Simplifying a hydrological ensemble prediction system with a backward greedy selection of members – Part 2: Generalization in time and space

Abstract

Abstract. An uncertainty cascade model applied to stream flow forecasting seeks to evaluate the different sources of uncertainty of the complex rainfall-runoff process. The current trend focuses on the combination of Meteorological Ensemble Prediction Systems (MEPS) and hydrological model(s). However, the number of members of such a HEPS may rapidly increase to a level that may not be operationally sustainable. This paper evaluates the generalization ability of a simplification scheme of a 800-member HEPS formed by the combination of 16 lumped rainfall-runoff models with the 50 perturbed members from the European Centre for Mediumrange Weather Forecasts (ECMWF) EPS. Tests are made at two levels. At the local level, the transferability of the 9th day hydrological member selection for the other 8 forecast horizons exhibits an 82 % success rate. The other evaluation is made at the regional or cluster level, the transferability from one catchment to another from within a cluster of watersheds also leads to a good performance (85 % success rate), especially for forecast time horizons above 3 days and when the basins that formed the cluster presented themselves a good performance on an individual basis. Diversity, defined as hydrological model complementarity addressing different aspects of a forecast, was identified as the critical factor for proper selection applications.

Simplifying a hydrological ensemble prediction system with a backward greedy selection of members – Part 1: Optimization criteria

“…It is a "pseudo test dataset" because there is a high probability that the data used in testing (the complete series) have been used in the BGS training process, becoming the indicator of an optimistic estimator of the selection (Diamantidis et al, 2000); however, we do emphasize that the first part of this research focuses on an analysis of scores in the BGS process with the subsequent integration of results, and the second phase presented in a companion paper (Brochero et al, 2011) shows a rigorous test of generalization in time and space.…”

“…This decision is justified on the variability within the ensemble forecasts as well as on the fact that the selection of hydrological members as a method of simplifying HEPS should be unique regardless of the forecast horizon. The companion paper (Brochero et al, 2011) assesses the transferability of the 9-day members' selection to other forecast horizons.…”

Simplifying a hydrological ensemble prediction system with a backward greedy selection of members – Part 1: Optimization criteria

“…Selected catchments are identified with the first three digits of each code used in Table 1. The other delimited basins are part of the study of results' generalization shown in Brochero et al (2011).…”

On the importance of training methods and ensemble aggregation for runoff prediction by means of artificial neural networks

“…Even if the ensemble is composed of ANN model outputs, such ANNs may have different architecture, may be trained by different optimization algorithms, or on different samples of data that are usually obtained by means of either bagging or boosting methods. Although in hydrology often the ensemble modelling is utilized by considering the whole ensemble simultaneously (Krzysztofowicz 2001, Cloke and Pappenberger 2009, Brochero et al 2011a, 2011b, Abaza et al 2013, the idea of aggregation of ensemble members to achieve a single forecast has also been widely discussed (Boucher et al 2009, Rathinasamy et al 2013, DeWeber and Wagner 2014. In this study we are concerned only with ensemble aggregation approaches, in fact two of the simplest of them, in which all ensemble members are of the same type.…”