Multi-Input Source Stress Analysis for Field Crops to Improve Yield Predictions

Farming requires placing bets on future weather risking upfront investments to get the best yield possible. In recent years, the weather has become more unpredictable and more extreme, which means that farmers are having to work harder and accept a higher degree of risk. Having tools available to quickly identify crops in stress is important, and remote sensing has proven to be useful for identifying crops in stress. However, remote sensing alone has difficulty in distinguishing between types of stress that result in similar physiological changes in the plant. Machine learning can integrate additional data sources, such as local weather and soil data, which can help distinguish types of stress especially early on when such information might be useful to farmers. Satellite platforms cannot, however, collect images often enough and with sufficient spatial and spectral resolution to build the models. Imagery from unmanned aircraft systems (UASs) can fill in these gaps and can be better coordinated with the required ground reference data collection. The goal of this research is to improve the accuracy of crop stress type identification and yield impact prediction from remote sensing data products by integrating supplemental environmental information into a single, machine learning-based model. This presentation will introduce the pipeline developed to automate frequency and resolution adjustments between data sources and complete data quality checking and cleaning processes. It will then introduce the Multi-Input Source Stress Analyzer (MISSA) and provide preliminary evaluations of MISSA as it is used to predict final yield and in-season biomass. This is an initial step in using MISSA to predict water stress during the growing season.