradley, C. A., Allen, T. W., Sisson, A. J., Bergstrom, G. C., Bissonnette, K. M., Bond, J., Byamukama, E., Chilvers, M. I., Collins, A. A., Damicone, J. P., Dorrance, A. E., Dufault, N. S., Esker, P. D., Faske, T. R., Fiorellino, N. M., Giesler, L. J., Hartman, G. L., Hollier, C. A., Isakeit, T., Jackson-Ziems, T. A., Jardine, D. J., Kelly, H. M., Kemerait, R. C., Kleczewski, N. M., Koehler, A. M., Kratochvil, R. J., Kurle, J. E., Malvick, D. K., Markell, S. G., Mathew, F. M., Mehl, H. L., Mehl, K. M., Mueller, D. S., Mueller, J. D., Nelson, B. D., Overstreet, C., Padgett, G. B., Price, P. P., Sikora, E. J., Small, I., Smith, D. L., Spurlock, T. N., Tande, C. A., Telenko, D. E. P., Tenuta, A. U., Thiessen, L. D., Warner, F., Wiebold, W. J. and Wise, K. A. 2019. Soybean yield loss estimates due to diseases in the United States and Ontario, Canada, from 2015 to 2019. Plant Health Prog. 22:483-495. doi.org/10.1094/PHP-01-21-0013-RS Breunig, M., Bittner, R., Dolezal, A., Ramcharan, A. and Bunkers, G. 2023. An assay to reliably achieve tar spot symptoms on corn in a controlled environment. bioRxiv. doi.org/10.1101/2023.01.12.523803 Broders, K., Iriarte-Broders, G., Bergstrom, G. C., Byamukama, E., Chilvers, M., Cruz, C., Dalla-Lana, F., Duray, Z., Malvick, D., Mueller, D., Paul, P., ,Plewa, D., Raid, R., Robertson, A. E., Salgado-Salazar, C., Smith, D., Telenko, D., VanEtten, K. and Kleczewski, N. M. 2022. Phyllachora species infecting maize and other grass species in the Americas represents a complex of closely related species. Ecol. and Evol. 12:e8832. doi.org/10.1002/ece3.8832 CABI. 2005. Sclerotinia sclerotiourm (Lib.) de Bary. In: Distribution Maps of Plant Diseases. Wallingford, U.K. CAB International. doi.org/10.1079/DMPD/2006650097 Caldwell, D. L., da Silva, C. R., McCoy, A. G., Avila, H., Bonkowski, J. C., Chilvers, M. I., Helm, M., Telenko, D. E. P. and Iyer-Pascuzzi, A. S. 2024. Uncovering the infection strategy of Phyllachora maydis during maize colonization: A comprehensive analysis. Phytopathology 114:1075-1087. doi.org/10.1094/PHYTO-08-23-0298-KC Cao, S., Song, J., Yuan, Y., Zhang, A., Ren, J., Liu, Y., Qu, J., Hu, G., Zhang, J., Wang, C., Cao, J., Olsen, M., Prasanna, B. M., San Vicente, F. and Zhang, X. 2021. Genomic prediction of resistance to tar spot complex of maize in multiple populations using genotyping-by- sequencing SNPs. Front. Plant Sci. 12:672525. doi.org/10.3389/fpls.2021.672525 Carisse, O. and Fall, M. L. 2022. Decision trees to forecast risks of strawberry powdery mildew caused by Podosphaera aphanis. Agriculture 11:29. doi.org/10.3390/agriculture11010029 Check, J. C., Byrne, A. M., Singh, M. P., Steinke, K., Widdicome, W. D. and Chilvers, M. I. 2023. Effects of nitrogen application rate and plant density on severity of tar spot of corn. Plant Health Prog. 24:416-423. doi.org/10.1094/PHP-12-22-0125-RS Check, J. C., Harkness, R. J., Heger, L., Sakalidis, M., Chilvers, M. I., Mahaffee, W. F. and Miles, T. D. 2024a. It’s a Trap! Part I: Exploring the applications of rotating-arm impaction samplers in plant pathology. Plant Dis. 108:1910-1922. doi.org/10.1094/PDIS- 10-23-2096-FE Check, J. C., Harkness, R. J., Heger, L., Chilvers, M. I., Mahaffee, W. F., Sakalidis, M. L. and 16 Miles, T. D. 2024b. It’s a Trap! Part II: An approachable guide to constructing and using rotating-arm air samplers. Plant Dis. 108:1923-1936. doi.org/10.1094/PDIS-01-24-0131- SR Chitrampalam, P., Qiu, C., Aldrich-Wolfe, L., Legn, Y., Zhong, S. and Nelson, B. 2015. Prevalence of inversion positive and inversion negative mating type (MAT) alleles and MAT heterokaryons in Sclerotinia sclerotiorum in the United States. Botany 93:497-505. doi.org/10.1139/cjb-2015-0035 Clarkson, J. P., Fawcett, L., Anthony, S. G. and Young, C. 2014. A model for Sclerotinia sclerotiorum infection and disease development in lettuce, based on the effects of temperature, relative humidity and ascospore density. PLoS ONE 9:e94049. doi.org/10.1371/journal.pone.0094049 Conrad, A. M. and Telenko, D. E. P. 2023. Efficacy of biocontrol agents Coniothyrium minitans and Bacillus amyloliquefaciens for managing Sclerotinia sclerotiorum in Indiana soybean. PhytoFrontiers 3:518-524. doi.org/10.1094/PHYTOFR-07-22-0080-R Crop Protection Network – Corn Disease Loss Calculator. 2025a. https://loss.cropprotectionnetwork.org/ Crop Protection Network – Soybean Disease Loss Calculator. 2025b. https://loss.cropprotectionnetwork.org/ Cubeta, M. A., Cody, B. R., Kohli, Y. and Kohn, L. M. 1997. Clonality in Sclerotinia sclerotiorum on infected cabbage in eastern North Carolina. Phytopathology 87:1000- 1004. doi.org/10.1094/PHYTO.1997.87.10.1000 da Silva, C. R., Check, J. C., MacCready, J. S., Alakonya, A. E., Beiriger, R., Bissonnette, K. M., Collins, A., Cruz, C. D., Esker, P. D., Goodwin, S. B., Malvick, D., Mueller, D. S., Paul, P., Raid, R., Robertson, A. E., Roggenkamp, E., Ross, T. J., Singh, R., Smith, D. L., Tenuta, A. U., Chilvers, M. I. and Telenko, D. E. P. 2021. Recovery plan for tar spot of corn, caused by Phyllachora maydis. Plant Health Prog. 22:596-616. doi.org/10.1094/PHP-04-21-0074-RP Dann, E. K., Diers, B. W. and Hammerschmidt, R. 1999. Suppression of Sclerotinia stem rot of soybean by lactofen herbicide treatment. Phytopathology 89:598-602. doi.org/10.1094/PHYTO.1999.89.7.598 Dillard, H. R., Ludwig, J. W., and Hunter, J. E. 1995. Conditioning sclerotia of Sclerotinia sclerotiorum for carpogenic germination. Plant Dis. 79:411-415. doi.org/10.1094/PD-79- 0411 Fall. M. L., Willbur, J. F., Smith, D. L., Byrne, A. M. and Chilvers, M. I. 2018a. Spatiotemporal distribution pattern of Sclerotinia sclerotiorum apothecia is modulated by canopy closure and soil temperature in an irrigated soybean field. Plant Dis. 102:1794-1802. doi.org/10.1094/PDIS-11-17-1821-RE Foster, A. J., Kora, C., McDonald, M. R. and Boland, G. R. 2011. Development and validation of a disease forecast model for Sclerotinia rot of carrot. Canadian Journal of Plant Pathol. 33:187-201. doi.org/10.1080/07060661.2011.563753 17 Ford, E. J., Miller, R. V., Gray, H. and Sherwood, J. E. 1995. Heterokaryon formation and vegetative compatibility in Sclerotinia sclerotiorum. Mycol. Res. 99:241-247. doi.org/10.1016/S0953-7562(09)80893-9 FRAC. 2019. Pathogen Risk List 2019. Fungicide Resistance Action Committee, CropLife International. Brussels, Belgium. Fu, M., Qu, Z., Pierre-Pierre, N., Jiang, D., Souza, F. L., Miklas, P. N., Porter, L. D., Vandemark, G. J. and Chen, W. 2024. Exploring the mycovirus Sclerotinia sclerotiorum hypovirulence-associated DNA virus 1 as a biocontrol agent of white mold caused by Sclerotinia sclerotiorum. Plant Dis. 108:624-634. doi.org/10.1094/PDIS-07-23-1458-RE Gauriau, O., Galarraga, L., Brun, F., Termier, A., Davadan, L. and Joudelat, F. 2024. Comparing machine-learning models of different levels of complexity for crop protection: A look into the complexity-accuracy tradeoff. Smart Agr. Technol. 7:100380. doi.org/10.1016/j.atech.2023.100380 Gongora-Canul, C., Jimenez-Beitia, F. E., Puerto-Hernandez, C., Avellandea, C., Kleczewski, N., Telenko, D. E. P., Shim, S., Solorzano, J. E., Goodwin, S. B., Scofield, S. R. and Cruz, C. D. 2023. Assessment of symptom induction via artificial inoculation of the obligate biotrophic fungus Phyllachora maydis (Maubl.) on corn leaves. BMC Research Notes 16:69. doi.org/10.1186/s13104-023-06341-y Gonzalez-Dominguez, E., Caffi, T., Rossi, V., Salotti, I. and Fedele, G. 2023. Plant disease models and forecasting: Changes in principles and applications over the last 50 years. Phytopathology 113:678-693. doi.org/10.1094/PHYTO-10-22-0362-KD Granke, L. L., Morrice, J. J., and Hausbeck, M. K. 2014. Relationships between airborne Pseudoperonospora cubensis sporangia, environmental conditions, and cucumber downy mildew severity. Plant Dis. 98:674‐681. doi.org/10.1094/PDIS-05-13-0567-RE Grau, C. R. and Hartman G. L. 2015. Compendium of Soybean Diseases, 5th edition. The American Phytopathological Society, St. Paul Hambleton, S., Walker, C. and Kohn, L. M. 2002. Clonal lineages of Sclerotinia sclerotiorum previously known from other crops predominate in 1999-2000 samples from Ontario and Quebec soybean. Can. J. Plant Pathol. 24:309-315. doi.org/10.1080/07060660209507014 Harteveld, D. O. C., Grant, M. R., Pscheidt, J. W. and Peever, T. L. 2017. Predicting ascospore release of Monilinia vaccinii-corymbosi of blueberry with machine learning. Phytopathology 107:1364-1371. doi.org/10.1094/PHYTO-04-17-0162-R Hartill, W. F. T. and Underhill, A. P. 1976. “Puffing” in Sclerotinia sclerotiorum and S. minor. New Zealand Journal of Botany 14:355-358. doi.org/10.1080/0028825X.1976.1042890 Helm, M., Sing, R., Hiles, R., Jaiswal, N., Myers, A., Iter-Pascuzzi, A. S. and Goodwin, S. B. 2022. Candidate effector proteins from maize tar spot pathogen Phyllachora maydis localize to diverse plant cell compartments. Phytopathology 112:2538-2548. doi.org/10.1094/PHYTO-05-22-0181-R Hernandez, M. and KC, A. N. 2024. Determining the timing of spore release by Botryosphaeriaceae speices in Oregon Vineyards. Plant Dis. 108:1033-1040. 18 doi.org/10.1094/PDIS-07-23-1359-RE Hirst, J. M. 1952. An automatic volumetric spore trap. Ann. Appl. Biol. 39:257-265. doi.org/10.1111/j.1744-7348.1952.tb00904.x Hock, J., Dittrich, U., Renfro, B. L. and Kranz, J. 1992. Sequential development of pathogens in the maize tar spot disease complex. Mycopathologia 117:157-161. doi.org/10.1007/BF00442777 Isard, S. A., Barnes, C. W., Hambleton, S., Ariatti, A., Russo, J. M., Tenuta, A., Gay, D. A. and Szabo, L. J. 2011. Predicting soybean rust incursions into the north American continental interior using crop monitoring, spore trapping, and aerobiological modeling. Plant Dis. 95:1346-1357. doi.org/10.1094/PDIS-01-11-0034 Johnson, D. A., Alldredge, J. R. and Hamm, P. B. 1998. Expansion of potato late blight forecasting models for the Columbia Basin of Washington and Oregon. Plant Dis. 82:642-645. doi.org/10.1094/PDIS.1998.82.6.642 Johnson, E. T., Dowd, P. F., Ramirez, J. L. and Behle, R. W. 2023. Potential biocontrol agents of corn tar spot disease isolated from overwintered Phyllachora maydis stromata. Microorganisms 11:1550. doi.org/10.3390/microorganisms11061550 Jones, S. J., Gent, D. H., Pethybridge, S. J. and Hay, F. S. 2011. Site-specific risk factors of white mould epidemics in bean (Phaseolus vulgaris) in Tasmania, Australia. N. Z. J. Crop Hortic. Sci. 40:147-159. doi.org/10.1080/01140671.2011.623707 Kabbage, M., Williams, B. and Dickman, M. B. 2013. Cell death control: The interplay of apoptosis and autophagy in the pathogenicity of Sclerotinia sclerotiorum. PLoS Pathog. 9:e1003287. doi.org/10.1371/journal.ppat.1003287 Kabbage, M., Yarden, O. and Dickman, M. B. 2015. Pathogenic attributes of Sclerotinia sclerotiorum: Switching from a biotrophic to necrotrophic lifestyle. Plant Sci. 233:53-60. doi.org/10.1016/j.plantsci.2014.12.018 Kader, K. A., Balasubramanian, P. M. and Chatterton, S. 2018. Influence of irrigation and plant canopy architecture on white mold of dry bean. Can. J. Plant Sci. 98:1280-1292. doi.org/10.1139/cjps-2018-0095 Khan, H. A., Mukhtar, M. and Bhatti, M. F. 2023. Mycovirus-induced hypovirulence in notorious fungi Sclerotinia: A comprehensive review. Braz. J. Microbiol. 54:1459-1478. doi.org/10.1007/s42770-023-01073-4 Kim, H. S., and B. W. Diers. 2000. Inheritance of partial resistance to sclerotinia stem rot in soybean. Crop Sci. 40: 55–61. doi.org/10.2135/cropsci2000.40155x King, J. E. and Polley, R. W. 1976. Observations on the epidemiology and effect on grain yield of brown rust in spring barley. Plant Path. 25:63-73. doi.org/10.1111/j.1365- 3059.1976.tb01926.x Koch, S., Dunker. S., Kleinhenz, B., Rohrig, M. and von Tiedemann, A. 2007. A crop loss- related forecasting forecasting model for Sclerotinia stem rot in winter oilseed rape. Phytopathology 97:1186-1194. doi.org/10.1094/PHYTO-97-9-1186 19 Kohli, Y., Brunner, L. J., Yoell, H., Milgroom, M. G., Anderson, J. B., Morrall, R. A. A. and Kohn, L. M. 1995. Clonal dispersal and spatial mixing in populations of the plant pathogenic fungus, Sclerotinia sclerotiorum. Mol. Ecol. 4:69–77. doi.org/10.1111/j.1365- 294X.1995.tb00193.x Kohli, Y., Morrall, R. A. A., Anderson, J. B. and Kohn, L. M. 1992. Local and trans-Canadian clonal distribution of Sclerotinia sclerotiorum on canola. Phytopathology 82:975-880. doi.org/10.1094/Phyto-82-875. Laurent, B., Douillet, A., Beslay, A., Bordes, J., Delmotte, F., Debord, C., Raynal, M. and the VISA network partners. 2022. The VISA network: A collaborative project between research institutes and vineyard owners to create the first epidemiological monitoring network of downy mildew epidemic based on aerial spore capture. BIO Web Conf. 50:04007. doi.org/10.1051/bioconf/20225004007 Lipps, S., Smith, D., Telenko, D., Paul, P., Kleczewski, N. and Jamann T. 2022. Identification of resistance for Phyllachora maydis of maize in exotic-derived germplasm. Crop Sci. 62:859-866. doi.org/10.1002/csc2.20709 Luis, J. M., Mehl, H. L., Plewa, D. and Kleczewski, N. M. 2023. Is Microdochium maydis associated with necrotic lesions in the tar spot disease complex? A culture-based survey of maize in Mexico and the Midwestern United States. Phytopathology 113:1890-1897. doi.org/10.1094/PHYTO-04-23-0109-R MacCready, J. S., Roggenkamp, E. M., Gdanetz, K. and Chilvers, M. I. 2023. Elucidating the obligate nature and biological capacity of an invasive fungal corn pathogen. Mol. Plant Microbe Interact 36:411-424. doi.org/10.1094/MPMI-10-22-0213-R Mahuku, G., Chen, J., Shrestha, R., Narro, L. A., Osorio Guerrero, K. V., Arcos, A. L. and Xu, Y. 2016. Combined linkage and association mapping identifies a major QTL (qRtsc8-1), conferring tar spot complex resistance in maize. Theor. Appl. Genet. 129:1217-1229. doi.org/10.1007/s00122-016-2698-y Mardones, M., Trampe-Jaschik, T., Oster, S., Elliot, M., Urbina, H., Schmitt, I. and Piepenbring, M. 2017. Phylogeny of the order Phyllachorales (Ascomycota, Sordariomycetes): among and within order relationships based on five molecular loci. Persoonia 39:74-90. doi.org/10.3767/persoonia.2017.39.04 Maublanc, A. 1904. Especes Nouvelles de champignons inferieurs. Bull. de la Societe Phytopathologique Francaise 20, 72. McCaghey, M., Willbur, J., Ranjan, A., Grau, C. R., Chapman, S., Diers, B., Groves, C., Kabbage, M. and Smith, D. L. 2017. Development and evaluation of Glycine max germplasm lines with quantitative resistance to Sclerotinia sclerotiorum. Front. Plant Sci. 8:1495. doi.org/10.3389/fpls.2017.0149 McCoy, A. G., Roth, M. G., Shay, R., Noel, Z. A., Jayawardana, M. A., Longley, R. W., Bonito, G. and Chilvers, M. I. 2019. Identification of fungal communities within the tar spot complex of corn in Michigan via next-generation sequencing. Phytobiomes 3:235-243. doi.org/10.1094/PBIOMES-03-19-0017-R 20 Mottaleb, K. A., Loladze, A., Sonder, K., Kruseman, G. and San Vicente, F. 2018. Threats of tar spot complex in the United States of America and its global consequences. Mitig. Adapt. Strategies Glob. Change 24:281-300. doi.org/10.1007/s11027-018-9812-1 Mueller, D. S., Pedersen, W. L. and Hartman, G. L. 2002. Effect of crop rotation and tillage system on Sclerotinia stem rot on soybean. Can. J. Plant Pathol. 24:450-456. doi.org/10.1080/07060660209507033 Munir, M., Wang, H., Dufault, N. S. and Anco, D. J. 2020. Early detection of airborne inoculum of Nothopassalora personata in spore trap samples from peanut fields using quantitative PCR. Plants 9:1327. doi.org/10.3390/plants9101327 Newlands, N. K. 2018. Model-based forecasting of agricultural crop disease risk at the regional scale, integrating airborne inoculum, environmental, and satellite-based monitoring data. Front. Environ. Sci. 6:63. doi.org/10.3389/fenvs.2018.00063 Netwon, H. C. and Sequeira, L. 1972. Ascospores as the primary infective propagule of Sclerotinia sclerotiorum in Wisconsin. Plant Dis. Rep. 56:798-802 Nieto-Lopez, E. H., Justo Miorini, T. J., Wulkop-Gil, C. A., Chilvers, M., Giesler, L. J. Jackson- Ziems, T. A., Kabbage, M., Mueller, D. S., Smith, D. L. Tovar-Pedraza, J. M., Willbur, J. F. and Everhart, S.E. 2023. Fungicide sensitivity of Sclerotinia sclerotiorum from U.S. soybean and dry bean, compared to different regions and climates. Plant Dis. 107:2395- 2406. doi.org/10.1094/PDIS-07-22-1707-RE Parbery, D. G. 1963. Studies on graminicolous species of Phyllachora Fckl. II. Invasion of the host and development of the fungus. Aust. J. Bot. 11:131-140. doi.org/10.1071/BT9630131 Parbery, D. G. 1967. Studies on graminicolous species of Phyllachora Fckl. V. a taxonomic monograph. Aust. J. Bot. 15:271-375. doi.org/10.1071/BT9670271 Pearson, R. C., Seem, R. C. and Meyer, F. W. 1980. Environmental factors influencing the discharge of basidiospores of Gymnosporangium juniperi-vinginianae. Phytopathology 70:262-266. doi.org/10.1094/Phyto-70-262 Peltier, A. J., Bradley, C. A., Chilvers, M. I., Malvick, D. K., Mueller, D. S., Wise, K. A. and Esker, P. D. 2012. Biology, yield loss and control of Sclerotinia stem rot of soybean. J. Integr. Pest Manag. 3:1-7. doi.org/10.1603/IPM11033 Pereyda-Hernandez, J., Hernandez-Morales, J., Sandoval-Islas, J. S., Aranda-Ocampo, S., De Leion-Garcia de Alba, C. and Gomez- Montiel, S. 2009. Etiology and management of tar spot (Phyllachora maydis Maubl.) of maize in Guerrero State, Mexico. Art Agro. 43:511- 519 Pethybridge, S. J., Brown, B. J., Kikkert, J. R. and Ryan, M. R. 2020. Rolled-crimped cereal rye residue suppresses white mold in no-till soybean and dry bean. Renewable Agriculture and Food Systems 35:599-607. doi.org/10.1017/S174217051900022X Pizolotto, C. A., Harrington, M., Brown, L., Murdock, M., Harrington, S., Marshall, J., Neher, O. T. and Woodhall, J. W. 2021. A real-time PCR assay for Erysiphe betae and its effectiveness when used with different spore trapping methods. Eur. J. Plant Pathol. 21 162:329-341. doi.org/10.1007/s10658-021-02405-6 Purdy, L. H. 1979. Sclerotinia sclerotiorum: History, diseases and symptomatology, host range, geographic distribution, and impact. Phytopathology 69:875-880. doi.org/10.1094/Phyto- 69-875 Rahman, A., Standish, J. R., D’Arcangelo, K. N., and Quesada-Ocampo, L. M.2021. Clade- specific biosurveillance of Pseudoperonospora cubensis using spore traps for precision disease management of cucurbit downy mildew. Phytopathology 111:312‐320. doi.org/10.1094/PHYTO-06-20-0231-R Ranjan, A., Westrick, N. M., Jain, S., Piotrowski, J. S., Ranjan, M., Kessens, R., Stiegman, L., Grau, C. R., Conley, S. P., Smith, D. L. and Kabbage, M. 2019. Resistance against Sclerotinia sclerotiorum in soybean involves re-programming of the phenylpropanoid pathway and up-regulation of anti-fungal activity targeting ergosterol biosynthesis. Plant Biotechnol. J. 17:1567-1581. doi.org/10.1111/pbi.13082 Reich, J. and Chatterton, S. 2023. Predicting field diseases caused by Sclerotinia sclerotiorum: A review. Plant Path. 72:3-18. doi.org/ 10.1016/S0261-2194(98)00035-0 Reich, J., McLaren, D., Kim, Y. M., Wally, O., Yevtushenko, D., Hamelin, R. and Chatterton, S. 2024. Predicting airborne ascospores of Sclerotinia sclerotiorum through machine learning and statistical methods. Plant Path. 73:1586-1601. doi.org/10.1111/ppa.13902 Ren, J., Wu, P., Huestis, G. M., Zhang, A., Qu, J., Liu, Y., Zheng, H., Alakonya, A. E., Dhliwayo, T., Olsen, M., San Vicente, F., Prasanna, B. M., Chen, J. and Zhang, X. 2022. Identification and fine mapping of a major QTL (qRtsc8-1) conferring resistance to maize tar spot complex and validation of production markers in breeding lines. Theor. Appl. Genet. 135:1551-1563. doi.org/10.1007/s00122-022-04053-8 Rogers, A., Jaiswal, N., Roggenkamp, E., Kim, H., MacCready. J. S., Chilvers, M. I., Scofield, S. R., Iyer-Pascuzzi, A. S. and Helm, M. 2024. Genome-informed trophic classification and functional characterization of virulence proteins from the maize tar spot pathogen Phyllachora maydis. Phytopathology 114:1940-1949. doi.org/10.1094/PHYTO-01-24- 0037-R Roper, M., Seminara, A., Bandi, M. M., Cobb, A., Dillard, H. R. and Pringle, A. 2010. Dispersal of fungal spores on a cooperatively generated wind. PNAS 107:17474-17479. doi.org/ 10.1073/pnas.1003577107 Ross, T. J., Allen, T. W., Shim, S., Thompson, N. M. and Telenko, D. E. P. 2024. Investigations into the economic returns resulting from foliar fungicides and application timing on management of tar spot in Indiana hybrid corn. Plant Dis. 108:461-472. doi.org/10.1094/PDIS-05-23-0932-RE Ross, T. J., Chilvers, M. I., Byrne, A. M., Smith, D. L., Mueller, B., Shim, S. and Telenko, D. E. P. 2023. Effects of disease resistance and fungicide application on management of tar spot on corn in the North Central United States. Plant Health Prog. 24:439-444. doi.org/10.1094/PHP-10-22-0103-RS Roth, M. G., Webster, R. W., Mueller, D. S., Chilvers, M. I., Faske, T. R., Mathew, F. M., 22 Bradley, C. A., Damicone, J. P., Kabbage, M. and Smith, D. L. Integrated management of important soybean pathogens of the United States in changing climate. 2020. J. Integ. Pest Manag. 11:1-28. doi.org/10.1093/jipm/pmaa013 Ruhl, G., Romberg, M. K., Bissonnette, S., Plewa, D., Creswell, T. and Wise, K. A. 2016. First report of tar spot of corn caused by Phyllachora maydis in the United States. Plant Dis. 100: 1496. doi.org/10.1094/PDIS-12-15-1506-PDN Saharan, G. S. and Mehta, N. 2008. Sclerotinia Diseases of Crop Plants: Biology, Ecology and Disease Management. Springer, Dordrecht, The Netherlands Salotti, I. and Rossi, V. 2023. A mechanistic model accounting for the effect of soil moisture, weather, and host growth stage on the development of Sclerotinia sclerotiorum. Plant Dis. 107:514-533. doi.org/10.1094/PDIS-12-21-2743-RE Schmidt, J. P., Lamb, J. A., Schmitt, M. A., Randall, G. W., Orf, J. H, and Gollany, H. T. 2001. Soybean varietal response to liquid swine manure application. Agron. J. 93:358-363. doi.org/10.2134/agronj2001.932358x Sexton, A. C., Whitten, A. R. and Howlett, B. J. 2006. Population structure of Sclerotinia sclerotiorum in an Austrailian canola field at flowering and stem-infection stages of the disease cycle. Genome 49:1408-1415. doi.org/10.1139/g06-101 Shah, D. A., Molineros, J. E., Paul, P. A., Willyerd, K. T., Madden, L. V. and De Wolf, E. D. 2013. Predicting Fusarium head blight epidemics with weather-driven pre- and post- anthesis logistic regression models. Phytopathology 103:906-919. doi.org/10.1094/PHYTO-11-12-0304-R Shah, D. A., De Wolf, E. D., Paul, P. A. and Madden, L. V. 2014. Predicting Fusarium head blight epidemics with boosted regression trees. Phytopathology 104:702-714. doi.org/10.1094/PHYTO-10-13-0273-R Shah, D. A., De Wolf, E. D., Paul, P. A. and Madden, L. V. 2023. Intro the trees: Random forests for predicting Fusarium head blight epidemics of wheat in the United States. Phytopathology 113:1483-1493. doi.org/10.1094/PHYTO-10-22-0380-R Shahovesis, F., Riahi Manesh, M. and del Rio Mendoza, L. E. 2022. Modeling risk of Sclerotinia sclerotiorum-induced disease development on canola and dry bean using machine learning algorithms. Sci. Rep. 12:864. doi.org/10.1038/s41598-021-04743-1 Singh, R., Shim, S., Telenko, D. E. P. and Goodwin, S. B. 2023. Parental inbred lines of the nested association mapping (NAM) population of corn show sources of resistance to tar spot in northern Indiana. Plant Dis. 107:262-266. doi.org/10.1094/PDIS-02-22-0314-SC Skelsey, P. 2021. Forecasting risk of crop disease with anomaly detection algorithms. Phytopathology. 111:321-322. doi.org/10.1094/PHYTO-05-20-0185-R Solorzano, J. E., Issendorf, S. E., Drott, M. T., Check, J. C., Roggenkamp, E. M., Cruz, C. D., Kleczewski, N. M., Gongora-Canul, C. C. and Malvick, D. K. 2023. A new and effective method to induce infection of Phyllachora maydis into corn for tar spot studies in controlled environments. Plant Methods 19:83. doi.org/10.1186/s13007-023-01052-8 23 Sun, P. and Yang, X. B. 2000. Light, temperature, and moisture effects on apothecium production of Sclerotinia sclerotiorum. Plant Dis. 84:1287-1293. doi.org/10.1094/PDIS.2000.84.12.1287 Telenko, D. E. P., Chilvers, M. I., Ames, K., Byrne, A. M., Check, J. C., Da Silva, C. R., Jay, W. S., Mueller, B., Ross, T. J., Smith, D. L. and Tenuta, A. U. 2022a. Fungicide efficacy during a severe epidemic of tar spot on corn in the United States and Canada in 2021. Plant Health Prog. 23:342-344. doi.org/10.1094/PHP-02-22-0012-BR Telenko, D. E. P., Chilvers, M. I., Byrne, A. M., Check, J. C., Da Silva, C. R., Kleczewski, N. M., Roggenkamp, E. M., Ross, T. J. and Smith, D. L. 2022b. Fungicide efficacy on tar spot and yield of corn in the Midwestern United States. Plant Health Prog. 23:281-287. doi.org/10.1094/PHP-10-21-0125-RS Thaning, C. and Nilsson, H.-E. 2008. A narrow range of wavelengths active in regulating apothecial development in Sclerotinia sclerotiorum. J. Phytopath. 148:627-631. doi.org/10.1111/j.1439-0434.2000.00546.x Thiessen, L. D., Keune, J. A., Neill, T. M., Turechek, W. W., Grove, G. G. and Mahaffee, W. F. 2016. Development of a grower‐conducted inoculum detection assay for management of grape powdery mildew. Plant Pathol. 65:238‐249. doi.org/10.1111/ppa.12421 Thiessen, L. D. 2024. Follow that spore! Airborne inoculum detection as a decision aid. PhytoFrontiers 4:5-13. doi.org/10.1094/PHYTOFR-06-23-0077-FI Timmer, L. W., Solel, Z., Gottwald, T. R., Ibanez, A. M. and Zitko, S. E. 1998. Environmental factors affecting production, release, and field populations of conidia of Alternaria alternata, the cause of brown spot of citrus. Phytopathology 88:1218-1223. doi.org/10.1094/PHYTO.1998.88.11.1218 Twengstrom, E., Kopmans, E., Sigvald, R. and Svensson, C. 1998a. Influence of different regimes on carpogenic germination of sclerotia of Sclerotinia sclerotiorum. J. Phytopathology. 146:487-493. doi.org/10.1111/j.1439-0434.1998.tb04610.x Twengstrom, E., Sigvald, R., Svensson, C. and Yuen, J. 1998b. Forecasting Sclerotinia stem rot in spring sown oilseed rape. Crop Prot. 17:405-411. doi.org/10.1016/S0261- 2194(98)00035-0 Uhm, J. Y. and Fujii, H. 1983. Heterothallism and mating type mutation in Sclerotinia trifoliorum. Phytopathology 73:569-572. doi.org/10.1094/Phyto-73-569 Uloth, M. B., Clode, P. L., You, M. P. and Barbett, M. J. 2015. Calcium oxalate crystals: an integral component of the Sclerotinia sclerotiorum/Brassica carinata pathosystem. PLoS One 10:e0122362. doi.org/10.1371/journal.pone.0122362 Van der Heyden, H., Dutilleul, P., Charron, J.-B., Bilodeau, G. J. and Carisse, O. 2021. Monitoring airborne inoculum for improved plant disease management. A review. Agron. Sustain. Dev. 41:40. doi.org/10.1007/s13593-021-00694-z Vieira, R. F., Paula Junior, T. J., Teixeira, H. and de S. Carneiro, J. E. 2010. White mold management in common bean by increasing within-row distance between plants. Plant Dis. 94:361-367. doi.org/10.1094/PDIS-94-3-0361 24 Wang, A., Shang, Z., Jiang, R., Zhang, M., Wang, J., Li, H., Zhang, B., Tang, H., Xu, F., Hu, X., Liu, W., Fan, J. R., Zhou, Y. and West, J. S. 2024a. Development and application of a qPCR-based method coupled with spore trapping to monitor airborne pathogens of wheat causing stripe rust, powdery mildew, and Fusarium head blight. Plant Dis. First Look. doi.org/10.1094/PDIS-03-24-0548-SR Wang, S.-Y., Zhang, Y.-J., Chen, X., Shi, X.-C., Herrera-Balandrano, D. D., Liu, F.-Q. and Laborda, P. 2024b. Biocontrol methods for the management of Sclerotinia sclerotiorum in legumes: A review. Phytopathology 114:1447-1457. doi.org/10.1094/PHYTO-01-24- 0006-RVW Webster, R. W., Roth, M. G., Reed, H., Mueller, B., Groves, C. L., McCaghey, M., Chilvers, M. I., Mueller, D. S., Kabbage, M. and Smith, D. L. 2021. Identification of soybean (Glycine max) check lines for evaluating genetic resistance to sclerotinia stem rot. Plant Dis. 105:2189-2195. doi.org/10.1094/PDIS-10-20-2193-RE Webster, R. W., Roth, M. G., Muller, B. D., Mueller, D. S., Chilvers, M. I., Willbur, J. F., Mourtzinis, S., Conley, S. P. and Smith, D. L. 2022. Integration of row spacing, seeding rates, and fungicide applications for control of Sclerotinia stem rot in Glycine max. Plant Dis. 106:1183-1191. doi.org/10.1094/PDIS-09-21-1931-RE Webster, R. W., Nicolli, C., Allen, T. W., Bish, M. D., Bissonnette, K., Check, J. C., Chilvers, M. I., Duffeck, M. R., Kleczewski, N., Luis, J. M., Mueller, B. D., Paul, P. A., Price, P. P., Robertson, A. E., Ross, T. J., Schmidt, C., Schmidt, R., Schmidt, T., Shim, S., Telenko, D. E. P., Wise, K. and Smith, D. L. 2023a. Uncovering the environmental conditions required for Phyllachora maydis infection and tar spot development on corn in the United States for use as predictive models for future epidemics. Sci. Reports 13:17064. doi.org/10.1038/s41598-023-44338-6 Webster, R. W., Mueller, B. D., Chilvers, M. I., Byrne, A. M., Boyse, J. F., Widdicombe, W. W., Mueller, D. S., Wiggs, S., Kandel, Y., Telenko, D. E. P., Ravellette, J., Shim, S. and Smith, D. L. 2023a. Integrating seeding rates and pesticide programs for managing Sclerotinia stem rot in Glycine max with nitrogen fertilizer applications. Plant Health Prog. 24:320-325. doi.org/10.1094/PHP-10-22-0102-RS Webster, R. W., Mueller, B. D., Conley, S. P. and Smith, D. L. 2023. Integration of soybean (Glycine max) resistance levels to sclerotinia stem rot into predictive S. sclerotiorum apothecial models. Plant Dis. 107:2763-2768. doi.org/10.1094/PDIS-12-22-2875-RE Weiss, A., Hipps, L. E., Blad, B. L. and Steadman, J. R. 1980. Comparison of within-canopy microclimate and white mold disease (Sclerotinia sclerotiorum) development in dry edible beans as influenced by canopy structure and irrigation. Agric. Meteorol. 22:11-21. doi.org/10.1016/0002-1571(80)90024-2 West, J. S. and Kimber, R. B. E. 2015. Innovations in air sampling to detect plant pathogens. Ann. Appl. Biol. 166:4-17. doi.org/10.1111/aab.12191 Willets, H. J. and Wong, J. A. L. 1980. The biology of Sclerotinia sclerotiorum, S. trifoliorum, and S. minor with emphasis on specific nomenclature. Bot. Rev. 46:101-165 Williams, B., Kabbage, M., Kim, H. J., Britt, R. and Dickman, M. B. 2011. Tipping the balance: 25 Sclerotinia sclerotiorum secreted oxalic acid suppresses host defenses by manipulating the host redox environment. PLoS Pathogens 7:e1002107. doi.org/10.1371/journal.ppat.1002107 Willbur, J. F., Ding, S., Marks, M. E., Lucas, H., Grau, C. R., Groves, C. L., Kabbage, M. and Smith, D. L. 2017. Comprehensive sclerotinia stem rot screening of soybean germplasm requires multiple isolates of Sclerotinia sclerotiorum. Plant Dis. 101:344-353. doi.org/10.1094/PDIS-07-16-1055-RE Willbur, J. F., Fall, M. L., Bloomingdate, C., Byrne, A. M., Chapman, S. A., Isard, S. A., Magarey, R. D., McCaghey, M. M., Mueller, B. D., Russo, J. M., Schlegel J., Chilvers, M. I., Mueller, D. S., Kabbage, M. and Smith, D. L. 2018a. Weather-based models for assessing the risk of Sclerotinia sclerotiorum apothecial presence in soybean (Glycine max) fields. Plant Dis. 102:73-84. doi.org/10.1094/PDIS-04-17-0504-RE Willbur, J. F., Fall, M. L., Byrne, A. M., Chapman, S. A., McCaghey, M. M., Mueller, B. D., Schmidt, R., Chilvers, M. I., Mueller, D. S., Kabbage, M., Giesler, L. J., Conley, S. P., Smith, D. L. 2018b. Validating Sclerotinia sclerotiorum apothecial models to predict Sclerotinia stem rot in soybean (Glycine max) fields. Plant Dis. 102:2592-2601. doi.org/10.1094/PDIS-02-18-0245-RE Willbur, J., McCaghey, M., Kabbage, M. and Smith, D. L. 2019a. An overview of the Sclerotinia sclerotiorum pathosystem in soybean: Impact, fungal biology, and current management strategies. Trop. Plant Pathol. 44:3-11. doi.org/10.1007/s40858-018-0250-0 Willbur, J. F., Mitchell, P. D., Fall, M. L., Byrne, A. M., Chapman, S. A., Floyd, C. M., Bradley, C. A., Ames, K. A., Chilvers, M. I., Kleczewski, N. M., Malvick, D. K., Mueller, B. D., Mueller, D. S., Kabbage, M., Conley, S. P. and Smith, D. L. 2019b. Meta-analytic and economic approaches for evaluation of pesticide impact on sclerotinia stem rot control and soybean yield in the North Central United States. Phytopathology 109:1157-1170. doi.org/10.1094/PHYTO-04-18-0124-R Wise, K. A., Smith, D., Freije, A., Mueller, D. S., Kandel, Y., Allen, T., Bradley, C. A., Byamukama, E., Chilvers, M., Faske, T., Friskop, A., Hollier, C., Jackson-Ziems, T. A., Kelly, H., Kemerait, B., Price, P., III, Robertson, A. and Tenuta, A. 2019. Meta-analysis of yield response of foliar fungicide-treated hybrid corn in the United States and Ontario, Canada. PloS One 14:e0217510. doi.org/10.1371/journal.pone.0217510 Xiao, K., Qiao, K., Cui, W., Xu, X., Pan, H., Wang, F., Wang. S., Yang, F., Xuan, Y., Li, A., Han, X., Song, Z. and Liu, J. 2023. Comparitive transcriptome profiling reveals the importance of GmSWEET15 in soybean susceptibility to Sclerotinia sclerotiorum. Front. Microbiol. 14:1119016. doi.org/10.3389/fmicb.2023.1119016 Yan, S., Loladze, A., Wang, N., Sun, S., Chilvers, M. I., Olsen, M., Burgeno, J., Petroli, C. D., Molnar, T., San Vicente, F., Zhang, X. and Boddupalli, M. P. 2022. Association mapping of resistance to tar spot complex in maize. Plant Breed. 141:745-755. doi.org/10.1111/pbr.13056 Young, C. S., Clarkson, J. P., Smith, J. A., Watling, M., Phelps. K. and Whipps, J. M. 2004. Environmental conditions influencing Sclerotinia sclerotiorum infection and disease 26 development in lettuce. Plant Path. 53:387-397. doi.org/10.1111/j.1365- 3059.2004.01018.x Zeng, W., Wang, D., Kirk, W. and Hao, J. 2012. Use of Coniothyrium minitans and other microorganisms for reducing Scleroinia sclerotiorum. Bio. Control 60:225-232. doi.org/10.1016/j.biocontrol.2011.10.009 Zhu, W., Wei, W., Fu, Y., Cheng, J., Xie, J., Li, G., Yi, X., Kang, Z., Dickman, M. B. and Jiang, D. 2013. A secretory protein of necrotrophic fungus Sclerotinia sclerotiorum that suppresses host resistance. PLoS One 8:e53901. doi.org/10.1371/journal.pone.0053901 27 CHAPTER 2: EFFECTS OF NITROGEN APPLICATION RATE AND PLANT DENSITY ON SEVERITY OF TAR SPOT OF CORN Summary With the recent introduction of tar spot to the Midwestern United States, questions regarding how agronomic choices affect tar spot development were posed by Michigan corn farmers. To investigate the role of nitrogen application rate and planting density on tar spot severity, field trials were conducted in 2019 and 2020. Plots of the nitrogen application rate trials were treated with either 0.5X, 1X or 1.5X the maximum return to nitrogen rates (90, 179 or 269 kg N per hectare, respectively). Plots of the planting density trials were planted at seeding rates of 69,000, 84,000, 99,000 or 114,000 seeds per hectare. Across six site years of field experiments, no significant (P > 0.05) relationship was found between nitrogen application rate and tar spot severity. Plant density had a significant (P > 0.05) inverse relationship with tar spot severity where a 41% decrease in area under the disease progress curve was seen for every additional 1,000 plants per hectare. To explore the feasibility of reducing seeding rate to manage tar spot, analyses were conducted to find the economically optimal planting density (EOPD) in the environments represented in these field trials. The EOPD ranged from 73,000 to 77,000 plants per hectare for two corn grain price points: US$150 to 300 per metric ton. These results demonstrate that nitrogen application rate is not a significant factor in tar spot development, and reducing seeding rate did not result in better production economics despite reducing tar spot severity. At all field experiment locations, hybrid susceptibility significantly (P < 0.05) reduced tar spot severity, establishing that hybrid selection is an important factor in tar spot management and should be carefully considered rather than diverging from existing agronomic practices. Source Check, J. C., Byrne, A. M., Singh, M. P., Steinke, K., Widdicombe, W. D. and Chilvers, M. I. 2023. Effects of nitrogen application rate and plant density on severity of tar spot of corn. Plant Health Prog. 24:416-423. doi.org/10.1094/PHP-12-22-0125-RS 28 CHAPTER 3: UNRAVELING THE ENVIRONMENTAL DRIVERS OF PHYLLACHORA MAYDIS SPORE RELEASE USING SPORE TRAPPING AND QPCR Abstract Phyllachora maydis, the causal agent of tar spot of corn, is an emerging disease in the United States and Canada. This study aims to improve our understanding of P. maydis spore release by utilizing spore trapping and quantitative PCR to assess the relationship between spore capture and environmental conditions. Burkard and rotating-arm air samplers were deployed in Michigan corn fields with natural disease pressure from 2021 to 2023. Correlation analysis and mixed-effects logistic regression were applied to examine the impact of environmental factors on spore capture. Through an exhaustive screening of candidate logistic regression models, results indicated that spore quantity is significantly negatively correlated with daily summaries of minimum temperature (P<0.05, τ=-0.24), mean temperature (P<0.05, τ=-0.25), maximum precipitation rate (P<0.05, τ=-0.33) and durations of temperature between 16.6 to 23ºC and relative humidity over 85% (P<0.001, τ=-0.27). Logistic regression models frequently incorporated temperature and humidity predictors, and the best performing model used daily averages of mean temperature and maximum relative humidity to discriminate presence and absence of spore detection, achieving a balanced accuracy of 85%. Across all site-years, spore traps did not detect P. maydis spores prior to the visible detection of tar spot symptoms but did detect spores before tar spot incidence reached 100%. Through this study, the environmental drivers of spore release were described to fill current knowledge gaps in the tar spot disease cycle. Additionally, a methodology for the capture and molecular quantification of airborne P. maydis spores is described which will benefit future tar spot epidemiology research. 29 Introduction Phyllachora maydis Maubl. is the causal agent of tar spot of corn (Zea mays L.), an emerging disease with significant economic impact in the Midwestern United States (Figure 1C and D). After the initial discovery of P. maydis in Illinois and Indiana in 2015 (Ruhl et al. 2016), tar spot confirmations have been made in 20 U.S. states and Ontario and Quebec, Canada (https://corn.ipmpipe.org/tar-spot/). Annual yield losses between 2018 to 2023 were estimated at 106 million bushels, valued at 521 million USD (Crop Protection Network 2024). Prior to its discovery in the U.S., P. maydis was first reported in Mexico in 1904 (Maublanc 1904) and has been documented throughout Central and South America (Mottaleb et al. 2019), where it is considered part of the tar spot disease complex caused by synergistic interactions with Monographella maydis E. Müll. & Samuels (syn. Microdochium maydis) and mycoparasite Coniothyrium phyllachorae Maubl. (Hock et al. 1992). However, in the U.S., P. maydis is currently the only member of the disease complex reported and is believed to be solely responsible for tar spot and yield losses (Caldwell et al. 2024; Luis et al. 2023; McCoy et al. 2019). The current description of the tar spot cycle is incomplete, primarily due to its recent introduction in the U.S. and difficulties observing this obligate biotrophic fungal pathogen in controlled environments. Much of what is known about the epidemiology of tar spot is based on empirical data collected in Mexico, which is considerably different in climate and corn production practices compared to the Midwestern U.S. (Hock et al. 1989; Hock et al. 1995). It has been demonstrated that P. maydis can overwinter in corn residue on the soil surface, which serves as primary inoculum in subsequent growing seasons (Groves et al. 2020; Kleczewski et al. 2019). From overwintered leaf debris, ascospores can be forcibly ejected from perithecia (Hock et al. 1995) or both ascospores and conidia can be exuded into gelatinous masses called cirrhi (Breunig et al. 2023; Parbery 1963a) to be spread via wind and rain splash. Spores of P. maydis are well suited for wind dispersal as evidenced by its rapid spread across the U.S. after its introduction (Kleczewski et al. 2020) and the top-down infection pattern in corn canopies of previously uninfected fields (Valle-Torres et al. 2020). Following contact with susceptible host tissue, an ascospore will form an appressorium to penetrate the leaf surface (Parbery 1963b). Intracellular hyphal growth facilitates colonization of host epidermal cells and leads to the formation of the clypeus, a shield-like structure of host 30 epidermis and melanized hyphae that protects the developing vegetative hyphae and fruiting bodies (i.e., perithecia and pycnidia) within the developing stroma (Caldwell et al. 2024; Parbery 1963b). Hyphal growth continues longitudinally and laterally into the leaf tissue, leading to stroma expansion and development on the abaxial epidermis (Caldwell et al. 2024). Pycnidia develop centrally to the clypeus and multiple perithecia develop surrounding the central pycnidium. After observing P. maydis development through histological studies, Caldwell et al. (2024) hypothesized that asexual conidia exuded from pycnidia act as spermatia by providing the complimentary mating-type idiomorph required for perithecia formation and production of asexual ascospores, resulting in this pattern of ascomata development in host leaf tissue. This hypothesis is also supported by MacCready et al. (2023) who demonstrated that P. maydis is heterothallic, needing complementary mating types for sexual reproduction by sequencing a mixed population of ascospores and conidia and finding a near equal mapping of reads to mat1-1 and mat1-2 mating type idiomorphs. A protocol for inoculation of corn by P. maydis also found that inoculation using primarily conidia led to poor tar spot symptom development (no symptom development or less than 10% incidence), while inoculating with ascospores led to 100% incidence (Breunig et al. 2023), suggesting ascospores are responsible for disease development and conidia act as spermatia in the disease cycle. Eventually, perithecia reach full maturity, swell, and force ascospores through the ostiole, completing the first cycle of inoculum production and infection in the polycyclic disease cycle. While much has been uncovered in recent years about the infection strategy of P. maydis, the optimal environmental conditions for pathogen and disease development are still poorly understood. In Mexico, P. maydis spore release during a tar spot epidemic was greatest when temperatures were between 17 to 23°C and relative humidity was over 85%, but a second peak in spore release was found when temperatures were over 23ºC and relative humidity < 70% (Hock et al. 1995). Ascospore germination and appressoria formation occurs increasingly between 10 and 20ºC and decreases at 25ºC in the presence of free water under laboratory conditions (Dittrich et al. 1991). Disease development is favored by temperatures between 17 to 22°C with moist conditions, including relative humidity over 75% and prolonged leaf wetness (Hock et al. 1995). However, recent epidemiological modeling research conducted across the Midwestern U.S. has found that tar spot development is negatively correlated with moisture variables (Webster et al. 2023). The reported length of the latency period from infection to sporulation 31 varies, likely due to differences in incubation conditions and experimental methods but has been reported between 10-21 days (Breunig et al. 2023; Dittrich et al. 1991; Gongora-Canul et al. 2023; Hock et al. 1995; Kleczewski et al. 2019; MacCready et al. 2023; Solorzano et al. 2023, 2024). Multiple protocols have been published for artificial inoculation with P. maydis in controlled environments to induce tar spot in corn (Breunig et al. 2023; Gongora-Canul et al. 2023). Their reproducibility between research groups is limited, potentially due to differences in inoculum sources, viability, and incubation conditions (Solorzano et al. 2023). These methods have the potential to be used to better understand the tar spot disease cycle, but they may be hindered by difficulty in carefully controlling experimental conditions, mimicking natural environments, natural variation of inoculum sources and the possible presence of multiple closely related species (Broders et al. 2022). Therefore, the role of the environment in the proliferation of P. maydis and the development of tar spot can be derived from empirical data collected during natural epidemics (Hock et al. 1995; Webster et al. 2023). To investigate the environmental triggers for spore release, spore traps have been used for decades (Granke et al. 2014; Hernandez and KC 2024; King et al. 1976; Pearson et al. 1980; Timmer et al. 1998). Impaction spore traps are particularly popular among botanical epidemiologists and include the Hirst-type and rotating-arm type spore traps (Figure 1E; Check et al. 2024a; Hirst 1952; West and Kimber 2015). Both are active sampling methods and use a power source to physically impact particles onto a sampling surface by either vacuum pressure or rotating the sampling surface at high speeds. In Mexico, Hock et al. (1995) monitored P. maydis spore release during a tar spot epidemic with a modified rotating-arm trap with a petroleum-jelly covered microscope slide and a Burkard spore trap. However, due to limitations in molecular identification at the time, spores were quantified by microscopic identification of P. maydis ascospores, excluding conidia in spore counts and possibly leading to misidentifications of ascospores due to their indistinct morphology as compared to some other fungal ascospores (Roggenkamp et al. 2024; West and Kimber 2015). As an example, Miller and Huhndorf (2005) demonstrated that ascospore morphology is extremely homoplastic and not useful for delimiting genera within the Sordariales. Spore traps are also valuable tools for providing timely data on the presence and development of inoculum throughout a growing season (Thiessen 2024). With airborne 32 inoculum surveillance and molecular detection tools, spores can be detected prior to the onset of visible disease symptoms, triggering scouting efforts or expediting disease management decisions (Munir et al. 2020; Pizolotto et al. 2021). Additionally, incorporating inoculum presence data into disease prediction models has the potential to increase prediction accuracy of fungicide application timing decision tools for in-season disease management (Newlands et al. 2018; Van der Heyden 2021). Regional air sampler networks are generally more common in regions with intense production of high value crops such as the wine-grape production regions of California, Oregon, and France and cucurbit-production regions in the Eastern U.S. (Coastal Viticulture Consultants (https://www.coastalvit.com); Laurent et al. 2022; Rahman et al. 2021; Theissen et al. 2016), but have been proposed for managing diseases of commodity crops such as soybean rust (Isard et al. 2011) and Fusarium head blight of wheat (Wang et al. 2024). There has also been renewed interest in inoculum surveillance systems in private industry with several companies offering pathogen detection systems to growers using various trapping, identification and quantification methods (Pollensense™ (www.pollensense.com); Root Applied Sciences (www.rootappliedsciences.com); Scanit Technologies, Inc (www.scanittech.com); Spornado (www.spornadosampler.com)). The potential for a regional spore trapping network to assist in disease management decisions for Midwestern corn producers can be justified by the economic value of losses due to tar spot in recent years and the volatility in year-to-year epidemic pressure (Mueller et al. 2020). Additionally, a TaqMan qPCR assay for the detection and quantification of P. maydis has been developed and validated by Roggenkamp et al. (2024). This assay targets the ITS region and was demonstrated to be species-specific when tested against a panel of closely related Phyllachora spp. and other common North American corn pathogens and endophytes. Using a spore standard curve prepared by a 10-fold dilution series of spore suspension, the Roggenkamp et al. 2024 assay was able to detect down to 152 spores. With the validation of molecular detection tools and spore trapping methods for P. maydis, the potential for a spore trapping network for early detection of P. maydis for Midwestern farmers and researchers exists. As tar spot continues to establish itself as a yield-limiting disease of corn in the U.S., elucidating the environmental requirements for spore release is critical to our understanding of the disease cycle and will aid in disease management by identifying potentially effective disease management practices and predicting disease epidemics. By validating the use of an existing P. 33 maydis qPCR assay for monitoring spore release, there is the potential for inoculum surveillance programs to help farmers with timely scouting and disease management efforts. The objective of this study was to describe the relationship between environmental conditions and spore capture through correlation analysis and mixed effects logistic regression modeling. Additionally, the ability of spore traps to detect P. maydis spore presence prior to visible tar spot symptoms was assessed and discussed. Materials and Methods Air sample and field data collection Rotating-arm spore traps (constructed as detailed by Check et al. 2024b) were deployed in Ingham and Van Buren counties, Michigan from 2021 to 2023 (Table 1). Rotating-arm spore traps used ER308/308L TIG welding rod (Grainger, Lake Forest, IL, U.S.) cut to 31.75-mm lengths as the sampling rods and were operated at 2400 rotations per minute (RPM). The theoretical sampling rate of the rotating-arm spore trap was 66 L/min based on the sampler design and speed (Aylor 2017; Check et al. 2024b). One Burkard 7-Day Recording Volumetric Spore Sampler (Burkard Manufacturing Co. Limited, Hertfordshire, U.K.) was deployed in all years at the Van Buren County location and in 2023 at the Ingham County location. During each site visit, Burkard samplers were confirmed to be intaking air at 10 L/min. The circumference of the Burkard spore trap drum was covered with 1.9 cm wide clear, polyester non-adhesive tape (Tape-Rite, New Hyde Park, New York, U.S.). Vacuum grease (Dow Corning, Midland, MI, U.S.) was used to coat the sampling surfaces and applied in a thin, even layer using gloved hands (Hirst 1952; Lacey and West 2006). Burkard and rotating-arm spore traps at the Van Buren County site were changed weekly. A single rotating-arm spore trap was changed every 3-4 days at the Ingham County location. Samplers were deployed immediately adjacent to naturally-infested fields. The height of the rotating-arm spore trap rods was 0.91 m (3 ft), and the height Burkard spore trap orifice was 0.46 m (1.50 ft). Heights were chosen to target initial spore release from infested residue on the soil surface. Spore traps were stationary throughout the growing season. Spore trapping experiments began in June and July prior to corn reproductive growth stages (VT/R1) and prior to tar spot detection in these fields (Abendroth et al. 2011). No other foliar diseases were present at appreciable levels near the spore traps. Air sampling continued until plants reached full physiological maturity (R6) and had senesced, at which time P. maydis would no longer have viable plant tissue to infect and reproduce. Rigorous 34 scouting was conducted weekly in fields to document the first incidence of visual tar spot symptoms (Table 1). Incidence data was collected from untreated check plots of hybrid and fungicide trials as the percentage of plants with any level of visible tar spot symptoms within the plot. Untreated checks were replicated four times within each trial. Plots were 5.18 m (17 ft) long by 3.05 m (10 ft) wide with 4 planted rows. Sample rods from rotating-arm spore traps were placed in 2-mL screw cap tubes, transported to the lab and stored at -20(cid:0)C until processing. Drums of the Burkard spore trap were swapped out in the field, transported back to the lab, processed immediately by cutting into 48 mm (24 hour) segments, placed into 2-mL screw cap tubes and stored at -20ºC until processing. DNA extraction from samples and positive controls Due to the obligate nature of P. maydis and the challenge of harvesting genetic material, positive controls were sourced from DH5ɑ Escherichia coli cells transformed with a pUC-GW- Kan plasmid vector (Azenta Life Sciences, South Plainfield, NJ, U.S.) containing the P. maydis partial internal transcribed spacer (ITS) 1 partial sequence, 5.8S ribosomal RNA and ITS 2 complete sequence, and the large subunit ribosomal RNA gene partial sequence (GenBank MG881847) (McCoy et al. 2018). The insert was 464 base pairs (bp), and the final size of the plasmid was 3,102 bp. DNA was extracted from the transformed plasmids using the Monarch® Plasmid Miniprep Kit (New England Biolabs, Ipswich, MA, U.S.). DNA concentration was quantified using the Quant-iT dsDNA broad range kit (ThermoFisher Scientific, Waltham, MA, U.S.). DNA was diluted to 106 plasmids per µL which amplified at 20 cycle thresholds (Ct). The DNeasy Plant Mini Kit (Qiagen, Venlo, Netherlands) was used for DNA extraction from air samples. Sample disruption protocols differed between Burkard and rotating-arm spore traps to accommodate their different characteristics and limitations (material, volume, quantity of vacuum grease). Samples were spiked with 10 ng salmon sperm DNA (Sigma Aldrich, St. Louis, MO, U.S.) prior to extraction to improve DNA recovery by acting as carrier DNA (Carisse et al. 2009a). To 2-mL screw cap tubes containing the air samples, 0.25 g of garnet lysing matrix (MP Biomedicals, Irvine, CA, U.S.) was added. Tubes containing the rotating- arm air samples, 10 ng of salmon sperm, garnet lysing matrix, and DNA extraction buffer were vortexed horizontally for 3 minutes at maximum speed (Vortex Genie 2, Scientific Industries, Bohemia, New York, U.S.) and centrifuged for 5 seconds at 10,000 RCF to dislodge particulate from the sampling surface and lyse spores. Using sterile forceps, the metal rods were removed 35 carefully. A ¼” ceramic sphere was added to the tubes and samples were homogenized in a FastPrep Tissue Homogenizer (MP Biomedicals, Irvine, CA, U.S.) at 6 m/s for 40 seconds. Samples containing spore suspensions alone were processed similarly. Tubes containing Burkard tape samples suspended in DNA extraction buffer were vortexed on a horizontal vortexer for 8 minutes, incubated on a 65(cid:0)C heat plate for 10 minutes and horizontally vortexed again for 8 minutes. Samples were then centrifuged at 11,000 RCF for 1 minute to settle particulates. To increase the solubility of the large volume of vacuum grease, 100 µL of chloroform was added to the tube, briefly vortexed and centrifuged at 20,000 RCF. The rest of the extraction process was followed according to the manufacturer’s protocol. Quantitative PCR with mock and environmental samples An exogenous internal control (EIC) was used to verify that air samples without amplification were true negatives and not due to the presence of PCR inhibitors in environmental samples (Haudenshield and Hartman 2011). EIC DNA was added at 1 fg/µL directly to the qPCR mastermix prior to distribution to reaction wells and amplified consistently at 29-31 cycle threshold (Ct). A standard curve was prepared using plasmid DNA in reaction with the EIC to confirm there was no interference between the two targets using 102 to 108 fg of P. maydis ITS target DNA (Supplemental Figure 1). Both PerfeCTa qPCR ToughMix and Multiplex mix (Quantabio, Beverly, MA, U.S.) were assessed for their performance in the presence of the ITS and EIC targets. To establish qPCR assay efficiencies, limits of detection, and demonstrate the relationship between spore quantity and Ct, standard curves were prepared for spores (1) in suspension, (2) on Burkard tapes, and (3) on rotating-arm spore trap rods. To prepare mock samples, ascospores of P. maydis were harvested by incubating infested leaf material in a 50-mL conical centrifuge tube of 45 mL of sterilized water on a shaking benchtop shaker table for 8 hours. The suspension was filtered through four layers of cheesecloth and ascospores were quantified using a hemocytometer. The spore suspension was then concentrated by centrifugation and resuspended in sterile water to a concentration of 106 spores per mL. The spore suspension was pipetted into 2-mL screw cap tubes, onto Burkard tapes, or on rods. For spore trap mock samples, the surfaces were coated with grease prior to pipetting and the suspension was allowed to evaporate fully from the sample surface before being placed into tubes. Mock samples were prepared for generation of standard quantification curves in triplicate as technical replicates and 36 inoculated with 10, 50, 500, 1,000, 5,000 and 10,000 ascospores. Samples were stored at -20(cid:0)C until processing and DNA extraction was performed using the same protocol as environmental samples. Linear regressions were modeled between the Ct and the log10 number of spores for each standard curve sample type (Burkard, rotating-arm, and spore suspension). The qPCR assay efficiency was calculated as: 𝐸𝑓𝑓𝑖𝑐𝑖𝑒𝑛𝑐𝑦 = 10 !" # $%&’( ) − 1 where slope was taken from the resulting linear regression (Bustin et al. 2009). The coefficient of determination (R2) between Ct and spore number was calculated and reported for each sample type. Primers, probes, and thermocycling conditions for a TaqMan-based qPCR assay targeting the ITS region of P. maydis used in this study were previously evaluated and characterized by Roggenkamp et al. (2024). Modifications were made to the reaction mix to include the use of the EIC (Supplemental Table 1). The qPCR were run in duplicate as technical replicates for all environmental samples. All qPCR in this study were performed on the Bio-Rad CFX96 Touch Real-Time PCR Detection System (Bio-Rad Laboratories, Hercules, CA, U.S.). Weather data sourcing Hourly air temperature (ºC) relative humidity (%), wind speed (m/s), precipitation rate (mm/hour) and cumulative precipitation (mm) were sourced using historical weather data services with a 4-km resolution (The Weather Company, IBM). Hourly data was used to calculate durations of relative humidity at 70%, 80% and 90% thresholds, the duration of relative humidity over 90% at night (10 p.m. to 6 a.m.) and durations of temperature and relative humidity conditions previously found to be associated with tar spot development (Webster et al. 2023) and P. maydis spore capture (Hock et al. 1995). Hourly weather data was summarized into daily maximums, minimums, means, sums and durations. Daily weather data was transformed into moving averages (MAs) across 2- and 3-day windows using the rollmean() function from the R package “zoo” v. 1.8 (Zeileis and Grothendieck 2005). Corresponding weather data and spore quantities, normalized by site-year to mitigate effects due to differences in the magnitude of capture by each trap type, were plotted to show trends between weather variables and spore capture. 37 Data analysis To calculate spore number from qPCR Ct values, the following equation was used: 𝑄𝑢𝑎𝑛𝑡𝑖𝑡𝑦 = 10[(,-"./-(01(’-)/4%&’(] where the intercept and slope correspond to the sample type’s standard curve linear regression. To compare the Burkard and rotating-arm spore trap, spore quantity was transformed to spore density (spores/m3) to adjust for differences in sampling rate and duration. The following equation was used for this transformation: 𝑆𝑝𝑜𝑟𝑒 𝑑𝑒𝑛𝑠𝑖𝑡𝑦 = 𝑄𝑢𝑎𝑛𝑡𝑖𝑡𝑦 𝑠𝑎𝑚𝑝𝑙𝑒 × 1000 𝐿 𝑚6 𝑆𝑎𝑚𝑝𝑙𝑖𝑛𝑔 𝑟𝑎𝑡𝑒 × 𝑆𝑎𝑚𝑝𝑙𝑖𝑛𝑔 𝑑𝑢𝑟𝑎𝑡𝑖𝑜𝑛 where sampling rate (L/min) and sampling duration (minutes/sample) is dependent on the spore trap type. A linear model was fit and a Spearman correlation analysis was performed to analyze the relationship between spore density captured by Burkard and rotating-arm spore traps. Prior to calculating spore density, daily Burkard spore captures were summed to match the window length of the rotating-arm spore trap samples. Data prior to the year and location’s first positive detection of either trap type were excluded in the linear model to avoid skewing the estimation of correlation strength between trap types. Spearman correlations between spore quantities and 1-, 2- and 3-day MAs were performed using the cor() function for the R package stats (R Core Team 2024). Prior to performing correlations, spore quantities were normalized separately for each site-year. Data only from years that had disease development and spore capture were used in correlation analyses. Correlations were assessed for their significance at the ⍺ = 0.05 level. Mixed effects logistic regression models were fitted to Burkard spore trap data from all years and locations. Burkard spore trap data was used alone because of the finer time scale and therefore greater number of observations available for modeling in comparison to rotating-arm spore traps. Independent variables included daily summaries (maximums, minimums and means) of weather variables. Only daily summaries of raw weather inputs (temperature, relative humidity, wind speed, precipitation, precipitate rate) were considered for model development as durations and summaries are correlated would be autocorrelated. Binary response data were generated based on the positive detection or absence of detection based on detection in samples by qPCR. Random effects included the trapping location and year. The strength and significance (P < 0.05) of correlations and the performance of single-variable logistic regression models for 38 each MA window determined which time window weather variable pool was used for model development. An automated and exhaustive screening of candidate logistic regression models was performed using the glmulti() function from R package “glmulti” v. 1.0.8 (Calcagno 2020) on the Michigan State University High Powered Computing Cluster. Models were ranked based on their Akaike’s Information Criteria (AIC) score. The maximum number of predictors in the candidate logistic regression models allowed was three and polynomial terms were not included in the variable pool. The top 100 models were output from the search and the best single-variable and multivariate models that did not contain multiple functions of the same weather variables were selected for further assessment. The selected models were assessed at probability thresholds from 0.05 to 0.95 by a 0.05 step based on their effect on balancing model specificity and sensitivity. Selected models were evaluated on Akaike information criteria (AIC), area under the receiver operating curve (AUROC), Cohen’s kappa coefficient (Kappa), Tjur’s coefficient of discrimination (Tjur’s R2) (Hughes et al. 2019), accuracy (%), balanced accuracy (%) (the average accuracy of predictions for both classes), specificity (%) and sensitivity (%) (Table 2). Geneious software v. 2024.0.7 was used for sequence alignment between qPCR primers, probes and GenBank accessions. Results qPCR standard curves and correlation between Burkard and rotating-arm air samples Limits of detection were consistent across sample types, all reaching 50 spores and amplifying at 36.25, 36.52 and 38.09 Ct for spores in suspension, Burkard samples, and rotating- arm spore trap samples, respectively (Figure 2). Inconsistent amplification was observed for 10 spores with only 41.66% of replicates amplifying before 40 Ct. Spore number and Ct were highly correlated with an R2 value of 0.9963, 0.9881 and 0.9755 and efficiencies varied between the sample types with 80.47%, 61.56% and 75.35% for Burkard, rotating-arm and spore suspension samples, respectively. The quantity of spores captured by each trap was moderately correlated (τ = 0.7) and the linear model relating capture by the two traps was statistically significant (P < 0.01) (Figure 3). The slope of the linear regression indicates that 21x more spores per m3 of air sampled were captured and detected by the rotating-arm spore trap in comparison to the Burkard spore trap. First P. maydis spore and tar spot detections Dates when experiments began ranged from late June to mid-July and are listed in Table 39 1. A total of 95 samples were collected from rotating-arm spore traps. A total of 324 samples were collected from Burkard spore traps, representing daily spore captures. Rotating-arm spore trap samples had 28% of samples with positive detections and 72% of samples with no amplification. Burkard spore trap samples had 18% of samples with positive detections and 82% of samples with no amplification. All samples that failed to amplify were confirmed as true negatives by the EIC. No positive spore trap samples were observed in 2022 or Ingham County in 2023 (Figure 4). In all years and locations, tar spot symptoms were visually confirmed in fields before the first P. maydis positive spore samples were detected. However, spores were detected prior to incidence reaching 100% in 2021 in Ingham and Van Buren counties and in Van Buren County in 2023. The delay between the first visible observation of tar spot symptoms and the first positive detections of P. maydis air samples ranged from 13 to 31 days, averaging 23 days. The maximum number of spores captured by both trap types corresponded closely in time at the Van Buren County location in 2021, but not in 2023, when the peak spore capture from the rotating-arm spore trap occurred at the end of the growing season (Figure 5). In all years where both trap types were deployed at the same field, the first positive detection of P. maydis by the Burkard and rotating-arm spore traps occurred within one week of each other. Correlation analysis among spore quantity and weather variables Maximum, mean and minimum of temperature, humidity, wind speed, precipitation rate, precipitation, hours of leaf wetness (over 90% humidity) at night, and durations of humidity and temperature ranges were examined for their statistical significance and correlation strength with spore quantity, normalized to the maximum by year and location using Spearman’s correlation analysis. Moving averages summarized across 1, 2 and 3 days were evaluated. Between 1-, 2- and 3-day MAs of weather data, 1-day MAs had the greatest number of significant correlations with spore quantity among weather variables that represented functions of weather conditions (Figure 6; Supplemental Table 2) and were therefore used for generating logistic regression candidate models. Minimum temperature (MinTemp) and mean temperature (MeanTemp) were significantly (P < 0.05) negatively correlated with normalized spore quantity for all MA windows. The correlation coefficients between spore quantity and minimum and mean temperature were as follows: -0.24 and -0.25 for 1-day MAs, -0.26 and -0.26 for 2-day MAs and -0.26 and -0.27 for 3-day MAs. Spore capture was also significantly negatively correlated with 1-day MAs of maximum precipitation rate (MaxPcp) with a coefficient of -0.33 but was not 40 significantly correlated in other MA windows. The duration of temperature between 16.6 and 23°C and relative humidity over 85% (HockPeak1) was previously demonstrated to coincide with peaks in P. maydis spore capture (Hock et al. 1995) but was significantly negatively correlated to spore capture in all MA windows (-0.07, -0.10, -0.14). Similarly, the duration of temperature between 17 and 25°C (WebsterTempDur) was significantly negatively correlated with spore capture for 2-day and 3-day MAs (-0.26, -0.25). No functions of relative humidity or wind speed were significantly correlated with spore capture among any MA windows (P > 0.05). For all MA windows evaluated here, there was a negative correlation between hours of leaf wetness at night and spore capture (-0.12, -0.08, and -0.11 for daily, 2- and 3-day MAs), but was not found to be statistically significant (P > 0.05). Logistic regression model development From the 100 candidate logistic regression models generated using the full variable pool, six were selected for further testing based on their predictors and AIC rank: LR1 = -1.48 – 0.16(MeanTemp) LR2 = -0.54 – 0.05(MeanHum) LR3 = 5.14 – 0.17(MeanTemp) – 0.07(MaxHum) LR4 = -2.41 – 0.17(MinTemp) + 0.49(MinWS) LR5 = 3.30 – 0.17(MinTemp) – 0.06(MaxHum) + 0.33(MinWS) LR6 = 4.17 – 0.14(MinTemp) – 0.07(MaxHum) – 0.03(MaxPcp) A temperature and relative humidity predictor was represented in all models except for the single variable models and LR4, which incorporated minimum temperature and wind speed. When models were expanded to include three predictors, models included either minimum windspeed or maximum precipitation rate. A probability threshold of 30% was selected based on the balance of model specificity and sensitivity (data not shown). Model performance was compared using the following metrics: Akaike’s Information Criteria (AIC), Area Under the Receiver Operating Characteristic curve (AUROC), Cohen’s kappa coefficient (Kappa), Tjur’s coefficient of discrimination (Tjur’s R2), accuracy (%), balanced accuracy (%), specificity (%) and sensitivity (%) (Table 2). Kappa ranged from 0.43 to 0.55 for all the selected models, indicating moderate agreement between the observations and model predictions (Landis and Koch 1977). All models achieved at least 0.77 AUROC, 77.42% balanced accuracy and a Tjur’s R2 of 0.58, demonstrating good discrimination between the two binary classes; positive P. maydis detection 41 and no detection. LR3 had the greatest AUROC (0.86), Kappa (0.55), Tjur’s R2 (0.71), accuracy (83.02%), balanced accuracy (85.56%), specificity (89.47%) and sensitivity (81.65%) of all the candidate models. LR3 scored and AIC of 205.66, which was marginally higher than the LR4 model that scored 205.35. In all the selected models, temperature, relative humidity and precipitation had a negative model coefficient for predicting the probability of positive P. maydis detection. Minimum wind speed was included in two of the selected models and had a positive model coefficient for predicting the probability of positive P. maydis detection. The response surface of LR3 was plotted for each site-year to visualize the interactions between mean temperature, maximum relative humidity and the probability of spore presence (Figure 7). Thus, as relative humidity increased, the probability of spore capture decreased, and reached 0% probability at 100% relative humidity. Similarly, as temperature decreased, the probability of spore presence increased and approached an upper asymptote around 15°C. Discussion The data generated in this study represents the first description of environmental conditions related to P. maydis spore release as observed through spore trapping and molecular detection. Previous P. maydis epidemiological studies have been conducted relating environmental conditions to the increase of tar spot severity in field plots in the U.S. (Webster et al. 2023), the success of P. maydis inoculation in controlled environments (Breunig et al. 2023; Gongora-Canul et al. 2023; Solorzano et al. 2023) and spore release using visual identification for quantification in Mexico (Hock et al. 1995). Previous studies in the U.S. focused solely on the development of tar spot, while the results from this study offer insight into the discrete process of spore release as it relates to environmental conditions, a currently under-studied part of the disease cycle. In this study, two new sample types were validated for use with a previously published qPCR assay for P. maydis detection (Roggenkamp et al. 2023): Burkard spore trap tapes and rotating-arm spore trap rods. On all sample types, a slightly lower spore LOD was achieved in this study (50 spores) than in the initial assay validation (152 spores). However, a decrease in qPCR efficiency was also observed (from 87.60% to 61.56-80.47%). The lower efficiency was not due to poor primer and probe design as the assay achieved 100% efficiency when used on non-environmental DNA (Supplemental Figure 1). There are multiple potential sources of 42 qPCR inhibitors in spore trap samples, such as the rotating-arm spore trap metal rods shedding ions during mechanical lysis process (Kuffel et al. 2020) or the presence of pollen, dust, and insects collected on the sampling surfaces (Mahaffee and Stoll 2016). However, the EIC consistently amplified in all samples (data not shown), demonstrating that there was no significant qPCR inhibition in spore trap samples. The methods used for processing samples determines DNA extraction efficiency, which directly impacts the sensitivity of the downstream qPCR analysis (Yang et al. 2021). Due to the nature and volume of samples, the methods used for sample homogenization and DNA extraction differed from the original study. This study and the previous study differ in their source of carrier DNA (salmon sperm DNA vs. corn leaf disks, respectively), lysis methods (vortexing, bead beating, and heating vs. bead beating alone) and DNA extraction methods (column-based vs. phenol-chloroform). While differences in qPCR efficiency are commonly seen between pure DNA and spore trap samples (Bello et al. 2021; Munir et al. 2020; Rahman et al. 2021; Wang et al. 2024), further optimization of laboratory methods will be necessary to achieve lower sensitivity and higher qPCR efficiency while remaining suitable for processing large numbers of samples for the development of an early detection system for tar spot. A recent study by Broders et al. (2022) aimed to characterize the genetic diversity within the genus Phyllachora by conducting a phylogenetic analysis on samples collected from diverse geographies and grass host species. Using the ITS and LSU gene regions, the authors identified five distinct genetic clusters corresponding to five ITS haplotypes. Genetic clusters 1, 2 and 3 included isolates collected from tar spot symptomatic corn in the Midwestern U.S. Genetic cluster 1 included only contemporary P. maydis isolates from the U.S., while genetic clusters 2 and 3 included contemporary and historical P. maydis isolates from North, Central and South America. Genetic cluster 3 also contained other Phyllachora species collected across the globe (i.e., P. vulgata P. sylvaticum, P. rottboellia, P. junci, P. heraclei, P. graminis, P. euphoribaceae, P. epicampis, P. diplocarpa, P. chaetocloae), suggesting that species within the genus Phyllachora are not limited to a singular host as previously understood and expected due to their obligate biotrophic lifestyle (Mardones et al. 2017; Parbery 1967). These results also support previous work by McCoy et al. (2019) who performed a fungal community network analysis using the ITS1 locus on leaf samples taken from a tar spot infested Michigan corn field. The authors found that 95% of corn leaf samples were coinfected with at least 2 distinct P. 43 maydis operational taxonomic units (OTUs), suggesting a significant degree of coinfection amongst OTUs and diversity in the ITS1 locus. The qPCR primers and probe targeting the ITS sequence used in this study differ in sequence from genetic clusters 4 and 5, but have 100% identify with genetic clusters 1, 2 and 3 (Broders et al. 2022). Therefore, the diversity of P. maydis currently present in the Midwestern U.S. should be detectable and quantifiable with the qPCR assay used in this study. Redesign of the qPCR primer and probe sequences will be necessary to produce an assay able to differentiate between genetic clusters 1, 2 and 3 for future study of P. maydis population genetics. The grouping of other Phyllachora species with P. maydis in genetic cluster 3 contradicts the specificity testing conducted by Roggenkamp et al. 2023 who used single nucleotide polymorphisms in the ITS region to design and validate the species-specific assay for P. maydis against P. graminis and P. vulgata. Based on the genetic clusters proposed by Broders et al. (2022), P. graminis isolates used by Roggenkamp et al. (2023) (GenBank OP831217 and OP831218) belong to genetic cluster 4, whereas Broders et al. (2022) found P. graminis grouping in both genetic clusters 3 and 4. The P. vulgata isolates used by Roggenkamp et al. (2023) (GenBank OP831211 and OP831215) do not align with any genetic cluster, while the isolate used by Broders et al. (2022) was grouped in genetic cluster 3. The isolates used in these studies do not include type specimens, and the identity of herbarium isolates is based on morphology and host species identity and therefore might be misleading. Future work will be needed to improve the genetic characterization of Phyllachora spp. to elucidate the presence of these possible subspecies through improved phylogenetic analyses with additional genetically informative loci (Mardones et al. 2017). This work will be crucial for understanding potential differences among subspecies in detection tools, host range/preference, virulence, and other epidemiologically-relevant factors. The support for comparing Burkard and rotating-arm spore trap performance in this study was limited by the low number of positive samples obtained by the rotating-arm spore traps. Dates of first P. maydis spore detection was similar between both spore trap types, and the presence and absence of spore detections by both trap types largely coincided. However, spore capture by the rotating-arm spore traps was multiple magnitudes greater than the capture by Burkard spore traps after adjusting for sampling rate and duration. Previous research has found conflicting results for comparing the quantity of spores captured by each trap type (Aylor 1993; 44 Crisp et al. 2013; Evenhuis et al. 1997; Sutton and Jones 1976; Torfs et al. 2019). Evenhuis et al. (1997) used a Burkard and rotating-arm spore trap to capture Mycocentrospora acerina and microscopic identification to quantify spore release in caraway fields. The authors found that while rotating-arm traps effectively captured ascospores, no ascospores were captured by the Burkard trap after nearly 1000 hours of sampling. A more recent study by Torfs et al. (2019) comparing Venturia inaequalis ascospore quantification by qPCR found that continuously running Burkard and rotating-arm spore traps were significantly correlated with a Pearson’s correlation coefficient of 0.66 and there was no significant difference in spore quantity between the trap types. Aylor (1993) used spore traps and microscopic identification to observe V. inaequalis spore release following wetting events and found that a Burkard spore trap consistently captured more spores than a rotating-arm spore trap across a range of wind speeds under field conditions. Performance of spore traps are highly influenced by their design and implementation, possibly obscuring biological explanations for differing results between spore trap types and individual studies. While understanding the difference in trap performance will be important for their implementation in experiments and pathogen surveillance networks, both trap types used in this study were able to first detect spores within one week of each other, which will be the critical factor in using spore traps to monitor epidemics. Additional work can be conducted to explore how spore trap placement determines the probability of detection of P. maydis from local and distant inoculum sources (Mahaffee et al. 2023). The delay in the first detection of P. maydis by spore traps in comparison to the detection of disease demonstrates that the spore traps used in this study were unable to detect initial spore flights. Detection of airborne inoculum prior to disease detection has been demonstrated in other pathosystems, leading to the development of early warning systems to initiate scouting efforts or fungicide sprays (Carisse et al. 2009b; Dhar et al. 2020; Thiessen et al. 2017). Based on a latency period of 14 days, the lag between first disease and spore detection suggests that spores were first detected during the second to third round of inoculum production. Field crops generally have higher action thresholds than fresh market crops, so capturing first spore flights is less critical in these production systems. Additionally, the average footprint per operation is much greater for field crops than specialty crops, in the Midwest making thorough scouting efforts challenging. Currently, studies on fungicide timing for optimal tar spot control have relied on growth stages to determine application timings (Ross et al. 2024), and a critical disease 45 threshold is unknown. Field observations suggest that incidence between 50% and 100% may be an effective economic threshold (Martin Chilvers, personnel communication), but this hypothesis remains untested. Although initial spore flights were not captured, positive spore detections were made before incidence measured 100%. Therefore, a P. maydis monitoring system that can detect spores prior to 100% incidence may still be an effective tool for helping farmers improve their reaction time to P. maydis incursions. Weekly monitoring of disease progression and spore capture will help elucidate the relationship between spore capture, epidemic development and yield loss to provide disease management recommendations or facilitate the creation of a risk advisement system using spore quantity data (Dhar et al. 2020; Newlands 2018). The development of logistic regression models demonstrated that spore release can be explained by functions of recent temperature and moisture conditions. The results from this study suggest that spore capture is negatively related to daily precipitation, minimum temperatures and mean temperatures. Surprisingly, the correlation coefficients and candidate linear regression models showed a negative relationship between relative humidity and spore capture. A previous study on P. maydis epidemiology by Hock et al. (1995) found that peaks in weekly spore catches coincided with two conditions: relative humidity greater than 85% and moderate temperatures (17 to 22ºC), and relative humidity less than 70% and high temperatures (over 23ºC). In this study, no significant (P > 0.05) correlation was found with the first set of conditions described by Hock et al. (1995), and a significant (P < 0.001) negative correlation was found with the second set of conditions. A recent study by Webster et al. (2023) utilized logistic regression modeling to predict the increase of tar spot severity and found that moisture variables (i.e., relative humidity, dew point, leaf wetness duration) were negatively correlated with tar spot development. Unlike the measurement of disease progression, spore trap data were affected by their interaction with environment and are differentially affected by spore trapping instruments. For example, the collection efficiency of a Burkard spore trap is very sensitive to wind speed while a rotating-arm spore trap is less sensitive (Frenz 2000; Jackson and Bayliss 2011). The environmental conditions experienced during these experiments may be unequally affecting spore trap performance, leading to biases between the two traps in spore quantities and the relationship between spore capture and environmental conditions. Currently, the dominant mechanism for P. maydis spore dispersal (rain splash vs. wind) is unidentified. With the spore trapping techniques used here, the proportion of spore release and dispersal through rain splash is 46 unaccounted for. While this study serves as an initial description of the environmental drivers of spore release in P. maydis, additional years and locations of spore trapping data will help substantiate or challenge the initial findings described here. A common limitation of using spore trapping to describe the relationship between spore release and environmental conditions is the difference in temporal scales in which spore capture (daily and multi-day windows) and weather (daily summaries) is observed compared to the instantaneous ejection of spores in nature (Check et al. 2024a). Additionally, in this study, weather data are taken from real-time mesoscale analysis weather data services with 4-km resolution and not from within the microclimate that is directly influencing spore release, potentially introducing incongruities between weather data used for modeling and experienced at the microclimate scale. Future studies on P. maydis spore dispersal would benefit from collecting air samples at a finer time scale, potentially looking at diurnal patterns of spore release (Guo and Fernando 2005; Carisse and Philion 2002; Gottwald and Bertrand 1982) to see the effect of nighttime moisture conditions, as highlighted as important in previous work for disease development (Webster et al. 2023), as well as collecting on-site weather data within the microclimate of the crop canopy in which P. maydis stroma are interacting with the environment. The results from this study contribute to the current understanding of P. maydis epidemiology and serves as the initial description of the interaction between spore release and environmental conditions as achieved through spore trapping and molecular quantification. P. maydis spore release is significantly negatively correlated with daily summaries of minimum temperature, mean temperature, maximum precipitation and durations of temperature between 16.6 to 23ºC and relative humidity over 85%), and can be predicted using daily measurements of maximum humidity and mean temperature. These models have the potential to be incorporated into a mechanistic model for predicting tar spot epidemics (Caffi et al. 2011; Salotti and Rossi 2023) or expanded upon to serve as a tar spot risk prediction tool (Dhar et al. 2020; Newlands 2018). Additionally, the use of spore trapping and qPCR as demonstrated here can be used to answer key questions about P. maydis lifestyle and tar spot management. Potential research avenues include the role of tillage in local inoculum survival (Forrer et al. 2021; Hofgaard et al. 2016; Schaafsma et al. 2005) or modeling horizontal and vertical spore dispersal gradients from inoculum sources (Renfroe-Becton et al. 2024; Eversmeyer and Kramer 1985; Roelfs 1972). Finally, improvements in field and laboratory methods to achieve greater detection 47 sensitivity will improve the feasibility of a P. maydis early detection spore trap network to guide growers on scouting and disease management. 48 Table 2.1: Years and locations where spore traps were deployed and start and end dates when the first and last sample was collected. Dates of first detections of tar spot symptoms and first detections of P. maydis spores by spore traps. All locations were fields planted with continuous corn rotations and natural tar spot pressure. Tables aNumber of samples collected at the year and location bDenotes no spore capture was observed 49 Table 2.2: Performance of binomial logistic regression models fit using daily Burkard spore trap data (n=324; 4 site years). Models predict the probability of spore capture based on daily summaries of weather conditions. aAkaike information criteria; a lower AIC score indicates a better fit to the data by the model bArea under the receiver operating characteristic curve; AUROC values range from 0 to 1 with values closer to 1 indicating better model performance cCohen’s kappa coefficient; Kappa values range from 0 to 1 with values closer to 1 indicating better agreement between observed values and model outputs dTjur’s coefficient of discrimination; A logistic regression analogue for coefficient of determination (R2) eBalanced accuracy, a measure of the average accuracy for both the majority and minority classes during classification fSpecificity, the proportion of predicted negatives to observed negatives gSensitivity, the proportion of predicted positives to observed positives 50 Figures Figure 2.1: Phyllachora maydis ascus under 400x magnification. Scale bar represents 20 μm (A). P. maydis conidia under 400x magnification. Scale bar represents 20 μm (A). Tar spot infection at the foliar (C) and field scale (D). Deployed rotating arm spore trap (left) and Burkard spore trap (right) at the Van Buren location in 2021 (E). 51 Figure 2.2: Standard curves for Burkard spore trap samples, rotating arm spore trap samples and spore suspensions. Each concentration is the product of three biological replicates and two technical qPCR replicates. Standard curves consisted of 3 biological replications and 2 technical replications of 50, 100, 500, 1,000, 5,000 and 10,000 ascospores. 52 Figure 2.3: Relationship between spores captured by the Burkard and rotating arm spore trap. All data from locations where both spore trap types and any positive detections within the year were used. 53 Figure 2.4: Spore quantity and disease incidence for all years and locations separated by trap type. 54 Figure 2.5: Phyllachora maydis spore capture and corresponding weather conditions for each location and year where positive spore detections were found. The number of spores were calculated using the standard curves for the corresponding sample type. Spore counts were normalized to each location, year and spore trap type for visualization of trends. Dates where rotating arm samples were collected are indicated by an asterisk. Burkard samples were collected for every date represented. 55 Figure 2.6: Correlation plots separated for 1 to 3-day moving average windows. Spore capture was normalized by dividing by the location and year’s maximum spore quantity prior to performing correlations (ScaleQuant). Moving averages of maximums, means and minimums of temperature (Temp), relative humidity (Hum), wind speed (WS), precipitation rate (MaxPcp), precipitation (Pcp), durations of relative humidity at 70%, 80% and 90% (Hum70, Hum80, Hum90) and durations of nighttime leaf wetness (WetNight) are represented. Various durations of specific conditions were included based on previous literature on the epidemiology of Phyllachora maydis: durations of temperature between 17 and 25 degrees C (WebsterTempDur), durations of temperature between 17 and 23 degrees C (HockTempDur), durations where temperatures ranged from 16.6 to 23 and relative humidity was over 85% (HockPeak1) and durations of temperature over 23.6 degrees and humidity less than 70% (HockPeak2). Spearmans’s rho is represented by color and shade. Significance levels are represented by the number of plot characters within each box with 1, 2, and 3 plot characters representing P < 0.05, 0.01 and 0.001, respectively. 56 Figure 2.7: Logistic regression model surfaces of the best performing model (LR3) for all locations with Burkard data. 57 LITERATURE CITED Abendroth, L. J., Elmore, R. W., Boyer, M. J. and Marlay, S. K. 2011. Corn Growth and Development. Iowa State University Extension, Ames, IA. Aylor, D. E. 1993. Relative collection efficiency of Rotorod and Burkard spore samplers for airborne Venturia inaequalis. Phytopathology 83:1116-1119. doi.org/10.1094/Phyto-83- 1116 Aylor, D. E. 2017. Aerial sampling. Pages 339-360 in: Aerial Dispersal of Pollen and Spores. D. E. Aylor, ed. American Phytopathological Society, St. Paul, MN. Bello, J. C., Sakalidis, M. L., Perla, D. E. and Hausbeck, M. K. 2021. Detection of airborne sporangia of Pseudoperonospora cubensis and P. humuli in Michigan using Burkard spore traps coupled to quantitative PCR. Plant Dis. 105:1373-1381. doi.org/10.1094/PDIS-07-20-1534-RE Breunig, M., Bittner, R., Dolezal, A., Ramcharan, A. and Bunkers, G. 2023. An assay to reliably achieve tar spot symptoms on corn in a controlled environment. bioRxiv. doi.org/10.1101/2023.01.12.523803 Broders, K., Iriarte-Broders, G., Bergstrom, G. C., Byamukama, E., Chilvers, M., Cruz, C., Dalla-Lana, F., Duray, Z., Malvick, D., Mueller, D., Paul, P. ,Plewa, D., Raid, R., Robertson, A. E., Salgado-Salazar, C., Smith, D., Telenko, D., VanEtten, K. and Kleczewski, N. M. 2022. Phyllachora species infecting maize and other grass species in the Americas represents a complex of closely related species. Ecol. and Evol. 12:e8832. doi.org/10.1002/ece3.8832 Bustin, S. A., Benes, V., Garson, J. A., Hellemans, J., Huggett, J., Kubista, M., Mueller, R., Nolan, T., Pfaffl, M. W., Shipley, G. L., Vandesompele, J. and Wittwer, C. T. 2009. The MIQE guidelines: Minimum information for publication of quantitative real-time PCR experiments. Clin. Chem. 55:4611-622. doi.org/10.1373/clinchem.2008.112797 Caffi, T., Rossi, V., Legler, S. E. and Bugiani, R. 2011. A mechanistic model simulating ascosporic infections by Erysiphe necator, the powdery mildew fungus of grapevine. Plant Pathol. 60:522-531. doi.org/10.1111/j.1365-3059.2010.02395.x Calcagno, V. 2020. glmulti: Model selection and multimodel inference made easy. R package version 1.0.8, https://CRAN.R-project.org/package=glmulti Caldwell, D. L., da Silva, C. R., McCoy, A. G., Avila, H., Bonkowski, J. C., Chilvers, M. I., Helm, M., Telenko, D. E. P. and Iyer-Pascuzzi, A. S. 2024. Uncovering the infection strategy of Phyllachora maydis during maize colonization: A comprehensive analysis. Phytopathology 114:1075-1087. doi.org/10.1094/PHYTO-08-23-0298-KC Carisse, O. and Philion, V. 2002. Meterological factors affecting periodicity and concentration of airborne spores of Bremia lactucae. Can. J. Plant Pathol. 2:184-193. doi.org/10.1080/07060660309506994 Carisse, O., Tremblay, D. M., Levesque, C. A., Gindro, K., Ward, P. and Houde, A. 2009a. Development of a TaqMan real-time PCR assay for quantification of airborne conidia of Botrytis squamosa and management of botrytis leaf blight of onion. Phytopathology 58 99:1273-1280. doi.org/10.1094/PHYTO-99-11-1273 Carisse, O., Bacon, R. and Lefebvre, A. 2009b. Grape powdery mildew (Erysiphe necator) risk assessment based on airborne conidium concentration. Crop Prot. 28:1036-1044. doi.org/10.1016/j.cropro.2009.06.002 Check, J. C., Harkness, R. J., Heger, L., Sakalidis, M. L., Chilvers, M. I., Mahaffee, W. F. and Miles, T. D. 2024a. It’s a Trap! Part I: Exploring the applications of rotating-arm impaction samplers in plant pathology. Plant Dis. 108:1910-1922. doi.org/10.1094/PDIS- 10-23-2096-FE Check, J. C., Harkness, R. J., Heger, L., Sakalidis, M. L., Chilvers, M. I., Mahaffee, W. F. and Miles, T. D. 2024b. It’s a Trap! Part II: An approachable guide to constructing and using rotating-arm air samplers. Plant Dis. 108:1923-1936. doi.org/10.1094/PDIS-01-24-0131- SR Crisp, H. C., Gomez, R. A., White, K. M. and Quinn, J. M. 2013. A side-by-side comparison of Rotorod and Burkard pollen and spore collections. Ann. Allergy Asthma Immunol. 111: 118-125. doi.org/10.1016/j.anai.2013.05.021 Crop Protection Network – Corn Disease Loss Calculator. 2024. https://loss.cropprotectionnetwork.org/ Dhar, N., Mamo, B. E., Subbarao, K. V., Koike, S. T., Fox, A., Anchieta, A., and Klosterman, S. J. 2020. Measurements of aerial spore load by qPCR facilitates lettuce downy mildew risk advisement. Plant Dis. 104:82‐93. doi.org/10.1094/PDIS-03-19-0441-RE Dittrich, U., Hock, J. and Kranz, J. 1991. Germination of Phyllachora maydis ascospores and conidia of Monographella maydis. Cryt. Bot. 2:214-218. Evenhuis, A., Verdam, B. and Zadoks, J. C. 1997. Splash dispersal of conidia of Mycocentrospora acerina in the field. Plant Pathol. 46:459-469. doi.org/10.1046/j.1365- 3059.1997.d01-42.x Eversmeyer, M. G. and Kramer, C. L. 1982. Vertical concentrations of fungal spores above wheat fields. Grana 26:97-102. doi.org/10.1080/00173138709428909 Forrer, H. R., Pflugfelder, A., Musa, T. and Vogelgsang, S. 2021. Low-cost spore traps: An efficient tool to manage Fusarium head blight through improved cropping systems. Agronomy 11:987. doi.org/10.3390/agronomy11050987 Frenz, D. A. 2000. The effect of windspeed on pollen and spore counts collected with the Rotorod sampler and Burkard spore trap. Ann. Allergy Asthma Immunol. 85:392-394. doi.org/10.1016/S1081-1206(10)62553-7 Gongora-Canul, C., Jimenez-Beitia, F. E., Puerto-Hernandez, C., Avellandea, C., Kleczewski, N., Telenko, D. E. P., Shim, S., Solorzano, J. E., Goodwin, S. B., Scofield, S. R. and Cruz, C. D. 2023. Assessment of symptom induction via artificial inoculation of the obligate biotrophic fungus Phyllachora maydis (Maubl.) on corn leaves. BMC Research Notes 16:69. doi.org/10.1186/s13104-023-06341-y Gottwald, T. R. and Bertrand, P. F. 1982. Patterns of diurnal and seasonal airborne spore 59 concentrations of Fusarium effusum and its impact on a pecan scab epidemic. Phytopathology 72:330-335. doi.org/10.1094/Phyto-72-330 Granke, L. L., Morrice, J. J. and Hausbeck, M. K. 2014. Relationships between airborne Pseudoperonospora cubensis sporangia, environmental conditions, and cucumber downy mildew severity. Plant Dis. 98:674‐68. doi.org/10.1094/PDIS-05-13-0567-RE Groves, C. L., Kleczewski, N. M., Telenko, D. E. P., Chilvers, M. I. and Smith, D. L. 2020. Phyllachora maydis ascospore release and germination from overwintered corn residue. Plant Health Prog. 21:26-30. doi.org/10.1094/PHP-10-19-0077-RS Guo, X. W. and Fernando, W. G. D. 2005. Seasonal and diurnal patterns of spore dispersal by Leptosphaeria maculans from canola stubble in relation to environmental conditions. Plant Dis. 89:97‐104. doi.org/10.1094/PD-89-0097 Haudenshield, J. S. and Hartman, G. L. 2011. Exogenous controls increase negative cell veracity in multiplexed, quantitate PCR assays for Phakopsora pachyrhizi. Plant Dis 95:343-352. doi.org/10.1094/PDIS-01-10-0023 Hernandez, M. and KC, A. N. 2024. Determining the timing of spore release by Botryosphaeriaceae species in Oregon vineyards. Plant Dis. 108:1033-1040. doi.org/10.1094/PDIS-07-23-1359-RE Hirst, J. M. 1952. An automatic volumetric spore trap. Ann. Appl. Biol. 39:257-265. doi.org/10.1111/j.1744-7348.1952.tb00904.x Hock, J., Kranz, J. and Renfro, B. L. 1989. El complejo “mancha de asfalto” de maíz: Su distribucción geográfica, requisitos ambientales e importancia económica en México. Rev Mex Fitopatol 7:129–135 Hock, J., Dittrich, U., Renfro, B. L. and Kranz, J. 1992. Sequential development of pathogens in the maize tar spot disease complex. Mycopathologia 117:157-161. doi.org/10.1007/BF00442777 Hock, J., Kranz, J. and Renfro, B. L. 1995. Studies on the epidemiology of the tar spot disease complex of maize in Mexico. Plant Path. 44:490-502 Hofgaard, I. S., Seehusen, T., Aamot, H. U., Riley, H., Razzaghian, J., Le, V. H., Hjelkrem, A. G. R., Dill-Macky, R. and Brodal, G. 2016. Inoculum potential of Fusarium spp. relates to tillage and straw management in Norwegian fields offspring oats. Front. Microbiol. 7:556. doi.org/10.3389/fmicb.2016.00556 Hughes, G., Choudhury, R. A. and McRoberts, N. 2019. Summary measures of predictive power associated with logistic regression models of disease risk. Phytopathology 109:712-715. doi.org/10.1094/PHYTO-09-18-0356-LE Isard, S. A., Barnes, C. W., Hambleton, S., Ariatti, A., Russo, J. M., Tenuta, A., Gay, D. A. and Szabo, L. J. 2011. Predicting soybean rust incursions into the north American continental interior using crop monitoring, spore trapping, and aerobiological modeling. Plant Dis. 95:1346-1357. doi.org/10.1094/PDIS-01-11-0034 Jackson, S. L. and Bayliss, K. L. 2011. Spore traps need improvement to fulfill plant biosecurity 60 requirements. Plant Pathol. 60: 801-810. doi.org/10.1111/j.1365-3059.2011.02445.x King, J. E. and Polley, R. W. 1976. Observations on the epidemiology and effect on grain yield of brown rust in spring barley. Plant Path. 25:63-73. doi.org/10.1111/j.1365- 3059.1976.tb01926.x Kleczewski, N. M., Donnelly, J. and Higgins, R. 2019. Phyllachora maydis, causal agent of tar spot on corn, can overwinter in Northern Illinois. Plant Health Prog. 20:198. doi.org/ 10.1094/PHP-04-19-0030-BR Kleczewski, N. M., Plewa, D. E., Bissonnette, K. M., Bowman, N. D., Byrne, J. M., LaForest, J., Dalla-Lana, F., Malvick, D. K., Mueller, D. S., Chilvers, M. I., Paul, P. A., Raid, R. N., Robertson, A. E., Ruhl, G. E., Smith, D. L. and Telenko, D. E. P. 2020. Documenting the establishment, spread, and severity of Phyllachora maydis on corn, in the United States. J. Integr. Pest Manag. 11:1-5. doi.org/10.1093/jipm/pmaa012 Kuffel, A., Gray, A. and Daeid, N. N. 2020. Impact of metal ions on PCR inhibition and RT- PCR efficiency. Int. J. Legal Med. 135:63-72. doi.org/10.1007/s00414-020-02363-4 Lacey, M. E. and West, J. S. 2006. Using a Burkard trap. In: The Air Spora. Springer, Boston, MA. doi.org/10.1007/978-0-387-30253-9_4 Landis, J. R. and Koch, G. G. 1977. The measurement of observer agreement for categorical data. Biometrics 33:159-174. doi.org/10.2307/2529310 Laurent, B., Douillet, A., Beslay, A., Bordes, J., Delmotte, F., Debord, C., Raynal, M. and the VISA network partners. 2022. The VISA network: A collaborative project between research institutes and vineyard owners to create the first epidemiological monitoring network of downy mildew epidemic based on aerial spore capture. BIO Web Conf. 50:04007. doi.org/10.1051/bioconf/20225004007 Luis, J. M., Mehl, H. L., Plewa, D. and Kleczewski, N. M. 2023. Is Microdochium maydis associated with necrotic lesions in the tar spot disease complex? A culture-based survey of maize in Mexico and the Midwestern United States. Phytopathology 113:1890-1897. doi.org/10.1094/PHYTO-04-23-0109-R MacCready, J. S., Roggenkamp, E. M., Gdanetz, K. and Chilvers, M. I. 2023. Elucidating the obligate nature and biological capacity of an invasive fungal corn pathogen. Mol. Plant Microbe Interact 36:411-424. doi.org/10.1094/MPMI-10-22-0213-R Mahaffee, W. F. and Stoll, R. 2016. The ebb and flow or airborne pathogens: Monitoring and use in disease management decisions. Phytopathology 106:420-431. doi.org/10.1094/PHYTO-02-16-0060-RVW Mahaffee, W. F., Margariaz, F., Ulmer, L., Bailey, B. N. and Stoll, R. 2023. Catching spores: Linking epidemiology, pathogen biology and physics to ground-based inoculum monitoring. Plant Dis. 107:13-33. doi.org/10.1094/PDIS-11-21-2570-FE Mardones, M., Trampe-Jaschik, T., Oster, S., Elliot, M., Urbina, H., Schmitt, I. and Piepenbring, M. 2017. Phylogeny of the order Phyllachorales (Ascomycota, Sordariomycetes): among and within order relationships based on five molecular loci. Persoonia 39:74-90. doi.org/10.3767/persoonia.2017.39.04 61 Maublanc, A. 1904. Especes Nouvelles de champignons inferieurs. Bull. de la Societe Phytopathologique Francaise 20, 72. McCoy, A. G., Roth, M. G., Shay, R., Noel, Z. A., Jayawardana, M. A., Longley, R. W., Bonito, G. and Chilvers, M. I. 2019. Identification of fungal communities within the tar spot complex of corn in Michigan via next-generation sequencing. Phytobiomes 3:235-243. doi.org/10.1094/PBIOMES-03-19-0017-R McCoy, A. G., Romberg, M. K., Zaworski, E. R., Robertson, A. E., Phibbs, A., Hudelson, B. D., Smith, D. L., Beiriger, R. L., Raid, R. N., Byrne, J. M. and Chilvers, M. I. 2018. First report of tar spot on corn (Zea mays) caused by Phyllachora maydis in Florida, Iowa, Michigan, and Wisconsin. Plant Dis. 102:1851. doi.org/10.1094/PDIS-02-18-0271-PDN Miller, A. N. and Huhndorf, S. M. 2005. Multi-gene phylogenies indicate ascomal wall morphology is a better predictor of phylogenetic relationships than ascospore morphology in the Sordariales (Ascomycota, Fungi). Mol. Phylogenet. and Evol. 35:60- 75 doi.org/10.1016/j.ympev.2005.01.007 Mottaleb, K. A., Loladze, A., Sonder, K., Kruseman, G. and San Vicente, F. 2018. Threats of tar spot complex in the United States of America and its global consequences. Mitig. Adapt. Strategies Glob. Change 24:281-300. doi.org/10.1007/s11027-018-9812-1 Mueller, D. S., Wise, K. A., Sisson, A. J., Allen, T. W., Bergstrom, G. C., Bissonnette, K. M., Bradley, C. A., Byamukama, E., Chilvers, M. I., Collins, A. A., Esker, P. D., Faske, T. R., Friskop, A. J., Hagan, A. K., Heiniger, R. W., Hollier, C. A., Isakeit, T., Jackson- Ziems, T. A., Jardine, D. J., Kelly, H. M., Kleczewski, N. M., Koehler, A. M., Koenning, S. R., Malvick, D. K., Mehl, H. L., Meyer, R. F., Paul, P. A., Peltier, A. J., Price, P. P., Robertson, A. E., Roth, G. W., Sikora, E. J., Smith, D. L., Tande, C. A., Telenko, D. E. P., Tenuta, A. U., Thiessen, L. D., and Wiebold, W. J. 2020. Corn yield loss estimates due to diseases in the United States and Ontario, Canada, from 2016 to 2019. Plant Health Prog. 21:238-247. doi.org/10.1094/PHP-05-20-0038-RS Munir, M., Wang, H., Dufault, N. S. and Anco, D. J. 2020. Early detection of airborne inoculum of Nothopassalora personata in spore trap samples from peanut fields using quantitative PCR. Plants 9:1327. doi.org/10.3390/plants9101327 Newlands, N. K. 2018. Model-based forecasting of agricultural crop disease risk at the regional scale, integrating airborne inoculum, environmental, and satellite-based monitoring data. Front. Environ. Sci. 6:63. doi.org/10.3389/fenvs.2018.00063 Parbery, D. G. 1963a. Studies on graminicolous species of Phyllachora Fckl. I. ascospores – Their liberation and germination. Aust. J. Bot. 11:117-130. doi.org/10.1071.BT9630117 Parbery, D. G. 1963b. Studies on graminicolous species of Phyllachora Fckl. II. Invasion of the host and development of the fungus. Aust. J. Bot. 11:131-140. doi.org/10.1071/BT9630131 Parbery, D. G. 1967. Studies on graminicolous species of Phyllachora Fckl. V. a taxonomic monograph. Aust. J. Bot. 15:271-375. doi.org/10.1071/BT9670271 Pearson, R. C., Seem, R. C. and Meyer, F. W. 1980. Environmental factors influencing the 62 discharge of basidiospores of Gymnosporangium juniperi-vinginianae. Phytopathology 70:262-266 Pizolotto, C. A., Harrington, M., Brown, L., Murdock, M., Harrington, S., Marshall, J., Neher, O. T. and Woodhall, J. W. 2021. A real-time PCR assay for Erysiphe betae and its effectiveness when used with different spore trapping methods. Eur. J. Plant Pathol. 162:329-341. doi.org/10.1007/s10658-021-02405-6 R Core Team. 2024. R: A language and environment for statistical computing. Vienna, Austria. https://www.R-project.org/ Rahman, A., Standish, J. R., D’Arcangelo, K. N., and Quesada-Ocampo, L. M.2021. Clade- specific biosurveillance of Pseudoperonospora cubensis using spore traps for precision disease management of cucurbit downy mildew. Phytopathology 111:312‐320. doi.org/10.1094/PHYTO-06-20-0231-R Renfroe-Becton, H., Kirk, K. R. and Anco, D. J. 2024. Measuring the distance and effects of weather conditions on the dispersal of Nothopassalora personata. Phytopathology 114: 549-557. doi.org/10.1094/PHYTO-05-23-0169-R Roelfs, A. P. 1972. Gradients in horizontal dispersal of cereal rust uredospores. Phytopathology 62:70-76. doi.org/10.1094/Phyto-62-70 Roggenkamp, E. M., Check, J. C., Biswal, A. K., Floyd, C. M., Miles, L. A., Nicolli, C. P., Shim, S., Salgado-Salazar, C., Alakonya, A. E., Malvick, D. K., Smith, D. L., Telenko, D. E. P. and Chilvers, M. I. 2024. Development of a qPCR assay for species-specific detection of the tar spot pathogen Phyllachora maydis. PhytoFront. 4:61-71. doi.org/ 10.1094/PHYTOFR-04-23-0050-FI Ross, T. J., Allen, T. W., Shim, S., Thompson, N. M. and Telenko, D. E. P. 2024. Investigations into economic returns resulting from foliar fungicides and application timing on management of tar spot in Indiana hybrid corn. Plant Dis. 108:461-472. doi.org/10.1094/PDIS-05-23-0932-RE Ruhl, G., Romberg, M. K., Bissonnette, S., Plewa, D., Creswell, T. and Wise, K. A. 2016. First report of tar spot of corn caused by Phyllachora maydis in the United States. Plant Dis. 100: 1496. doi.org/10.1094/PDIS-12-15-1506-PDN Salotti, I. and Rossi, V. 2023. A mechanistic model accounting for the effect of soil moisture, weather, and host growth stage on the development of Sclerotinia sclerotiorum. Plant Dis. 107:514-533. doi.org/10.1094/PDIS-12-21-2743-RE Schaafsma, A. W., Tamburic-Ilincic, L. and Hooker, D. C. 2005. Effect of previous crop, tillage, field size, adjacent crop, and sampling direction on airborne propagules of Gibberella zeae/Fusarium graminearum, Fusarium head blight severity, and deoxynivalenol accumulation in winter wheat. Can.J. Plant Pathol. 27:217‐224. doi.org/10.1080/07060660509507219 Solorzano, J. E., Subbaiah, A., Floyd, C. M., Kleczewski, N. M. and Malvick, D. K. 2024. Foliar inoculation of Phyllachora maydis into corn induces infection and local spread in field environments. Plant Dis. 108:2672-2676. doi.org/10.1094/PDIS-02-24-0367-SC 63 Solorzano, J. E., Issendorf, S. E., Drott, M. T., Check, J. C., Roggenkamp, E. M., Cruz, C. D., Kleczewski, N. M., Gongora-Canul, C. C. and Malvick, D. K. 2023. A new and effective method to induce infection of Phyllachora maydis into corn for tar spot studies in controlled environments. Plant Methods 19:83. doi.org/10.1186/s13007-023-01052-8 Sutton, T. B. and Jones, A. L. 1976. Evaluation of four spore traps for monitoring discharge of ascospores of Venturia inaequalis. Phytopath. 66:453-456. doi.org/10.1094/Phyto-66-453 Thiessen, L. D. 2024. Follow that spore! Airborne inoculum detection as a decision aid. PhytoFront. 4:5-13. doi.org/10.1094/PHYTOFR-06-23-0077-FI Thiessen, L. D., Neill, T. M. and Mahaffee, W. F. 2017. Timing fungicide application intervals based on airborne Erysiphe necator concentrations. Plant Dis. 101:1246-1252. doi.org/10.1094/PDIS-12-16-1727-RE Thiessen, L. D., Keune, J. A., Neill, T. M., Turechek, W. W., Grove, G. G. and Mahaffee, W. F. 2016. Development of a grower‐conducted inoculum detection assay for management of grape powdery mildew. Plant Pathol. 65:238‐249. doi.org/10.1111/ppa.12421 Timmer, L. W., Solel, Z., Gottwald, T. R., Ibañez, A. M. and Zitko, S. E. 1998. Environmental factors affecting production, release, and field populations of conidia of Alternaria alternata, the cause of brown spot of citrus. Phytopathology 88:1218-1223. doi.org/10.1094/PHYTO.1998.88.11.1218 Torfs, S., Van Poucke, K. V., Campenhout, J. V., Ceustermans, A., Croes, S., Bylemans, D., Van Hemelrijck, W., Keulemans, W. and Heungens, K. 2019. Venturia inaequalis trapped: molecular quantification of airborne inoculum using volumetric and rotating arm samplers. Eur. J. Plant Pathol.155:1319-1332. doi.org/10.1007/s10658-019-01858-0 Valle-Torres, J., Ross, T. J., Plewa, D., Avellaneda, M. C., Check, J., Chilvers, M. I., Crus, A. P., Dalla Lana, F., Groves, C., Gongora-Canul, C., Henriquez-Dole, L., Jamann, T., Kleczewski, N. M., Lipps, S., Malvick, D., McCoy, A. G., Mueller, D. S., Paul, P. A., Puerto, C., Schloemer, C., Raid, R. N., Robertson, A., Roggenkamp, E. M., Smith, D. L., Telenko, D. E. P. and Cruz, C. D. 2020. Tar spot: An understudied disease threatening corn production in the Americas. Plant Dis. 104:2541-2550. doi.org/10.1094/PDIS-02- 20-0449-FE Van der Heyden, H., Dutilleul, P., Charron, J.-B., Bilodeau, G. J. and Carisse, O. 2021. Monitoring airborne inoculum for improved plant disease management. A review. Agron. Sustain. Dev. 41:40. doi.org/10.1007/s13593-021-00694-z Wang, A., Shang, Z., Jiang, R., Zhang, M., Wang, J., Li, H., Zhang, B., Tang, H., Xu, F., Hu, X., Liu, W., Fan, J. R., Zhou, Y. and West, J. S. 2024. Development and application of a qPCR-based method coupled with spore trapping to monitor airborne pathogens of wheat causing stripe rust, powdery mildew, and Fusarium head blight. Plant Dis. First Look. doi.org/10.1094/PDIS-03-24-0548-SR Webster, R. W., Nicolli, C., Allen, T. W., Bish, M. D., Bissonnette, K., Check, J. C., Chilvers, M. I., Duffeck, M. R., Kleczewski, N., Luis, J. M., Mueller, B. D., Paul, P. A., Price, P. P., Robertson, A. E., Ross, T. J., Schmidt, C., Schmidt, R., Schmidt, T., Shim, S., Telenko, D. E. P., Wise, K. and Smith, D. L. 2023. Uncovering the environmental 64 conditions required for Phyllachora maydis infection and tar spot development on corn in the United States for use as predictive models for future epidemics. Sci. Reports 13:17064. doi.org/10.1038/s41598-023-44338-6 West, J. S. and Kimber, R. B. E. 2015. Innovations in air sampling to detect plant pathogens. Ann. Appl. Biol. 166:4-17. doi.org/10.1111/aab.12191 Wickham, H. 2016. ggplot2: Elegant graphics for data analysis. Springer-Verlag New York. https://ggplot2.tidyverse.org Yang, Y., Zhou, Q., Zahr, K., Harding, M. W., Feindel, D. and Feng, J. 2021. Impact of DNA extraction efficiency on the sensitivity of PCR-based plant disease diagnosis and pathogen quantification. Eur. J. Plant Pathol. 159:583-591. doi.org/10.1007/s10658-020- 02189-1 Zeileis, A. and Grothendieck, G. 2005. zoo: S3 infrastructure for regular and irregular time series. J. Stat. Softw. 14:1-27. doi.org/10.18637/jss.v014.i06 65 Supplementary tables and figures Table A2.1: qPCR assay reaction mix. APPENDIX a5’FAM and 3’BHQ1-labeled b5’HEX and 3’BHQ1-labeled 66 Figure A2.1: Exogenous internal control and qPCR mastermix testing. 67 CHAPTER 4: MULTI-CROP SCLEROTINIA SCLEROTIORUM APOTHECIA PREDICTION MODELS FOR IRRIGATED ENVIRONMENTS ARE IMPROVED BY ON-SITE WEATHER MONITORING AND SUPERVISED MACHINE LEARNING Abstract Sclerotinia sclerotiorum causes Sclerotinia stem rot, or white mold, on multiple economically important crops in Michigan. Soybean farmers and crop consultants in the Midwestern U.S. currently use S. sclerotiorum apothecia prediction models to inform fungicide application timing to optimize disease control and economic return. However, current models have not been validated for use in dry bean or potato and do not account for the effects of irrigation on apothecia development. To improve S. sclerotiorum apothecia prediction, on-site weather data were collected and used to generate new binomial logistic regression (LR) and supervised machine learning (ML) models for irrigated soybean, dry bean and potato fields. The ML algorithms investigated included decision trees, random forests and support vectors machines. Decision tree classification models outperformed LR and other ML models, achieving 77% accuracy on testing data. Accuracy increased to 89% when on-site weather data were included, indicating that on-site weather monitoring may be required to reliably predict apothecia presence in irrigated environments. Feature importance analysis identified row shading (the distance the plant canopy extends into the row) as critical for prediction accuracy. The minimum row shading required to trigger apothecia development varied slightly between crop types and row spacings, from 0.15 to 0.21 m. Apothecia density peaked when soil temperatures were 21.51°C and volumetric water content were 11.43% and 19.58%. Additionally, a rapid increase in apothecia presence was observed after canopy closure reached 87%. Future model testing and validation will be required prior to deployment as a decision aid for farmers and crop consultants. 68 Introduction The necrotrophic fungal plant pathogen Sclerotinia sclerotiorum (Lib.) de Bary is one of the most destructive plant pathogens due to its global distribution (CABI 2005; Purdy 1979) and extensive host range (Boland and Hall 1994). Sclerotinia sclerotiorum is especially problematic in the North Central United States, where cool, moist conditions and intensive crop production create a conducive environment for its development and persistence (Saharan and Mehta 2008). The diseases caused by S. sclerotiorum have a variety of names but are known as Sclerotinia stem rot (SSR) on soybean (Glycine max), and white mold on dry bean (Phaseolus vulgaris) and potato (Solanum tuberosum). Sclerotinia stem rot is characterized by the development of fluffy, white mycelia on aerial plant parts and melanized sclerotia in and outside of the host plant organs (Peltier et al. 2012; Purdy 1979). Decades of research using both field and controlled environment experiments have been conducted to describe the conditions required for S. sclerotiorum to complete its lifecycle (Twengstrom et al. 1998a; Ferraz et al. 1999; Sun and Yang 2000; Hao et al. 2002; Wu and Subbarao 2008; Fall et al. 2018a, 2018b). The lifecycle of S. sclerotiorum is anchored by the survival of sclerotia in soil to serve as the site of myceliogenic and carpogenic germination. Following harvest, sclerotia are deposited onto the soil surface and may be incorporated into the soil through tillage where they may survive for up to 5 years to serve as the primary inoculum source in following growing seasons (Adams and Ayers 1979). Following a conditioning period, the length of which depends on soil temperature, moisture, and sclerotium size (Dillard et al. 1995), sclerotia in the top 5 cm of soil germinate to form apothecia, the sexual reproductive structures of S. sclerotiorum (Wu and Subbarao 2008) (Figure 3.1A, 3.1B). In response to changes in relative humidity, ascospores are forcibly ejected in synchrony from their asci to cooperatively generate air flow and escape the surrounding still air layer (Hartill and Underhill 1976; Roper et al. 2010). To initiate infection, ascospores that land on vulnerable senescent tissue of nearby host plants secrete cell wall degrading enzymes and oxalic acid to quickly break down host tissue by modulating the environment to optimize enzyme activity and suppress host defense responses (Hegedus and Rimmer 2005; Kabbage et al. 2015). Successful infection is favored by cool temperatures and the presence of free water from rain, irrigation, fog or dew on plant surfaces (Abawi and Grogan 1979; Fall et al. 2018a; Peltier et al. 2012), which is encouraged by dense 69 vegetative growth in the crop canopy (Fall et al. 2018b). Sclerotinia sclerotiorum can also infect its hosts through myceliogenic germination of sclerotia for basal stem infection of nearby plants, but this method of infection is less epidemiologically important than carpogenic infection in many hosts (Abawi and Grogan 1975; Newton and Sequeira 1972; Willets and Wong 1980). Infection leads to stem colonization, producing water-soaked lesions, wilting and eventual plant death (Figure 3.1C, 3.1D). Sclerotial formation is triggered in response to nutrient deprivation (Christias and Lockwood 1973), which contributes to pathogen persistence and inoculum load in following seasons. Despite the wealth of existing literature describing the conditions needed throughout the lifecycle of S. sclerotiorum, the prediction of its life stages and resulting disease severity remains a difficult task (Reich and Chatterton 2023). Sclerotinia stem rot management is highly integrative, using cultural, genetic, biological and chemical methods (Johnson and Atallah 2006; Pethybridge et al. 2019a; Webster et al. 2022; Webster et al. 2023a). In-season management often relies on fungicide applications made during bloom to protect flowers from infection by ascospores, and application timing is a critical factor in the efficacy of chemical control (Willbur et al. 2019). Host flowering lasts several weeks longer than the active period of fungicidal compounds, leaving the host vulnerable to infection without repeated applications. Therefore, a single well-timed fungicide application is critical for effective and economical SSR control in soybean as multiple applications are often not affordable in lower value crops such as soybean and dry bean (Kandel et al. 2021). In high- value specialty crops such as potato, multiple fungicide applications for SSR control are more common in disease management programs (Budge and Whipps 2001; Pethybridge et al. 2019b). However, both systems can benefit from reduced chemical use and input costs by informing farmers of high-risk periods of apothecia presence to optimize fungicide timing and improve disease control (Foster et al. 2011; Clarkson et al. 2004). The development and severity of SSR is determined by the elements of the disease triangle: the presence of the pathogen (ascospore-bearing apothecia), the susceptibility of the host (presence of flowers or other senescent tissue), and conducive weather conditions (Agrios 2005). Without the use of predictive models to summarize disease risk based on these interacting factors, it can be difficult to determine the risk of infection, leading to ineffective or excessive fungicide sprays. The economic impact of SSR and its unique biological characteristics (i.e., monocyclic disease cycle, limited host susceptibility window, strong environmental influence) 70 have made it a target for prediction modeling efforts. Many decision support tools have been developed for SSR management across host plant species, including canola (Koch et al. 2007; Turkington et al. 1991, 1993; Twengstrom et al. 1998b), carrot (Foster et al. 2011), dry bean (Harikrishnan and del Rio 2008; Jones et al. 2011), lettuce (Clarkson et al. 2014) and soybean (Willbur et al. 2018a). Despite this, there is still a lack of confidence in disease prediction tools by farmers for SSR management. Reich and Chatterton (2023) performed a scoping review of current S. sclerotiorum prediction modeling research and found that environmental parameters important for driving S. sclerotinia development are often not considered when generating prediction models, including factors such as irrigation, soil type and row spacing that modulate the relationships between the environment and pathogen. Without accounting for the effects of irrigation on soil temperature and moisture, predictive models are unable to consider the effects of differential irrigation practices on apothecia development. Additionally, the logistic regression model framework that is commonly used in plant disease epidemiology cannot capture the complex interactions and interdependencies that underly biological processes, such as apothecia development (Hao et al. 2003; Wu and Subbarao 2008; Fall et al. 2018). Logistic regression is less fit to account for these complexities without the development of large polynomial models which are at high risk of overfitting (Dillard et al. 1995; Sun and Yang 2000; Sperschneider 2019). While supervised machine learning algorithms were adopted in the late 1990s in plant disease epidemiology (De Wolf and Francl 1997; 2000), they have gained popularity in recent years (Landschoot et al. 2012; Skelsey 2021; Webster et al. 2023). Machine learning (ML) models can benefit plant disease prediction modeling as they are able to account for complex non-linear relationships that may underly biological processes, such as apothecial development. However, logistic regression models are easily interpretable while ML models are often not, and ML models are similarly prone to overfitting when increasing in complexity. Therefore, both approaches are still valuable in plant disease forecasting to strike a balance between model accuracy, complexity and interpretability. Adoption of in-field data collection and reporting technology will aid the advancement of precision agriculture efforts (Shafi et al. 2019). In-field data collection using sensors, data- logging and remote communication systems to monitor weather, soil, and plant properties have applications in prescriptive irrigation management (Dong et al. 2024; Jimenez et al. 2022), optimizing planting, fertilization and harvest timing (Khan et al. 2022; Lavanya et al. 2020) 71 and pest and disease forecasting (Chen et al. 2022; Delfani et al. 2024; Khattab et al. 2019; Kim et al. 2018; Liu et al. 2022). While regional weather station networks and remote sensing techniques can inform prediction models for crop management recommendations, they are unable to account for site-specific factors that introduce discrepancies between observed and predicted weather variables, sometimes leading to unreliable model outputs. Therefore, on-site weather monitoring can offer advantages to farmers when paired with decision support systems to further optimize crop productivity and economics. Previously, a SSR decision support system, “Sporecaster,” was developed and validated to advise soybean farmers in the North Central U.S. of high-risk periods of S. sclerotiorum apothecia presence to guide fungicide spray timing (Willbur et al. 2018a, 2018b). When observing a differential effect of temperature between rainfed and irrigated environments, the authors chose to develop two separate models: one for each environment type. While the rainfed model is still performing satisfactorily, the irrigated model requires improvement (personal communication, Damon Smith, Table A3.1). Improved accuracy of S. sclerotiorum apothecia prediction in irrigated environments would benefit North Central U.S. soybean farmers and crop advisors who are already using predictive modeling tools to aid in fungicide timing decisions. Expanding the user base by developing a multi-crop model could significantly enhance the impact of predictive modeling on disease management across the region. Furthermore, the number of irrigated acres in the Midwestern U.S. has grown steadily in the past 20 years across crop species, and this trend is expected to continue as precipitation events become more erratic under climate change (Dong et al. 2023), emphasizing the need to support irrigated production systems. This study employed logistic regression and supervised machine learning to develop new models for the prediction of S. sclerotiorum apothecia in irrigated environments for soybean, dry bean and potato crops in Michigan. On-site weather monitoring systems were used to record soil temperature and moisture to capture the effects of irrigation on S. sclerotiorum apothecia development. The objectives of this work were four-fold: i) develop predictive models for apothecia presence using logistic regression and machine learning techniques; ii) assess model performance before and after incorporating soil moisture and temperature features; iii) compare the performance of the original irrigated model for irrigated environments, updated logistic regression models, and proposed machine learning models for their ability to accurately predict 72 apothecia presence; and iv) identify the most influential weather parameters on apothecia development through feature importance analysis of decision tree models. Materials and Methods Data collection Apothecia monitoring. From 2021 to 2023, research and commercial fields were monitored weekly or twice weekly for S. sclerotiorum apothecia. The soybean research fields were planted in 6-row plots with a seeding rate of 444,789 seeds per hectare (180,000 seeds per acre). Each soybean plot measured 4.27 m (14 ft) by 1.52 m (5 ft) with 0.38 m (15 in) row spacing. Soybean plots with 0.76 m (30 in) row spacing were created by removing alternating rows of two adjacent soybean plots with 0.38 row spacing. The dry bean research fields were planted in 4-row plots with a seeding rate of 321,237 seeds per hectare (130,000 seeds per acre) besides for one dry bean field in Montcalm in 2021 (coordinates 42.20, -84.99) which was planted at 227,337 seeds per hectare (92,000 seeds per acre). Each dry bean plot measured 7.32 m (24 ft) by 2.01 m (6.6 ft) with 0.51 m (20 in) row spacing. The potato research fields were planted in 2-row plots at a seed spacing of 25.4 cm (10 inch) for a rate of 45,305 seed potatoes per hectare (18,342 seed potatoes per acre) and measured 6.10 m (20 ft) by 1.74 m (5.7 ft). Scouting began at least two weeks prior to crop flowering and continued until apothecia were no longer found, plants had reached full maturity, or fields became unnavigable due to vine or mold growth, whichever came first. Dates when apothecia scouting began and ended are included in Table 1. Using a meter stick laid in the center of two rows, apothecia were counted between the two adjacent rows. In commercial fields, counts were initiated at least 30 m (100 ft) from field borders and repeated every 20 paces from the previous scouting point in a W- or Z- shaped sampling pattern. In research fields, counts were made in plots for 20 observations per location visit. Differences in row spacing caused differences in area scouted for each crop type. Therefore, apothecia density (the number of apothecia per m2) was calculated based on the row spacing and was used to represent the apothecia presence at a location on the day of scouting. Estimations of apothecia density and corresponding risk (%) predictions from the original irrigated model are visualized in Figure A3.1. Apothecia density was used to create a binary response variable based on a threshold of 0.18 apothecia per m2 with positive and negative binary responses coded as above and below the threshold, respectively (Willbur et al. 2019). 73 Independent variables. On-site soil conditions were monitored at each apothecia scouting location using the Low-Cost Monitoring System (LOCOMOS) platform developed by the Michigan State University Irrigation Lab (Dong et al. 2024) (Figure 3.1E). The LOCOMOS were equipped with SoilWatch10 soil moisture sensors (Pino-Tech, Stargard, Poland), DS18B20 temperature probes (Analog Device, Wilmington, MA, U.S.), a 12V solar panel, a 12V 7A battery and a solar battery charge controller. Sensor readings took place every 15 minutes. Sensor readings, data storage and remote communications were controlled using a Particle Boron microcontroller. Data was transferred using embedded 4G LTE modem and stored to the LOCOMOS server hourly. The remaining weather variables (air temperature, relative humidity, wind speed and dew point) were sourced using historical weather data services with a 2.5-km resolution (The Weather Company, IBM). Hourly data was used to find the daily means, minimums, maximums and sums of weather variables and calculate the daily duration of air temperature between 10 and 25ºC (ATD) and relative humidity over 86% (RHD) (Clarkson et al. 2007, 2014; Willbur et al. 2019). Thirty-day moving averages (MAs) of daily weather summaries were then calculated (Twengstrom et al. 1998b; Willbur et al. 2019). Soil temperature and moisture data were handled similarly. Percent canopy closure was measured during each location visit using Canopeo (https://canopeoapp.com/; Patrignani and Ochsner 2015; Figure 3.1F) on a smartphone. To measure canopy closure, the camera was positioned between two adjacent rows and photos were framed by the center of the planted rows. In 2021, three measurements were taken during apothecia scouting and averaged to represent percent canopy closure at the location that day. In 2022 and 2023, measurements were taken at each observation point and averaged to represent the canopy closure at that scouting location. Canopy closure (%) was multiplied by row spacing (m) to create the row shading (m) feature, summarizing the two metrics and representing the between-row shading effect. The soil type of each location was taken from USDA-NRCS Web Soil Survey (USDA-NRCS 2025). Soil type was coded as an ordinal category with levels 0, 1, and 2 representing sand, sandy loam and loam soils. Model development and assessment All model hyperparameter tuning, training, testing, assessment and other analyses were done in Python v.3.12.4 using scikit-learn v.1.3.0 (Pedregosa et al. 2011). 74 Feature selection and dimensionality reduction. Feature selection and dimensionality reduction were employed to condense the pool of potential features while preserving relevant information from independent data. All non-categorical independent variables were rescaled using maximum absolute rescaling. Two different methods were employed: feature selection by correlation analysis and dimensionality reduction by Principal Component Analysis (PCA). Feature selection by correlation analysis was achieved by performing Point-Biserial correlation analysis between the binary response variable and all potential weather variables and row shading. Weather variables with the strongest and most highly significant correlations with the response variable were chosen as potential model features. Unique functions of the weather variables were chosen based on their correlation strength. A PCA was performed on all continuous weather variables, as it reduces collinearity among predictors while capturing variability in the data and condensing the total number of features. Principal component loadings were extracted for the reduced feature pools (Figure A3.2). The number of principal components (PCs) was determined by the number of eigenvectors needed to explain 90% of the variance of all independent variables after maximum absolute rescaling (Figure A3.3). Repeating model development across two feature pools was done to quantify any advantage gained in predictive ability at the cost of sacrificing model interpretability. Data splitting, stratification and cross-validation scheme. Data was split (80:20) between training and testing sets. Splits were stratified to maintain equal proportions of positive and negative binary responses between the data sets. Model training was performed using k-fold cross validation with 5 folds and 20% of training data per fold. Models were developed using two different pools of variables: (1) variables readily available from gridded online historical weather data services (i.e. air temperature, relative humidity, rainfall, dew point, and windspeed) (RTMA) and (2) variables that were taken from on-site LOCOMOS stations (i.e. soil temperature and moisture) in addition to historical weather data services (RTMA+LOCOMOS). Repeating model development across the two feature pools was done to quantify the change in predictive accuracy by employing on-site weather data collection. Logistic regression. Sequential forward, backward, forward floating and backward floating selection methods were used for binomial logistic regression model development to determine selection method and feature number using MLxtend v.0.23.1 (Raschka 2018). Sequential selection method and feature number were chosen based on model parsimony and the 75 accuracy of candidate models on training data using the previously described cross-validation scheme. Following feature selection, hyperparameter tuning was done using an exhaustive search of all combinations of provided hyperparameter values using the GridSearchCV function from scikit-learn. Hyperparameters resulting in the greatest accuracy on training data were used for tuning the final model (Table A3.2). Selected features and their coefficients were extracted after final model selection. Supervised machine learning algorithms. The supervised classification ML algorithms used were decision tree (DT), random forest (RF) and support vector machine (SVM). Selected modeling approaches represented a diversity of transparent (LR and DT) and opaque (RF and SVM) ML models (Assis et al. 2025). DT models classify observations by performing a hierarchal sequence of logical tests using the associated features in internal nodes with the resulting class labels assigned in leaf nodes (Kingsford and Salzberg 2008). RF models are ensembles of DT models that are generated by training on randomized subsets of data (Breiman 2001). In RF models, the final class label is determined by the majority outcome of the member decision trees. RFs can decrease the probability of misclassification by reducing overfitting. SVM models classify observations by optimizing a separating hyperplane to discriminate between classes in high-dimensional space (Noble 2006). These algorithms were chosen because they have demonstrated value in classification tasks for plant disease prediction in other pathosystems, such as Athelia rolfsii on peanut (Sanjel et al. 2024), Fusarium graminearum on wheat (Shah et al. 2023) and Magnaporthe grisea on rice (Kaundal et al. 2006). ML model hyperparameter tuning was conducted using GridSearchCV and hyperparameters returning the greatest accuracy on training data were chosen for tuning the final model. The hyperparameters considered for each machine learning algorithm are summarized in Table A3.2. Automated machine learning benchmarks. Automated machine learning (AutoML) was used to establish a benchmark for performance of supervised classification machine learning algorithms to compare to results from subjective hyperparameter tuning. AutoML was performed using TPOT v.0.11.7 (Le et al. 2020) on the Michigan State University High Performance Computing Cluster using Python version 3.6.4. TPOT is built on scikit-learn and uses genetic programming to develop machine learning pipelines by automating data preprocessing, feature selection and hyperparameter tuning for different ML base models. Accuracy was used to score 76 models during pipeline optimization. Data prior to normalization and feature selection were fed into the AutoML pipeline. The pipeline optimization process was iterated for ten generations. Fit and performance metrics. Logistic regression, ML and AutoML models were assessed by the accuracy on each of the cross-validation folds of training data and the average accuracy across all folds. Testing data withheld during model training was used for all fit and performance assessments. Model performance on test data was evaluated using area under the receiver operating characteristic (AUROC) curve, accuracy (%), sensitivity (%), specificity (%) and precision (%). These metrics were derived from confusion matrices and calculated as 𝐴𝑐𝑐𝑢𝑟𝑎𝑐𝑦 = 𝑇𝑃 + 𝑇𝑁 𝑇𝑃 + 𝑇𝑁 + 𝐹𝑃 + 𝐹𝑁 𝑆𝑒𝑛𝑠𝑖𝑡𝑖𝑣𝑖𝑡𝑦 = 𝑆𝑝𝑒𝑐𝑖𝑓𝑖𝑐𝑖𝑡𝑦 = 𝑇𝑃 𝑇𝑃 + 𝐹𝑁 𝑇𝑁 𝑇𝑁 + 𝐹𝑃 𝑃𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛 = 𝑇𝑃 𝑇𝑃 + 𝐹𝑃 where TP, TN, FP, FN represent true positives, true negatives, false positives, false negatives. Accuracy was used to assess the correct classification of all predictions. Sensitivity and precision were used to assess the correct classification of the positive classes and specificity for negative binary classes. Model performance was not assessed separately for each crop type as data stratification prioritized preserving proportions of the binary response variable instead of crop type. The original irrigated model was also assessed for fit and performance on the same set of observations as new models. The original irrigated model is as follows: 𝐿𝑜𝑔𝑖𝑡(𝜇) = −2.38(𝑅𝑜𝑤 𝑠𝑝𝑎𝑐𝑖𝑛𝑔) + 0.65(𝑀𝑎𝑥𝐴𝑇) + 0.38(𝑀𝑎𝑥𝑅𝐻) − 52.65 where row spacing is represented by a binary value for narrow (<= 0.38 m) and wide (> 0.38 m) row spacing and MaxAT and MaxRH represent 30-day MAs of maximum air temperature and relative humidity (Willbur et al. 2018a). Probabilities were calculated using the logit equation: 𝑃𝑟𝑜𝑏𝑎𝑏𝑖𝑙𝑖𝑡𝑦 = 1 1 + 𝑒"7&89-(:) 77 Finally, the probability thresholds determined during model validation were used to assign binary outcomes for apothecial presence predictions (Willbur et al. 2018b). Narrow row spacing observations used a 35% threshold and wide row spacing observations used a 10% threshold. Observations and predictions were assessed similarly as described for LR, ML and AutoML models. As a resource for those interested in using the original irrigated model for crops it was not originally developed for, the model was assessed on the total data pool (all training and testing data) to demonstrate its value for dry bean and potato white mold management (Table A3.1). Feature importance Feature importance was extracted for the correlation-based DT models to assess the contribution of individual features to model accuracy. Feature importance is measured as the reduction in scoring criterion (i.e., entropy) across all nodes in the DT, where entropy represents the impurity of the data set, calculated as: ; 𝐻 = − S 𝑝9 9<# 𝑙𝑜𝑔=(𝑝9) Where m represents the number of classes and pi represents the probability of an observation belonging to class i (Kingsford and Salzberg 2008). Feature importance is a built-in property from scikit-learn DecisionTreeClassifier() function and represents the total reduction of entropy (i.e., information gain) by each feature normalized to the size of the data set. Selected epidemiology investigations Based on model performances and feature importances, a selection of features was investigated further for their relationships with apothecia density. To study the relationship between the apothecia density and canopy closure, the cumulative sum of the binary apothecia response variable as a function of canopy closure was prepared. This was used to identify an important inflection point in canopy closure for apothecia development. The relationship between row shading and apothecia density was similarly prepared as a scatterplot and the first observations of apothecia density above the binary threshold were marked for each categorical row spacing. A Gaussian model and bimodal Gaussian were fit to mean soil temperature and moisture data, respectively, based on their distribution, to analyze the relationships between apothecia density and these environmental factors. Amplitude (A), mean (μ) and standard deviation (σ) for each model were extracted. 78 Code and data availability Figures were generated using matplotlib v.3.8.4, seaborn v.0.13.2, and graphviz v.2.50.0 Data, code, models and AutoML outputs generated in this study is available at https://github.com/checkjill/SSR-predictive-modeling. Results Scouting efforts from 2021 to 2023 resulted in 2,952 individual observations from commercial and research soybean (n=1,568), dry bean (n=514) and potato fields (n=870) (Table 3.1). These data were summarized into 152 daily observations of field-scale apothecial density. Nearly equal proportions of the binary classification response variable, defined as observations above or below 0.18 apothecia per m2 (Willbur et al. 2018a), were observed (56% and 44%, respectively). Apothecia were observed at all locations except for two commercial soybean fields in 2022, Ionia and Eaton. The lack of apothecia observations likely reflects an absence of sufficient SSR field history rather than unfavorable local conditions as apothecia were observed at other locations in the same year. Therefore, these data were excluded from model development. By conducting Point-Biserial correlation analysis between all potential model features and the binary classification response variable, the model development feature pool was reduced to six weather variables and row shading, the distance shading from the plant canopy extends into the row (Figure 3.2). The six weather variables included 30-day MAs of daily maximum RHD, sum ATD, sum of soil moisture, sum of rainfall, maximum windspeed, and maximum soil temperature. Of all the potential features, row shading was the most strongly correlated with apothecia presence (rpb = 0.49), followed by the maximum RHD (rpb = 0.40), sum of ATD and (rpb = 0.37) and sum of soil moisture (rpb = 0.30). Sum of rainfall was only weakly, positively correlated with apothecia presence (rpb = 0.12). Maximum soil temperature and maximum wind speed were negatively correlated with apothecia presence (rpb = -0.38 and rpb = -0.05, respectively). All variables included in the potential feature pool were significantly correlated with the binary apothecia response (P < 0.0001) except for sum of rainfall and maximum wind speed. The RTMA only data set for model development consisted of row shading, maximum RHD, sum of ATD, sum of rainfall and maximum wind speed, while the RTMA+LOCOMOS data set for model development additionally included sum of soil moisture and maximum soil temperature. 79 Logistic regression modeling using the correlation-based feature pool resulted in the following models: 𝐿𝑅>?@A = −0.67(𝑆𝑜𝑖𝑙 𝑡𝑦𝑝𝑒) + 5.30(𝑀𝑎𝑥 𝑅𝐻𝐷) + 2.03(𝑆𝑢𝑚 𝐴𝑇𝐷) 𝐿𝑅>?@AB7C,C@C4 = −0.66(𝑆𝑜𝑖𝑙 𝑡𝑦𝑝𝑒) + 2.16(𝑆𝑢𝑚 𝑠𝑜𝑖𝑙 𝑚𝑜𝑖𝑠𝑡𝑢𝑟𝑒) Both models incorporated soil type, while the model developed using RTMA data only incorporated 30-day MAs of maximum RHD and sum of ATD and the model developed using RTMA+LOCOMOS data incorporated 30-day MAs of sum of soil moisture. The logistic regression models were the worst performing of all modeling approaches except for PCA-based feature selection using RTMA+LOCOMOS data, where the LR model performed similarly to the DT model (Table 3.2). Logistic regression models developed from the correlation-based feature pool performed particularly poorly in their ability to discriminate between true and false positives (precision), scoring 53% and 54% for RTMA and RTMA+LOCOMOS data sets, respectively. No model performance metrics were improved by including LOCOMOS data in the logistic regression models, with overall accuracy dropping 8% when incorporating 30-day MAs of soil moisture in place of RHD and ATD. However, accuracy was increased by 7% in the PCA-based logistic regression models after incorporating LOCOMOS data. AutoML was used to establish a benchmark for ML model performance. The AutoML models achieved 77% and 83% accuracy when tested on RTMA and RTMA+LOCOMOS data, respectively. However, both AutoML models exhibited a severe imbalance between sensitivity and specificity with an unacceptable bias towards negative binary class predictions. Despite the inflated specificity scores by the AutoML search, all other model performance metrics were able to be met or surpassed by the ML models. Among the ML models developed by using the correlation-based RTMA only feature pool, the DT model performed the best with an accuracy of 77%, sensitivity of 69%, specificity of 86% and precision of 85%. Neither the LR, RF, SVM or AutoML model trained on the same data set had superior performance in any metric. Similarly, among the models developed using the correlation-based RTMA+LOCOMOS feature pool, the DT model outperformed all other models in accuracy and sensitivity but had slightly lower specificity (84%) compared to the RF (87%), SVM (85%), and AutoML (90%) models and slightly lower precision (84%) compared to the RF and SVM models (85%). However, the DT model’s sensitivity (94%) far exceeded all other models (59-73%). 80 Of models developed using the PCA-based RTMA only feature pool, the RF and SVM models both had the greatest overall accuracy (77%). However, the balance between sensitivity and specificity was better in the SVM model (75 and 78%) than the RF model (71 and 81%). Improvement in model accuracy after incorporating LOCOMOS was only possible in the LR and DT models and not the RF and SVM models. Performance was greatly improved in the DT model after incorporating LOCOMOS data, with a 17%, 29% and 5% increase in accuracy, sensitivity and specificity, respectively. The original irrigated model was tested similarly to the newly developed models to demonstrate improvement in predictions by using different modeling approaches and on-site weather data. The original irrigated model accuracy on testing data was 57%. The only model that performed worse than the original irrigated model was the logistic regression correlation- based model developed using RTMA+LOCOMOS data, which performed worse in accuracy (57% and 56%) and specificity (59% and 53%) but better in sensitivity (50% and 59%) and precision (31% and 53%). Across all modeling algorithms, feature selection/dimensionality reduction methods and data sources, the DT model developed using the correlation-based RTMA+LOCOMOS feature pool performed best. This DT was visualized to show the logical tests performed to determine class predictions (Figure 3.3). The values used in internal node logical tests represent weather data rescaled from 0 to 1 from the minimum to maximum raw values instead of actual feature values. The root node of the DT is a logical test on maximum RHD (<= 0.883), followed by two internal nodes with logical tests on row shading. Observations with limited maximum RHD (<= 0.883) were classified as negative based on a row shading threshold of (<=0.837). Under high maximum RHD (>0.883), limited row shading (<=0.335) still resulted in the absence of apothecia. These root and initial internal nodes are then followed by hierarchal logical tests on maximum soil temperature, sum of ATD, soil type, sum of soil moisture and maximum wind speed. Of the two different approaches to improve prediction accuracy of apothecia presence, incorporating on-site soil moisture and temperature data contributed more to improving model performance than using PCA for dimensionality reduction. Models developed using the RTMA+LOCOMOS feature pool saw an average 4.11% improvement in accuracy over models developed using RTMA only, with only 2 models seeing a decrease in accuracy (LR developed 81 with correlation-based feature pool and RF developed with PCA-based feature pool) and one model seeing equal performance (SVM developed with PCA-based feature pool). In contrast, there was an average 0.50% difference in accuracy between correlation- and PCA-based models, which is largely skewed by the 21% increase in accuracy in the RTMA+LOCOMOS logistic regression model. As DTs were the most accurate models developed from the correlation-based feature pool, feature importance was extracted to investigate the contribution of each weather variable to the prediction accuracy. For both the RTMA and RTMA+LOCOMOS models, row shading and maximum RHD were the most important features (0.32, 0.24 and 0.23, 0.26, respectively) (Figure 3.4). For the RTMA model, the remaining features ranked by importance were sum of ATD (0.23), sum rainfall (0.09), maximum WS (0.07) and soil type (0.05). For the RTMA+LOCOMOS model, the remaining features ranked by importance were maximum soil temperature (0.17), sum of soil moisture (0.12), soil type (0.10), maximum WS (0.08), and sum of ATD (0.04). Rainfall was not incorporated into the RTMA+LOCOMOS model and therefore had a feature importance of 0. As row shading was found to be particularly important during feature importance analysis, canopy closure, row shading and apothecia density data were plotted to visualize their relationships and identify critical thresholds across different production systems (Figure 3.5). At 87% canopy closure, the cumulative sum of the binary apothecia response variable increased rapidly, indicating that the majority of positive cases of apothecia presence occurred after 87% canopy closure. Between 0.38-, 0.51-, 0.76- and 0.86-m row spacings, the first observation of the binary classification value occurred when row spacing reached 0.15, 0.18, 0.21, and 0.18 m (5.91, 7.09, 8.27, 7.09 in), respectively. With the improvement predictions through the inclusion of soil moisture and temperature, the relationship of these factors with apothecia density were explored further (Figure 3.6). Models were fit to the LOCOMOS-sourced soil condition variables, mean soil temperature and mean soil moisture, and apothecia density. Initially, separate models for each soil type were considered, but not all models were able to be fit prior to reaching the maximum number of iterations (n=1,000). The apothecia density and mean soil temperature Gaussian model coefficients were 1.95, 21.51 and 0.95, for A, µ and σ, respectively, indicating the greatest apothecia density occurred when 30-day MA soil temperatures were 21.51°C. The apothecia 82 density and mean soil moisture bimodal Gaussian model coefficients were 1.63, 11.43, and 1.31 for A, µ and σ, respectively for the first peak and 4.86, 19.58 and 0.43 for the second peak, indicating that the greatest apothecia density occurred when 30-day MA soil moisture was 11.43 and 19.58%, but peak apothecia density occurred at 19.58% soil moisture. However, both models were skewed by observations from locations with loamy sand soils in 2021, as these observations clustered tightly together and had the greatest estimations of apothecia density. However, the first peak of the bimodal Gaussian model relating apothecia density and mean soil moisture consists primarily of observations from 2021 from locations with sandy soils. Discussion Prediction models can further the goal of implementing integrated pest management strategies by informing decision-makers of high disease risk periods to potentially optimize disease control and economic returns while promoting sustainable use of pesticides (Rossi et al. 2023). The current S. sclerotiorum apothecia prediction model for soybean SSR management for irrigated environments was previously tested and validated in commercial fields in Iowa, Michigan and Wisconsin (Willbur et al. 2018b). However, a lack of irrigated commercial locations hindered the ability to assess model performance. In efforts to improve apothecia prediction for SSR management in irrigated environments across diverse host species, on-site weather data collection and machine learning (ML) pipelines were explored. Through these means, multiple potential apothecial prediction models were proposed, with the most promising model achieving 89% accuracy on independent validation data. Similar to the original irrigated model, the logistic regression (LR) model developed using the RTMA feature pool variables selected air temperature and relative humidity features (Willbur et al. 2018a). However, unlike the original irrigated model, durations of optimal temperature ranges and high relative humidity were chosen over daily minimums, maximums and averages, which may better summarize the conditions needed for apothecia development, as previous work has found that extended durations (~30 days) of conducive conditions are needed for S. sclerotiorum sclerotia conditioning (Clarkson et al. 2007; Twengstrom et al. 2008). Additionally, the most important features of the decision (DT) model developed with RTMA data were maximum relative humidity duration over 86% (RHD) and sum of air temperature duration between 10 and 25ºC (ATD), consistent with the features selected for the LR developed using the same data pool. When RTMA+LOCOMOS features were considered, sum of ATD 83 dropped significantly in the ranking of important features and was replaced by the maximum soil temperature as the third most important feature. Overall, the ranking of feature importances for the best performing models is consistent with what is established about SSR epidemiology and S. sclerotiorum biology (Clarkson et al. 2004; Foster et al. 2011; Fall et al. 2018b; Koch et al. 2007; Twengstrom et al. 1998b; Willbur et al. 2018a Wu and Subbarao 2008), indicating that the models are capturing the relevant information required for accurate predictions. Unsurprisingly, rainfall was not correlated with apothecia presence as the contribution of irrigation to precipitation was not included in the measurements of rainfall. Additionally, rainfall was not included in the best performing model, demonstrating the lack of predictive power of rainfall measurements alone in irrigated environments. Following a scoping review of the literature on the prediction of diseases caused by S. sclerotiorum, Reich and Chatterton 2023 hypothesized that the frequency of precipitation events may be more useful in the prediction of apothecia than precipitation volume due to a lack of relationship found between apothecia counts and measurements of precipitation volumes in previous studies (Foster et al. 2011; Willbur et al. 2018). This is supported by Twengstrom et al. 1998a, who found that irrigating soybean fields daily (5 mm/day) resulted in more apothecia than irrigation weekly (35 mm/week) despite resulting in the same application volume. In the present study, rain buckets were used to measure the frequency and volume of rainfall and irrigation events. However, managing rain buckets in research and production fields proved challenging due to the need for height adjustments as crop height increased and the requirement for agricultural equipment to move through the fields. As a result, the in-field rain buckets yielded unreliable data on precipitation frequency and volume and were excluded from the analysis. Instead, soil moisture and temperature were used to account for the local effects of irrigation and were shown to improve model predictions in comparison to RTMA data alone. Although the relationships between soil moisture, soil temperature and S. sclerotiorum apothecia production have been well studied in controlled environments (Dillard et al. 1995, Hao et al. 2002; Sun and Yang, 2000; Wu and Subbarao 2008), their inclusion in the development of prediction models using empirical field data is relatively rare (Reich and Chatterton 2023). In previous prediction modeling efforts, daily soil temperature was used to predict the severity index of Sclerotinia rot of carrot, where observations with mean soil temperatures below 19°C were assigned the greatest risk level (Foster et al. 2011). However, in 84 the present study, a mean soil temperature of 21.51°C was found to coincide with the greatest apothecia density. A previous study describing apothecia development in narrow and wide row spaced soybeans modeled the relationship between soil temperature and apothecia density, finding very similar estimations of optimal soil temperatures for apothecia development (21.53- 23.52°C) (Fall et al. 2018b). Although Twengstrom et al. 1998a found that 30 days of continuous soil moisture was critical for apothecia development, the lesser importance of soil moisture in comparison to soil temperature found here agrees with previous modeling efforts. Foster et al. 2011 found that there was no significant correlation between soil moisture (measured as soil matric potential) and number of apothecia found in carrot plots, but still included soil moisture as a risk factor, where observations with soil matric potential less than -20 kPa was assigned the greatest risk level. However, in this study, a significant positive correlation was found with soil moisture, and it was included as the fourth most important feature in the best performing model. Further investigations are warranted to understand the role of soil moisture on apothecia development, and possible interactions with mediating factors, such as soil texture, host crop species, planting density and irrigation practices (Reich and Chatterton 2023). The results from these modeling efforts support previous work emphasizing the importance of canopy closure in apothecia development (Fall et al. 2018b; Foster et al. 2011). The Sclerotinia rot of carrot prediction model developed by Foster et al. (2011) used a 95% canopy closure threshold to assign risk points for the presence of S. sclerotiorum apothecia. In contrast, Fall et al. 2018b identified 40% canopy closure as a critical threshold for apothecia development regardless of soybean row spacing. In the present study, a similar critical threshold was found where the first observation of apothecia density above the binary threshold across all crop types and row spacings occurred when canopy closure was 42.12%. However, the risk of apothecia presence increased significantly around 87% canopy closure, similar to the risk threshold established by Foster et al. 2011 for carrot. In their study, Fall et al. 2018b also found that distance from the planted row was eight times more influential on apothecia density than canopy closure according to ordinary least squares regression analysis. Therefore, our approach used the distance that shade from the plant canopy extends into the row (row shading) as the measure of crop development, rather than canopy closure. In the two best-performing models using the correlation-based feature pool, row shading was either the most important (DT RTMA only) or second most important (DT RTMA+LOCOMOS) feature, confirming the importance of 85 the distance of shading into the row found by Fall et al. 2018b and contributing to the accuracy of the models developed herein. Although using a canopy closure threshold would be easier to incorporate into the user interface of a decision support tool, row shading was considered more suitable than canopy closure for a multi-crop model. Row shading represents a more consistent measurement between crops, whereas percent canopy closure is dependent on row spacing. Users can be prompted to input their row spacing (as categorical, 0.38 m (15”), 0.51 m, (20”), 0.76 m (30”), 0.86 m (34”)) and their estimated canopy closure (as a percent), either visually estimated or measured via Canopeo. From these measurements, row shading can be calculated and input into the predictive model. When applying machine learning tools, there is often a sacrifice in interpretability to achieve greater predictive power (Assis et al. 2025). Interpretability can be important for models intended for use by farmers, as transparent and explainable models are more likely to be accepted and perceived as trustworthy (Gardezi et al. 2023). In this study, PCA was employed for dimensionality reduction in effort to improve predictions at the expense of interpretability. However, PCA-based models showed no substantial improvement over correlation-based models. Additionally, opaque random forest and support vector machine models did not outperform the transparent DT models. Therefore, the implementation of robust but interpretable ML models stand to improve SSR management. Determining how much improvement these new approaches offer will require field validation across a wide geographic area and economic assessment of a complete decision support system. Instead of improving predictions through methods that limit model interpretability, predictions were largely improved by the incorporation of on-site weather data, which presents different challenges for real-world deployment. Weather monitoring sensors can be costly and require regular calibration, maintenance, and power and communication systems to support the feed of local data into prediction models (Ojha et al. 2015). However, the LOCOMOS used in the present study are designed specifically to be cost-effective and suited for deployment in agricultural settings (Dong et al. 2024). Multiple Internet of Things (IoT) systems have been developed to support plant disease prediction (Chen et al. 2022; Delfani et al. 2024; Khattab et al. 2019; Kim et al. 2018; Liu et al. 2022), but there is still a low adoption of these technologies in agriculture (Hundal et al. 2023). Similar IoT systems have been deployed to support prediction models in non-agricultural environments, such as natural disaster event prediction 86 (Esposito et al. 2022) and water quality monitoring (Lakshmikantha et al. 2021). This suggests that agriculture and pest management could significantly benefit from the broader adoption of these emerging technologies, unlocking potential for smarter, more sustainable practices. Conclusions and future directions Supervised ML models proved effective at accurately predicting the presence of apothecia in soybean, dry bean and potato fields in Michigan, particularly DT models. On-site weather (LOCOMOS) systems monitoring soil temperature and moisture were required to significantly improve predictions in comparison to remotely accessible data alone. Row shading was an important feature in model performance, lending an ability to estimate the risk of apothecia presence across diverse cropping systems, host architectures and row spacings. Model interpretability was not sacrificed to improve apothecia presence predictions. Although DT models are not as easily interpretable as LR models, DTs are still considered transparent ML models, and weather and row shading data did not require transformation into PCs to improve accuracy. While the original irrigated model was 57% accurate, the proposed DT model using on-site measurements of soil temperature and moisture achieved an accuracy of 87% across host crops. However, comparisons between models are hindered by an unfair advantage new models have, as the locations and crop types used for training and testing ML models were not used for the original irrigated model development. Therefore, the comparisons between previous and newly developed models should be approached with some caution. To build upon the work presented here, additional data will be required to represent a greater diversity in geographies, S. sclerotiorum field histories and host crop architectures to test model performance before the deployment of an improved decision support system for SSR management in irrigated environments. Future validation will be required to assess the ability of these models to improve SSR control and economic returns through on-farm trials comparing model fungicide timing recommendations to standard growth-stage timed trials (Caffi et al. 2012; Foster et al. 2011; Small et al. 2015; Willbur et al. 2018). Additional site-specific factors can be incorporated during model validation to improve the performance of these models. Integrating host resistance into the original irrigated model demonstrated that adjusting action thresholds based on plant host resistance level improved prediction accuracy of SSR development (Webster et al. 2023b). The data used for model development is largely representative of research fields with high historical SSR pressure. Incorporating a feature 87 representing different levels of SSR field history has the potential to improve final disease control and production economics by adjusting LR thresholds and ML predictions determining fungicide application timing. In all, this work demonstrates a robust methodology for the prediction of soilborne fungal pathogens utilizing ML tools, quantifies the value of on-site weather monitoring for soilborne disease prediction, demonstrates the utility of transparent ML models for accurate disease prediction, and paves the way for future improvements on a risk prediction tool for Midwestern soybean, dry bean and potato farmers suffering losses from SSR. 88 Table 3.1: Apothecia scouting location information. All apothecia scouting was conducted in research and commercial fields in Michigan. Tables aExpressed as latitude, longitude and rounded to the hundredth to protect cooperator anonymity bNumber of observations taken at a location and year. The number of daily estimates of apothecia density is represented in parentheses cSoil types were taken from the National Cooperative Soil Survey (www.ncrs.usda.gov) 89 Table 3.2: The performance metrics for candidate logistic regression and supervised machine learning models for predicting S. sclerotiorum apothecia presence on hold out test data. Correlation analysis and principal coordinates analysis were used for feature selection and dimensionality reduction. Auto machine learning (AutoML) was used to create a benchmark for model performance. Model development was conducted on two different data pools, data sourced exclusively from real-time mesocale analysis (RTMA) and RTMA data supplemented with soil moisture and temperature sourced from on-site weather monitoring systems. The performance of the original irrigated model was also assessed. aArea under the receiver operating characteristic curve bCalculated as the ratio of the sum of true positives and true negatives to all predictions cCalculated as the ratio of true positives to the sum of true positives and false negatives dCalculated as the ratio of true negatives to the sum of true negatives and false positives eCalculated as the ratio of true positives to the sum of true positives and false negatives 90 Figures Figure 3.1: A single Sclerotinia sclerotiorum apothecia on the soil surface (A). A cluster of S. sclerotiorum apothecia originating from a single sclerotia (B). Wilting due to S. sclerotiorum infection in a soybean field (C). Characteristic fluffy, white mycelial growth on S. sclerotiorum infected soybean stems (D). A Low-Cost Monitoring System (LOCOMOS) deployed in a dry bean field (E). An image of a healthy soybean canopy taken during canopy closure measurements (F). 91 Figure 3.2: Feature pool determined by point biserial correlation analysis. Features represent 30- day moving averages. A unique function of each potential weather variable was chosen based on correlation strength and significance. Significance is indicated by asterisks with 1, 2, and 3 asterisks representing statistical significance at P < 0.05, 0.001, and 0.0001, respectively. Row shading represents the distance into the planted row that was shaded due to canopy closure. Acronyms are as follows: RHD (duration of relative humidity over 86%), ATD (duration of air temperature between 10 and 25°C) and WS (wind speed). 92 Figure 3.3: Decision tree model developed used RTMA+LOCOMOS data with a variable pool selected by correlation analysis. Each internal node represents a logical test on one of the independent variables. Internal and leaf node colors are represented by the purity of the resulting pool of observation. Orange represents observations of apothecia below the apothecia density threshold (0.18 apothecia per m2) and blue represents observations above the threshold. All continuous independent weather variables were normalized using maximum absolute rescaling prior to model development. Categorical soil type was recoded as 0, 1 and 2 representing loam, loamy sand and sand soil types. This model represents the best performing model by overall accuracy (89%) among the models developed in this study. Row shading represents the distance into the planted row that was shaded due to canopy closure. Acronyms are as follows: RHD (duration of relative humidity over 85%), ATD (duration of air temperature between 10 and 25°C) and WS (wind speed). 93 Figure 3.4: Feature importance for the best performing correlation-based models for discriminating between observations with apothecia presence above and below the apothecia density threshold (0.18 apothecia per m2). Feature importance represents the total decrease in information gain by the feature in the model. Row shading represents the distance into the planted row that was shaded due to canopy closure. Acronyms are as follows: RHD (duration of relative humidity over 85%), ATD (duration of air temperature between 10 and 25°C) and WS (wind speed). 94 Figure 3.5: Top: The relationship between canopy closure (%) and the cumulative sum of the apothecia binary response normalized to all positive binary responses. The red, dotted vertical 95 Figure 3.5 (cont’d) line represents 90% canopy closure. Bottom: The relationship between apothecia density for each crop and row spacing category. A red, dotted vertical line indicates the first observation of apothecia density over the binary threshold (0.18 apothecia per m2). 96 Figure 3.6: Left: the relationship between 30-day moving averages of soil temperature (°C) and apothecia density (apothecia per m2). A Gaussian model was fit to the data to represent the mean, standard deviation and amplitude corresponding to the peak in apothecia density. Right: the relationship between 30-day moving averages of soil moisture (%) and apothecia density. A bimodal Gaussian model was fit to the data to represent the means, standard deviations and amplitudes corresponding to the peaks in apothecia density. 97 LITERATURE CITED Abawi, G. S. and Grogan, R. G. 1975. Source of primary inoculum and effects of temperature and moisture on infection of beans by Whetzelinia sclerotiorum. Phytopathology 65:300- 309. doi.org/10.1094/Phyto-65-300 Abawi, G. S. and Grogan, R. G. 1979. Epidemiology of diseases caused by Sclerotinia species. Phytopathology 69:899-904. doi.org/10.1094/Phyto-69-899 Adams, P. B. and Ayers, W. A. 1979. Ecology of Sclerotinia species. Phytopathology 69:896- 899. doi.org/10.1094/Phyto-69-896 Agrios, G. N. 2005. Plant Pathology, 5th ed. Elsevier Academic Press, Burlington, MA. Assis, A., Dantas, J. and Andrade, E. 2025. The performance-interpretability trade-off: a comparative study of machine learning models. J. Reliab. Intell. Environ. 11:1. doi.org/10.1007/s40860-024-00240-0 Boland, G. J. and Hall, R. 1994. Index of plant hosts of Sclerotinia sclerotiorum. Can. J. Plant Pathol. 162:93-108. doi.org/10.1080/07060669409500766 Breiman, L. 2001. Random Forests. Mach. Learn. 45:5-32. doi.org/10.1023/A:1010933404324 Budge, S. P. and Whipps, J. M. 2001. Potential for integrated control of Sclerotinia sclerotiorum in glasshouse lettuce using Coniothyrium minitans and reduced fungicide application. Phytopathology 91:221-227. doi.org/10.1094/PHYTO.2001.91.2.221 CABI. 2005. Sclerotinia sclerotiorum (Lib.) de Bary. In: Distribution Maps of Plant Diseases. Wallingford, U.K. CAB International. doi.org/10.1079/DMPD/20066500971 Chen, C. J., Li, Y. S., Tai, C. Y., Chen, Y. C. and Huang, W. M. 2022. Pest incidence forecasting based on Internet of Things and Long Short-Term Memory Network. Appl. Soft Comput. 124:108895. doi.org/10.1016/j.asoc.2022.108895 Christias, C. and Lockwood, J. L. 1973. Conservation of mycelial constituents in four sclerotium-forming fungi in nutrient-deprived conditions. Phytopathology 63:602-605. doi.org/10.1094/Phyto-63-602 Clarkson, J. P., Phelps, K., Whipps, J. M., Young, C. S. Smith, J. A. and Watling, M. 2004. Forecasting sclerotinia disease on lettuce: Toward developing a prediction model for carpogenic germination of sclerotia. Phytopathology 94: 268-279. doi.org/10.1094/PHYTO.2004.94.3.268 Clarkson, J. P., Phelps, K., Whipps, J. M., Young, C. S., Smith, J. A. and Watling, M. 2007. Forecasting sclerotinia disease on lettuce: A predictive model for carpogenic germination of Sclerotinia sclerotiorum sclerotia. Phytopathology 97:621-631. doi.org/10.1094/PHYTO-97-5-0621 Clarkson, S. J., Fawcett, L., Anthony, S. G. and Young, C. 2014. A model for Sclerotinia sclerotiorum infection and disease development in lettuce, based on the effects of temperature, relative humidity and ascospore density. PLoS ONE 9:94049. doi.org/10.1371/journal.pone.0094049 98 De Wolf, E. D. and Francl, L. J. 1997. Neural networks that distinguish infection periods of wheat tan spot in an outdoor environment. Phytopathology 97:83-87. doi.org/10.1094/PHYTO.1997.87.1.83 De Wolf, E. D. and Francl, L. J. 1999. Neural network classification of tan spot and Stagonospora blotch infection periods in a wheat field environment. Phytopathology 90: 108-113. doi.org/10.1094/PHYTO.2000.90.2.108 Delfani, P., Thuraga, V., Banerjee, B. and Chawade, A. 2024. Integrative approaches in modern agriculture: IoT, ML and AI for disease forecasting amidst climate change. Precis. Agric. 25:2589-2613. doi.org/10.1007/s11119-024-10164-7 Dillard, H. R., Ludwig, J. W., and Hunter, J. E. 1995. Conditioning sclerotia of Sclerotinia sclerotiorum for carpogenic germination. Plant Dis. 79:411-415. doi.org/10.1094/PD-79- 0411 Dong, Y., Christenson, C., Kelley, L. and Miller, S. 2023. Trends and future of agricultural irrigation in Michigan and Indiana. Irrigation and Drainage 73: 346-358. doi.org/10.1002/ird.2862 Dong, Y., Werling, B., Cao, Z. and Li, G. 2024. Implementation of an in-field IoT system for precision irrigation management. Front. Water. 6:135397. doi.org/10.3389/frwa.2024.1353597 Esposito, M., Palma, L., Belli, A., Sabbatini, L. and Pierleoni, P. 2022. Recent advances in Internet of Things solutions for early warning systems: A review. Sensors. 22:2124. doi.org/10.3390/s22062124 Fall, M. L., Boyse, J. F., Wang, D., Wilbur, J. F., Smith, D. L. and Chilvers, M. I. 2018a. Case study of an epidemiological approach dissecting historical soybean Sclerotinia stem rot observations and identifying environmental predictors of epidemics and yield loss. Phytopathology 108:469-478 doi.org/10.1094/PHYTO-12-16-0446-R Fall. M. L., Willbur, J. F., Smith, D. L., Byrne, A. M. and Chilvers, M. I. 2018b. Spatiotemporal distribution pattern of Sclerotinia sclerotiorum apothecia is modulated by canopy closure and soil temperature in an irrigated soybean field. Plant Dis. 102:1794-1802. doi.org/10.1094/PDIS-11-17-1821-RE Ferraz, L. C. L., Café Filho, A. C., Nasser, L. C. B. and Azevedo, J. 1999. Effects of soil moisture, organic matter and grass mulching on the carpogenic germination of sclerotia and infection of bean by Sclerotinia sclerotiorum. Plant Pathol. 48:7-82. doi.org/10.1046/j.1365-3059.1999.00316.x Foster, A. J., Kora, C., McDonald, M. R. and Boland, G. R. 2011. Development and validation of a disease forecast model for Sclerotinia rot of carrot. Canadian Journal of Plant Pathol. 33:187-201. doi.org/10.1080/07060661.2011.563753 Gardezi, M., Joshi, B., Rizzo, D. M., Ryan, M., Prutzer, E., Brugler, S. and Dadkhah, A. 2023. Artificial intelligence in farming: Challenges and opportunities for building trust. Agron. J. 116:1217-1228. doi.org/10.1002/agj2.21353 99 Hao, J. J., Subbarao, K. V. and Duniway, J. M. 2002. Germination of Sclerotinia minor and S. sclerotiorum sclerotia under various soil moisture and temperature combinations. Phytopathology 93:443-450. doi.org/10.1094/PHYTO.2003.93.4.443 Harikrishnan, R. and del Rio, L. E. 2008. A logistic regression model for predicting risk of white mold incidence on dry bean in North Dakota. Plant Dis. 92:42-46. doi.org/10.1094/PDIS- 92-1-0042 Hartill, W. F. T. and Underhill, A. P. 1976. “Puffing” in Sclerotinia sclerotiorum and S. minor. New Zealand Journal of Botany 14:355-358. doi.org/10.1080/0028825X.1976.10428909 Hegedus, D. D. and Rimmer, S. R. 2005. Sclerotinia sclerotiorum: When “to be or not to be” a pathogen? FEMS Microbiol. Lett. 251:177-184. doi.org/10.1016/j.femsle.2005.07.040 Hundal, G. S., Laux, C. M., Buckmaster, D., Sutton, M. J. and Langemeier, M. 2023. Exploring barriers to the adoption of Internet of Things-based precision agriculture practices. Agriculture 12:163. doi.org/10.3390/agriculture1301016 Jimenez, A. F., Cardenas, P. F. and Jimenez, F. 2022. Intelligent IoT-multiagent precision irrigation approach for improving water use efficiency in irrigation systems at farm and district scales. Comput. Electron. Agric. 192:106635. doi.org/10.1016/j.compag.2021.106635 Johnson, D. A., and Atallah, Z. K. 2006. Timing fungicide applications for managing sclerotinia stem rot of potato. Plant Dis. 90:755-758. doi.org/10.1094/PD-90-0755 Jones, S. J., Gent, D. H., Pethybridge, S. J. and Hay, F. S. 2011. Site-specific risk factors of white mould epidemics in bean (Phaseolus vulgaris) in Tasmania, Australia. N. Z. J. Crop Hortic. Sci. 40:148-159. doi.org/10.1080/01140671.2011.623707 Kabbage, M., Yarden, O. and Dickman, M. B. 2015. Pathogenic attributes of Sclerotinia sclerotiorum: Switching from a biotrophic to necrotrophic lifestyle. Plant Sci. 233:53-60. doi.org/10.1016/j.plantsci.2014.12.018 Kandel, Y. R., Hunt, C., Ames, K., Arneson, N., Bradley, C. A., Byamukama, E., Byrne, A., Chilvers, M. I., Giesler, L. J., Halvorson, J., Hooker, D. C., Kleczewski, N. M., Malvick, D. K., Markell, S., Potter, B., Pedersen, W., Smith, D. L., Tenuta, A. U., Telenko, D. E. P., Wise, K. A., and Mueller, D. S. 2021. Meta-analysis of soybean yield response to foliar fungicides evaluated from 2005 to 2018 in the United States and Canada. Plant Dis. 105:1382-1389. doi.org/10.1094/PDIS-07-20-1578-RE Kaundal, R., Kapoor, A. S. and Raghava, G. P. S. 2006. Machine learning techniques in disease forecasting: A case study on rice blast prediction. BMC Bioinform. 7:485. doi.org/10.1186/1471-2105-7-485 Khan, A. A., Faheem, M., Bashir, R. N., Wechtaisong, C. and Abbas, M. Z. Internet of Things (IoT) assisted context aware fertilizer recommendation. IEEE Access 10:129505-129519. doi.org/10.1109/ACCESS.2022.3228160 Khattab, A., Habib, S. E. D., Ismail, H., Zayan, S., Fahmy, Y. and Khairy, M. M. 2019. An IoT- based cognitive monitoring system for early plant disease forecast. Comput. Electron. Agric. 166:105028. doi.org/10.1016/j.compag.2019.105028 100 Kim, S., Lee, M. and Shin, C. 2018. IoT-based strawberry disease prediction system for smart farming. Sensors 18:4051. doi.org/10.3390/s1811405 Kingsford, C. and Salzberg, S. L. 2008. What are decision trees? Nat. Biotech. 26:1011-1013. Koch, S., Dunker, S., Kleinhenz, B., Rohrig, M. and von Tiedemann, A. 2007. A crop-loss related forecasting model for Sclerotinia stem rot in winter oilseed rape. Phytopathology 97:1186-1194. doi.org/10.1094/PHYTO-97-9-1186 Lakshmikantha, V., Hiriyannagowda, A., Manjunath, A., Patted, A., Basavaiah, J. and Anthony, A. A. 2021. IoT based smart water quality management system. Glob. Transit. Proc. 2:182-186. doi.org/10.1016/j.gltp.2021.08.062 Landschoot, S., Waegeman, W., Audenaert, K., Vandepitte, J., Haesaert, G. and De Baets, D. 2012. Toward a reliable evaluation of forecasting systems for plant diseases: A case study using Fusarium head blight of wheat. Plant Dis. 96:889-896. doi.org/10.1094/PDIS-08- 11-0665 Lavanya, G., Rani, C. and Gaenshkumar, P. 2020. An automated low cost IoT based fertilizer intimation system for smart agriculture. Sustain. Comput. Inform. Syst. 28:100300. doi.org/10.1016/j.suscom.2019.01.002 Le, T. T., Fu, W. and Moore, J. H. 2020. Scaling-tree based automated machine learning to biomedical big data with a feature set selector. Bioinformatics 36:250-256. doi.org/10.1093/bioinformatics/btz470 Liu, Z., Nawaz Bashir, R., Iqbal, S., Shadid, M. M. A., Tausif, M. and Umer, Q. 2022. Internet of Things (IoT) and machine learning model of plant disease prediction – blister blight for tea plant. IEEE 10:44934-44944. doi.org/10.1109/ACCESS.2022.3169147 Newton, H. C. and Sequeira, L. 1972. Ascospores as the primary infective propagule of Sclerotinia sclerotiorum in Wisconsin. Plant Dis. Rep. 56:798-802 Noble, W. S. 2006. What is a support vector machine. Nat. Biotech. 24:1565-1567. Ojha, T., Misra, S. and Singh Raghuwanshi, N. S. Wireless sensor networks for agriculture: The state-of-the-art in practice and future challenges. Comput. Electron. Agr. 118:66-84. doi.org/10.1016/j.compag.2015.08.011 Patrignani, A. and Ochsner, T. E. 2015. Canopeo: A powerful new tool for measuring fractional green canopy cover. Agron. J. 107:2312-2320. doi.org/ 10.2134/agronj15.0150 Pedregosa, F., Varoquaux, G., Gramfort, A., Michel, V., Thirion, B., Grisel, O., Blondel, M., Prettenhofer, P., Weiss, R., Dubourg, V., Vanderplas, J., Passos, A., Cournapeau, D., Brucher, M., Perrot, M. and Duchesnay, É. 2011. Scikit-learn: Machine Learning in Python. JMLR 12:2825-2830. doi.org/10.5555/1953048.2078195 Peltier, A. J., Bradley, C. A., Chilvers, M. I., Malvick, D. K., Mueller, D. S.., Wise, K. A. and Esker, P. D. 2012. Biology, yield loss and control of Sclerotinia stem rot of soybean. J. Integr. Pest Manag. 3:1-7. doi.org/10.1603/IPM11033 101 Pethybridge, S. J., Brown, B. J., Kikkert, J. R. and Ryan, M. R. 2019a. Rolled-crimped cereal rye residue suppresses white mold in no-till soybean and dry bean. Renew. Agr. Food Syst. 35:599-607. doi.org/10.1017/S174217051900022X Pethybridge, S. J., Gugino, B. K. and Kikkert, J. R. 2019b. Optimizing fungicide timing for management of white mold in processing snap bean in New York. J. Crop Prot. 125:104883. doi.org/10.1016/j.cropro.2019.104883 Purdy, L. H. 1979. Sclerotinia sclerotiorum: History, diseases and symptomatology, host range, geographic distribution, and impact. Phytopathology 69: 875-880. doi.org/10.1094/Phyto- 69-875 Raschka, S. 2018. MLxtend: Providing machine learning and data science utilities and extensions to Python's scientific computing stack. J. Open Source Softw. 3:638. doi.org/10.21105/joss.00638 Reich, J. and Chatterton, S. 2023. Predicting field diseases caused by Sclerotinia sclerotiorum: A review. Plant Pathol. 72:3-18. doi.org/10.1111/ppa.13643 Roper, M., Seminara, A., Bandi, M. M., Cobb, A., Dillard, H. R. and Pringle, A. 2010. Dispersal of fungal spores on a cooperatively generated wind. PNAS 107:17474-17479. doi.org/ 10.1073/pnas.1003577107 Rossi, V., Caffi, T., Salotti, R. and Fedele, G. 2023. Sharing decision-making tools for pest management may foster implementation of Integrated Pest Management. Food Secur. 15:1459-1474. doi.org/10.1007/s12571-023-01402-3 Saharan, G. S. and Mehta, N. 2008. Sclerotinia Diseases of Crop Plants: Biology, Ecology and Disease Management. Springer, Dordrecht, The Netherlands Sanjel, S., Colee, J., Barocco, R. L., Dufault, N. S., Tillman, B. L., Punja, Z. K. Seepaul, R. and Small, I. M. 2024. Environmental factors influencing stem rot development in peanut: Predictors and action thresholds for disease management. Phytopathology 114:393-404. doi.org/10.1094/PHYTO-05-23-0164-R Shafi, U., Mumtaz, R., Garcia-Nieto, J., Hassan, S. A., Zaidi, S. A. R. and Iqbal, N. 2019. Precision agriculture techniques and practices: From considerations to applications. Sensors 19:3796. doi.org/10.3390/s19173796 Shah, D. A., De Wolf, E. D., Paul, P. A. and Madden, L. V. 2023. Into the trees: Random forests for predicting Fusarium head blight epidemics of wheat in the United States. Phytopathology 113:1483-1493. doi.org/10.1094/PHYTO-10-22-0380-R Skelsey, P. 2021. Forecasting risk of crop disease with anomaly detection algorithms. Phytopathology. 111:321-322. doi.org/10.1094/PHYTO-05-20-0185-R Sperschneider, J. 2019. Machine learning in plant-pathogen interactions: empowering biological predictions from field scale to genome scale. New Phytol. 228:5. doi.org/10.1111/nph.15771 102 Sun, P. and Yang, X. B. 2000. Light, temperature, and moisture effects on apothecium production of Sclerotinia sclerotiorum. Plant Dis. 84:1287-1293. doi.org/10.1094/PDIS.2000.84.12.1287 Turkington, T. K. Morrall, R. A. A. and Gugel, R. K. 1991. Use of petal infestations to forecast sclerotinia stem rot of canola: Evaluation of early bloom sampling, 1985-90. Can. J. Plant Pathol. 13:50-59. doi.org/10.1080/07060669109500965 Turkington, T. K. and Morrall, R. A. A. 1993. Use of petal infestation to forecast Sclerotinia stem rot of canola: The influence of inoculum variation over the flowering period and crop density. Phytopathology 83:682-689. doi.org/10.1094/Phyto-83-682 Twengstrom, E., Kopmans, E., Sigvald, R. and Svensson, C. 1998a. Influence of different regimes on carpogenic germination of sclerotia of Sclerotinia sclerotiorum. J. Phytopathology. 146:487-493. doi.org/10.1111/j.1439-0434.1998.tb04610.x Twengstrom, E., Sigvald, R., Svensson, C. and Yuen, J. 1998b. Forecasting Sclerotinia stem rot in spring sown oilseed rape. Crop Prot. 17:405-411. doi.org/10.1016/S0261- 2194(98)00035-0 USDA-NRCS. 2025. Web Soil Survey. Online publication. https://websoilsurvey.nrcs.usda.gov/app/ Webster. R.W., Roth, M. G., Mueller, B. D., Mueller, D. S., Chilvers, M. I., Willbur, J. F., Mourtzinis, S., Conley, S. P. and Smith, D. L. 2022. Integration of row spacing, seeding rates, and fungicide applications for control of sclerotinia stem rot in Glycine max. Plant Dis. 106: 1183-1191. doi.org/ 10.1094/PDIS-09-21-1931-RE Webster, R. W., Muller, B. D., Chilvers, M. I., Byrne, A. M., Boyse, J. F., Widdicombe, W. W., Mueller, D. S., Wiggs, S., Kandel, Y., Telenko, D. E. P., Ravellette, J., Shim, S. and Smith, D. L. 2023a. Integrating seeding rates and pesticide programs for managing Sclerotinia stem rot in Glycine max with nitrogen fertilizer applications. Plant Health Prog. 24:320-325. doi.org/10.1094/PHP-10-22-0102-RS Webster, R. W., Mueller, B. D., Conley, S. P., Smith, D. L. 2023b. Integration of soybean (Glycine max) resistance levels to sclerotinia stem rot into predictive Sclerotinia sclerotiorum apothecia models. Plant Dis. 107:2763-2768. doi.org/10.1094/PDIS-12-22- 2875-RE Willbur, J. F., Fall, M. L., Bloomingdate, C., Byrne, A. M., Chapman, S. A., Isard, S. A., Magarey, R. D., McCaghey, M. M., Mueller, B. D., Russo, J. M., Schlegel J., Chilvers, M. I., Mueller, D. S., Kabbage, M. and Smith, D. L. 2018a. Weather-based models for assessing the risk of Sclerotinia sclerotiorum apothecial presence in soybean (Glycine max) fields. Plant Dis. 102:73-84. doi.org/10.1094/PDIS-04-17-0504-RE Willbur, J. F., Fall, M. L., Byrne, A. M., Chapman, S. A., McCaghey, M. M., Mueller, B. D., Schmidt, R., Chilvers, M. I., Mueller, D. S., Kabbage, M., Giesler, L. J., Conley, S. P., Smith, D. L. 2018b. Validating Sclerotinia sclerotiorum apothecial models to predict Sclerotinia stem rot in soybean (Glycine max) fields. Plant Dis. 102:2592-2601. doi.org/10.1094/PDIS-02-18-0245-RE 103 Willbur, J. F., Mitchell P. D., Fall, M. L., Byrne, A. M., Chapman, S. A., Floyd, C. M., Bradkey, C. A., Ames, K. A., Chilvers, M. I., Kleczewski, N. M., Malvick, D. K., Mueller, B. D., Mueller, D. S., Kabbage, M., Conley, S. P. and Smith, D. L. 2019. Meta-analytic and economic approaches for evaluation of pesticide impact on Sclerotinia stem rot control and soybean yield in the north central United States. Phytopathology 109:1157-1170. doi.org/10.1094/PHYTO-04-18-0124-R Willets, H. J. and Wong, J. A. L. 1980. The biology of Sclerotinia sclerotiorum, S. trifoliorum, and S. minor with emphasis on specific nomenclature. Bot. Rev. 46:101-165 Wu, B. M. and Subbarao, K. V. 2008. Effects of soil temperature, moisture and burial depths on carpogenic germination of Sclerotinia sclerotiorum and S. minor. Phytopathology 98:1144-1152. doi.org/10.1094/PHYTO-98-10-1144 104 Supplementary tables and figures APPENDIX Table A3.2: Performance metrics for the original irrigated model on all crop types used to develop multi-crop models. aCalculated as the ratio of the sum of true positives and true negatives to all predictions bCalculated as the ratio of true positives to the sum of true positives and false negatives cCalculated as the ratio of true negatives to the sum of true negatives and false positives dCalculated as the ratio of true positives to the sum of true positives and false negatives 105 Table A3.2: Hyperparameters optimized for supervised classification machine learning algorithms using GridSearchCV. 106 Figure A3.1: Apothecia density and original irrigated model risk (%) at each location throughout 107 Figure A3.1 (cont’d) the scouting window. Thresholds for apothecia density (apothecia per m2) and original irrigated model spray risk are represented by dashed lines. 108 Figure A3.2: PCA loadings for RTMA-based models (left) and RTMA+LOCOMOS-based models (right). Acronyms are as follows: AT (air temperature), DP (dew point), RH (relative humidity), WS (wind speed), RH86 (duration of relative humidity over 86%), ATD (duration of air temperature between 10 and 25°C), RF_IBM (rainfall), SM_IR (soil moisture under irrigation) and ST (soil temperature under irrigation). 109 Figure A3.3: Scree plots for maximum absolute rescaling determining the number of principal components using a 90% explained variance cut off. 110 CHAPTER 5: CONCLUSIONS Conclusions Effective management strategies for field crop diseases in Michigan are essential for the environmental and economic sustainability of the state’s agricultural industry. Disease management can be improved by identifying agronomic and environmental risk factors of pathogen proliferation and disease development. This dissertation aimed to describe risk factors for two major diseases of field crops in Michigan and use this knowledge to create tools for plant disease management As tar spot of corn became established in the northern U.S. in the previous decade, information regarding the role of agronomic factors in disease development became of critical significance to farmers and other stakeholders. In this dissertation, field trials revealed that nitrogen application rate had no significant effect on disease development, while low planting populations were associated with greater disease development. However, the economic analysis revealed that the economic penalty from increasing planting rate did not offset the benefit from reduced disease severity. With this information, farmers can now make better informed agronomic decisions without concern of increasing their risk of tar spot development. Additionally, these field trials demonstrated the importance of planting partially resistant corn hybrids for managing tar spot, which remains the most effective management strategy. Ongoing studies aim to describe how low planting populations affect the corn canopy microclimate to better understand this phenomenon. To effectively communicate these findings, this work has been presented at extension events throughout the state and summarized in an article published by the Crop Protection Network. The development of early warning systems is a key advancement in modern integrated pest management strategies. This dissertation combined molecular detection technologies with spore trapping to provide the first description of environmental drivers of P. maydis spore release in the United States. Results aligned with current tar spot prediction modeling efforts, confirming spore release and disease development are strongly influenced by temperature, with a preference for cooler conditions. Describing conditions that lead to P. maydis spore release contributes to our current understanding of the tar spot disease cycle, which is crucial to identifying risk factors leading to severe tar spot outbreaks. These initial findings provide a foundation for a tar spot early warning system, which has the potential to be integrated into an empirical or mechanistic 111 prediction model for tar spot epidemic development. An early warning system integrating spore trapping and predictive modeling would provide farmers with valuable tools for disease management. Future experiments exploring how different spore trapping technologies, placements and sample processing methods improve the capture and recovery of P. maydis will help achieve earlier pathogen detection to increase the feasibility of a tar spot early warning system. Additionally, this research resulted in a collaborative effort to create and disseminate resources for constructing low-cost rotating-arm spore traps in the form of journal articles and workshops to improve the affordability and accessibility of these technologies. Prediction models have the potential to improve plant disease management optimizing disease control, improving production epidemics and reducing unnecessary fungicide use. Midwestern soybean growers already utilize prediction models to inform fungicide application timing for Sclerotinia stem rot management. To improve these tools for irrigated environments and expand their application to potato and dry bean crops, this research employed machine learning (ML) and Internet-of-Things (IoT) weather monitoring systems. The newly developed models outperformed existing prediction models and extended their utility beyond soybean farmers. Additionally, there is potential to validate these models for additional irrigated cropping systems with similar production practices and host canopy architectures, such as peanut in the southern U.S. and pulse crops in the central U.S. Even though model improvements were observed without IoT integration, incorporating soil moisture and temperature data significantly increased prediction accuracy. While both logistic regression (LR) and ML models improved predictions compared to the original irrigated model, ML provided a substantial advantage. This study demonstrated the value of integrating on-site weather data collection and ML-based prediction models, and achieved such without sacrificing interpretability, which is important for farmer trust and adoption. Implementing these models will require the maintenance of IoT systems and stable communications networks to feed site-specific weather data into a decision support system to output risk predictions. One possible avenue to simplify this process is the use of wireless soil sensors, which have been deployed in other irrigation management systems. Exploring the balance between maximizing predictive power and managing the complexities of hosting computationally expensive ML models and integrating IoT systems should be prioritized in future validation efforts. Overall, insights gained offer valuable guidance for other plant disease prediction model efforts. Future field trials across diverse geographies will be required to 112 validate this tool prior to deployment as a decision support tool and will be critical for finding the balance between system complexity and predictive power. In conclusion, this dissertation expands current knowledge on the epidemiology and management of diseases affecting Michigan field crops. Collectively, corn, soybean, potato and dry bean represent over six million acres of Michigan farmland. Thus, the disease management strategies and predictive tools that were developed in this dissertation for tar spot of corn and Sclerotinia stem rot on soybean, potato and dry bean have contributed valuable knowledge and resources to Michigan farmers and those beyond our state boundaries. This work also serves as a foundation for advancing modern agricultural practices by integrated early disease warning systems and machine-learning based prediction tools. 113 List of published and submitted works from the duration of my PhD * Denotes first authorship † Denotes co-first authorship Valle-Torres, J., Ross, T. J., Plewa, D., Avellaneda, M. C., Check, J., Chilvers, M. I., Cruz, A.P., Dalla Lana, F., Groves, C., Gongora-Canul, C., Henriquez-Dole, L., Jamann, T., Kleczewski, N., Lipps, S., Malvick, D., McCoy, A. G., Mueller, D. S., Paul, P. A., Puerto, C., Schloemer, C., Raid, R. N., Robertson, A., Roggenkamp, E. M., Smith, D. L., Telenko, D. E. P. and Cruz, C. D. 2020. Tar spot: An understudied disease threatening corn production in the Americas. Plant Dis. 104:2541-2550. Plant Disease Feature Article Rocco da Silva, C.†, Check, J.†, MacCready, J. S.†, Alakonya, A. E., Beriger, R., Bissonnette, K. M., Collins, A., Cruz, C. D., Esker, P. D., Goodwin, S. B., Malvick, D., Mueller, D. S., Paul, P., Raid, R., Robertson, A. E., Roggenkamp, E., Ross, T. J., Singh, R., Smith, D. L., Tenuta, A. U., Chilvers, M. I. and Telenko, D. E. P. 2021. Recovery plan for tar spot of corn, caused by Phyllachora maydis. Plant Health Prog. 22:596-616 Editor’s Pick Telenko, D. E. P., Chilvers, M. I., Ames, K., Byrne, A. M., Check, J. C., Da Silva, C. R., Jay, W. S., Mueller, B., Ross, T. J., Smith, D. L. and Tenuta, A. U. 2022. Fungicide efficacy during a severe epidemic of tar spot on corn in the United States and Canada. Plant Health Prog. 23:342-344. Telenko, D. E. P., Chilvers, M. I., Bryne, A., Check, J. C., Rocco da Silva, C. Kleczewski, N., Roggenkamp, E., Ross, T. J. and Smith, D. L. 2022. Fungicide efficacy on tar spot and yield of corn in the Midwestern United States. Plant Health Prog. 23:281-287. Editor’s Pick Check, J. C.*, Aime, M. C., Byrne, J. and Chilvers, M. I. First report of southern rust (Puccinia polysora) on corn (Zea mays) in Michigan. 2022. Plant Dis. 106:2262 Check, J. C.*, Byrne, A. M., Singh, M. P., Steinke, K., Widdicombe, W. D. and Chilvers, M. I. 2023. Effects of nitrogen application rate and plant density on severity of tar spot of corn. Plant Health Prog. 24:416-423. 2023 Best Student Paper Award Solorzano, J. E., Issendorf, S. E., Drott, M. T., Check, J. C., Roggenkamp, E. M., Cruz, C. D., Kleczewski, N. M., Gongora-Canul, C. C. and Malvick, D. K. 2023. A new and effective method to induce infection of Phyllachora maydis into corn for tar spot studies in controlled environments. Plant Methods. 19:83. Webster, R.W., Nicolli, C., Allen, T. W., Bish, M. D., Bissonette, K., Check, J. C., Chilvers, M. I., Kleczewski, N., Mueller, B. D., Price, P. P., Paul, P., Robertson, A. E., Ross, T. J., Schmidt, C., Schmidt, R., Schmidt, T., Shim, S., Telenko, D. E. P., Wise, K. and Smith, D. L. 2023. Uncovering the environmental conditions required for Phyllachora maydis infection and tar spot development on corn in the United States for use as predictive models for future epidemics. Sci. Rep. 13:17064. Check, J. C.†, Harkness, R. J. †, Heger, L.†, Chilvers, M. I., Mahaffee, W. F., Sakalidis, M. L. and Miles, T. D. 2024. It’s a trap! Part I: Exploring the application of rotating-arm 114 impaction samplers in plant pathology. Plant Dis. 108:1910-1922. Plant Disease Feature Article Check, J. C.†, Harkness, R. J.†, Heger, L.†, Chilvers, M. I., Mahaffee, W. F., Sakalidis, M. L. and Miles, T. D. 2024. It’s a trap! Part II: An approachable guide to constructing and using rotating-arm air samplers. Plant Dis. 108:1923-1936. Editor’s Pick Roggenkamp, E. M., Check, J. C., Biswal, A. K., Floyd, C. M., Miles, L. A. Nicolli, C. P., Sujoung, S., Salgado-Salazar, C., Alakonya, A. E, Malvick, D. K., Smith, D. L., Telenko, D. E. P. and Chilvers, M. I. 2024. Development of a qPCR assay for species-specific detection of the tar spot pathogen Phyllachora maydis. PhytoFrontiers 4:61-71. Most- Read Article of 2024 Wade, C., Check, J. C., Chilvers, M. and Dong, Y. 2025. Monitoring leaf wetness dynamics in corn and soybean fields using an IoT (Internet of Things)-based monitoring system. Smart Agr. Technol. 11:100919. Check, J. C.*, Jacobs, J., Phillips, P., Roggenkamp, E., Willbur, J. and Chilvers, M. 202X. Unraveling the environmental drivers of Phyllachora maydis spore release and dispersal using spore trapping and qPCR. Phytopathology. Accepted Rana, S., Ali, N., Cao, Z., Check, J., Chilvers, M., Willbur, J., Werling, B., Liu, Y. and Dong, Y. 202X. Comparative analysis of machine learning models to restore gaps in multivariate time series leaf wetness sensor data. Comput. Electron. Agric. In review. Check, J. C.*, Bales, S., Dong, Y., Smith, D. L., Webster, R. W., Willbur, J. F. and Chilvers, M. I. 202X. Multi-crop Sclerotinia sclerotiorum apothecia prediction models for irrigated environments are improved by on-site weather monitoring and supervised machine learning. Phytopathology. In review. 115