Modelling Crop Yield Prediction with Random Forest and Remote Sensing Data

Ensemble learning methods combined with remote sensing data can optimize yield forecasting and provide real-time insights for decision-making. In predictive agriculture, having predictive accuracy over crop yield is essential for managing food security and adapting to climate change. This study aims to integrate satellite remote sensing data into agro-climatic region farms for yield prediction using machine learning with the Random Forest algorithm. The implementation approach utilizes MODIS and Sentinel 2 satellites, which provide multispectral imagery and NDVI/EVI estimates in conjunction with Precipitation data, Land Surface Temperature, and altimetry data. Supervised learning occurred in the training phase, requiring historical crop yield datasets sequentially divided into train/test datasets. During the validation phase, accuracy was according to relevance metrics established by R in conjunction with RMSE and MAE. A performance evaluation was conducted on the other baseline models, SVR and linear regression, and improved accuracy performance was showcased when utilizing random forest. The results have demonstrated the significance of applying ensemble learning techniques augmented with remote sensing data towards operational crop yield forecasting. This work strengthens the remote sensing technology for precision agriculture by developing an Earth observation-based yield estimating methodology that is observable, scalable, and straightforward.