From calibration to real‐time operations: an assessment of three precipitation benchmarks for a Swiss river system

Abstract

ABSTRACTIn hydrometeorological prediction systems, meteorological real‐time data and forecast data from numerical weather prediction models are needed as input for hydrological and hydraulic models. This paper evaluates the application of a validated hydrological–hydraulic model chain by using real‐time operationally available precipitation‐forcing datasets as a benchmark. The analyses should detect the problems occurring once the prediction system will leave the calibration environment and need to be reconfigured for real‐time deployment. The precipitation‐forcing benchmarks used for 2013 were rain gauge measurements, quantitative radar estimates, quantitative precipitation estimates (QPE) and a radar–rain gauge merged precipitation product generated by spatio‐temporal co‐kriging (with external drift). The discharge was simulated in a transnational river basin in southern Switzerland and northern Italy. The best discharge simulation results were obtained when rain gauge data from automatic measurement stations were used as meteorological input for the models, because these were used for the model calibration. The modelled radar QPE greatly underestimated the rainfall volume, and thus the simulated discharge, because the raw radar time series were too short for calibration. Therefore, quantile mapping was applied to post‐process the radar QPE. Quantile mapping, as applied in this work with relatively short available radar‐only precipitation records, appears to be unsuitable for operational use, but adequate as a post‐processing method for past data series. Combining the rain gauge and radar QPE improved simulations significantly compared with the original radar QPE. A lower performance was found in the Italian region because the meteorological stations were not considered there for processing this product.