Hydrological ensemble forecasting in mesoscale catchments: Sensitivity to initial conditions and value of reforecasts

Abstract

Hydrological forecasts can benefit from model climatologies compiled from long, consistent, retrospective forecasts (reforecasts). Typical areas of application include the estimation of return periods for extreme events, the detection of model deficiencies, such as systematic errors or the calibration of current forecasts using the reforecasts, and suitable observations. One difficulty when creating long reforecasts is the availability of good states for the model initialization, as a sufficiently dense network of meteorological observations is rarely available in most catchments for a long period. With this study, the creation of forecast and reforecasts in such catchments is motivated by considering two aspects: first, a comparison where it is shown that hydrological forecasts benefit most from high‐quality initial conditions on the basis of local observations, however, less accurate initial conditions using ERA‐interim reanalysis are also sufficient to provide skillful hydrological forecasts useful for many users. Second, we demonstrate that hydrological reforecasts compiled without long‐term meteorological observations have an additional value, e.g., they allow for more skillful runoff forecasts. Utilizing those reforecasts can compensate for forecast errors induced by less accurate initializations. For this study, hourly hydrological reforecasts, based on the PREcipitation‐Runoff‐EVApotranspiration HRU Model (PREVAH) were used, with an 18 year long, five‐member global ensemble reforecast data set from the European Centre for Medium Range Weather Forecasts (ECMWF) Variable Resolution Ensemble Prediction System (VarEPS) as forcing for lead times up to 10 days. Three mesoscale catchments of different characteristics in the Swiss Alps were analyzed.