Project start: 01.01.2017

Duration: 2 years

Funding: Bayerisches Staatsministerium für Umwelt und Gesundheit

Project lead: Harald Kunstmann

Involved scientists: Patrick Laux, Manuel Lorenz

Summary

In the context of global warming, concerns about the impact of climate change at regional and local levels are growing. Even high-resolution regional climate models are usually unable to reproduce observed small-scale climate features, especially in mountainous regions. Significant deviations in simulated meteorological fields, e.g. Precipitation and temperature do not allow for direct use of the model output in subsequent performance models in hydrology, agriculture, or other disciplines. We investigate the possibility of applying copula-based multivariate statistics for downscaling and bias correction of regional climate modeling outputs.

In contrast to conventional approaches of correlation-based statistical downscaling, Copula-based methods allow a very flexible consideration of the dependency between local, small-scale climate properties and regional or global relationships. The approach makes it possible to model dependencies between variables without being bound to normal distributions. In addition, the dependence on variables can be analyzed independently of the edge distributions. Our analysis focuses on the dependence structure between observed and dynamically modeled fields of precipitation and temperature.

For the first time, we apply the copula-based analysis to continuous time series by applying an ARMA-GARCH transformation to the time series to obtain iid records (independent and identically distributed).

Various theoretical copula families are studied and fit tests are performed to make the optimal choice. Based on the derived theoretical Copula models, stochastic simulations are performed to quantify the uncertainties of the results obtained.

In the course of the project, the developed new method for local refinement and bias correction of current RCM simulations for Germany and the Alpine region will be applied. A comparison of the results with conventional bias correction methods such as quantile mapping must now show the benefits of the copula-based approach.