This is an article coauthored with Dr. Jen Burney and Dr. Ran Godlblatt at the UC San Diego’s School of Global Policy and Strategy and UC San Diego’s Big Pixel Initiative. We used high 3-band and medium multi-spectral resolution satellite images to map tree cover in the Brazilian Sertāo. We used random forest-based supervised machine learning to determine the classification accuracy of this method and compared the results with both types of images. The results show that using high-resolution imagery with only RGB bands yields better results than medium-resolution imagery with 11 bands such as Landsat 8.
These results can help firms and policy makers decide the technology to use when assessing land cover with satellite imagery and evalute the trade-offs of their costs.