The combination of precision agriculture technology and big data allows farmers to improve the sustainability in crop production by improving profit and minimizing environmental loss. However, knowledge about managing sub-field yield variations over the years (yield stability) is less understood, so the actual adoption rate of precision technology is low, especially in small farms. The overarching goal of this dissertation was to evaluate the spatial and temporal variability of maize and soybean yields related to landscape characteristics and management in the U.S. Midwest (Chapter 1).Chapter 2 aims to develop a yield stability map using satellite images for ten maize/soybean main production states in the Corn Belt. This study uses Landsat images over eight years and common land unit polygons as the primary dataset to classify each field into stable high, stable low, and unstable areas. I further quantified nitrogen (N) balance and loss and associated monetary and environmental loss for each yield stability class. Across a wide range of crop growth environments, this large-scale analysis has shown reliable and consistent subfield yield patterns that could improve fertilizer application.Chapter 3 investigates the potential of these low-yielding areas in supporting biofuel crop (switchgrass) production to replace current crops and subsequent improvement in soil and climate change. I also proposed planting a cover crop, rye, to evaluate the effect on the soil nitrate level. The switchgrass and cover crop production were simulated by the Systems Approach to Land Use Sustainability (SALUS) model, which also modeled soil organic matter dynamics under the proposed and current (continuous maize or maize/soybean rotation) scenarios. Benefits such as reduced soil nitrogen, improved soil structure, and higher heterogeneity of landscape composition demonstrate the potential of yield stability maps to support sustainable agricultural production. In Chapter 4, I combined the yield stability maps with remote sensing images to improve the maize yield prediction at subfield scale. In this study, I first made a further division in the unstable area by including the topographic features and modeled cumulative crop water stress. The modeled water stress was combined with Landsat Analysis Ready Dataset (ARD) to predict maize yield using a random forest algorithm. Results showed that incorporating the modeled water stress with vegetation index derived from Landsat images produced the highest accuracy in estimated maize yield compared to satellite images alone. This work emphasized the yield stability map's ability to capture subfield spatial variations of soil and plant available water stress information.Chapter 5 compares the accuracy in yield level estimation between yield stability maps and high-resolution Planet images (3-m, bi-weekly) and analyzes the temporal changes of spatial variation of crop conditions during the growing season. The yield stability maps could predict similar maize yield levels in stable zones compared to satellite images. The temporal analysis in the spatial variation indicated greater changes in the field at the beginning or end of the season, while spatial patterns of crop growth dynamics remain stable in the mid-season. Early- or late-season changes did not always impact the final yield. This analysis provided an in-depth analysis of the in-season image so farmers could better monitor crops in real-time. Conclusions from these research projects and recommendations on interpretation yield and satellite images for sustainable production are presented in Chapter 6.
Read
