Dynamical downscaling of hybrid seasonal predictions

Summer heatwaves have been increasing in frequency in the past decades over Europe (Rousi et al. 2022), and so has their impact on public health and the productive system. Predicting these events on a subseasonal-to-seasonal timescale would be crucial to mitigate their negative effects (C. J. White et al. 2017). This thesis proposes a methodology integrating ensemble subsampling and dynamical downscaling to improve the representation of summer temperatures, focusing on the city of Bologna and the neighbouring rural areas. The main aspect concerns the preparation of a seasonal forecast dataset to initialize the downscaling model. The procedure combines publicly available datasets to ensure easy replicability. Through sensitivity tests, I identify an adequate configuration of the WRF model to downscale a subset of members from the seasonal forecast. The dynamical downscaling reduces the monthly-averaged two-metre temperature BIAS and MAE across all locations considered, with greater benefit observed in the urban locations. Introducing the sea surface temperature update in WRF reduces the temperature BIAS over marine areas, but further corrections would be needed to address it fully. The performance of simple statistical corrections is also explored, highlighting the potential of combining the dynamical downscaling with the Mean and Variance Adjustment technique within a hybrid parallel approach. This work also acknowledges the limitations of the current setup while outlining the possible future steps. These include implementing a process-informed ensemble subsampling and improving the ocean-atmosphere coupling. Additionally, I present the recommended metrics for a more comprehensive evaluation of the results.