Spatiotemporal Assessment of Rice Growth and Yield Using Remote Sensing Derived Indices and GIS in Batticaloa District, Sri Lanka
Janushika S* and Dayawansa NDK
ABSTRACT
Rice yield is significantly influenced by climatic variability, water availability, and agronomic practices. Effective monitoring of paddy growth and yield assessment is essential to ensure food security. This study aimed to monitor rice crop growth and predict yield in the Batticaloa District, Sri Lanka, using remote sensing and Geographic Information Systems (GIS). The analysis focused on the Yala season due to frequent cloud cover during the Maha season. Sentinel-2 Level-2A imagery from the 2023 and 2024 Yala seasons was utilized to derive the Normalized Difference Vegetation Index (NDVI), Normalized Difference Water Index (NDWI), and Red-edge index for assessing crop health and stress variations. Paddy cultivation areas were delineated through supervised classification of satellite images, supported by ground truth data collected via field surveys and farmer interviews. The relationships between vegetation indices and yield were examined using regression models. Multi-temporal analysis of NDVI, NDWI, and Red-edge indices revealed a distinct pattern, with peak values occurring approximately eight weeks after planting. The NDVI-based yield prediction model achieved an R² of 0.70, while the Red-edge model yielded an R² of 0.69, demonstrating strong correlations between vegetation indices and yield. Predicted yields for the 2024 Yala season were approximately 5046 kg/ha (NDVI) and 5005 kg/ha (Red-edge), compared to the observed yield of 4497 kg/ ha. The Root Mean Square Error (RMSE) values for the NDVI and Red-edge models were 12.21% and 11.29%, respectively. These results highlight the effectiveness of remote sensing and GIS in monitoring rice growth and estimating yield, and underscore the potential of integrating such approaches with advanced technologies to promote precision agriculture in Sri Lanka. Future studies should aim to improve prediction accuracy using higher-resolution imagery, enhanced ground truth datasets, and machine learning techniques
