Influence of rainfall observation network on model calibration and application

Abstract

Abstract. The objective in this study is to investigate the influence of the spatial resolution of the rainfall input on the model calibration and application. The analysis is carried out by varying the distribution of the raingauge network. A meso-scale catchment located in southwest Germany has been selected for this study. First, the semi-distributed HBV model is calibrated with the precipitation interpolated from the available observed rainfall of the different raingauge networks. An automatic calibration method based on the combinatorial optimization algorithm simulated annealing is applied. The performance of the hydrological model is analyzed as a function of the raingauge density. Secondly, the calibrated model is validated using interpolated precipitation from the same raingauge density used for the calibration as well as interpolated precipitation based on networks of reduced and increased raingauge density. Lastly, the effect of missing rainfall data is investigated by using a multiple linear regression approach for filling in the missing measurements. The model, calibrated with the complete set of observed data, is then run in the validation period using the above described precipitation field. The simulated hydrographs obtained in the above described three sets of experiments are analyzed through the comparisons of the computed Nash-Sutcliffe coefficient and several goodness-of-fit indexes. The results show that the model using different raingauge networks might need re-calibration of the model parameters, specifically model calibrated on relatively sparse precipitation information might perform well on dense precipitation information while model calibrated on dense precipitation information fails on sparse precipitation information. Also, the model calibrated with the complete set of observed precipitation and run with incomplete observed data associated with the data estimated using multiple linear regressions, at the locations treated as missing measurements, performs well.

Influence of Spatial Precipitation Sampling on Hydrological Response at the Catchment Scale

“…Results of the simulation confirmed the most important findings of the analysis at event scale, in agreement with Bardossy et al (2008). Fig.…”

“…As an example, Schuurmans et al (2007) showed that the spatial variability of daily rainfall has a major effect on discharge and spatial distribution of groundwater level and soil moisture content of the catchment. More recently, studies based on continuous simulations have also been carried out (Bardossy et al 2008), confirming that an excessive reduction of rain gauges, up to a certain threshold number, makes model performances worse. Meselhe et al (2009), using a physically based and conceptual hydrologic model, showed that an increase in the rain gauge density or the rainfall data temporal resolution caused a significant improvement of the hydrograph estimation.…”

Rainfall disaggregation for hydrological modeling: Is there a need for spatial consistence?

“…Krajeski et al (1991) also conclude that for the analysis of spatial problems, fully-distributed models may be more suitable and recommend those for further studies. Bárdossy and Das (2008) point out that with an increasing spatial resolution of the applied rainfall-10 runoff model, the sensitivity of for example the rain gauge density and hence the spatial rainfall patterns may increase as well. The rainfall-runoff simulations were carried out with two models, the semi-distributed HBV model and the fullydistributed WaSiM-model.…”

“…WaSiM) lead to numerical diffusion and hence to a "smudging" of the areal rainfall, resulting in less differences in runoff statistics. Other investigations raise the question if spatial rainfall patterns can be transferred sufficiently into runoff with semi-distributed models and thus with a coarse spatial resolution (Krajeski et al, 20 1991, Obled et al, 1994, Bárdossy and Das, 2008.…”

Analysis of the Influence of Rainfall Spatial Uncertainty on Hydrological Simulations Using the Bootstrap Method

“…Rainfall is regarded as the key driving input of hydrological models and it is impossible to produce accurate runoff predictions if forced with inaccurate rainfall data [11]. The impact of spatial and temporal error in predicting rainfall on predicted flow has been highlighted by many researchers [12][13][14][15]. Currently, rainfall data from rainfall stations are still indispensable because the station data are regarded to be relatively accurate and more reliable than the radar data at the point where a station locates [16,17].…”