A non-stationary gridded weather generator conditioned on large-scale weather circulation patterns for Central Europe

For many flood risk assessments at large spatial scales, long-term meteorological data (e.g. precipitation, temperature) with spatially coherent representation are needed. This is where a regional weather generator comes into play. Meteorological fields for a specific region are strongly dependent on weather circulation patterns (CP) at larger scales. Additionally, there is evidence that these fields covariate with the average regional surface temperature (ART). With future climate change, such changes in both CP and ART should be included in weather generators.

This study presents the development of such a non-stationary gridded weather generator conditioned on large-scale weather circulation patterns for Central Europe. The reanalysis dataset ERA5 (1^o x 1^o) is used for weather type classification. The E-OBS gridded observational dataset (0.25^ox 0.25^o) is used to parameterize the meteorological fields, such as precipitation and temperature (minimum, maximum, average). The spatial and temporal dependence is represented by the multivariate auto-regressive model. Daily precipitation amount is modelled by the extended generalized Pareto distribution and daily temperature is modelled by the transformed normal distribution. Both fields are conditioned on CP and allow to covariate with ART. In this way, the regional weather generator is capable of capturing “between-type” and “within-type” climate variability and can be used to generate long synthetic data for flood risk assessment in present and future periods.