license: cc-by-4.0
tags:
- geospatial
- satellite-imagery
- remote-sensing
- earth-observation
- climate
- extreme-weather
- drought
- heat-cold-wave
- agriculture-monitoring
- agriculture
- yield
- crop-yield
pretty_name: CropClimateX
size_categories:
- n>1T
task_categories:
- time-series-forecasting
- tabular-regression
- tabular-classification
- image-feature-extraction
CropClimateX
- Repository: TBA
- Paper: TBA
The database includes 15,500 small data cubes (i.e., minicubes), each with a spatial coverage of 12x12km, spanning 1527 counties in the US. The minicubes comprise data from multiple sensors (Sentinel-2, Landsat-8, MODIS), weather and extreme events (Daymet, heat/cold waves, and U.S. drought monitor maps), as well as soil and terrain features, making it suitable for various agricultural monitoring tasks. It integrates crop- and climate-related tasks within a single, cohesive dataset.
In detail, the following data sources are provided:
Uses
The dataset allows various tasks, including yield prediction, phenology mapping, crop condition forecasting, extreme weather event detection/prediction, sensor fusion, pretraining on crop areas, and multi-task learning.
Dataset Structure
The data is stored in ZARR format in 12x12km minicubes, allowing compression, grouping, and automatically applying offsets and scaling when loading with Xarray. Xarray is recommended for reading the files to utilize the full functionality. Each modality is located in a different folder. The folder contains the grouped minicubes of each county.
The following counties are covered by the dataset (a) all data except Sentinel-2 (b) only Sentinel-2:
Dataset Creation
The minicube locations were optimized using a genetic and a sliding grid algorithm (details are provided in the paper). The following data providers were used to gather the data: Planetary Computer, Google Earth Engine, SentinelHub, and Copernicus Data Space Ecosystem. The APIs were accessed with terragon.
Citation
If you use this dataset, please consider citing: TBA