Location: Hydrology and Remote Sensing LaboratoryTitle: The efficiency of data assimilation
|NEARING, G.S. - Goddard Space Flight Center|
|YATHEENDRADAS, S. - Goddard Space Flight Center|
|ZHAN, X - Collaborator|
|LIU, J. - National Aeronautics And Space Administration (NASA)|
|CHEN, F. - Science Systems Associate|
Submitted to: Water Resources Research
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 6/1/2018
Publication Date: 7/17/2018
Citation: Nearing, G., Yatheendradas, S., Crow, W.T., Zhan, X., Liu, J., Chen, F. 2018. The efficiency of data assimilation. Water Resources Research. 54(9):6374-6392. https://doi.org/10.1029/2017WR020991.
Interpretive Summary: Data assimilation is a powerful tool by which geophysical observations are used to constrain dynamic model predictions. It is currently applied to integrate soil moisture observations with models and produce best-possible estimates of soil moisture conditions for various applications (e.g., agricultural drought monitoring and numerical weather prediction). However, the mathematics applied in these systems is based on a series of approximations regarding the statistical nature of errors (affecting both the observations and the model) and the linearity of the dynamic model. Using information theory, this paper provides the first rigorous, quantitative description of how these various approximations degrade the eventual accuracy of soil moisture predictions made by a data assimilation system. As such, it provides a critical roadmap for improving these systems. The results of this paper will eventually be used by weather forecasters, drought monitoring agencies and water resource managements to improve the quality of soil moisture information available for their forecasts or assessments.
Technical Abstract: Data assimilation is the application of Bayes’ theorem to condition the states of a dynamical systems model on observations. Any real-world application of Bayes’ theorem is approximate, and therefore we cannot expect that data assimilation will preserve all of the information available from models and data. We outline a framework for measuring data assimilation-relevant information in models, remote sensing observations, and evaluation data in a way that allows us to quantify information loss during imperfect data assimilation. This facilitates quantitative analysis of tradeoffs between improving (expensive) remote sensing observing systems vs. investing in improved data assimilation capabilities. This also allows us to systematically decompose various error components in the application of Bayes’ theorem, and thereby to quantify specific deficiencies and areas for improvement in a particular data assimilation application. We demonstrate this information-theoretic framework to the problem of assimilating remote sensing soil moisture retrievals from AMSR-E into a land surface model, and we find that the Ensemble Kalman Filter uses only a very small fraction of the available information.