Paddy Field Mapping Using Multi-Source Remote Sensing and Random Forest in a Small-Scale Area of Tegallalang, Bali

Abstract

Paddy field mapping is essential to distinguish paddy fields from other vegetation or other cropland. However, mapping in long term faces challenges because of short lifespan of the plants causes land to change rapidly from one phase to another. This study proposes an approach to extract paddy field area using combination of optical vegetation index and synthetic aperture radar (SAR). Sentinel 1 – SAR and Sentinel-2 Surface Reflectance (SR) were used to identify paddy field with high spatial resolution. Combination of Enhanced Vegetation Index (EVI) and VV (vertical-vertical) and VH (vertical-horizontal) polarization were used to tackle cloud coverage. This approach applying Random Forest algorithm using Google Earth Engine platform in Tegallalang Sub-District as small scale pilot project. Overall accuracy (OA) and Kappa Coefficient obtained was 96% and 0.94, respectively. This accuracy performance was relatively higher compared to minimum cartographic standard 85%. This result showed that the generated land use map has high level validity. This study expected to be able to generally applied in the geographically small and complex terraced areas of the Subak system.