The weather generator (WG) as currently used in GRADE1 shows several methodical imperfections, potentially leading to either under- or overestimation of large precipitation extremes. This is caused by the relatively short lengths of the observational records used as source data for the weather generator. A potential solution is to replace the presently used observational dataset with much longer datasets, generated by weather and/or climate models like RACMO and SEAS5. The goal of this study is to remove the bias from the RACMO and SEAS5 datasets and compare them with the current source dataset and the timeseries calculated by the WG either using the current source dataset as input or RACMO and SEAS5 datasets respectively,, with special attention on their performance to describe meteorological (high) extremes. The dataset comparison is executed for the Rhine, Meuse and Vecht basins, which differ in size and consequently in hydrological response time.
A qualitative comparison of the datasets based on several climate variables shows good similarities for all basins. Quantitative differences have been corrected by means of quantile mapping, leading to comparable results in all catchments and enabling use of RACMO and SEAS5 data in quantitative analysis as well.
Comparison of extreme multi-day precipitation conditions between the datasets shows excellent climatological agreement up to return periods of 65 years (i.e. within the length of the available observations). An important finding is that, for larger return periods, multi-day precipitation extremes are higher in the RACMO and SEAS5 dataset compared to the WG associated to the observational dataset (except the 10-day precipitation extremes in the Rhine basin). This strongly hints on underestimation by WG based on observational records. The underestimation decreases for sums over more days and increases for larger return periods. The effect is visible in all basins, but is more pronounced in smaller basins (like the Vecht). If this underestimation is passed on to the hydrological results, the current GRADE procedure may lead to a false sense of security regarding flood hazard.
Comparison of summer and winter extremes between the dataset shows considerably larger summer precipitation extremes in RACMO and SEAS5 compared to the WG associated to the observational dataset for all basins, particularly in the Vecht basin. Annual extremes in RACMO and SEAS5 consist of a larger fraction of summer events than the WG associated to the observational dataset indicates. This implies that the assumption that extreme precipitation events mainly occur during winter has to be reconsidered.
Spatial and temporal comparison of the datasets shows that RACMO is able to generate intense precipitation events in summer. Methodically the WG is not capable of simulating the extreme daily peaks that characterise these events, as the concept limits the extreme 1-day extremes to the observational events. This implies that the effect of extreme summer events (which seems to be essential in smaller basins) on hydrology can better be analysed using the RACMO dataset than the WG. Additionally, the RACMO (and SEAS5) dataset allow for spatial analysis of extreme events, which is impossible using the WG dataset.
A preliminary analysis of the hydrological discharges of RACMO and SEAS5 in the Meuse basin shows that the WG results associated to the observational dataset also underestimate hydrological high extremes. A first visual inspection shows over an order in magnitude difference in return period between RACMO and the WG result slightly smaller difference between SEAS5 and the WG. Based on the meteorological results, the differences are expected to be even larger in the Vecht basin. This shows the importance of further investigation of the hydrological results of RACMO and SEAS5 in all basins.
Regarding the results of this study, extra attention is required for the magnitude of summer events, both in hydrological and meteorological perspective.
L. van Voorst, H. van den Brink. Improving the GRADE weather generator by using synthetic datasets from RACMO and SEAS5.
KNMI number: TR-398, Year: 2022, Pages: 98