README.md

metadata

license: cc-by-4.0
tags:
  - climate
  - weather
  - extreme events
  - machine learning
task_categories:
  - image-segmentation
  - image-to-image
configs:
  - config_name: default
    data_files:
      - split: train
        path:
          - data/weather/coldwave/coldwave_records.csv
          - data/weather/heatwave/heatwave_records.csv
          - data/weather/tropicalCyclone/tropicalCyclone_surface_records.csv
          - data/weather/expcp/expcp_metadata.csv
          - data/weather/storm/storm_metadata.csv
          - data/eo/fire/training/training_index.csv
          - data/eo/fire/flood/training_index.csv
      - split: test
        path:
          - data/weather/coldwave/coldwave_records_test.csv
          - data/weather/heatwave/heatwave_records_test.csv
          - >-
            data/weather/tropicalCyclone/tropicalCyclone_surface_records_test.csv
          - data/weather/expcp/expcp_metadata.csv
          - data/weather/storm/storm_metadata.csv
          - data/eo/fire/validation/validation_index.csv
          - data/eo/flood/validation/validation_index.csv

The ExEBench Datasets are collected from weather and earth observation data from diverse sources.

Category	Heatwaves	Cold waves	Tropical cyclones	Storms	Extreme precipitation	Fires	Flood
Data type	Weather	Weather	Weather	EO	EO	EO	EO
Data source	EmDat, ERA5, ISO-3	EmDat, ERA5, ISO-3	EmDat, ERA5, IBTrACSv04	TASSRAD19	TRMM 3B42 V7, IMERG half-hourly Final Run	HLS Burn Scars	UrbanSAR-Floods
Sensor/variable	t₂m (max), land-, soil-, topography- masks	t₂m (min), land-, soil-, topography- masks	mslp, u10, v10; z, u, v at levels,land-, soil-, topography- masks	Radar prec. rate, noise mask	Precip Radar & TRMM Imager	Landsat/Sentinel-2, burn masks	Sentinel-1, flood masks
Spatial resolution	0.25°	0.25°	0.25°	500 m	0.1°	30 m	20 m
Temporal resolution	Daily	Daily	Hourly	5-minute	Half-hourly	N.A	N.A
Spatial coverage	Global	Global	Tropics	Trentino South Tyrol	Global	Contiguous US	Global
Temporal coverage	2019–2023	2019–2023	2019	2010–2019	2020–2023	2018–2021	2016–2023
Event number	55	9	95	931	1092	N.A	11
Frame size	CHW (C=4)	CHW (C=4)	CHW & CDHW (C=3, D=5)	LCHW (C=1, 480×480)	LCHW (C=1, 50×50)	CHW (6×512×512)	CHW (8×512×512)
Train split	4,844	338	12,993	810	796	540	405
Test split	366 frames	221 frames	2,438 frames	121 sequences	296 sequences	264 pairs	285 pairs
Task	Image-image prediction	Image-image prediction	Trajectory tracking	Video prediction	Video prediction	Segmentation	Segmentation (change detection)