Environmental OSINT for conflict zones: a Sentinel Hub and HDX workflow

Why environmental OSINT matters

Armed conflict degrades soils, water, forests, and air long after the fighting subsides. Access to conflict zones is usually restricted, so satellites and public humanitarian datasets are often the only way to document what is happening at scale. The methodology below is designed for small newsrooms, university teams, and humanitarian analysts working without a commercial imagery budget.

It complements the country-level work on Humanity Centered Data, including the environmental impact of armed conflict hub and the Sudan, Ukraine, and Yemen environmental dashboards.

Sentinel Hub Custom Script

// Sentinel Hub Custom Script: dNBR (burn severity)
// Requires two Sentinel-2 L2A scenes as inputs.
function setup() {
  return {
    input: [{ bands: ["B08", "B12"], units: "REFLECTANCE" }],
    output: { bands: 1, sampleType: "FLOAT32" },
    mosaicking: "ORBIT",
  };
}
function evaluatePixel(samples) {
  const pre  = samples[samples.length - 1]; // earliest scene
  const post = samples[0];                  // latest scene
  const nbrPre  = (pre.B08  - pre.B12)  / (pre.B08  + pre.B12);
  const nbrPost = (post.B08 - post.B12) / (post.B08 + post.B12);
  return [nbrPre - nbrPost]; // dNBR: >0.27 = moderate/high burn
}

Python with rasterio and HDX admin boundaries

import rasterio, numpy as np

with rasterio.open("s2_pre.tif") as pre, rasterio.open("s2_post.tif") as post:
    nir_pre,  swir_pre  = pre.read(1),  pre.read(2)
    nir_post, swir_post = post.read(1), post.read(2)

def nbr(nir, swir):
    return (nir - swir) / (nir + swir + 1e-6)

dnbr = nbr(nir_pre, swir_pre) - nbr(nir_post, swir_post)

# Zonal stats against an HDX admin-boundary GeoJSON
import geopandas as gpd
from rasterstats import zonal_stats
admin = gpd.read_file("hdx_admin2.geojson")
stats = zonal_stats(admin, dnbr, affine=pre.transform,
                    stats=["mean", "max"], nodata=np.nan)
admin["burn_mean"] = [s["mean"] for s in stats]
admin.to_file("burn_by_admin2.geojson", driver="GeoJSON")

Requires rasterio, geopandas, and rasterstats. Admin polygons come from the HDX Common Operational Dataset (COD-AB) for the country of interest.

• Change-detection map. Pre and post NBR side by side with a dNBR panel and a legend keyed to burn-severity thresholds.
• Admin-level table. Zonal statistics of dNBR or NDVI change per admin2, joined to affected-population counts from HDX.
• Incident overlay. ACLED events plotted over the change raster in the same time window to show co-location of violence and damage.
• Reproducibility panel. Bounding box, sensor, acquisition dates, index formula, cloud mask threshold, and links to raw scenes.