基于星地融合数据与随机森林算法的青海省降水量时空分布特征分析

Spatial and Temporal Distribution of Precipitation in Qinghai Province Based on Star-ground Fusion Data and Random Forest Algorithm

针对低密度气象站点分布区降水量空间信息提取精度不高及卫星遥感降水产品空间分辨率粗糙问题,以2020年青海省逐月GPM产品为例,在地理信息技术支持下生成由地形(DEM、坡度、坡向、曲率、起伏度)、地表覆被(NDVI、NPP)、海陆位置(经度、纬度)和水汽等10个因子构成的环境协变量集,运用随机森林(Randomforest,RF)算法对GPM数据进行精细化融合,并基于56个气象站点资料对融合精度进行独立验证。结果表明,该融合模型能克服GPM产品分辨率粗糙问题,实现区域降水量时空尺度精细化;协变量对模型精度有重要影响,在地形起伏度小、大气水含量低的地区具有更高融合精度;研究区降水量时空分布不均衡,降水中心在1~8月呈逆时针移动,在9~12月呈顺时针移动。

To address the problem of low precision of spatial information extraction and rough spatial resolution of precipitation products from satellite remote sensing in the distribution area of low-density meteorological stations,taking the monthly GPM products in Qinghai Province in 2020 as an example,this study generated an environmental covariate set consisting of 10 factors such as topography(DEM,slope,slope direction,curvature,undulation),surface cover(NDVI,NPP),land and sea position(longitude,latitude)and water vapor with the support of geographic information technology.The random forest(RF)algorithm was used to refine and fuse the monthly GPM products in Qinghai Province in 2020,and the fusion model was independently validated based on the data from 56 meteorological stations.The results show that the fusion model can overcome the problem of coarse resolution of GPM products and realize the refinement of regional precipitation at spatial and temporal scales;The covariates have an important influence on the model accuracy,and the RF method has higher accuracy in areas with low topographic relief and atmospheric water content;The spatial and temporal distribution of precipitation in the study area is uneven,with the precipitation center moving counterclockwise from January to August and clockwise from September to December.