Datensatz

Zusammenfassung

Scenario-neutral and robust adaptation methods assess the vulnerability of climate-sensitive systems against a range of plausible climate conditions, independent of the socioeconomic scenarios typically used in climate modeling. Stochastic weather generators facilitate such analyses by enabling fast and flexible simulation of meteorological time series based on historical observations. Long-term changes in climate conditions are often described via corresponding changes in summary statistics or climate indices. However, adjusting stochastic weather generators to produce simulations consistent with perturbed summary statistics is challenging, especially for more complex statistics and weather generator models. We refer to this problem as climatology matching. In this work, we make two key contributions: First, we develop a flexible framework for stochastic weather generation based on Generalized Additive Models for Location, Scale, and Shape (GAMLSS). The proposed weather generator is capable of efficiently and accurately simulating daily temperature (mean, minimum, and maximum) and precipitation time series over multi-decadal time scales after being calibrated on historical data. Second, we propose a fully probabilistic formulation of the climatology matching problem, to which we apply techniques from the field of simulation-based inference (SBI). We evaluate our approach using weather station data from the German Weather Service and demonstrate its potential for scenario-neutral impact assessment by simulating realistic daily meteorological time series under various climate change conditions. Our method provides an efficient and flexible framework for stress-testing climate impact models with the potential to enhance the robustness of scenario-neutral impact assessments.