基于Sentinel-1/2动态耦合移栽期特征的水稻种植模式识别

及时准确地获取水稻种植模式变化对于有效防控“非粮化”和完成“双碳”目标具有重要意义。现有研究多基于固定时间窗口挖掘水稻生长期特征，且多使用的是单一卫星影像数据，难以应用于大范围水稻制图。本文通过Sentinel-1/2数据构建动态窗口提取移栽期光学/雷达特征，利用其耦合关系实现水稻种植模式制图。将该算法应用于湖南和江西两省水稻制图。基于1402个地面参考点位对水稻提取结果进行验证，总体精度达92.80%；在县域尺度上，湖南和江西两省水稻制图面积与农业统计数据也具有高度一致性，R²达0.85以上。相比于用固定窗口进行水稻特征提取，该方法具有较强的鲁棒性和迁移能力，为实现更大范围作物制图提取提供新的思路和参考依据。2018—2021年江西省水稻制图结果表明，水稻总种植面积减少9.47%，约3460 km²，水稻种植强度从1.62下降至1.49；在种植模式上，“双改中”趋势明显，双季稻种植面积锐减21.61%，其中约84%改种中稻。

Mapping Paddy Rice Planting Patterns based on Sentinel-1/2

Accurate and timely spatiotemporal change information of rice planting patterns is significant for effectively preventing "no-grain" and achieving carbon peak and carbon neutrality goals. However, it is challenging to establish a rice mapping method at large spatial domain. This study developed a novel paddy rice planting mapping method based on dynamic coupling optical/radar features of transplanting period. The proposed algorithm was applied to paddy rice mapping in Jiangxi and Hunan province. The derived paddy rice planting map was evaluated using 1402 ground reference sites, and it had an overall accuracy of 92.80%. The paddy rice planting area was also highly consistent with the agricultural census data (R² > 0.85) at the county level. Compared with rice feature extraction using a fixed window, the proposed method has strong robustness and migration ability, and provides a new idea and reference for crop mapping at large spatial domain. The result showed that the paddy rice planting area in Jiangxi province decreased by 3460 km² (9.47%) from 2018 to 2021. The rice cropping intensity in Jiangxi province had decreased by 0.13 due to the change of double-cropping rice to medium rice. The double-cropping rice planting area decreased by 21.61%, with 84% shifted to single cropping rice.