Hydrological models

[ladislavgaal]

Dear SWICCA colleagues, I have a question related to one of the hydrological models used in the SWICCA database. Our client (the Slovak Hydrometeorological Institute) raised concerns related to the reliability of the VIC hydrological model in the Central European settings – namely, the model yields suspiciously low river discharges (and accordingly, derived statistics such as the 100-year flood) for the Danube River in Bratislava. Does anyone have any similar experience in other regions?

The problem in more detail. Our local case study aims at estimating the 100-year flood (i) on the basis of the observed data from the past and (ii) on the basis of the 33 simulated data series from the future. The ratio of the estimates of the 100-year flood return level based on future simulations vs. past observations is defined as the indicator of the climate change. Our results indicate that the overall performance of the 33 data sets are balanced: increase in the 100-year flood is expected to occur 14 times, decrease 12 times, while no change (i.e. when the change in absolute value is less than 5%) is expected in 7 cases. Nevertheless, one gets a different picture when inspecting the results stratified by the hydrological models: increases dominate in case of two hydrological models (HYPE and Lisflood), while the VIC model only yields decrease in the estimate of the 100-year flood.

The representative our client raised some concerns related to the performance of the VIC model. It is suspicious the estimates of the 100-year flood based on the VIC outputs are the lowest one overall. This induces questions whether the VIC model is appropriate for modelling river discharges in the Central European settings, particularly for the Danube River. Our client furthermore stressed that they have no similar objections against the other two hydrological models since they are familiar with them. The HYPE model, or more precisely, the HBV model is used in the operational practice at the Slovak Hydrometeorological Institute, while the Lisflood model is also known to them since it is used within the international project EFAS (www.efas.eu).

Note that our case study only focuses on a single site (Bratislava), so our findings described above certainly cannot be generalized, but right for this reason I am turning to the rest of the SWICCA community with the question whether they encountered some similar feelings during their work with the data from different hydrological models.

[FulcoLudwig]

There is two issues here:
1. Is the VIC model reliable for your location
2. The client does not trust the VIC model but does trust the result of the other two models.

It is hard to say if the VIC model is reliable but an analyses by Roudier et al. (2016) judged that all three models have sufficient performance to be included in the flood analyses. Our analyses in Roudier et al. also showed that Lisflood showed historically the best performance. This is not surprising given this model is developed specifically for floods in Europe. If we look at the future, the European patterns for the future are relatively similar for all three models in terms of predicting more floods throughout central Europe (see figure S3 in Roudier et al. 2016). So what you are finding is probably specific for your site. In general all three models give increases in in 1/10 and 1/100 year floods throughout most of Europe. So the VIC models could be non reliable but the conclusion that because the model gives a different result it should be discarded is not valid. However the larger scale pattern is different (more floods) and the fact that the other two models do show increases makes it likely that floods will increase and raises some question about the VIC results. The main differences of VIC compared to E-Hype and LisFlood are in how the models estimate snowmelt. If the increase in floods in E-Hype and Lisflood are mainly in spring due to earlier snow melt this could explain the difference between the models.
It is important to add a note here on HBV: this is a good model for floods under current conditions and some flood monitoring but has some limitations for climate change impact assessments. Due to the simple snow melt routine and the way it estimated evaporation.
A second important point is the trust of the client. Your client indicates that he trusts the other two models more. This is a concern you need to take serious. Given that one models gives a different results, the overall trends across multiple assessments indicate more floods in central Europe and the fact that your client trust the other two models I would recommend to do analyses with E-Hype and Lisflood only.

Let me know if you any additional questions
send me an email (fulco.ludwig@wur.nl) if can’t access the paper and I will send you a copy.

Roudier, R., J.C.M. Andersson, C. Donnelly, L. Feyen, W. Greuell & F. Ludwig 2016. Projections of future floods and hydrological droughts in Europe under a +2°C global warming, Climatic Change 135, 341-355
link to paper: https://link.springer.com/article/10.1007/s10584-015-1570-4

[ladislavgaal]

Dear Fulco,
many thanks for your reply. I found it really useful and encouraging.
For a long time I have been hesitating whether we I have the right to completely neglect the results of the hydrological model VIC or, in spite of all concerns, I have to include them into the final summary, causing then considerable bias in the final summary and messages for the policy makers… Your reply assured me that the first choice is the better one. The final conclusions of the our case study were, therefore, based on the results of the Lisflood and E-HYPE models, while it was stressed that the results of the VIC model were ignored for two fundamental reasons: i) the client’s trust in the other two hydrological models, and ii) the trends indicated by the VIC model contradict the findings from multiple assessments of future trends in extreme river discharges in Central Europe (specifically in the Upper Danube Basin).

• This reply was modified 6 years, 10 months ago by ladislavgaal.

[Author]

Posts