Farmland trees in India (2018-2022)

This dataset presents a detailed analysis of individual tree changes within farmlands across India for the years 2018 to 2022. Utilizing PlanetScope satellite images, isolated trees were mapped for each year. We used the farmland class of WorldCover to keep only trees falling into this class, which can cause unexpected patterns (e.g. large trees not mapped as they were not mapped as farmland).

Each tree has the detection confidence of each year as an attribute. The change confidence is the aggregated confidence over 5 years and can be used as a measure of uncertainty in the detection. Trees that have detection confidence values below 0.5 are likely shrubs or misclassification, or the image quality was low. If a tree was detected in both 2018 and 2019, but not in 2020-2022, it has likely disappeared.

File system and download

The dataset contains about 0.5 billion trees, saved in the 353 files in the format of geopackage as gpkg, grouped into 89 zip folders for zenodo ingestion. The spatial coverage of files follows a grid system with identical gridIDs, which can be located in the file name, e.g., ps2_PSScene_2018-2022_gridIDs_195_308_000_0000_composite_lshm_0_0_0.gpkg. The filename and the associated group name can be found in the file of india_grid_bygroup.geojson. The geojson file can be opened via QGIS. One can find the group name using the 'Identify Feature' tool for a target grid covering your area of interest. It turns for example, id: 85,24, file: ps2_PSScene_2018-2022_00085_00024_260_391_000_0000_composite_lshm_0_0_0.gpkg, and groupname: group_62.zip.

Web Viewer

We also developed a viewer to explore the tree detection confidence: https://rs-cph.projects.earthengine.app/view/tree.

License

Any usage must be solely for Noncommercial education or scientific research purposes, and publication in academic or scientific research journals. Licensee agrees that all such publications must include an attribution that clearly and conspicuously identifies Planet Labs PBC.