<b>A </b><b>spatiotemporal geostatistical downscaling-integration framework</b><b> </b><b>of </b><b>rain gauge and satellite-based precipitation datasets in the Pearl River basin, China</b>

High-resolution and -precision precipitation datasets are important for hydrological modeling, drought and flood monitoring, and water resource management. Remote sensing provides spatially continuous precipitation datasets for wide areas; however, their use in hydrology studies of local regions and watersheds has been limited because of the associated coarse resolution and low precision. Based on the geostatistical theory, a spatiotemporal downscaling-integration framework of rain gauge and satellite-based precipitation datasets has been proposed in this study. First, the area-to-point kriging (ATPK) interpolation with the scale effect considered is used to downscale the Tropical Rainfall Measuring Mission (TRMM) product. The comparison experiment between the downscaled TRMM precipitation estimates and the rain gauge observations shows that the R² is 0.854, and the mean error (ME) and root mean square error (RMSE) are 5.44 mm and 50.11 mm, respectively. The accuracy of the downscaled TRMM precipitation estimates is slightly higher than the original TRMM data. The downscaling process causes no significant estimation error, while preserving the spatial pattern of the original TRMM precipitation data. Then, a spatiotemporal regression kriging (STRK) model is constructed for estimation and generation of monthly precipitation datasets at a spatial resolution of 1 km for the Pearl River Basin from 2001–2013, by integrating rain gauge data, downscaled TRMM precipitation results and multi-source raster datasets of auxiliary variables such as DEM and NDVI. The cross-validation results show that R² reaches 0.872, and the ME and RMSE are -4.13 mm and 46.64 mm, respectively. The STRK model is found to enable exploitation of the large-scale coverage of remote sensing products and the high-precision characteristics of rain gauge observations. These findings can facilitate the characterization of the spatiotemporal pattern of precipitation and improve small-scale hydrological modeling.高分辨率高精度降水数据集对水文模拟、旱涝监测与水资源管理具有重要意义。遥感技术可为大范围区域提供空间连续的降水数据集，但受限于其较粗的空间分辨率与较低的精度，该类数据在局地流域水文研究中的应用受到限制。本研究基于地统计理论，提出了一种融合雨量站观测与卫星降水数据集的时空降尺度整合框架。首先，采用考虑尺度效应的面到点克里金（area-to-point kriging, ATPK）插值方法对热带降雨测量任务（Tropical Rainfall Measuring Mission, TRMM）产品进行降尺度处理。将降尺度后的TRMM降水估算结果与雨量站观测数据开展对比实验，结果显示决定系数（R²）为0.854，平均误差（mean error, ME）与均方根误差（root mean square error, RMSE）分别为5.44 mm与50.11 mm。降尺度后的TRMM降水估算精度略高于原始TRMM数据，且降尺度过程未引入显著估算误差，同时完整保留了原始TRMM降水数据的空间分布格局。随后，本研究整合雨量站数据、降尺度后的TRMM降水结果以及数字高程模型（digital elevation model, DEM）、归一化植被指数（normalized difference vegetation index, NDVI）等多源辅助变量栅格数据集，构建了时空回归克里金（spatiotemporal regression kriging, STRK）模型，用于生成2001–2013年珠江流域1 km空间分辨率的逐月降水数据集。交叉验证结果显示，决定系数（R²）可达0.872，平均误差（ME）与均方根误差（RMSE）分别为-4.13 mm与46.64 mm。研究表明，时空回归克里金模型可充分利用遥感产品的大范围覆盖特性与雨量站观测数据的高精度特征。本研究结果可为降水时空分布特征的精细化刻画提供支撑，并助力小尺度水文模拟研究。