Rice cropping sequence mapping in the tropical monsoon zone via agronomic knowledge graphs integrating phenology and remote sensing

Abstract

Rice cropping sequence mapping via multitemporal remote sensing and agronomic techniques provides critical geoinformatics for agroecosystem modeling. The East Asian tropical monsoon region is an important rice-growing area, and the regional food security depends on efficient rice mapping. Frequent cloud cover during the rainy season leads to insufficient available optical remote sensing images that cover the growth stages of rice, which renders remote sensing-derived rice identification difficult. Thus, we proposed a simple and efficient strategy from an agronomic knowledge graph perspective in which specific phenological events of rice and Landsat/Sentinel-2 time series were employed to extract different rice cropping sequences on the Leizhou Peninsula (China). Then, a control group and five experimental groups were established by integrating spectral features via pixel- and object-based random forest (RF) algorithms. The results revealed that five key phenological events could be obtained for rice cropping sequence identification in the study area. The overall accuracy of the pixel-based classification results ranged from 83.48 % to 92.49 %, whereas that of the object-based classification results ranged from 84.98 % to 92.80 %. These findings indicate that efficient rice cultivation mapping via optical remote sensing data requires the selection of specific time windows corresponding to phenological events to benefit rice cultivation monitoring and regional agroecosystem sustainability.