Deliberate agricultural and food-system destruction mapping in conflict

Satellite time-series can distinguish deliberate crop destruction from drought or seasonal die-back by combining NDVI anomaly timing, SWIR char signatures and SAR-detected structure collapse, building on published UNOSAT and FAO assessments of Yemen, Syria and Ukraine.

Sensors

Sentinel-2 MSI: 10 m resolution in visible and near-infrared bands, 20 m in SWIR; 5-day revisit at the equator with both satellites. NDVI time-series detects sudden greenness loss; SWIR bands 11 and 12 (1610 nm and 2190 nm) resolve char signatures from field burning with high contrast against healthy or dry-senescent vegetation.
Sentinel-1 SAR C-band: 6 m (IW mode) to 20 m resolution; 6-day repeat at mid-latitudes, cloud-independent. Detects collapse of grain silos, storage sheds and irrigation pump stations through backscatter change and coherence loss. Cannot penetrate reinforced concrete roofs but reliably flags structural footprint loss.
Landsat 8/9 OLI: 30 m multispectral, 100 m thermal; 8-day revisit per satellite, 4-day combined. Longer archive (Landsat 5 onward) enables multi-year baseline construction for distinguishing conflict-period anomalies from historic drought cycles. Thermal band adds evidence of residual heat from recent burning.
Planet SuperDove: 3 m resolution, daily revisit over most land areas on tasking. Eight spectral bands including red-edge and NIR. Useful for confirming field-level damage and resolving ambiguity between mechanically harvested and burned fields at sub-parcel scale, though archive depth and coverage vary by licence.

Why timing relative to harvest is the key variable

Crop destruction that occurs at peak greenness, two to six weeks before expected harvest, is the single strongest indicator that damage is deliberate rather than incidental. Drought stress, disease and frost all produce NDVI decline, but they follow gradual, spatially continuous trajectories that respect soil-moisture gradients and topography. A field burned in July in a Mediterranean climate, when NDVI should be near its seasonal maximum, produces an abrupt single-scene drop of 0.4 to 0.7 NDVI units with a spatial boundary that is rectangular or follows field cadastre lines rather than contour lines. That pattern is not nature.

FAO and UNOSAT assessments of Yemen (2015 onward) and Syria documented exactly this signature: destruction concentrated in breadbasket governorates during the wheat and barley growing season, with timing inconsistent with any plausible natural cause. The Ukraine conflict has added a third large dataset, with Sentinel-2 time-series showing abrupt NDVI loss across Kherson and Zaporizhzhia oblasts in areas where satellite-derived crop calendars predicted imminent harvest.

What a SWIR char signature gives away

Fresh char from field burning has a distinctive spectral fingerprint in the shortwave infrared. Sentinel-2 band 12 (2190 nm) returns very low reflectance from charred organic matter, while band 11 (1610 nm) shows moderate reflectance. The ratio between these bands, combined with a collapsed NIR signal, produces a burn index value that separates char from dry bare soil, which tends to have higher and more uniform SWIR reflectance across both bands. The Normalised Burn Ratio (NBR = (NIR - SWIR) / (NIR + SWIR)) has been used in published wildfire mapping since the 1990s; the same physics applies to deliberate field burning, with the added interpretive step of checking whether the fire perimeter respects field boundaries.

One honest limit: char signatures fade within two to four weeks as wind redistributes ash and bare soil is exposed. Analysis must be conducted on imagery acquired within days of the event, or the spectral evidence degrades to an ambiguous bare-soil signal. This places a premium on near-real-time tasking and archive triage rather than retrospective analysis alone.

Silos, irrigation and the infrastructure layer

Destroying crops in the field is fast and visible. Destroying storage and water infrastructure is slower but more durable in effect. A grain silo that collapses denies food security for months; a breached irrigation headwork can render fields unproductive for years. Sentinel-1 SAR detects both.

Coherence change detection compares two SAR acquisitions six days apart. A standing metal-roofed silo produces a stable, high-coherence return; after collapse or demolition, coherence drops sharply and backscatter intensity changes in a manner inconsistent with weather or vegetation phenology. Published UNOSAT methodology for Syria applied this approach to storage facilities in Aleppo and Raqqa governorates. The method works at 6 m resolution in Interferometric Wide Swath mode, sufficient to resolve individual farm buildings of 15 m or larger footprint. Smaller structures, such as pump houses and field-level irrigation gates, sit at or below reliable detection thresholds and require corroboration from optical imagery or ground reporting.

Irrigation canal breaches are detectable indirectly: downstream fields that were irrigated in prior seasons lose their irrigated-crop spectral signature within one growing cycle, producing a spatial pattern of agricultural abandonment that follows canal service areas rather than rainfall isohyets.

Distinguishing deliberate destruction from the noise

The analytical challenge is not detection; it is attribution. Fires occur naturally in dry seasons. Crop failure from drought is common in the same geographies where conflict occurs. Three tests, applied together, raise confidence that destruction is deliberate.

First, spatial selectivity: deliberate burning tends to respect administrative or ethnic boundaries and to spare fields associated with one community while destroying adjacent ones. Second, timing: destruction during peak growing season or immediately pre-harvest has no agronomic explanation. Third, co-location with conflict events: cross-referencing agricultural damage polygons against ACLED conflict event data or UNOSAT incident databases identifies whether damage clusters around documented military activity rather than following drought probability maps derived from CHIRPS rainfall data.

None of these tests is individually conclusive. Together they build a documented, auditable evidentiary chain. Honest analysts should note that cloud cover in tropical or monsoon-affected conflict zones (parts of Myanmar, the Sahel) can create gaps of two to four weeks in Sentinel-2 coverage, during which evidence may be lost. Sentinel-1 SAR partially compensates but cannot recover spectral char signatures.

From pixels to accountability

The output of this analysis is not a map for its own sake. It is a dated, geo-referenced evidentiary record: which fields, which facilities, on which dates, with what spectral and structural evidence. That record is the input to humanitarian response planning, sanctions documentation and, increasingly, international criminal proceedings where satellite imagery has been admitted as corroborating evidence.

Satellize applies the same NDVI anomaly and burn-index methods it developed for the Kingdom of Tonga crop-estimation programme to conflict-affected agricultural zones, adding the SAR coherence layer and the conflict-event cross-reference that civilian food-security work does not require. Outputs are delivered as GIS-compatible polygon layers with attached metadata fields for date of acquisition, sensor, confidence class and the specific spectral index values that triggered the alert. Analysts at receiving organisations can interrogate the evidence directly rather than accepting a summary conclusion.

What this method cannot do

No satellite-based method can establish intent from orbit. It can establish fact: vegetation was destroyed here, on this date, in a pattern inconsistent with natural causes. Intent is a legal determination that requires this evidence in combination with command-structure documentation, communications intercepts and witness testimony.

Resolution also imposes limits on livestock facility assessment. Individual animals are not visible at Sentinel-2 or Sentinel-1 resolution. Facility destruction is detectable; whether animals were present is not. Planet SuperDove at 3 m can resolve vehicle presence and structural damage at larger livestock compounds, but even daily revisit does not guarantee cloud-free coverage on the specific day of an event. The evidentiary record will always have gaps. Documenting those gaps explicitly is as important as documenting what was observed.

Typical figures

Optical spatial resolution	10 m (Sentinel-2 visible/NIR), 20 m (Sentinel-2 SWIR), 30 m (Landsat 8/9), 3 m (Planet SuperDove)
SAR spatial resolution	6 m (Sentinel-1 IW mode ground range), 20 m (Sentinel-1 IW mode azimuth)
Revisit cadence	5 days (Sentinel-2 combined); 6 days (Sentinel-1 at mid-latitudes); 8 days per Landsat satellite, 4 days combined; daily (Planet SuperDove, subject to tasking)
Key spectral bands	NIR (842 nm), SWIR1 (1610 nm), SWIR2 (2190 nm) for NDVI and NBR; C-band 5.4 GHz for SAR coherence and backscatter
Minimum detectable field size	Approximately 0.5 ha at Sentinel-2 10 m; 0.1 ha at Planet SuperDove 3 m. Sub-field damage within larger parcels may be missed at Sentinel-2 resolution.
Char signature persistence	2 to 4 weeks before spectral contrast with bare soil becomes unreliable; analysis must be time-critical
Cloud limitation	Optical sensors fully blocked by cloud; Sentinel-1 SAR unaffected. In high-cloud regions, optical gaps of 2 to 4 weeks are common in growing season
Archive depth	Sentinel-2 from 2015; Landsat from 1972 (Landsat 1) enabling multi-decade baseline; Sentinel-1 from 2014
Delivery formats	GeoPackage or Shapefile polygon layers with per-feature metadata; GeoTIFF change-magnitude rasters; PDF evidence reports with scene thumbnails and index values

Analytics Satellize can run

Harvest-season NDVI anomaly map	Z-score deviation from 5-year Sentinel-2 NDVI baseline, masked against drought probability from CHIRPS rainfall anomaly to isolate conflict-related signal	GIS polygon layer with anomaly magnitude, date of detection and confidence class; updated per new cloud-free acquisition
Burn-scar classification with field-boundary overlay	Normalised Burn Ratio (NBR) thresholding on Sentinel-2 SWIR bands, intersected with cadastral or NDVI-derived field boundaries to test spatial selectivity	Classified raster and polygon export; field-by-field burn status table with acquisition date
Storage and silo structure-collapse detection	Sentinel-1 SAR coherence change detection (6-day interferometric pairs) combined with backscatter intensity delta; thresholds calibrated against UNOSAT published methodology	Point layer of flagged structures with pre/post backscatter values, coherence score and date range
Irrigation abandonment propagation map	Multi-season NDVI trajectory analysis downstream of detected canal breaches; comparison of irrigated-crop spectral signatures against canal service-area polygons derived from topographic data	Annual GIS layer showing progressive loss of irrigated agriculture by canal segment
Conflict-event cross-reference report	Spatial join of agricultural damage polygons against ACLED or UNOSAT incident databases; statistical test for spatial clustering relative to conflict events versus drought probability contours	PDF report with maps, statistical outputs and explicit confidence statements; suitable for submission to humanitarian or legal bodies
Longitudinal food-system damage chronology	Time-ordered stack of optical and SAR change events across a defined area of interest, assembled into a dated evidence timeline from archive imagery	Interactive timeline dataset with linked scene thumbnails and index values per event; GeoPackage archive

Sources

Who does the work

We can get this done for you. Satellize runs its own analyst desk and a strong science team. You do not buy a data feed and work out what it means; our people source the imagery, run the analysis described on this page, and hand you the answer with its confidence limits stated. Discuss this requirement.