saifeldinkhedir-coder/ecomine-observatory

Open-source multi-sensor satellite screening for the environmental footprint of mining across the Middle East and Africa. Google Earth Engine, GPL-3.0. # EcoMine Observatory **Open-source, multi-sensor screening tools for the environmental footprint of mining in the Middle East and Africa.** Built on Google Earth Engine and the opengeos stack (`geemap`, `leafmap`). All pixel computation happens on Earth Engine servers, so the tools run on ordinary hardware with no GPU. Author: Seifeldin Alkhedir (ORCID 0000-0003-0821-2991), University of Khartoum. Licence: GPL-3.0. Version: 0.1.0. > **Status: v0.1.0 — early release, single-site validation.** The tools run > and produce maps, change indicators, and — as of this release — a first > supervised-classifier accuracy figure. That figure (IoU 0.97 on the mine > class) is measured at **one site only** (Kangala, South Africa) and has not > been tested for transfer to any other site. Several thresholds are still > hand-chosen rather than derived from data, and are declared as such in the > provenance output. Treat every number as a screening result, not a verdict, > and read the caveats in `docs/RESULTS_LOG.md` before citing anything. --- ## What this is A set of transparent, reproducible tools that: - Pull imagery from **five sensors** for a defined area: **Sentinel-2** (10 m optical), **Sentinel-1** (C-band radar, cloud-piercing), **Landsat 8/9** (30 m optical, used for cross-validation), **SRTM** (terrain), and **Sentinel-5P** (atmospheric - *regional screening only*, since TROPOMI's 5.5 x 3.5 km pixel is larger than a single mine site; the regional protocol was verified in earlier work but its code is **not shipped in v0.1.0**) - Compute core spectral indices — NDVI, NDWI, MNDWI, BSI, NDMI — plus radar features (VV, VH, VV−VH) and terrain features (elevation, slope), 16 in all - Produce **change indicators** between two epochs: vegetation loss (VLI), surface exposure (SEI), water disturbance (WDI), an optional radar disturbance index (RDI, `--radar`), and a composite summary (MEI) - Attach **machine-readable provenance** to every number: which sensor, which date …