'7‘ Il'ri y ‘.h.|lv"i5. ‘hi l ll1WWW?"lflfllifilflil'liflfllsu'l'fllfl'lflflflililllfl j “m” 3 1293 10570 9665 This is to certify that the thesis entitled Predictive Mode1s for a Microprocessor-Based On-line Pest Management System presented by ASI F NASEEM has been accepted towards fulfillment of the requirements for %.ZN':/ Major professor Date Wifl 0-7639 OVERDUE FINES: _ 25¢ per My per item . (AI-“W ‘ W: Place in book return to remove ""lu‘mr ':- charge from circulation record PREDICTIVE MODELS FOR A MICROPROCESSOR-BASED ON-LINE PEST MANAGEMENT SYSTEM BY Asif Naseem A THESIS Submitted to Michigan State University in partial fuifiiiment of the requirements for the degree of MASTER OF SCIENCE Department of Eiectricai Engineering and Systems Science 1980 e 5‘ I" . I I : . P- ’ / w' .l , 2 .3 “ w' ABSTRACT PREDICTIVE MODELS FOR A MICROPROCESSOR-BASED ON-LINE PEST MANAGEMENT SYSTEM By Asif Naseem This thesis describes a multi-species phenology modeling system. The system contains a library of pest models in a generalized format -so that a particular pest species may be modeled by providing the system with a set of data that includes information on the various life stages, their duration in physiological units, and reproduction function. The modeling system is structured to receive synchronizing information from the field. Although emphasis was placed on pest phenology while formulating the modeling system, it, also, incor- porates factors accounting for losses and additions. This flexi- bility makes it possible to successively refine a model when further data becomes available. This system contains, as two special cases, a model for prediction of infection caused by disease powdery mildew and a phenology model for apple tree development. The powdery mildew disease prediction model operates on climatic data and predicts the mildew infection severity level. The apple tree phenology model predicts the phenophases of the fruit and vegetative development. This system resides in a microprocessor-based intelligent field instrument as the resident program. And this instrument can be installed in a farm to gather on-line weather data which serves as input to the modeling system. Growers can interact with the instru- ment via a keyboard and alphanumeric display to acquire pest manage- ment information. ACKNOWLEDGEMENTS No individual has ever influenced my educational growth as much as Dr.P.D.Fisher has. Without his constant inspiration, guidance and direction this project would not have been realized. I extend my sincere thanks to the following persons for their supportive efforts in this project: Dr. Alan L. Jones for his assistance in developing the powdery mildew infection prediction algorithm, Dr. Brian A. Croft for his guidance in the development of the Pest Modeling System, and Dr. Robert C. Seem for providing the necessary data on apple tree phenology. I also gratefully acknowledge use of Dr. Thomas J. Manetsch’s work in aggrigative distributed delay modeling and the valuable suggestions and incisive comments made by Dr. Eric Goodman and Dr. Ram L. Tummala. Sincere thanks are also due to Ginny Mrazek for typing this thesis and Charles Dorcey for preparing the figures. This research was supported in part by the U.S. Department of Agri- culture under Grant No.901-15-45. ii TABLE OF CONTENTS Chapter l. INTRODUCTION 2. MODEL DEVELOPMENT 2.1 A PEST MANAGEMENT SYSTEM 2.2 DEVELOPMENT OF A GENERALIZED PEST MODEL .2.2.l OBJECTIVES 2.2.2 MODEL DEFINITION 2.2.3 MATHEMATICAL FORMULATION 2.3 SUMMARY 3. IMPLEMENTATION USING AITICROPROCESSOR-BASEB SYSTEM 3.] INTRODUCTION 3.2 THE INSTRUMENT 3.2.1 CONTROL SECTION 3.2.2 INPUT SECTION 3.2.3 OUTPUT SECTION 3.2.4 SYSTEM CLOCK 3.3 GENERAL SOFTWARE ORGANIZATION 3.3.1 START UP AND INITIALIZATION 3.3.2 INTERRUPT SERVICE 3.3.3 KEYBOARD SERVICE ROUTINE 3.3.4 RAINFALL SERVICE ROUTINE 3.3.5 RTC SERVICE ROUTINE iii (A) 0030‘. ll 21 23 23 25 25 27 28 28 29 29 3T 3T 3T 35 Chapter PEST MODELING SYSTEM 4. 4.1 4.2 4.3 4.4 4.5 INTRODUCTION DATA ACQUISITION 4.2.1 ENVIRONMENTAL DATA 4.2.2 BIOLOGICAL DATA ALGORITHM MODELING THE COOLING MOTH DATA STORAGE POWDERY MILDEN AND TREE PHENOLOGY MODELS 5.1 5.2 'POWDERY MILDEN 5.1.1 DISEASE CYCLE 5.1.2 PREDICTION ALGORITHM 5.1.3 DATA STORAGE TREE PHENOLOGY MODEL 5.2.1 CLUSTER AND TERMINAL DEVELOPMENTS 5.2.2 ALGORITHM 5.2.3 DATA STORAGE CONCLUSION REFERENCES iv 60 6O 6O 61 68 68 69 73 73 76 78 LIST OF TABLES Table 4.1 4.2 5.1 5.2 5.3 5.4 Stage Durations for Codling Moth Relationship Between Three Biofix Points and Egg Hatch in Both Generations of Codling Moth Powdery Mildew Daily Severity Values Powdery Mildew Infection Levels Phenophases of Cluster Development Penophases of Terminal Development 57 63 66 7O 71 LIST OF FIGURES E1993. 2.l A Simple Operational On-Line Pest Management System 2.2 Pest Maturity Distributions 2.3 Compartmentalization of Life Cycle of a With-in Season Insect Pest Species 2.4 Life Cycle of Codling Moth and AsSociated Processes and a Preliminary Phenology Model for Codling Moth 2.5 The Erlang Family of Density Functions 2.6 Decomposition of a kth-Order Distributed Delay 2.7 kth-Order Distributed Delay Process with Losses 3.l Block Diagram of the Various Sections of the Instrument and their Interconnections 3.2 General Software Organization 3.3 Interrupt Service Routine 3.4 Keyboard Interrupt Routine 3.5 RTC Service Routine 4.1 Insect Development as a Function of Temperature 4.2 Approximate Temperature Curve for a Day 4.3 Degree-Day Determination for a Day 4.4 Temperature vs Degree-Day Curve 4.5 Degree-Hour Determination from the Curve Obtained From Ten Minute Averages 4.6 DMCAL Routine vi 10 12 14 16 17 26 3O 32 33 34 38 40 41 43 44 46 Figure 4.7 4.8 4.9 4.10 4.11 5.1 5.2 5.3 5.4 5.5 Flowchart of DHCAL Routine Flowchart of DDCAL Routine A Simplified Flow Diagram of PMS General Flow Diagram of Subroutine DELAY Relationship of Pheromone Trap Catch to Oviposition Data Acquisition Intervals for Powdery Mildew Algorithm Data Evaluation for Determination of Mildew Severity Level for One Day A General Flow Diagram of Powdery Mildew Algorithm Relationship of Cluster and Terminal Phenophases and the Natural Logarithm of Cumulative GDH Flowchart of the Routine TREMDL vii 65 67 72 74 CHAPTER 1 INTRODUCTION Pest control problems have long existed and so have pest manage- ment programs. For the past three decades organosynthetic pesticides have been used widely all over the world. While pesticides have been useful and effective, they are not free of problems. Their intensive and unilateral use poses possible serious side effects on pests, their natural enemies, and man and his environment. In recent years attempts have been made to form a middle ground between chemical and non-chemical control strategies, thus intro- ducing the concept of "integrated control" [l]. Considerable progress has been made along three lines: 1) understanding the life system of pests and crops; 2) use of biological control; 3) mathematical modeling. Mathematical modeling has brought about revolutionary advances in physical sciences. The application of this tool to pest manage- ment has met with great success [4]. The key to this success lies in the fact that mathematical models offer many advantages in the sense that they are more ordered, more flexible and relatively easier to manipulate and simulate. Mathematical models may be simulated over a centralized computing facility and the data obtained may be used to design control strategies. When a centralized computing facility is used for simulation, it is very important that reasonably accurate weather information is available to the system. Weather variations among farms usually demand increased number of monitoring sites within the area. Rapid dissemination of information is also 1 very important in order for the action programs to be effective. A microprocessor-based intelligent field instrument can effectively eliminate these problems at a farm level. Such an instrument may be either dedicated to gather real-time climatic data or it may have a complete management algorithm as a resident program so that in addition to acquiring on-line weather data, the instrument is also capable of prediction on the basis of the resident predictive model. In the former case the acquired data may be communicated to a cen- tralized computer-based Integrated Pest Management (IPM) system which sends back recommendations to the field based on the received in- formation combined with regional data. . This thesis describes a multi-species phenology modeling system that resides as the resident program in a self-contained microprocessor- based field instrument. This stand-alone instrument keeps track of the time and date, acquires real-time environmental data at reg- ular intervals, updates predictive models and issues pest management recommendations to growers. Chapter 2 presents the pest management problem and a comparative discussion on two routes that may be adopted in developing pest models, namely, population abundance and pheno- logical approach. It also presents the mathematical formulation of the biological processes governing pest's lives. The instrument's general hardware and software organization are discussed in Chapter 3. Chapter 4 presents the development of a multi-species Pest Modeling System (PMS). The operation of this system is illustrated through a specific example. A predictive model for powdery mildew infection severity level and an apple tree phenology model are presented in Chapter 5. Chapter 6 presents a summary and conclusions. CHAPTER 2 MODEL DEVELOPMENT Pest management has been defined as ”the reduction of pest prob- lems by actions selected after the life systems of the pests are understood and the ecological as well as economic consequences of these actions have been predicted, as accurately as possible, to be in the best interest of mankind" [l3]. It involves information acquisition, prediction, decision making and taking action to meet the pest problem. 2.l A PEST MANAGEMENT SYSTEM Figure 2.l Shows the basic components of a pest management sys- tem. Agro-ecosystem is the actual system to be controlled. It pro- vides a description of the key insect population, natural enemies and all associated pest species. The biological monitoring component supplies data on densities of each pest species in time. Real-time climatic data like rain, humidity, temperature and, perhaps, some forecasts are provided by the environmental monitoring element. These data serve as input to each pest species life system. In order to be able to recommend a particular course of action, some kind of prediction mechanism is needed. The input to this component consists of information received from above mentioned sources. 'The output is a suggested course of action. Management strategies and subse- quent action programs are based upon recommendations received from the prediction and recommendation algorithm. A feedback loop from the agro-ecosystem regulates the application of pest management strategies. Emwmxm ucoeocmcaz “no; ocw4-:c Pcccwuagomc apne_m < —.N mezowd z<¢ooma zouhu< a mu_ouh<¢hm hzuzucmouu-o¢o< 5:38: 5:56.282. 1-111. uz_¢o-zoz a zo_»u_om¢a szu The effectiveness of management strategies and action programs is directly related to the validity and accuracy of prediction. For prediction, there is a trade off between monitoring the develop- ment of an insect pest and forecasting its expected development based on some kind of prediction algorithm. The greater the predictability the lesser the need to monitor. In the most primitive system the grower uses his intuition and experience to forecast and then weigh the relative merits of potential actions. The outcome of this process is a recommendation as to what should be done. In a more advanced system the grower might use a guide published by experienced pro- fessionals. In a much more sophisticated system forecasting the insect development is based upon computer simulation or mathematical models. One specific approach to model development is described below. Faced with a pest situation, the pest control system designer outlines a model in his mind or creates it by taking a sample plot of the field. He can then simulate control strategies on his model and compare the outcome with the desired results. This process helps him in designing control strategies. Mathematical models, in this context, offer many advantages in the sense that they are more or- dered, more flexible and relatively easier to manipulate and simu- late. No model is ever true in any absolute sense; sOme are better than others at representing events in the real world. This follows from the fact that all models are actually abstractions of the real world. Three of the most important and desirable features of models are a) generality b) realism and c) accuracy [4] . a) Generality. It is the measure of variety of Situations to which model could be applied satisfactorily. b) Realism. It is a function of how closely the structure of the model mimics the processes in the real world. c) Accuracy. It is a measure of degree Of prediction. An ideal model would combine each of these features to an equally high degree. In pest control, one is concerned basically with ac- curacy. The primary test Of a pest control model would be to see if it will lead to a control strategy which will produce, within certain limits, predictable results. As noted earlier, one desired feature of a model is generality so that life systems of various pest species can be Simulated using the same model. This makes it possible for models to be used as research tools. Predictive Extension Timing Estimator (PETE) is one such modeling system [12]. It is a computer based multi-species, phenology modeling system. This modeling system has a generalized format enabling the re- searcher to construct models for various pest species by providing appropriate data. 2.2 DEVELOPMENT OF A GENERALIZED PEST MODEL The following discussion deals with the development of a gen- eralized pest control model employing "system science approach". 2.2.1 OBJECTIVES Two types of insect pest life models may be constructed depend- ing upon the kind and extent of information needed out of the model. The first, "Population Abundance Model", is employed where the object is to keep track of the actual distributions along the various stages in the model. And the second, the ”phenology model”, emphasizes the timing of the model [l2]. The latter approach is employed where prediction of timing for key events is the prime interest. As is to be expected, the population abundance model requires precise in- formation, such as absolute values of age specific mortality rates, reproductive rates, migration rates, etc. This makes the construc- tion of such models fairly complicated and time consuming. Infor- mational needs of phenological models are not as strict, since these models are insensitive to the shape of maturity distributions. Approximate data may be adequate for such models. In many pest management activities phenological models adequately serve the pur- pose. Figure 2.2 illustrates the informational needs involved in the two types of models. Figure 2.2-a compares the shape of dis- tributions involved in a model which predicts the timing of initial events (broken line) with a hypothetical population distribution (solid line). Figure 2.2-b depicts a more involved shape of dis- tribution associated with a model which is to predict timing of peak events as well. More accurate and specific information is required for prediction of relative densities as shown in Fig. 2.2-c. Shape of the maturity distribution shown in Fig. 2.2-d is associated with an abundance model which, as mentioned earlier, requires exact in- formation for prediction of absolute densities and hence, is the most demanding one. So far as the prediction of initial event timing is concerned the phenology model represented in Fig. 2.2-a can func- tion reliably as well as the rest of the three which involve succeed- ingly higher level of complexity. One important fact to be noted mucm>w Pmowuwgu mumowncm mzogg<.m=owuznmgumwo xuwgzpmz “mom N.~ mezmwd 2: 3V ‘ A 22582:. E382 < 5:23: 2:3“; ¢ .IIIII. rzmpse m... w e t t: « « A3 m u . r. I I I I I I I. » ET < hzm>w 50¢ :35 .55.”. from these figures is that despite incorporating crude data and severe distributional errors, phenology models can adequately be used for timing purposes. Broadly, the Objectives are to construct a generalized multi- species phenology model and define a minimum biological data set required to model a particular pest species. 2.2.2 MODEL DEFINITION For extension implementation, model formulation should be gen- eral enough so as to be applicable to a wide class of organisms. Also, models should be structured such that it is possible to refine them when further data become available. Such a structured pheno- logical model, through refinement and incorporation of more accurate data, can consequently become an abundance model. This purpose can be accomplished by compartmentalizing the processes of development, reproduction, etc. into various recognizable developmental units such as eggs, larvae, etc. These developmental units serve as basic components of the model. The choice of these components is arbitrary so that the components having homogeneous rate functions can be grouped together into one. Functionally distinct instars can be considered as separate components [l,4,5,l2]. The generalized block diagram of Fig. 2.3, shows the concept of dividing species life system into "stages". Maturity distribution can be tracked within these stages. "Stage losses" represent pro- cesses like predation, death, emigration etc. In case of multivoltine species successive generations are linked through reproduction. An important consideration to be made while formulating the model is the fact that the maturity distribution curve flattens out with time, 10 mevh saw: mocuwgm> copumpsaom m:_muogo=H Anv mmpumqm “mom uu¢mc~ commmm cw-;u_z a do opozu adv; do copuc~__aacoaugcgsou Aav m.m oezmwd Aav I - >552: _ u _ < _ N I. A Any zo~p<¢mzum zo_»u=oomau¢ Willy. hxmz macs mmoo uu oru do :agocpo xuope A m oezcwd zc_huum zc__umm haaz_ baahao lIJ m¢m=o___azou . dzu .A-——_.¢<¢m>wx A. A 3 dozfizcu muacuz. uu_>¢w. _ _ . 1L WAW AW Id! Id». - _ E ._....,__ A 3?; he; . zc__uum soehzcu 27 tated by the three switches provided on the board. Through proper operation of these switches, the processor can be placed in one of the three modes: reset, run and single step. 3.2.2. INPUT SECTION The keyboard, environmental sensors and signal conditioners in conjunction with the interrupt and service request control from the input section. The environmental sensors include the temperature sensing circuitry, rainfall and leaf wetness sensors. The temperature sensing circuitry includes two precision therm- istors, a monostable multivibrator, a flip flop and four CMOS trans- mission gates.' One of the thermistors is used to measure dry bulb temperature while the other thermistor measures wet bulb temperature. In conjunction with a temperature stable capacitor, one of the precision thermistors forms a linear time-varying constant which is incorporated within a monostable to produce a temperature-controlled pulse width. Depending Upon whether dry-bulb or wet-bulb temperature needs to be measured, under program control the processor selects one therm- istor by enabling appropriate transmission gates. Then the processor triggers the monostable multivibrator and the temperature-controlled pulse is applied to the processor DMAOUT line upon which the processor executes successive DMAOUT instructions. One of the microprocessor's 16-bit registers keeps account of the DMAOUT instructions executed. At the end of the DMA cycle the count in this register is propor- tional to the temperature being measured. With some algebraic manipulation and a look-up table, this count is converted into the actual temperature. 28 The rainfall sensor consists of a tipping bucket rain gauge that produces a 100 ms pulse upon each accumulation of one hundredth of an inch of rain. This pulse calls for processor attention via an interrupt. Upon acknowledging this request, the processor ap- propriately updates the rainfall account. A plastic tube, on which two non-connecting wires spaced 0.1 inch apart are wound, constitute a leaf wetness sensor. This sensor simu- lates a leaf and sits at a representative spot on a tree. It presents the interpreting Circuitry with a resistance which corresponds to the degree of leaf wetness. A flip flop is set when the resistance of the sensor drops below a threshold indicating that the leaves are wet. The processor can then, under program control, inquire this flip flop as to whether the leaves are wet or dry. A weather resistant elastomer-action twelve-key keyboard also forms a part of the input section. This keyboard provides the user- machine interaction link. Depression of any key on this keyboard produces an interrupt service request via a keyboard encoder Chip. Software routines determine the subsequent actions. 3.2.3. OUTPUT SECTION A four-digit alphanumeric liquid crystal display along with the associated driving cirCuitry form the output section. All of the accumulated data and predictions are presented to the user via this low-power, high-contrast LCD. 3.2.4. SYSTEM CLOCK The timing information to the modeling system is provided by an on-board real time clock (RTC). This RTC synchronizes various timed activities by requesting interrupt service at precisely timed intervals. 29 3.3. GENERAL SOFTWARE ORGANIZATION Figure 3.2 shows general software organization which determines the operational characteristics of the instrument. The software is organized to enable the system to perform the following functions: data acquisition, data processing and evaluation, and operator inter- action (keyboard and display control). The start-up and instrument initialization function is respon- sible for performing the hardware/software initialization and placing the system in an operational mode. At this point, the system essen- tially waits for the occurrence of external events (interrupts). Upon this occurrence the interrupt service routine determines as to which of the four interrupt routines must be executed. The clock and rainfall interrupt routines form the data acquisition function. These data are then transmitted to the modeling system where they are processed and evaluated. The keyboard interrupt routine performs the output (display) function. The fourth interrupt routine is provided to facilitate software debugging and trouble shooting. 3.3.1 START UP AND INITIALIZATION The start up function initiates the program execution after the instrument has been placed in the RUN mode through proper opera- tion of the RESET and RUN Switches. Then the instrument is initial- ized through software by setting up the microprocessor registers, clearing all buffers, setting up the date and time, setting up the temperature, humidity and leaf wetness locations. Following the start up, the WAIT function is executed. This function is executed simply by a jump to itself instruction until an interrupt occurs. The program execution returns to this loop also after completion of an interrupt service routine. .30 :o_um~_cnceo memzuwom Pmemcmu N.m mezcwd >ucdozu=a “Hep zc_hu_eu¢a 33.: :3 a 9:582 88.. meow “Sade _ mz_h=c¢ uu_>emm _ E mu_>¢um aa¢wm _ Y _ _Y mu_>eum A a¢ux _ pa=¢¢upz_ _ hzurazhmz_ a a: h¢oo womncfi A.¢ mgzowm 22:22:— ..5 cagmuzzb "O. 9.533%; éhzwtmcam>mo «was: “Nx 9.523%: 45.2995..ng «mg.— “C. ‘III “12.53.33. mx 8. C. .1 \ \ . \ _ \ _ \ \ u>¢=u . eup~5u 43:”? Sun a a \ z \ I \ I 39 4.2.1 ENVIRONMENTAL DATA The daily temperature follows an approximate Sine function with a maximum and a minimum value. This is shown in Fig. 4.2. The system must acquire this daily temperature curve so as to be able to determine the amount of heat accumulated over the day which serves as one input to the modeling system. The amount of accumulated heat is expressed in terms of degree-days above a threshold (kl). Figure 4.3 shows how to determine the number of degree-days from the daily temperature curve. Three developmental thresholds (kl, k2, k3) are shown. Por- tions of the daily temperature curve are integrated to obtain the number of degree-days accumulated that will be used to enhance the model. The shaded areas Show this integration. This integration is explained for the following four cases: Let T = Temperature and DD = Degree-Days. £E$_Ll= T < kl The temperature is below the lower developmental threshold. Hence, DD = 0. CASE 2: k1< T< k2 The temperature is between the lower and upper developmental thresh- olds. Hence, DD = T - k1. 40 Ace m god m>gzu mezpwewasoh mumswxoeqa< N.¢ mezmwd AIIIIIII.>§= u:» go m:_h IIIIII 2.xh dl--- SUfllVUBdHBI x k3 The temperature is beyond the threshold of inhibition. The heat accumulated beyond this threshold causes insect seizure. Hence, DD = 0. It is apparent from the above that the determination of degree-days is closely related to the three developmental thresholds. It is, therefore, important that correct values of these thresholds are known. Figure 4.4 relates the degree-day accumulation to temperature and the three developmental thresholds. PMS obtains the daily temperature curve by sampling temperature at timed intervals. The determination of the degree-days proceeds as follows. PMS directs the hardware to measure temperature once every minute. After an interval of ten minutes, an average of the ten values is calculated and retained. This value corresponds to a ten-minute average value in Fig. 4.5. Then it is determined whether this temperature value contributes to degree-days or not. This is 43 o>ezu Awoiooeooo m> oezumgmaswp ¢.A we:o_d I 5.353%: Q 8. C. «— SAND-333930 -— 44 momoem>< museum: go» see; umcvmuno m>e=u ecu seem :oAumc_Egmuoo ezoziomgaoo m.¢ mesmAD .AIIImeSs.: ezhiill m e . n N _ w¢<¢u>< uh=z_:-zmh «I---3801V83dH31--- 45 accomplished in the routine DMCAL. This routine compares the tem- perature value with the three developmental thresholds and decides in favor of one of the four cases discussed earlier. The routine accumulates the amount of heat if the temperature value does con- tribute to the "useful" heat. It also keeps account of the number of ten-minute intervals that contribute to the "useful" heat. A general flowchart of DMCAL is shown in Fig. 4.6. This routine is called every ten minutes. At the end of every hour (which corre- sponds to six ten-minute intervals) the hourly average is calculated by dividing the heat accumulated over the last six ten-minute inter- vals by the number of intervals that contributed to this heat. 'This average is expressed as degree-hours indicating thereby the heat accumulated over a period of one hour. The flow diagram of the software routine which accomplishes this task is shown in Fig. 4.7. This routine also accumulates the degree-hours and keeps track of the number of hours which contribute to degree-hours. This data is used to compute the degree-days at the end of the day (midnight). This computation is performed by the routine DDCAL which calculates the average of the accumulated degree-hours over the number of hours that contributed to the degree-hours. This routine is called once every twenty four hours (see Fig. 4.8). Referring back to Fig. 4.5, the shaded portion represents the amount of "useful" heat. Portions of the curve are shaded differently to emphasize successive calcu— lation of degree-hours. 4.2.2 BIOLOGICAL DATA A particular pest species may be modeled by providing the PMS with three sets of biological data. In addition, synchronizing 46 GET 10 MINUTE AVERAGE TEMP . DM10 8 O L MIO'TEMP-Kl DM10 8 K2 - K1 L DMSUM-DMSUM + DM10 DM COUNTER CNTM Figure 4.6 DMCAL Routine GET DM COUNTER (ddM) Is onM-vo IFS 2 ND Uh ui+‘”5m‘ Niao CNTH' DH COUNTER CNTH Figure 4.7 Flowchart of DHCAL Routine 48 GET III COUNTER Own” IS ? NO . DH 00 m 00 ‘ 0 a Figure 4.8 Flowchart of DDCAL Routine 49 information from the field is also required that corresponds to the biological fix points. The three-sets of data are listed below: 1) number of life stages of the pest Species being modeled; 2) duration of each developmental stage in units of accumulated heat (degree-days); 3) founder population distribution. Specifically, this provides the system with the initial overwintered distributions along the various stages of the model. BIOlogical FIX points or BIOFIX points are defined as the ini- tial critical events. For a multivoltine pest species, different BIOFIX points are defined as follows. BIOFIX l - first adult trapped with no significant interrUption in catches thereafter; BIOFIX 2 - peak trap catch in the first generation; BIOFIX 3 - peak trap catch in the second generation; Etc. 4.3 ALGORITHM Figure 4.9 represents a simplified flow diagram of PMS. For details on the PMS software refer to the PMS documentation [19]. The PMS is initialized by providing the system with the three sets of biological data discussed in the previous section._ The first two blocks of Fig. 4.9 represent this initialization. After the system has been initialized for a particular pest species and the date/time set up, the system essentially waits for the triggering information from the field which is typically the first BIOFIX point. While waiting for this information, the system regularly updates the date and time. The synchronizing information (BIOFIX) must 50 accompany the cumulative amount of heat expressed in terms of degree- days expected at the BIOFIX. Upon receiving this information, PMS updates the model to the point where BIOFIX occurred. At this point, the model is synchronized with the field. PMS then activates the environmental data acquisition system at regular intervals and com- putes the degree-days accumulated beyond the BIOFIX point. Once every twenty four hours (at midnight) iterative software routines are activated to update the model. The number of degree-days ac- quired over the day determines the number of iterations through the iterative routines. This activity is represented by the inner loop of the flow diagram of Fig. 4.9. Results of this activity are stored in the output buffer. This activity is repeated until the end of the season. A more detailed flow diagram of the main iterative routine is shown in Fig. 4.10. This routine implements the mathematical equa- tions developed for the basic building block in Chapter 2. One iteration through this generalized routine updates one life stage. This routine is successively called as many times as there are number of stages in the pest life cycle to update all the stages [19]. 4.4 MODELING THE CODLING MOTH This section illustrates the operation of PMS for a particular pest species. We Choose the COOLING MOTH as an example. The codling moth, Laspeyresia pomonella, is the most important of three tortricid apple pests which attack the fruit and feed inter- nally. In most areas, the codling moth is multivoltine, i.e., it produces multiple generations within a single season. In Michigan, it goes through two generations, with part of the first generation 51 POPULATION PARAM ER INPUT INITIAL POPULATION SET UP MNTH. DAY COUNTER 1 WAIT FOR SYNCH. INPUT FROM THE FIELD SYNCHRONIZING INPUT FROM THE FIELD (BIOFIX) UPDATE DEVELOP- ————' MENTAL UNIT k ___LLIEE_iIA§EI____ COMPUTE STAGE LOSSES I.E. MORTALITIES,ETC. NO UPDATE MNTH, DAY COUNTER .L COMPUTE REPRODUCTION RATE 1 UPDATE THE OUTPUT BUFFER Figure 4.9 A Simplified Flow Diagram of PMS DELAY VIN, VOUT, R, DEL, DELP, OT, K I IB.,+ DEL-DELP KXDI ‘ ‘ 2xDTxK c“DEEP- F IDT-CxMaxIB.0] DELP I DEL J30 3 R(K)=R( K)+Ax(VIN-BXR( K)) NO LR(I)-R(J)+Ax[R(I+l )-DxR(I)J] YES Figure 4.10 General Flow Diagram of Subroutine DELAY 53 entering diapause until the following spring. The seasonal life cycle of this pest was shown in Chapter 2. The proportion of dia- pausing first generation larvae can vary from year to year depending on the climatic conditions, but it averages normally between 50% and 60%. The growth rate of codling moth is directly related to temper- ature and, to a lesser extent, other climatic factors. The develop- ment begins at approximately 50°F which can be taken as the common lower developmental threshold for the immature stages of moth's life. Each stage has a specific heat requirement to complete development and transform to the next stage. Thermal requirements for full development of the life stages have been established throUgh research [14]. These requirements, expressed in terms of degree-days are shown in Table 4.1. Values listed in the column labeled STAGE DURA- TION represent the minimum number of degree-days spent in each stage. Values in the next column indicate the first appearance of the stage after x degree-days. A full one generation cycle is completed after about 1000 degree-days. The moth growth is slower during the early season and towards the end of summer than the middle of the season. This is expected due to the higher temperatures during the month Of July. The average number of days spent in various stages are shown in the last column of Table 4.1. The Degree-Day values listed in Table 4.1 form one set of input data to PMS. Pheromone traps are used to acquire information on biofix points. At the beginning of the season, the trap is used to obtain an esti- mate of the first adult emergence. This corresponds to the first biofix point. A typical relationship of pheromone trap catch to 54 Table 4.1 Stage Durations For Codling Moth[l4] STAGE DURATION CUMMULATIVE STAGE DURATION DEGREE-DAYS (DD. k,-SO°F) SINCE JAN. 1 (DAYS) Overwi nteri ng 82 Larvae Pupae 256 82 18 Spring Adults 50 342 Eggs 158 392 6 Larvae 532 550 25 Pupae 256 1082 14 Sumner Adults 50 1342 Eggs 158 1392 8 Larvae 1550 55 :5 83.3035 op :33 an: ocosoeogm do 3:28:30: *— NOIlISOd IAO . ...aum . 22 29:3 031... 88 :2. .35.. w ...e aL=e_d >§ @ 11-— HDlVD dVUl ATXBBM —— 56 oviposition is shown in Fig. 4.11. It can be seen from this figure that trap catches increase and peak when approximately 50% of the spring emergence is complete. Trap catches drop off when oviposition begins to increase. Catches drop to a minimum when first generation oviposition reaches a maximum. During the second generation, trap catches increase with oviposition, but, often, oviposition reaches a maximum before maximum of trap catch occurs. The arrows in the figure indicate the three biofix points. Table 4.2 gives the relationship between the three biofix points and egg hatch in both generations. First egg hatch can be expected 243 degree-days after the first catch. The average degree-day totals beginning January 1 for first catch, first egg hatch, peak catch, 50% egg hatch in the first and 50% egg hatch and peak trap catch in the second generation are written diagonally across from the upper left to the lower right corner. The information presented here is sufficient to construct a phenology model for codling moth on PMS. This information can be provided to PMS as the following three sets of data. 1) Model parameters: Order of the delay. k = 15 represents an appropriate density function. 2) Population parameters: a) number of stages in moth's life cycle; N = 5; b) developmental duration of the five life stages expressed as number of degree-days above 50°F. Table 4.1 lists these values; .332. can an...“— soé ~53 5 do 3:33.. o .333 3...: 3.28.. 332. a: 2,322.3 a ...:2. 2.93.. 332, 8 3322.3 .5 57 a 3 N S GB: a: use 8:2 882 a8: .95 3.. m 3 Hum 0 as: 18 dos. 882 4.3.8. 5:: 8w «3. N am: an: ONNN a8. 52: 8“ as 9 .68 ea new 5:3 5.5: m u we. as. as 5:: 8m. 5.: m aae~ =u~

_oe_ae :04 do eewaeeeon Azedvo Appe_e=: a>_ue_ae em_= do eo_oaL=ou Azexvo "ado: asuweom_< sow__z Aeouzoa god upm>eoch cowuwmwzcu< mama _.m oe:m_d _l a: meamfidlivllllv. A V s: a :35: 5.: a J A ”:52“: hszzzx sea N. .e. :— .ea _ .e;.e pazzzz GEN. _ _ N. — . _ Av. 1 as 52;: A E. 52 63 Table 5.1 Powdery Mildew Daily Severity Values (a) (b) I ( ) i HOURS 0; AVERAGE TEMPERATURE HOURS OF c - MILDREW SE9) mags: (D(HRH) TAv (°F) RH 40%<(D(LRH)) VERITY LEVEL (S) D(HRH)>7 TAV<40° or TAV>89° D(LRH) 2 4 D D(HRH)<7 TAV<4O° or TAV>89° D(LRH) 24 O D(HRH)<7 TAV<4O° or TAV>89° D(LRH) < 4 I D(HRH)<7 4O°89° D(LRH) <4 I D(HRH)<7 40°7 4O°STAv<88° D(LRH) < 4 3 (a) D(HRH) denotes hours of RH greater than preceding spore rel ease. (b) TAV denotes average air temperature for 24 hours from 1 p.m. each day. (c) D(LRH) denotes hours of RH lesser than 40% during each day. (d) Powdery Mildew severity values for each day. 95% from 10 p.m. to 6a.m. each night 64 or more, then the mildew severity level is none (S = O). Severity level is highest (S = 3) if the daily average temperature is within the range 400 to 88°F, early morning relative humidity was 95% or higher for seven hours or more and the duration of afternoon low humidity was less than four hours. Mildew severity level determi- nation process is Shown in the flow diagram of Fig. 5.2. This de- cision making process is embedded in the powdery mildew infection prediction algorithm. After daily severity level is determined, the sum of the severity values for the last five days is calculated. This sum is, then, used to determine the actual infection level. Four infection levels are listed in Table 5.2. Infection level is NONE if the sum is less than seven and is SEVERE if it is twelve or greater. I The above severity level determination process was based on the assumption of fair weather conditions. In case of rainfall and resulting leaf wetness the following adjustments are made in the daily severity values. 1) If the leaves are wet for a period of seven hours or more from 9 p.m. to 6 p.m. the preceding night, then the severity value is reduced by l. 2) For a rainfall of 1/2“ or more in any day, the severity value for the following day does not exceed 1. After June 15, adjustments in the severity values are discontinued. A general flow diagram of this algorithm is shown in Fig. 5.3. The data acquisition system for the powdery mildew algorithm is the same as described in the previous chapters. 65 Ana mco do; Am>m4 AHAEw>mm 3oc__z do cowum:AELouoo god :o_um:_m>m open ~.m we:m_d Wm mm; N.m__ .nm# can Nam N .msLGV .uoqugva mm» mm» ~. .9: ma Aummmavc mu» :25 ~. .mz: c v .Nm — r. um cum 3.35. E ~. oamv> (a) I 14 _ M 2 TP - -13.2 + 1.9 1n(GDH) 2 12, E d 5 '53 10 _ K P." A i l 1 1 7 8 9 10 ll 12 (b) _ Loge Accum. GDH -—-> Figure 5.4 a) Relationship of Cluster Phenophase (CP) and the Natural Logarithm of Cumulative GDH With a Base temperature of 42 F [9] b) Reltionship of Terminal Phenophase (TP) and the Natural Logarithm of Cumulative GDH with a Base Temperature of 42° F [9] 73 until the final phenophases (Fig. 5.4-b). Phenophase-GDH relationships on the linear portions of the Fig. 5.4 can be expressed mathematically. CLUSTER PHENOPHASE CP = 72.5 + 9.5 Ln(GDH) TERMINAL PHENOPHASE TP = -l3.2 + 1.9 Ln(GDH) 5.2.2 ALGORITHM The two equations relating the floral and terminal development to heat are implemented in the routine TREMDL. A general flow dia- gram of this routine is shown in Fig. 5.5. This routine is called once every hour. The hourly average air temperature is compared to the base temperature (42°F) to determine if any degree-hours have accumulated. If the average temperature is below 42°F, no degree- hours have accumulated during the last hour, otherwise the number of GDH is equal to the difference of the average temperature and the base temperature. This value is then added to the degree hours accumulated thus far since the start of the season. This cumulative value is then used to determine the present cluster and terminal phenophases. The routine then updates the output buffers. In case of loss of memory the current phenophase can always be calcualted by providing the model with the current accumulated value of GDH. 5.2.3 DATA STORAGE The TREMDL regularly updates a short history buffer every hour and a long history buffer every twenty four hours. The short history buffer contains upto-the-hour information for the current day while the long history buffer contains daily data for the last seven days. Figure 5.5 74 GET HOURLY AVERAGE TEMP. TAV TAV ; 42°F "0 2 YES GDH a O GU'I'TAV-42 t.e ACCUMULATE GDH (CGU-l) l CALCULATE CLUSTER B TERMINAL PHENOPHASES LOOK UP THE CATAGORI ES l . UPDATE OUTPUT BUFFERS RETURN Flowchart of the Routine TREMDL 75 Every time the long history buffer is updated the data for the earliest day is lost. Both of the buffers-are set up to contain the following set of information. 1) DATE: current date; 2) GDHC: accumulated degree hours for Cluster; 3) CATC: cluster phenophase; 4) GDHT: accumulated degree hours for terminal; 5) CATT: terminal phenophase. The user can aceess and display this information via the instrument keyboard [19]. CHAPTER 6 CONCLUSION The purpose of this research effort has been to develop, design, implement and evaluate a microprocessor-based multispecies pest phe- nology modeling system. Two additional algorithms have, also, been included in the system. One algorithm predicts powdery mildew in- fection severity level while the other models the phenological de- velopment of the apple tree. Various physical considerations had to be taken into account while designing this field instrument. It was to be installed at a representative location in an orchard to monitor climatic condi- tions for the purpose of updating various models. It had to be designed to operate thrOUgh the extreme weather conditions of an uncontrolled environment. The fact that it had to be battery powered demanded that the electronic Circuitry be as less power consumptive as possible. The pest phenology modeling system was constrained to have a generalized format so that a particular pest species may be modeled by providing the system with a minimum set of biological information. Powdery mildew disease prediction algorithm was to compile an infection period based on current and past weather states and predict the mildew infection severity level. The third algorithm was to predict pheno- phases (growth stages) of floral and terminal development of apple tree. The three algorithms have been implemented and, currently, are being tested on two prototype instruments in the laboratory. The PMS, powdery mildew and tree model software resides in 8k of ROM. Prediction data is stored in 1k of RAM. Development of ’ 76 77 this software was greatly facilitated by the Software Development Package and a Simulator provided by RCA Solid State Division [17]. This package resides on a Cyber 750, the computing system at MSU. Once installed in the field, this microprocessor-based instru- ment will assist growers in the implementation of pest management strategies through the prediction of critical events in pest pop- ulation dynamics. It will help minimize the use, and optimize the timing, of pesticide sprays leading to reliable and effective control of a variety of economic pests. This microprocessor-based instrument will be tested in the field and evaluated during the 1981 growing season. Results of the field tests will be reported thereafter. Presently, traditional methods are employed in structuring the models. Because a microprocessor-based predictive field instrument has distinct features like real-time environmental data acquisition capability and the ability to process and evaluate the data more frequently at low cost, the future research will focus on exploring alternative modeling methodologies which make more effective use of the advantages offered by a microprocessor-based system. 10. 11. 12. REFERENCES Berryman, A.A., and L.V. Piehaar. "A Powerful Method of In- " vestigating the Dynamics and Management of Insect Populations, Env. Ent., Vol. 3, No. 2, April 1974, pp. 199-206. Croft, B.A. "Tree Fruit Pest Management," in: Introduction to Insect Pest Management. R.L. Metcalf and W.H. Luckmann, Eds., Wiley-Interscience, NY, 1975, pp. 471-507. Croft, B.A., J.L. Howes, and S.M. WelCh. "A Computer-based Extension Pest Management System," Env. Ent., 1975, Vol. 5, pp. 20-34. Conway, G.R., and G. Murdie. "Population Models as a Basis for Pest Control," 12th Symp. British Ecol. Soc., Blackwell Sci. PubR., Oxford, England. Coulman, B.A., S.R. Reice, and R.L. Tummala. "Population Model- ing: A Systems Approach," Science, Washington D.C., Vol. 175, ' pp. 518-521. Fisher, P.D., and S.L. Lillevik. "Monitoring System Optimizes Apple-Tree Spray Cycle," Electronics, Nov. 1977. Fisher, P.D., A.L. Jones, and D.L. Neuder. "Microprocessors Improve Pest Management Practices. Int. Survey of Practice and Experience," Infotech Int. Ltd.,Maiden Head, England, 1979. pp. 89-102. Haynes, D.L., R.K. Brandenberg, and P.D. Fisher. "Environmental Monitoring Network for Pest Management Systems," Env. Ent., Oct. 1973, pp. 889-899. Seem, R.C., and M. Szkolnik. "Phenological Development of Apple Trees," Prediction, 1979, pp. 16-20. Manetsch, T.J. "Time Varying Distributed Delays and Their Use in Aggregative Models of Large Systems," IEEE Tran. on Systems, Man, and Cybernetics,.Vol. SMC-6, No. 8, Aug. 1976. Manetsch, T.J., and G.L. Park. '"System Analysis and Simulation with Application to Economic and Social Systems," IEEE Tran. on Systems, Man, and Cybernetics, Nov. 1977. Welch, S.M., B.A. Croft, J.F. Brunner, M.F. Michels. "PETE: An Extension Phenology Modeling System for Management of a Mgltispecies Pest Complex," Env. Ent., 1978, Vol. 7, pp. 487- 78 13. 14. 15. 16. 17. 18. 19. 79 Ruesink, W.G. "Status of Systems Approach to Pest Management," Ann. Rev. Entomol., 1976, Vol. 21, pp. 27-44. "Management of Codling Moth in Michigan," Research Report, 337 Farm Science, Mich. State Univ. Agr. Exp. Station, Jan. 1978. User manual for the CDP1802 COSMAC Microprocessor. MPM-201, 1976. RCA Solid State Div. Somerville, NJ. COSMAC Microprocessor Product Guide. MPG-180, 1977. RCA Solid State Div. Somerville, NJ. Timesharing manual for the RCA CDP1802 COSMAC Microprocessor. MPM-202, 1976. RCA Solid State Div. Somerville, NJ. Evaluation kit manual for the RCA CDP1802 COSMAC Microprocessor. MPM-203, 1976. RCA Solid State Div. Somerville, NJ. Pest Modeling System Documentation. MICHIGAN STRTE UN IV. LIBRARIES lllllllll llllllllllllHHIIHHllllllllHlllllllllllll 31 9310 709665 2 '5