Potential of Pre-harvest WorldView-3 Satellite data in Predicting Soybean Yield and Compositions Using Machine Learning Algorithms
Abstract
Preharvest soybean yield and seed composition prediction can provide prescriptive information for agriculture management practices, grain policy-making and food security. Increased advancement of remote sensing technology in precision agriculture enabled acquiring imagery with high spatial, spectral and temporal resolution at a relatively low cost. Artificial intelligence applications accelerated the data extracting, processing and interpreting steps from those images. This study investigates the potential of WorldView-3 imagery under different machine learning frameworks in predicting yield and seed compositions from the soybean's canopy signals. The WorldView-3 images were collected at the reproductive stage five (R5) during the 2017 and 2021 growing seasons over a soybean test site at Bradford Research Center near Columbia, Missouri. The combination of canopy spectral and texture features extracted from processed WorldView-3 Images via Python programming is then used as inputs for the prediction. Multi_task machine learning (MTL), which predicts the yield and compositions simultaneously, was compared with Automated Machine Learning (AotoML), where targets were predicted separately to examine the potential of WorldView-3-based imagery data in estimating soybean seed composition and yield.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2022
- Bibcode:
- 2022AGUFM.B45I1835D