The spatial and temporal resolutions of hydrological models are generally different from resolutions used in meteorological and climate models. This means that a translation is required from one scale to the other. The intensity of exchange processes such as precipitation and evaporation are by no means constant within the grid size of meteorological forecasting models (typically ~ 20 km). Also, within a single time-step, whether this is a month, day or hour, the intensity of precipitation in particular can be highly variable. Downscaling of precipitation is of particular interest for flood forecasting, operational management and climate change assessment studies. A fundamental aspect for downscaling rainfall is that the predicted rainfall is not exact but is often presented through a probability density function.
The present project is initiated to test downscaling techniques for rainfall, and to assess the benefit to the water management field in terms of improved system performance. We limit ourselves to the assessment of the information supplied by KNMI¿s Limited Area Numerical Weather Prediction Model HIRLAM for the benefit of the water management within the water boards. 13 months of HIRLAM precipitation forecasts in representative grid points were compared with five measured rainfall time series for the Rijnland district water board. Using 3 hours of rainfall forecasts and 10 rainfall classes, the comparison was made by applying the FBI, ETS and TSS skill scores. The results were very similar for the five representative stations. They show a very high percentage of false alarms, which could be halved if 24-hour aggregated forecast data were used, while on the other hand heavy rainfall is clearly underpredicted. It is likely that the forecast skill for extreme rainfall can be more satisfactory, however, in cases where NWPs are tuned to more extreme situations, or effective downscaling techniques are used. |
