The auto-tuned land data assimilation system (ATLAS)

Abstract

Land data assimilation systems are commonly tasked with merging remotely sensed surface soil moisture retrievals with information derived from a soil water balance model driven by observed rainfall. The performance of such systems can be degraded by the incorrect specification of parameters describing modeling and observation errors. Here the Auto-Tuned Land Data Assimilation System (ATLAS) is introduced to simultaneously solve for all parameters required for the application of a simple land data assimilation system to integrate satellite-based rainfall and soil moisture retrievals for drought monitoring applications. The approach is based on combining a triple collocation (TC) strategy with the statistical analysis of filtering innovations and designed to leverage the simultaneous availability of satellite-based soil moisture products acquired from both active and passive microwave remote sensing. A number of variants of the ATLAS approach - each based on a different strategy for leveraging TC and innovation analysis within an adaptive filtering framework - are derived and evaluated through a synthetic twin experiment. In addition, a preliminary real data analysis is conducted using actual satellite-based products and evaluated against independent ground-based observations. Results illustrate the potential of ATLAS to improve the analysis of soil moisture anomalies using data products derived from the Global Precipitation Measurement (GPM) and the NASA Soil Moisture Active/Passive missions. Key Points It is difficult to optimally parameterize a land data assimilation system Existing approaches are biased by auto- and cross-correlated errors We derive a new approach which is more robust to nonideal errors ©2013. American Geophysical Union. All Rights Reserved.