Anais do XXI Simpósio Brasileiro de Sensoriamento Remoto

This study explores the use of deep learning models to predict soybean yield in the eastern Pampas of South America, utilizing open-source data from remote sensing, weather, and topography. Soybean fields across Uruguay, Argentina, Brazil, and Paraguay were analyzed over four growing seasons (2021-2024), totaling 1,519 fields. We aimed to develop a predictive model that operates independently of any ground data from the paddock being predicted, ensuring flexibility and scalability across various agricultural regions. Two prediction scenarios were evaluated: yield at harvest time and one month prior. Four deep learning architectures—Recurrent Neural Networks (RNN), Convolutional Neural Networks (CNN), Multilayer Perceptrons (MLP), and Transformers (TRN)—were trained and tested. The CNN achieved the best performance at harvest with an RMSE of 486 kg/ha, while the RNN excelled in early prediction with an RMSE of 496 kg/ha. These results highlight the potential of deep learning for field-level yield prediction and agricultural decision-making.