Information on rice planting areas is critically important for food and water security, as well as for adapting to climate change. Mapping rice globally remains challenging due to the diverse climatic conditions and various rice cropping systems worldwide. Synthetic Aperture Radar (SAR) data, which is immune to climatic conditions, plays a vital role in rice mapping in cloudy, rainy, low-latitude regions but it suffers from commission errors in high-latitude regions. Conversely, optical data performs well in high-latitude regions due to its high observation frequency and less cloud contamination but faces significant omission errors in low-latitude regions. An effective integrated method that combines both data types is key to global rice mapping. Here, we propose a novel adaptive rice mapping framework named Rice-Sentinel that combines Sentinel-1 and Sentinel-2 data. First, we extracted key phenological phases of rice (e.g., the flooding and transplanting phase and the rapid growth phase), by analyzing the characteristic V-shaped changes in the Sentinel-1 VH curve. Second, we identified potential flooding signals in rice pixels by integrating the VH time series from Sentinel-1 with the Land Surface Water Index (LSWI) and Enhanced Vegetation Index (EVI) from Sentinel-2, utilizing the generated phenology phases. Third, the rapid growth signals of rice following its flooding phase were identified using Sentinel-2 data. Finally, rice fields were identified by integrating flooding and rapid growth signals. The resultant rice maps in six different case regions of the world (Northeast and South China, California, USA, Mekong Delta of Vietnam, Sakata City in Japan, and Mali in Africa) showed overall accuracies over 90 % and F1 scores over 0.91, outperforming the existing methods and products. This algorithm combines the strengths of both optical and SAR time series data and leverages biophysical principles to generate robust rice maps without relying on any prior ground truth samples. It is well-positioned for global applications and is expected to contribute to global rice monitoring efforts.

中文翻译：

---

通过整合光学和雷达数据改进基于物候的水稻映射算法


有关水稻种植面积的信息对于粮食和水安全以及适应气候变化至关重要。由于全球气候条件多样，水稻种植系统多样，全球水稻地图仍然具有挑战性。合成孔径雷达 （SAR） 数据不受气候条件的影响，在多云、多雨、低纬度地区的水稻测绘中起着至关重要的作用，但在高纬度地区则存在佣金误差。相反，光学数据由于观测频率高、云污染少，在高纬度地区表现良好，但在低纬度地区面临严重的遗漏误差。将这两种数据类型相结合的有效综合方法是全球稻米制图的关键。在这里，我们提出了一个名为 Rice-Sentinel 的新型自适应水稻绘图框架，它结合了 Sentinel-1 和 Sentinel-2 数据。首先，我们通过分析 Sentinel-1 VH 曲线中特征性的 V 形变化，提取了水稻的关键物候阶段（例如，洪水和移栽阶段以及快速生长阶段）。其次，我们通过将 Sentinel-1 的 VH 时间序列与 Sentinel-2 的地表水指数 （LSWI） 和增强型植被指数 （EVI） 相结合，利用生成的物候阶段，确定了水稻像素中的潜在洪水信号。第三，使用 Sentinel-2 数据确定了水稻在洪水阶段后的快速生长信号。最后，通过整合洪水和快速增长信号来识别稻田。在世界六个不同案例地区（东北部和华南地区、美国加利福尼亚州、越南湄公河三角洲、日本坂田市和非洲马里）生成的水稻地图显示总体准确率超过 90%，F1 分数超过 0。91，优于现有方法和产品。该算法结合了光学和 SAR 时间序列数据的优势，并利用生物物理原理生成强大的水稻地图，而无需依赖任何先前的地面实况样本。它为全球应用做好了准备，有望为全球水稻监测工作做出贡献。