Improving the resolution of atmospheric data with neural networks
Atmospheric models and observations often have a limited resolution that does not fully capture small-scale processes. A neural network developed at LTE can provide estimates of the high-resolution field underlying an observed low-resolution image consisting of precipitation or cloud data. While it’s impossible to know exactly what the underlying high-resolution image is, our network can generate multiple possible high-resolution for each low-resolution input. This allows us to also estimate the uncertainty in the high-resolution images. In the future, we aim to use these results to improve the quality of atmospheric observations and weather forecasts.
Publication
Researchers: Jussi Leinonen