A Sentinel-2 Dataset for River Gauge Estimation Across Germany
26,862 Sentinel-2 true-color image chips for river gauge estimation.
- Source
- Zenodo
- Published
- 2026
- Country
- Germany
- Region
- Europe
- Format
- multiple formats
- Access
- Open access
- License
- cc-by-4.0
- Themes
- Remote Sensing & Satellites, Fresh Water
About this dataset
This dataset contains 26,862 Sentinel-2 true-color image chips for river gauge estimation. Each image annotated with the information about the location, the measuring station for the gauge level, the measurement value and a class for each level relative to the median of the station. Each image is provided as a 224 × 224 pixel RGB PNG file and covers a 512 m × 512 m geographic footprint.
In addition to the image chips, the raw data retrieved from the Copernicus Data Space Ecosystem Sentinel Hub Process API is included in the dataset in TIF format. The TIF files include the bands B02, B03, B04, B08, B11, B12, SCL and dataMask. The bands B08, B11 and B12 refer to the near-infrared, short-wave infrared 1 and short-wave infrared 2 bands, respectively. The Scene Classification MAP (SCL) can be used to classify pixels as water, cloud, snow, etc. The dataMask can be used to filter out pixels with invalid or missing data.
The geographic locations of the samples are based on a list of river gauge measuring stations from PegelOnline, which publish their measurements. For each measuring station, a 512 m × 512 m geographic footprint with the station at the center is created and requested from the Copernicus API.
The annotations include the measured value in centimeters, the station median, the 20% and 80% quantile, the interquartile range, and the station percentile. The measurements are not standardized across different stations. Therefore, the measurements should be normalized and estimated relative to the station median. The station percentile and the median of the station can be used for the normalization. The 20% and 80% quantiles split the measurements into the three classes low, normal, and high, with the high class above the 80% quantile, the low class below the 20% quantile and the normal class between the 20% and 80% quantile.
Image chips were generated from the Sentinel-2 Level-2A imagery using bands B04, B03, and B02 as red, green, and blue channels. Imagery was retrieved through the Copernicus Data Space Ecosystem Sentinel Hub Process API for the period from January 1, 2022 to August 20, 2026. The Collection was generated using a maximum cloud-cover threshold of 5% and least-cloud-cover mosaicking.
How the data was collected
This dataset contains 26,862 Sentinel-2 true-color image chips for river gauge estimation. Each image annotated with the information about the location, the measuring station for the gauge level, the measurement value and a class for each level relative to the median of the station. Each image is provided as a 224 × 224 pixel RGB PNG file and covers a 512 m × 512 m geographic footprint.
In addition to the image chips, the raw data retrieved from the Copernicus Data Space Ecosystem Sentinel Hub Process API is included in the dataset in TIF format. The TIF files include the bands B02, B03, B04, B08, B11, B12, SCL and dataMask. The bands B08, B11 and B12 refer to the near-infrared, short-wave infrared 1 and short-wave infrared 2 bands, respectively. The Scene Classification MAP (SCL) can be used to classify pixels as water, cloud, snow, etc. The dataMask can be used to filter out pixels with invalid or missing data.
The geographic locations of the samples are based on a list of river gauge measuring stations from PegelOnline, which publish their measurements. For each measuring station, a 512 m × 512 m geographic footprint with the station at the center is created and requested from the Copernicus API.
The annotations include the measured value in centimeters, the station median, the 20% and 80% quantile, the interquartile range, and the station percentile. The measurements are not standardized across different stations. Therefore, the measurements should be normalized and…
Image…