This repo contains scripts for the supervised classification of land cover in Iraq, Lebanon, and Syria.
In modis-bootstrap, sources which enable bootstrapping MODIS land cover classifications onto Landsat imagery are provided. These follow a workflow derived from https://doi.org/10.1016/j.rse.2017.05.024. The python scripts sitting at the root of this directory clean, extract, and process data ready for training. At the top of each script, a short description of its purport is provided. The gee-scripts directory contains javascript necessary to export all imagery from Google Earth Engine which is required. The gridsearch contains helper functionality for randomly traversing hyperparameter search spaces in a few popular classifier algorithms.
In binary-classification, a series of Google Earth Engine javascript files are provided (each, followed by a link to the script in Google Earth Engine) which enable an interactive approach to land classification that is a fair bit simpler than the bootstrapping technique and can be iteratively improved through the Earth Engine interface. Supplementary scripts, contained in the supplementary directory, support export of EVI changes, night light changes, and deforestation estimates