Interpreting Results: OHRFC
For the OHRFC study, QPF clearly improved the stage height forecasts as measured by the mean absolute error for all lead times. But we hope the discussion of this case demonstrated that verification measures can provide much more useful information than just a simple measure of error. In this case we explored the trends with lead time of error, bias, correlation and skill. We looked at those parameters for different sets of basins defined by their response time characteristics. And we made some comparison between forecasts with QPF input and those without, including the relative impacts runtime modifications within the hydrologic model.
With those additional scores we learned that increasing negative bias with lead time was most likely the reason for increasing absolute errors and decreasing correlation with lead time. We also saw that QPF input appears to reduce but not eliminate these negative trends associated with longer lead times. In fact the QPF input appears to be more important than runtime modification for reducing bias in the longer lead times. This suggests that QPF input increases skill in the longer lead times. Finally, we observed through many of the statistics that a reduction in forecast performance with lead time occurs more quickly with fast-response basins than with slow response basins.
Without the variety of verification scores, we would not know that the negative trend in error and correlation with increasing lead time is due mainly to a negative bias. We also would not have confirmed that this impact is greater for the fast response basins. Furthermore, the scores suggest that the QPF input may be more important to hydrologic forecast performance than runtime modifications. With this information the users and developers of the hydrologic models are better equipped to account for limitations of the forecasts and subsequently improve the forecast system.
