A 30-m annual paddy rice dataset in Northeastern China during period 2000-2023.

Here, we used (1) hierarchical screening principle to select ground samples, (2) the linear interpolation and Whittaker smoothing Landsat5/7/8 time series data and (3) enhanced spectral-feature sets to map annual 30-meter rice distribution in Northeastern China since the 21st century by using random forest method via Google Earth Engine (GEE) platform. In summary, the resultant maps have high overall accuracy (OA) ranging from 0.93 to 0.99, and the satellite estimates corresponded well with statistics for most cities (R2≥0.7, p<0.01). In addition, our results had higher accuracy than that of similar crops mapping datasets. This is the first effort in China to reconstruct paddy rice pattern at the 30-m resolution along such a detailed and long time-series.

本研究采用（1）分层筛选原则选取地面样本，（2）对陆地卫星5/7/8号（Landsat5/7/8）时间序列数据开展线性插值与惠特克平滑处理，（3）采用增强光谱特征集，依托谷歌地球引擎（Google Earth Engine, GEE）平台，通过随机森林算法绘制了21世纪以来中国东北地区的年度30米分辨率水稻分布。综上，所得制图成果的总体精度（Overall Accuracy, OA）介于0.93至0.99之间，多数城市的卫星反演结果与统计数据吻合度较高（决定系数R²≥0.7，p<0.01）。此外，本研究所得精度优于同类作物制图数据集。这是国内首次以30米分辨率，基于如此长时序、高细节的时间序列重建水田水稻分布格局。