Remote Sensing-based Paddy Field Classification using Machine Learning Algorithms on Sentinel-1 and Sentinel-2 Satellite Data

Farmers can report their paddy fields affected by disasters for insurance claims using mobile applications that allow submitting photos and polygons drawn on satellite base maps of their agricultural land. However, farmers tend to falsely classify their non-agricultural land as paddy fields, which poses a challenge in accurately identifying disaster-affected areas. Therefore, developing a reliable method for accurately identifying agricultural fields is essential to improve the identification of damaged regions. This study used remote sensing data from Sentinel-1 and Sentinel-2 satellites. We applied Random Forest (RF) and Classification and Regression Trees (CART) machine learning (ML) algorithms on satellite data to classify paddy and non-paddy fields. In addition, we incorporated the Normalized Difference Vegetation Index (NDVI) as a graphical indicator preceding the implementation of ML algorithms to improve the classification accuracy in Sentinel-2 images. Our results showed that the RF algorithm applied to the Sentinel-2 image achieved the highest classification accuracy, with a model accuracy of 97.19% and overall accuracy of 93.49%. The CART algorithm on Sentinel-2 performed with marginally lower accuracy than RF. However, both RF and CART algorithms exhibited inferior performance on Sentinel-1 imagery, with an accuracy of around 85%. The findings suggest the potential of using ML algorithms on remote sensing data for accurate and efficient paddy field classification.