Characterizing Precipitation Forcing Uncertainty in Land Data Assimilation Using an Ensemble-Based Bayesian Approach

Abstract:

Ensemble-based data assimilation techniques are often applied to land surface models in order to estimate components of terrestrial water and energy balance. Precipitation forcing uncertainty is the principal source of spread among the ensembles which is required for utilizing information in observations to correct model priors. Precipitation fields may have both position and magnitude errors. However current approaches to uncertainty of precipitation forcing in land data assimilation systems often do no more than apply multiplicative errors to precipitation fields. In this presentation we show that an ensemble-based Bayesian approach can be used to produce stochastic replicates of precipitation fields that are conditioned on precipitation observations. We proposes a new ensemble-based approach to characterize uncertainties (in both magnitude (intensity) and phase (location)) associated with precipitation retrieval from space-born instruments. Unlike previous studies, this method derives the error likelihood using an archive of historical measurements and provides an ensemble characterization of measurement error. The ensemble replicates are generated using a stochastic method and are intermittent in space and time. The replicates are concisely described with a low-dimensional and problem-specific set of attributes. The attributes are derived using a novel dimensionality reduction scheme that takes advantage of principle component analysis. A non-parametric importance sampling is formulated in terms of the attribute vectors to solve this Bayesian sampling problem. Results indicate that our ensemble estimation approach is able to provide an improved description of precipitation uncertainties by giving a realistic posterior ensemble that is narrower than the prior.