Performance and robustness of probabilistic river forecasts computed with quantile regression based on multiple independent variables

摘要

This study applies quantile regression (QR) to predict exceedanceprobabilities of various water levels, including flood stages, withcombinations of deterministic forecasts, past forecast errors and rates ofwater level rise as independent variables. A computationally cheap techniqueto estimate forecast uncertainty is valuable, because many national floodforecasting services, such as the National Weather Service (NWS), onlypublish deterministic single-valued forecasts. The study uses data from the82 river gauges, for which the NWS' North Central River Forecast Centerissues forecasts daily. Archived forecasts for lead times of up to 6 daysfrom 2001 to 2013 were analyzed. Besides the forecast itself, this study usesthe rate of rise of the river stage in the last 24 and 48 h and theforecast error 24 and 48 h ago as predictors in QR configurations. Whencompared to just using the forecast as an independent variable, adding thelatter four predictors significantly improved the forecasts, as measured bythe Brier skill score and the continuous ranked probability score. Mainly,the resolution increases, as the forecast-only QR configuration alreadydelivered high reliability. Combining the forecast with the other fourpredictors results in a much less favorable performance. Lastly, the forecastperformance does not strongly depend on the size of the training data setbut on the year, the river gauge, lead time and event threshold that arebeing forecast. We find that each event threshold requires a separateconfiguration or at least calibration.