We have developed a purely convolutional neural network to increase the spatial resolution of climate models, which is necessary to provide climate change scenarios at the local scale. The training of this machine learning method is conducted using PyTorch, an optimized tensor library for deep learning. We downscale ERA5 reanalysis data, initially at 0.25 degrees resolution, to a finer 0.05 degrees resolution, using the CERRA dataset as ground truth. Utilising surface-level variables along with those at 850hPa, 700hPa, and 500hPa, we generate downscaled outputs for probability distributions of precipitation and air surface temperature. This approach enables the creation of high-resolution climate data, crucial for detailed analysis and modelling in various scientific and environmental applications. The results have been assessed against HCLIM a regional climate model at high resolution, showing comparable results. This result is key to ensuring that the signal of a coarse-resolution global climate model can use these methods to provide regional information of the climate. The method is evaluated against CERRA, and against HCLIM, an even higher-resolution regional dynamic climate model (3 km) divided into three European subregions. The statistical difference between CNN and CERRA is typically significantly lower than the statistical difference between HCLIM and CERRA. Results are promising, so that the new method will be used and further developed to complement regional dynamic-thermodynamic climate models. The high-resolution datasets that will be generated with this machine learning method will be used to assess how different global warming levels, rapid changes in the climate system and the occurrence of tipping points impact the climate at regional scales. We will focus on Europe and polar regions, assessing changes in the mean state, climate variability and extreme events.  Using results, generated in Alvis will allow us to create high-resolution climate information to understand the regional threads of reaching different warming levels. All of this as part of the EU Horizon project OptimESM. The development and testing of this machine learning algorithm have been carried out in Berzelius, and currently, we are applying to computing time in Alvis to continue our project.