Towards crop yield estimation at a finer spatial resolution using machine learning methods over agricultural regions

Abstract

Reliable yield estimation is crucial for food security and agricultural production especially in the intensively agricultural region. This study constructed a gridded yield estimation framework by driving machine learning models with remote sensing vegetation index and meteorological forcing. Among eight machine learning methods, support vector machine (SVM), k-nearest neighbor regression (KNN), and Gaussian process regression (GPR) models outperformed the others. Precipitation, temperature, and the fraction of photosynthetically active radiation are key factors for yield estimation. The yield estimation at county level and regional level were further conducted to explore the scale effect (estimation accuracy varies with spatial resolution). Different scales hold diverse spatial variability information. Finer scales that are more representative of spatial variability generally result in the better accuracy. This study demonstrates that a more accurate yield estimation can be achieved at a finer grid level, thus providing guidelines for agricultural planting structure.