MOSAIKS

mosaiksteam@gmail.com Combining satellite imagery with machine learning (SIML) has the potential to address global challenges by remotely estimating socioeconomic and environmental conditions in data-poor regions, yet the resource requirements of SIML limit its accessibility and use. The mission of MOSAIKS is to make SIML more accessible by making the process simpler and easier. Using MOSAIKS, you can make predictions in areas of interest in five steps: 1. Download MOSAIKS features from this API for the areas where you have labels. 2. Merge the features spatially with your own ground truth information (called “labels”) 3. Run a regression of your labels on the MOSAIKS features 4. Evaluate performance 5. Make predictions in a new area of interest, downloading additional features as necessary. We’ve found that MOSAIKS, though simple, works well across diverse prediction tasks (e.g. forest cover, house price, road length). And, it’s fast; MOSAIKS achieves accuracy competitive with deep neural networks at orders of magnitude lower computational cost (Rolf et al., 2021). Additional tutorial materials on how to use MOSAIKS can be found at mosaiks.org. ## Downloading features The native resolution features are organized using a 0.01 x 0.01 degree latitude-longitude global grid, centered at .005 degree intervals. Features have been created from a 2019 Quarter 3 composite image of the earth from Planet Labs . You will generally receive features in a tabular **.csv** format. Each row represents a unique grid cell (or administrative unit), with the first two columns representing latitude and longitude coordinates (or the administrative unit code), and subsequent columns representing *K* features (for now, there are *K* = 4000 features). ## Obtaining features using the Coarsened Global Grids We offer MOSAIKS features for the globe at coarsened resolutions that are easy to download. The advantage of using these files, is that they provide rich information globally and are relatively small in file size. For many users intending to experiment with the platform, these grid files may be a great place to start. Currently, we offer 1 x 1 degree, 0.25 0.25 degree, and 0.1 x 0.1 degree coarsened grids. These aggregations are available with area weights as well as population weights. **Proceed to **[**Coarsened Global Grids**]() ## Obtaining features using the Administrative Region Aggregations We offer MOSAIKS features that are aggregated to the country (ADM0), state/province (ADM1), and county/municipality (ADM2) levels. A significant amount of administrative data is only available when aggregated up to these political units. For many users using label data for ADM units, these files may be all that is needed. Just as with the Global Grids, these administrative unit aggregations are available with area weights as well as population weights. These data are also what is used to produce the results of [Sherman et al., (2026)]() For more information on administrative unit aggregations, see [Sherman et al., (2026).]() **Proceed to **[**Administrative Region Aggregations**]() ## Dense grid methods (advanced users) More advanced users, may want native resolution grid files (0.01 x 0.01 degree resolution). Users can query for these files using directly using Redivis. More information on these query methods will be added soon. Data download limits may apply. For questions, contact [mosaiksteam@gmail.com](). When referring to the MOSAIKS methodology or when generating MOSAIKS features, please reference “A generalizable and accessible approach to machine learning with global satellite imagery.” Nature Communications (2021) You can use the following Bibtex: `@article{article, author = {Rolf, Esther and Proctor, Jonathan and Carleton, Tamma and Bolliger, Ian and Shankar, Vaishaal and Ishihara, Miyabi and Recht, Benjamin and Hsiang, Solomon}, year = {2021}, month = {07}, pages = {}, title = {A generalizable and accessible approach to machine learning with global satellite imagery}, volume = {12}, journal = {Nature Communications}, doi = {10.1038/s41467-021-24638-z}}`