Bayesian Methods for Hydrometeorological Modeling and Forecasting (Invited)
Abstract
Efficient management of water resources requires skillful modeling and forecasting of hydrometeorology at the river basin scale. This includes but is not limited to basin precipitation for input into hydrologic models, downscaling precipitation from grid scale to a set of decision points and combining streamflow or precipitation from multiple model sources. Estimation of these variables in space and time, which is necessary for improved hydrometeorologic forecasting in the basins, is typically done from limited observations, thus requiring a good quantification of uncertainties. Emerging Bayesian methods offer attractive alternatives to traditional approaches to the above problems given their robust estimation of the variables and their attendant uncertainties. In this paper, we provide a sampling of Bayesian methods with three diverse applications. In the first, gridded variable such as precipitation from a climate model or satellite-based estimates need to be blended with point (station) observations to obtain improved gridded estimates of precipitation and the uncertainties, useful for driving hydrologic models and natural hazard mitigation. A Spatial Bayesian Hierarchical model is proposed for this problem and we apply this to blending precipitation over Central America and Upper Colorado River Basin. The second application consists of combining information from multiple sources - specifically, ensembles of precipitation are available at several grid points surrounding a location where a skillful estimate is desired. For this, Bayesian Model Averaging is applied to obtain a combined posterior probability distribution of precipitation. Last but not least, streamflow forecasts are generated from a number of models (dynamical and statistical) in the Colorado River Basin each with different skill and uncertainty - a robust combination method is desired that combines these forecasts to obtain an improved forecast. A Bayesian model combination applied for this problem results in a posterior distribution of streamflow forecasts that combines the different forecasts. These methods offer a sampling of applications and can be easily extended to other situations, including near term estimation of precipitation by blending model and radar observations, or combining multi-model ensembles of precipitation from the National Multi Model Ensemble (NMME) suite for applications such as hydrologic modeling.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2013
- Bibcode:
- 2013AGUFM.H11K..06R
- Keywords:
-
- 1816 HYDROLOGY Estimation and forecasting;
- 1872 HYDROLOGY Time series analysis;
- 1833 HYDROLOGY Hydroclimatology;
- 1840 HYDROLOGY Hydrometeorology