Integrating the Sentinel-1, Sentinel-2 and topographic data into soybean yield modelling using machine learning

Abstract

It is crucial to accurately and timely estimate crop yield within field variability for sustainable management and precision farming applications. Various Earth observation systems have been developed for crop monitoring and yield prediction. However, there is a need for further research that integrates multiplatform data, advances in satellite technology, and data processing to apply this knowledge to agricultural practices. The integration of satellite imagery and environmental data has been used increasingly in recent years to predict crop yields using machine learning techniques. In recent years, VIs derived from optical satellites, particularly Sentinel 2 (S2), have gained popularity, but their availability is affected by weather conditions. On the other hand, the backscatter data from Sentinel 1 (S1) is less commonly used in agriculture due to its complex interpretation and processing, but it is not influenced by the weather. This study aims to improve the accuracy of yield predictions by combining remote sensing data with environmental variables. The use of satellite data S1 and S2 was used to identify the optimal phenological period, and a training model was developed using four machine learning techniques, including Random Forest Regression (RF), K Nearest Neighbor (KNN), Multiple Linear Regression (MLR) and Decision Tree (DT). The results showed that RF provided the highest values among the four techniques. The validation process using RF demonstrated high accuracy rates, with R2 ranging from 0.41 to 0.89, the mean square error of the root (RMSE) ranging from 0.122 to 0.224 t/ha, and the mean absolute error (MAE) ranging from 0.089 to 0.163 t/ha. The integration of satellite data S1 and S2 with topographical information may be useful for monitoring, mapping, and forecasting crop yields on small and fragmented farmlands. This approach can provide farmers, agricultural businesses, and policymakers with accurate and timely predictions of crop yield, which can facilitate decision making and provide early warnings for potential crop losses.