A systematic review of sensitivities in the Swedish flood-forecasting system

Since the early 1970s operational flood forecasts in Sweden have been based on the hydrological HBV model. However, the model is only one component in a chain of processes for production of hydrological forecasts. During the last 35 years there has been considerable work on improving different parts of the forecast procedure and results from specific studies have been reported frequently. Yet, the results have not been compared in any overall assessment of potential for improvements. Therefore we formulated and applied a method for translating results from different studies to a common criterion of error reduction. The aim was to quantify potential improvements in a systems perspective and to identify in which part of the production chain efforts would result in significantly better forecasts. The most sensitive (> 20% error reduction) components were identified for three different operational-forecast types. From the analyses of historical efforts to minimise the errors in the Swedish flood-forecasting system, it was concluded that 1) general runoff simulations and predictions could be significantly improved by model structure and calibration, model equations (e.g. evapotranspiration expression), and new precipitation input using radar data as a complement to station gauges; 2) annual spring-flood forecasts could be significantly improved by better seasonal meteorological forecast, fresh re-calibration of the hydrological model based on long time-series, and data assimilation of snow-pack measurements using georadar or gamma-ray technique; 3) short-term (2 days) forecasts could be significantly improved by up-dating using an auto-regressive method for discharge, and by ensembles of meteorological forecasts using the median at occasions when the deterministic forecast is out of the ensemble range. The study emphasises the importance of continuously evaluating the entire production chain to search for potential improvements of hydrological forecasts in the operational environment.