This repository contains the data for the paper "Local Off-Grid Weather Forecasting with Multi-Modal Earth Observation Data". The paper presents a novel multi-modal deep learning method that downscales gridded weather forecasts, such as ERA5 and HRRR, to provide accurate off-grid predictions. The model leverages both gridded data and local weather station observations from MADIS to make predictions that reflect both large-scale atmospheric dynamics and local weather patterns. The model is evaluated on a surface weather prediction task and shows significant improvement over baseline methods, including ERA5 interpolation, HRRR analysis and HRRR forecast and a multi-layer perceptron. Use the following citation when the data or the model are used:> Yang, Q.; Giezendanner, J.; Civitarese, D. S.; Jakubik, J.; ,Schmitt E.; Chandra, A.; Vila, J.; Hohl, D.; Hill, C.; Watson, C.; Wang, S.; Local Off-Grid Weather Forecasting with Multi-Modal Earth Observation Data. arXiv, October 2024. https://doi.org/10.48550/arXiv.2410.12938 The following data is available:- Shapefile of the Northeastern United States (NE-US, extracted from NWS)- Shapefile containing the location and number of observations (2019-2023) of the MADIS stations in NE-US- Processed hourly averaged MADIS data for the NE-US (2019-2023)- ERA5 data for the NE-US (2019-2023), gridded and interpolated- HRRR data for the NE-US (2019-2023), gridded and interpolated For MADIS, ERA5 and HRRR, the following variables are available:- u and v component of wind vector at 10 meters above ground- temperature at 2 meters above ground- dewpoint at 2 meters above ground