Rice fields classification through spectral-temporal data fusion during the rainy and dry seasons using Sentinel-2 optical images in Subang Regency, West Java, Indonesia

The most accurate method for rice fields mapping involves a phenological approach using optical remote sensing and a multisource data integration approach. However, these approaches do not consider the two rice growing periods in tropical regions, which are the rainy and dry seasons. During the rainy season, the optical remote sensing data are affected by clouds and haze. On the other hand, during the dry season, rainfed rice fields are not planted with rice. Therefore, this study proposed a new scheme for rice fields classification in the tropical regions using data fusion between different seasonal periods. Three data fusion scenarios based on reflectance fusion, temporal feature fusion, and information fusion from remote sensing data during the rainy and dry seasons were analyzed. The results showed that the accuracy of rice fields classification increased by using the proposed scheme, rather than a single period. The best fusion scenario was the information fusion strategy with the highest increase in precision accuracy, from 92.72% in reflectance fusion and 93.17% in temporal feature fusion to 94.99%. This strategy distinguished the rice fields from the fish pond and other seasonal crops such as sugar plantations.