Development of a Copula-Based Weather Generator for Assessment of Climate Impact on the Hydrodynamic and Ecologic State of Highly Sensitive Aquatic Systems Using the Example of Lake Constance
Final Report Abstract
While it is apparent that climate change is an increasingly pressing issue, assessment of its influence on complex systems like large lakes is associated with uncertainties. We started to address this problem by developing an autoregressive weather generator VG already in an earlier DFG-funded project (BodenseeKlima), but were convinced that more sophisticated methods – that also advance the state of the art – are required. This lead to the adoption of copulas, replacing the linear models in VG with general dependencies in our newly developed weather generator Weather-Cop. Being presented with more than a few variables, we dared to deal with rain in a backwards-thanusual way, exploiting the dependencies between precipitation- and non-precipitation variables. This is a pre-processing step that enables the use of unified models for precipitation occurrence and amount. We see this approach as a first step that needs to be studied further in order to identify efficient precipitation generators. Obviously, in order to be useful at other locations, more variables than precipitation have to be available. Generating precipitation with a weather generator that was originally designed to provide input to lake models, broadens its area of application. As more complex models require more data to avoid over-fitting, we were determined to achieve parameter parsimony. This was addressed by a unique combination of decorrelation by vine copula and phase randomization. Phase randomization was also key for parameter-free extension to multisite weather generation. By comparing our weather generators with widely used ones in terms of entropy, we gained insight into shortcomings of all studied weather generation models. These deficiencies point to possible new directions for developing precipitation generation methods that balance order and disorder in strong rainfall events. With weather generators that are based on resampling techniques enjoying increased popularity, we found it to be important to point out its limitations. Our results show that especially when dealing with changed climate, underestimating variability is unavoidable.
Publications
- (2015, April). Generating wind fields that honour point observations and physical conservation laws. In EGU General Assembly Conference Abstracts (Vol. 17)
Schlabing, D., & Bárdossy, A.
- (2017 September). Comparing five weather generators in terms of entropy In International Workshop on Stochastic Weather Generators for Hydrological Applications (SWGen-Hydro)
Schlabing, D., & Bárdossy, A.
