卫星反演降水与地面观测资料的变分同化技术研究

Abstract: The high quality precipitation has plays significant in global water cycle. Considering the limitation relying only on stations distribution in homogeneity and data-sparse regions, such as Tiber Plateau, the South area, and depopulated zone, the combination of satellite retrieval product and gauge stations will be studied for the merged precipitation products. Firstly, to adjust the system bias, we have defined the gauge-CMORPH differences, which are smoothed over a history period data centering at the target day. The satellite system error which has its physically fixed the saturation determined by retrieval algorithm itself has been successfully adjusted through our approach. Secondly, the study is to develop a new technique through a two-dimensional variational analysis (2D-VAR) to blend the precipitation information from multiple sources for the construction of a high-quality precipitation analysis. Finally, Ground Cross Validation (GCV) experiments in which each affected station is withdrawal at once indicated that the variatoinal analysis can particularly be beneficial and subsequent diagnosis of heavy rainfall events. It also reveals that the precipitation analysis field has the ability to improve the accuracy of rainfall estimation and capture the spatial pattern agreements in relatively Tibet Plateau and data-sparse regions.

高质量、高可靠性的降水产品是开展全球水循环研究的基础和必要支撑。考虑到我国的常规站站点东密西疏，站点分布不均匀，青藏高原与西部区域站点稀疏，无人区没有观测等局限性，卫星可以提供高时空分辨率、覆盖范围广的降水估计，弥补雨量计观测的不足，本研究拟结合卫星反演降水产品和地面观测资料进行变分同化研究。首先，以地面观测为基准，基于“历史资料统计量”方法，扩大收索历史资料的空间信息和时间长度，有效订正卫星反演降水产品的系统误差。其次，根据变分同化理论，进行多源降水数据的尺度归一化研究，进一步开发卫星反演降水产品与地面观测资料的同化算法，从而实现多源降水观测资料的优势互补和有效结合。最后，广义交叉验证对合成产品进行独立性验证，给出产品的质量状况和分区域适用性评价，重点优化合成产品在青藏高原及西部站点稀疏地区的应用效果。

本研究开发了多种观测降水数据变分同化算法，实现了卫星反演降水与地面观测资料的优势互补和有效融合，研制出一套中国区域逐小时、10km×10km高时空分辨率、高质量的合成降水产品。并利用交叉验证方法验证了融合产品在不同站网密度、不同区域的精度和适用性评估，研究表明融合产品在检测夏季对流性强降水事件和对极端天气事件的研究具有重要意义。主要研究成果包括：1）设计变分算法，讨论单个降水要素的泛函代价函数，为了获得最优的观测值，转化为求泛函函数极值的问题，即基于分析值与实测值在最小二乘意义上求最小的思想，选用拟牛顿最小化算法，迭代更新最优降水分析场。2）其次，提出基于“历史资料统计量”，依据卫星格点上与地面格点上降水量大小比例线性比，扩大“历史资料”的空间搜索范围和时间尺度上资料的长度来有效订正卫星非独立的系统偏差。3）同时，提出加入调整因子来放大山地对观测误差的影响，假设相对误差是随地形复杂度线性增大，使用格点内标高的标准偏差来表达地形的复杂度，本文作者用降水量本身代表标高验证结果显示此方法从降雨特点分布更为合理，有效的订正了卫星反演降水产品微弱降水高估、强降水低估现象。4）利用交叉验证方法，独立验证产品的空间分布特征、不同降水量级下、不同区域的精度和适用性评价，独立验证了产品能较好的反应温暖地区的弱降水和极强降水事件。