Stochastic Weather Generators for spatial-temporal generation of temperature and rain over France
par
Fokko Du Cloux
La Doua
Stochastic weather generators are efficient statistical models producing synthetic weather series by replicating key statistical properties without the computational cost of physical models. However, for applications requiring simulation over a large area with many locations, challenges arise due to non-stationarity over time and spatial-temporal dependencies. In this presentation, we first introduce a temperature generator combining a single-site mean-variance decomposition and tools from geostatistics to produce a daily stochastic weather generator for temperature with arbitrary spatial resolution over France. Then, we tackle the more complex precipitation variable. The rain occurence is modeled through a Hidden Markov Model with spatially correlated Bernoulli emissions, to break down the complexity in different states. This model allows us to retrieve some “known” weather types through the maximum a posteri sequence of states. For the rain intensity, we use appriopriate Extended Generalized Pareto distributions in each class, and model the space-time dependency using the same geostatistics tools as in the temperature model. Both generators are evaluated for their ability to reproduce spatial-temporal events, like heat waves or dry spells according to their spatial extent.