Resources

We have worked to develop multiple ways to access MOSAIKS features:

• Global Administrative Units (Planet imagery) You can download features aggregated to country, province, or municipality (ADM0/ADM1/ADM2 boundaries), as described in Sherman et al. (2023). These features are based on Planet imagery. These files are relatively small in size and can be used at these resolutions or to downscale administrative data. To download them, visit the ADM Aggregations page.

• Coarsened Global Grids (Planet imagery) You can download features aggregated to a coarsened grid, which may be useful for many global- or continent-level analyses. These files are relatively small in size and can be used when label data are also coarse. To download them, visit the Coarsened Global Grids .

• Global 0.01 x 0.01 degree grid (Planet imagery) You can download features for a complete and dense grid of global land areas via Redivis. These features are based on on quarterly mosaics from Planet’s Surface Reflectance Basemaps produce from 2019 Q3. Because the complete data set is large (multiple TB), you will need to query for custom subsamples of the imagery. To download them, register at Redivis and access the `mosaiks_2019_planet` table (see python demonstration notebook).

• USA grid from Rolf et al (Google basemap imagery) You can download features for a set of locations across the United States, as described in Rolf et al. (Nature Communications, 2021). These features are based on imagery from the Google Earth base map. You can download the features from the Code Ocean Capsule associated with that manuscript (the capsule will also allow you to replicate the analysis of that paper on a remote machine). The Github repository for that analysis is here.

• Recompute MOSAIKS features (Landsat & Sentinel imagery) You can recompute MOSAIKS features yourself using Microsoft’s Planetary Computer (Github repo which currently supports Gaussian random convolutional features). This approach will not provide the benefit of pre-computed features, since you will recompute features on-the-fly every time, but the massive compute power of the Planetary Computer makes this relatively fast and cheap for users.

More Links:

• A Python notebook showing how to query native resolution grid features.

• A slidedeck describing the Rolf et al. paper.

• An R notebook illustrating the MOSAIKS pipeline

• More resources are currently in development. Please check back soon or contact mosaiksteam@gmail.com with questions.
  

Don’t forget to see our Tutorial Page here, which has an example Python notebook that we walk through in the video.

If you are looking for new data sets that we create using MOSAIKS (not features), we will be posting those here.