A New Postprocessing Method for Improved Ensemble Forecasts

Postprocessing methods are usually applied to the forecast ensemble of hydrological and hydrometeorologial variables as to reduce the uncertainties arising from climatology, model structure and parameters, and initial conditions at the forecast date. A commonly used statistical technique in bias correction of hydrologic forecasts, Quantile Mapping (QM), is a blind-matching method being ineffective sometimes to well-correct the forecasts towards the observed values. To approximate the effectiveness of QM procedure before stepping into the forecast mode, this study proposes an auxiliary variable called failure index (γ). While the downsides of QM is addressed, an alternative postprocessor based on copula functions is introduced which build a multivariate joint distribution of observations and model predictions. A set of 2500 hypothetical forecast ensembles with parametric marginal distributions of simulated and observed variables are postprocessed with both QM and the proposed multivariate postprocessor. Deterministic forecast skills show that the proposed copula-based postprocessing is more effective than the QM method in improving the forecasts. It is found that the performance of QM is highly correlated with the failure index, unlike the multivariate postprocessor. In probabilistic metrics, the proposed multivariate postprocessor generally outperforms QM. Further evaluation of techniques is conducted for river flow forecast of Sprague River Basin in southern Oregon. Results show that the multivariate postprocessor performs better than the QM technique; it reduces the ensemble spread and is a more reliable approach for improving the forecast.