Scale-Problems in Assimilating of Passive Microwave Observation into Coupled Models

本项目探讨了无源l波段卫星观测在中尺度集水区与完全耦合的陆地系统模式的综合数据同化中的应用和价值。这种模式的陆地成分的模式分辨率通常为100米，大气成分的模式分辨率通常为1公里，这比卫星观测的分辨率要小得多，通常为几十公里。基于集合的数据同化需要通过观测算子从地面系统模型生成综合观测值，并将其与生成分析集合的观测值进行比较。由于模型状态不一定包括观测算子所需的所有信息，这在很大程度上是一个辐射传递模型，因此缺失的信息必须从外部数据中推断出来。该项目的主要目标是开发一种合适的观测算子，它能够以最好的方式模拟l波段卫星观测，并将其用于利用地面系统建模平台（TerrSysMP， Shrestha等人，2014年）和并行数据同化框架（PDAF， Nerger等人，2013年）在数据同化背景下利用这些观测。第一阶段主要关注灵活观测算子的编制及其与实际观测值的比较，第二阶段将进一步改进观测算子对植被的表示及其操作化，但主要关注其用于数据同化。这包括(a)量化由数据同化模型生成的观测值与虚拟现实和真实现实观测值之间的偏差，(b)数据同化实验，量化这些观测值在其他观测值（如降水）下的价值，以及(c)预处理和过滤方法，以更好地利用大尺度观测值的信息内容。

This project explores the use and value of passive L-band satellite observations for ensemble-based data assimilation with fully-coupled terrestrial system models for mesoscale catchments. Model resolutions are typically of the order of 100 m for land components and 1 km for the atmospheric component of such models, which is much smaller than for the satellite observations with typically tens of kilometers. Ensemble-based data assimilation requires the generation of synthetic observations from the terrestrial system model via an observation operator, which are compared with observations for the generation of the analysis ensemble. Since the model state does not necessarily include all information required for the observation operator, which is in large parts a radiative transfer model, missing information must be inferred from external data.The main objectives of the project are the development of a suitable observation operator, which is able to mimic L-Band satellite observations in the best possible way, and its use for the exploitation of such observations in the data assimilation context with the Terrestrial Systems Modeling Platform (TerrSysMP, Shrestha et al. 2014) coupled to the Parallelized Data Assimilation Framework (PDAF, Nerger et al. 2013). While Phase I focused mainly on the compilation of a flexible observation operator and its comparison with real observations, Phase II will pursue further improvements concerning the representation of vegetation by the observation operator and its operationalization, but mainly concentrate on its use for data assimilation. This includes (a) the quantification of biases between observations generated from the data assimilation model and observations both from the virtual and true reality, (b) data assimilation experiments which quantify the value of such observations given other observations (e.g. precipitation), and (c) pre-processing and filtering methods to better exploit the information content of the large-scale observations.