Trading Space for Time in Bayesian Framework

Development of an adequate water security strategy demands the ability to predict implications of human activity on water resources. This calls for methods that allow dealing with hydrologic non-stationarity while estimating the corresponding uncertainty. In our work, we use regionalized information, as contained in regionalised streamflow response indices, in a Bayesian framework to constrain hydrological models, thus trading space for time when dealing with non-stationarity. The main challenge of this approach is 1) to specify corresponding likelihood functions, and 2) to specify a prior. In our approach, the likelihoods are derived explicitly by taking account of the inter-index error covariance structure, so that regional information is neither neglected nor double-counted. Meanwhile a prior is shown to play a significant role when hydrologic behavioral evidence is scarce, and thus might lead to a biased parameter sampling. US catchments taken from the MOPEX database are used to test the methodological development. The results show the method can significantly reduce uncertainty in runoff prediction, although the most valuable indices vary with catchment characteristics. In some cases indices can even reduce the quality of predictions and uncertainty bounds. Also, the influence of prior choice on prediction precision and accuracy is estimated, and recommendations for a prior choice in regionalization are provided.