V3I9P26

Copernicus Sentinel-2 Imagery for Improving the Accuracy of Paddy Rice Yield Estimation 

Sa’ad Ibrahim^1*

Abstract

Accurate spatial data on rice planting areas and yield distribution is crucial for food security and for improving livelihoods. However, obtaining precise spatial information on paddy rice yields remains challenging. This research explores the combination of Copernicus Sentinel-2 satellite imagery with artificial intelligence methods, including both machine learning (ML) and deep learning (DL) models, to improve the accuracy of rice yield estimation. Six algorithms were used to enhance model performance: Random Forest (RF), Support Vector Machines (SVM), Multilayer Perceptron (MLP), Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), and Recurrent Neural Networks (RNN). The study tested four input scenarios based on field rice yield data: (1) using only reflectance bands, (2) using only vegetation indices, (3) combining bands and vegetation indices with Shuttle Radar Topography Mission (SRTM) data, and (4) applying feature selection on all input variables from 10-meter resolution multispectral Sentinel-2 imagery to capture crop health and growth patterns. Findings show that DL models, particularly CNN and MLP, delivered higher accuracy in yield prediction compared to traditional ML regression methods. The best results were achieved with the mutual information regression spectral feature selection method for CNN (RMSE = 982.84 kg/ha, Bias = 12.96) and MLP (RMSE = 984.04 kg/ha, Bias = 27.18). This study offers a strong framework for integrating remote sensing and AI to aid spatial planning, resource management, smart agriculture, and enhance food security and livelihoods.

Keywords:

Sentinel-2; Rice Yield Estimation; Deep Learning; Remote Sensing