The frequencies of extreme climate events are projected to be continuously increasing, which poses great challenges to crop production. Excessive precipitation has been the second largest threat to the U.S. crop production, next to droughts. Timely crop information at the field level is critical for building climate-resilient cropping systems, yet such timely monitoring framework for individual farm fields is lacking. The goal of my dissertation research project is to develop an innovative deep learning-based modeling framework to assess the excessive precipitation’s effects on field-level crop yield in a near real-time manner.
The project designs a framework from dataset generation, crop mapping, yield estimation, to yield gap and variation analysis. First, we have developed a novel hybrid deep learning-based data fusion model to generate near real-time high-resolution imagery. Second, we have devised a phenology-guided in-season crop mapping model for timely crop identification. To achieve near real-time yield estimation, a transfer learning-based crop yield estimation model for crop yield forecasting will be developed. Field-level yield loss, yield gap, and yield variability under weather extremes will then be assessed. The proposed framework will advance the ability to utilize remote sensing data to forecast crop yield of farm fields in a near real-time fashion, contributing to more sustainable agricultural management and food security.