We make detailed, high resolution maps of African croplands by combining human and machine intelligence with new satellite imaging capabilities.
Are there any new fields?
Were any fields lost?
Did field sizes change?
A custom procedure for converting daily, high- resolution imagery collected by small satellites into cloud-free seasonal composite images
A web-based platform that allows rapid collection of training data, with built-in quality control procedures for reducing labeling errors.
We train a model to identify crop fields in the imagery, iteratively refining it with new labels collected from areas where its predictions are least certain
An algorithm converts predicted cropland probabilities into individual field boundaries.