Predicting extreme precipitation events is one of the main challenges of climate science in this decade.
Despite the computational power available nowadays, current state-of-the-art Global Climate Models’ (GCMs) spatial resolution is still too coarse to correctly represent and predict small-scale phenomena such as convection, therefore precipitation prediction is still imprecise.
For this reason, downscaling techniques play a crucial role, both for the understanding of the physical mechanisms behind precipitation onset and development, and for applications like hydrologic studies, risk prediction and emergency management.

Taking advantage of Deep Learning techniques, we exploit a conditional Generative Adversarial Network (cGAN) to train a generative model able to perform precipitation downscaling.
This model, a deep Convolutional Neural Network (CNN), takes as input the precipitation field at the scale resolved by GCMs, adds random noise, and outputs a possible realization of the precipitation field at higher resolution, preserving the statistical properties of the input field.
Also, being conditioned by the coarse-scale precipitation, the spatial structure of the produced small-scale field is consistent with the one prescribed by the input GCM prediction.
To assess the skill of our model, we try to reconstruct the daily total precipitation field over the Taiwan region, starting from a coarsened version of the ERA5 reanalysis dataset.
Results show the good ability of the model in capturing the features of the fine-scale precipitation field.
In addition, compared to other downscaling techniques, our model has the advantage of being computationally inexpensive at run time and easily generalizable to any geographical domain.