多源多尺度降雨观测数据融合方法及其水文应用研究

Precipitation is a key driver for hydrological modeling. Conventional point-scale in situ measurements are able to provide accurate observation of precipitation directly but their sparse coverage makes them difficult for regional studies. Satellite remote sensing shows complementing capabilities of covering a large spatial extent, but it is cumbersome to capture the local-scale precipitation variability due to its coarse resolution. Therefore, developing new data fusion methods to effectively merge different sources of precipitation data (e.g., in situ and satellite observations) seems promising, as it can provide spatially and temporally consistent precipitation dataset which can further improve the hydrological predictability. In this study, based on ground gauge and satellite-based precipitation data, we propose to develop a new data fusion framework to merge multi-source and multi-scale dataset, and further evaluate its performance for the hydrological application in the typical basins of the upper Yangtze River. .Objectives of the study are four folds: (1) improve the error model and its parameterization method of satellite based precipitation data to give a better characterization of the likelihood function for true precipitation value estimation with given satellite data; (2) bridge the scale gap between the large-scale (satellite) and point-scale (gauge) observations through a newly developed downscaling technique; (3) formulate a Bayesian hierarchical model to assimilate multi-source information (i.e. likelihood information retrieved from the error model, cross-scale information from downscaling approach and prior information from gauge measurements); (4) drive the state-of-art distributed hydrological model using the enhanced forcing after data fusion, and assess the hydrological predictability with different spatial resolution and error levels contained in the fused data. .This study will provide a novel theoretical framework to better characterize the spatial structure of precipitation and reduce its uncertainty through the integration of available information across different scales. In addition, the use of this proposed framework guarantees a high-resolution, continuous and self-consistent forcing dataset, which will further form a practical basis for the improved monitoring and modeling of the hydrologic cycle.

降雨是水文模拟的关键输入变量。传统站点观测直接准确但空间代表性不足，卫星遥感观测空间连续但局地精度有限，因此发展多源观测融合方法对于准确获悉降雨时空分布、提高水文监测和预报能力具有重要意义。本项目拟以长江上游典型流域为试点区域，基于卫星遥感和地面站点观测发展多源多尺度降雨观测数据融合方法并进行水文应用评估。首先，改进卫星遥感观测的误差模型及其参数化方案，更好地估计不同雨量和环境因素下遥感降雨观测的似然信息；然后，发展尺度降解模型来考虑卫星遥感与站点观测间的尺度差异；基于贝叶斯统计推断原理，同化遥感降雨观测的似然信息和尺度降解信息，并利用站点观测先验信息，发展多源多尺度降雨观测数据融合方法；最终通过耦合分布式水文模型，评估融合后降雨输入不确定性及其空间分辨率信息对水文预报能力的影响。本研究可为综合利用多源观测信息来提高流域多尺度水循环过程的监测及模拟能力提供理论基础和实践支撑。

降雨是水文模拟的关键输入变量。传统站点观测直接准确但空间代表性不足，卫星遥感观测空间连续但局地精度有限，因此发展多源观测技术及其融合方法对于准确获悉降雨时空分布、提高水文监测和预报能力具有重要意义。本项目以淮河中上游为典型研究区域，面向流域洪水预报应用需求比较了卫星遥感、地面雷达、雨量站点及数值天气模式的多源多尺度降雨估测技术适用性，在此基础上对多源多尺度降雨栅格数据误差特征量化、误差模型改进以及多源降雨数据信息融合方法进行了初步探讨。主要研究内容包括：一、全面评估了多卫星融合以及多雷达拼图定量降雨栅格数据在研究区域精度及其误差特征；二、改进提出了基于位移伽马分布的卫星遥感降雨误差模型，可以同时考虑降雨的探测误差及估值误差；三、发展了基于贝叶斯统计推断原理多源降雨数据融合模型，并对该模型应用效果进行了初步检验；四、拓展了面向对象的多源栅格降雨数据的空间信息挖掘新思路，并初步验证了其在卫星遥感与数值模型观测降雨信息挖掘中的作用。研究表明多源降雨观测信息具有各自相对优势，其相对有效信息取决于自身估测方法限制、误差模型结构以及选择的误差评估指标或方式，通过数据融合可有效降低综合估测误差。本研究可为综合利用多源降雨观测信息来提高流域多尺度水循环过程的监测及模拟能力提供理论基础和实践支撑。