Value of river discharge data for global-scale hydrological modeling

This paper investigates the value of observed river discharge data for global-scale hydrological modeling of a number of flow characteristics that are e.g. required for assessing water resources, flood risk and habitat alteration of aquatic ecosystems. An improved version of the WaterGAP Global Hydrology Model (WGHM) was tuned against measured discharge using either the 724-station dataset (V1) against which former model versions were tuned or an extended dataset (V2) of 1235 stations. WGHM is tuned by adjusting one model parameter (gamma) that affects runoff generation from land areas in order to fit simulated and observed long-term average discharge at tuning stations. In basins where gamma does not suffice to tune the model, two correction factors are applied successively: the areal correction factor corrects local runoff in a basin and the station correction factor adjusts discharge directly the gauge. Using station correction is unfavorable, as it makes discharge discontinuous at the gauge and inconsistent with runoff in the upstream basin. The study results are as follows. (1) Comparing V2 to V1, the global land area covered by tuning basins increases by 5% and the area where the model can be tuned by only adjusting gamma increases by 8%. However, the area where a station correction factor (and not only an areal correction factor) has to be applied more than doubles. (2) The value of additional discharge information for representing the spatial distribution of long-term average discharge (and thus renewable water resources) with WGHM is high, particularly for river basins outside of the V1 tuning area and in regions where the refined dataset provides a significant subdivision of formerly extended tuning basins (average V2 basin size less than half the V1 basin size). If the additional discharge information were not used for tuning, simulated long-term average discharge would differ from the observed one by a factor of, on average, 1.8 in the formerly untuned basins and 1.3 in the subdivided basins. The benefits tend to be higher in semi-arid and snow-dominated regions where the model is less reliable than in humid areas and refined tuning compensates for uncertainties with regard to climate input data and for specific processes of the water cycle that cannot be represented yet by WGHM. Regarding other flow characteristics like low flow, inter-annual variability and seasonality, the deviation between simulated and observed values also decreases significantly, which, however, is mainly due to the better representation of average discharge but not of variability. (3) The choice of the optimal sub-basin size for tuning depends on the modeling purpose. While basins over 60 000 km(2) are performing best, improvements in V2 model performance are strongest in small basins between 9000 and 20 000 km(2), which is primarily related to a low level of V1 performance. Increasing the density of tuning stations provides a better spatial representation of discharge, but it also decreases model consistency, as almost half of the basins below 20 000 km(2) require station correction.