Sup3rWind Data (CONUS)

Description

This data contains paired European Centre for Medium-Range Weather Forecasts Reanalysis version 5 (ERA5) and the Wind Integration National Dataset Toolkit (WTK) images for 2007 and 2010 over two regions in the US, with domain sizes ~800x800 (latitudes from 25.89 to 41.58, and longitudes from -97.25 to -77.85) and ~600x1600 (latitudes from 32 to 46.68, and longitudes from -126.81 to -85.35), respectively. This is created using tools from the NREL-rex and NREL-sup3r packages (both linked as resources below). A Lambert projection is used for WTK. The data includes two variables: the u and v components of the wind velocity at 10m from the surface. The training and validation splits consist of ERA5 at 30-km and WTK at 6-km spatial resolution from the year 2007. This 6-km WTK dataset is created by coarsening the WTK grid from its original 2-km resolution to 6-km resolution. The 30-km ERA5 is realigned, i.e. regrided to the 6-km WTK coarsened grid using inverse distance weighted interpolation. The year 2010 is used for testing, including two sets of test data with (1) ERA5 at 30-km and WTK at 6-km spatial resolution and (2) ERA5 at 30-km and WTK at 2-km spatial resolution. All of them have a temporal resolution of 1-hour. This data allows training machine learning models to downscale from low-resolution (LR) ERA5 to high-resolution (HR) WTK with an upsampling factor (the ratio of the size of the HR grid to the LR grid) of 5x and testing it on the same 5x factor as well as a higher upsampling factor of 15x. Please refer to the "Project Preprint" resource linked below for more details on the dataset and experiments. The work detailed in the "Sup3rWind Preprint" resource below also performs ERA5 to WTK downscaling.

Resources

Name Format Description Link
0 Super Resolution for Renewable Energy Resource Data With Wind From Reanalysis Data (Sup3rWind) and Application to Ukraine preprint, a related work detailing similar dataset/experiment. https://arxiv.org/pdf/2407.19086
0 On the Effectiveness of Neural Operators at Zero-Shot Weather Downscaling preprint, detailing the the dataset and experiments https://arxiv.org/pdf/2409.13955v1
57 Train/valid/test splits for region 1 with domain size ~800x800 (latitudes from 25.89 to 41.58, and longitudes from -97.25 to -77.85). https://data.openei.org/files/6210/region1_800x800%20%281%29.zip
57 Train/valid/test splits for region 2 with domain size ~600x1600 (latitudes from 32 to 46.68, and longitudes from -126.81 to -85.35). https://data.openei.org/files/6210/region2_600x1600.zip
21 REsource eXtraction Tool (rex) repository. https://github.com/NREL/rex
21 The Super-Resolution for Renewable Resource Data (sup3r) software uses generative adversarial networks to create synthetic high-resolution wind and solar spatiotemporal data from coarse low-resolution inputs. https://github.com/NREL/sup3r/

Tags

  • data
  • processed-data
  • weather
  • deep-learning
  • united-states
  • dataset
  • downscaling
  • wind
  • climate
  • computational-science
  • model
  • wtk
  • windspeed
  • sup3rwind
  • conus
  • machine-learning
  • super-resolution
  • neural-operators
  • energy

Topics

Categories