Within-field soybean yield prediction: integrating crop growth modelling, remote sensing and machine learning

Summary

This thesis presents methods for predicting and explaining within-field variability in soybean yield using process-based crop growth models, machine learning methods, and openly available remote sensing imagery. This thesis assessed the soybean cropping system from different regions of the main agricultural areas. The methods used capitalize on the broad availability of satellite data to compensate for the lack of input data required to operate crop growth models. This thesis further investigates the advantage of leveraging the strengths of the physical models and machine learning algorithms by combining both methods. The methods evaluated in this thesis contribute to the development of decision-making tools towards sustainable crop management practices. It provides insights into cropping systems (spatial variability of biophysical variables such as biomass, crop evapotranspiration and yield) that may provide the foundation for developing objective metrics to quantify the environmental impact of agriculture systems.