Title: Data Assimilation to Mitigate Structural Error in Land Surface Models
Presenting Author: Grey Nearing
Organization: NASA GSFC
Co-Author(s): Christa Peters-Lidard

Abstract:
Computational simulation models are subject to errors and uncertainty because they spatially and temporally aggregate complex micro-scale physical processes [Beven and Freer, 2001], and data assimilation mitigates these errors by conditioning model simulations on observations [Reichle, 2008]. Classically, this involves updating model states, however the classical approach is limited because states are local and transient, which means that information from observations used to update state estimates cannot be used to improve model predictions in times or locations outside of a small window around when and where observations were collected. We have developed a new method of data assimilation that updates not only the state of the model, but also the structure of the model equations used to approximate aggregated micro-scale physical processes [Nearing, 2013], and thereby improves the model's performance during a forecasting period (e.g., see Figure). The capability will directly compliment existing LIS data assimilation and parameter estimation tools that are the focus of current and past AIST projects.