Scientific Verification of Deterministic River Stage Forecasts

Abstract

One element of a complete verification system is the ability to determine why forecasts behave as they do. This paper describes and demonstrates an operationally feasible method for conducting this type of diagnostic verification analysis. Hindcasts are generated using different configurations of the forecast system and then the skill of the generated hindcasts is compared. The hindcasts and comparisons are constructed to isolate individual elements of the forecast process. The approach is used to evaluate the role of model calibration, model initial conditions, and precipitation forecasts in generating skill for deterministic river forecasts. The authors find that calibration and initial conditions provide skill for the short lead-time forecasts, with precipitation forecasts providing the majority of the skill in forecasts of high stages at longer lead times. At all lead times, this study shows model calibration is essential, as the calibration makes forecasts reliable.