A hybrid graphtemporal Deep learning framework for crop yield prediction using multisource satellite and Agronomy data

Crop yield prediction plays a pivotal role in modern agriculture by aiding farmers and stakeholders in making informed decisions about crop management and resource allocation. This study presents a machine learning-based approach to predict crop yield using multi-source synthetic agronomic and satellite data. The dataset comprises key variables such as field characteristics, land surface temperature (LST), soil moisture, rainfall, temperature, nitrogen levels, and vegetation indices (NDVI, EVI). Two machine learning algorithms, Linear Regression and Random Forest Regression, were employed to model the relationship between these factors and the predicted crop yield. The performance of both models was evaluated using metrics such as R² (Coefficient of Determination), MAE (Mean Absolute Error), and RMSE (Root Mean Square Error). The results show that Random Forest Regression outperforms Linear Regression, achieving a test R² of 96.35% and significantly lower error metrics. The Linear Regression model, while less accurate, demonstrated moderate predictive power with a test R² of 69.26%. The findings emphasize the potential of machine learning models in crop yield forecasting and their applications in precision agriculture, thereby contributing to more efficient farming practices and better resource management.

Keywords: Crop Yield Prediction, Machine Learning, Random Forest Regression, Linear Regression, Agronomic Data, Satellite Data, NDVI, EVI, R², MAE, RMSE, Precision Agriculture.