An automatic rice mapping method based on an integrated time-series gradient boosting tree using GF-6 and sentinel-2 images

Timely and accurate mapping of paddy rice cultivation based on remote sensing technology is crucial and valuable for ensuring food security and sustainable environmental management. In most relevant studies, rice mapping was conducted using time-series images, but conventional rice mapping methods are not specifically designed for time-series data, making it difficult to extract the deep information contained in these data. To address these problems, in this paper, an automatic rice mapping method based on an integrated time-series gradient boosting tree (Auto-ITSGBT) is proposed using GF-6 WFV and Sentinel-2 MSI data. This method accounts for the local and overall shape features of time-series curves, and fully exploits the information related to phenological characteristics between time-series data. The proposed rice mapping method is tested and validated in three typical rice-producing areas, which are located in different provinces of China characterized by diverse climate conditions, planting times or topographies. The results show that the overall accuracy and Kappa coefficient of the method exceeded 95% and 0.93, respectively, at all study sites, respectively. Our method performs better than the existing competing methods, with an overall accuracy improvement of 2% to 4%. To identify the rice planting areas as early as possible, rice mapping was conducted by reducing the number of images one by one. The rice distribution map was obtained in mid-July with an overall accuracy of at least 90%, thus obtaining a spatial distribution map of rice with high accuracy before harvesting.