Sparse Local Implicit Image Function for sub-km Weather Downscaling
This provides a more accurate method for high-resolution weather prediction, which is incremental as it builds on implicit neural representations for a specific domain.
The paper tackled the problem of downscaling weather variables like temperature and wind to sub-kilometer resolution using sparse station data, achieving up to 50% better accuracy for temperature and 10-20% for wind compared to baselines.
We introduce SpLIIF to generate implicit neural representations and enable arbitrary downscaling of weather variables. We train a model from sparse weather stations and topography over Japan and evaluate in- and out-of-distribution accuracy predicting temperature and wind, comparing it to both an interpolation baseline and CorrDiff. We find the model to be up to 50% better than both CorrDiff and the baseline at downscaling temperature, and around 10-20% better for wind.