Agricultural land abandonment and food-system disruption in conflict areas
Conflict disrupts agriculture through displacement, deliberate destruction, and fear of movement. Time-series analysis of Sentinel-2 NDVI and Sentinel-1 backscatter can detect uncultivated fields at 10 m resolution, separating war-driven abandonment from drought or voluntary fallow.
Sensors
- Sentinel-2 MSI: 10 m resolution in red and near-infrared bands used to compute NDVI; 5-day revisit at equator with both satellites. The primary tool for detecting absence of seasonal green-up in fields as small as roughly 0.1 ha. Cloud cover is a persistent constraint in humid theatres.
- Sentinel-1 SAR C-band: 6-day repeat in IW mode at 10 m range resolution. Detects tillage and crop emergence as backscatter increases independent of cloud or smoke. Particularly useful in Syria and Yemen where optical gaps from cloud are less common but dust haze is problematic.
- MODIS Terra/Aqua: 250 m NDVI products (MOD13Q1, MYD13Q1) at 16-day composites provide a 20-plus year baseline stretching back to 2000. At 250 m, a field must be at least roughly 6 ha to avoid severe mixed-pixel contamination. Useful for regional trends and anomaly baselines, not field-level attribution.
- Landsat-9 OLI-2: 30 m resolution with a 16-day revisit, bridging the scale gap between MODIS and Sentinel-2. The continuous archive back to 1972 (Landsat 1 through 9) is irreplaceable for establishing pre-conflict baselines in areas where Sentinel-2 data begins only in 2015.
What a missing harvest looks like from orbit
A cultivated field has a rhythm. Bare soil after ploughing gives way to low green cover, a NDVI peak during the vegetative maximum, then senescence. That seasonal pulse is legible in multispectral time series. When conflict disrupts farming, the pulse stops: NDVI stays near bare-soil values through a growing season that should show a peak of 0.5 to 0.7 or higher for healthy cereal crops. The absence is the signal.
SAR adds a complementary view. Tillage roughens the soil surface, raising C-band backscatter by several decibels relative to smooth fallow. Crop emergence adds volume scattering. When neither event occurs in a season where both should, the backscatter time series stays anomalously flat. Published work on Syria (Eklund et al., Remote Sensing of Environment, 2017) used exactly this combination to map abandonment at governorate scale, finding that irrigated area in some provinces fell by more than half between 2010 and 2014.
Three causes that look similar from space and why the distinction matters
Abandonment driven by displacement, deliberate agricultural destruction (scorched-earth tactics, irrigation infrastructure bombing), and drought-induced crop failure can all produce the same flat NDVI signature in a single season. Getting the cause wrong leads to the wrong humanitarian response.
Drought can be ruled out, or confirmed, by cross-referencing with rainfall estimates from GPM IMERG or CHIRPS and with the Vegetation Condition Index derived from the long MODIS baseline. If rainfall is near-normal but NDVI is suppressed, drought is not the primary driver. Deliberate destruction leaves physical traces: Sentinel-1 coherence loss over irrigation infrastructure, or optical evidence of burned field margins. Displacement-driven fallowing tends to be spatially correlated with population movement data from UNHCR or mobile-network sources. No single satellite layer resolves the ambiguity alone; the analysis is always a convergence of evidence.
This matters operationally. Food-system disruption from displacement calls for supply-chain intervention and cash transfers. Destruction of irrigation infrastructure calls for engineering assessment and reconstruction funding. Conflating them wastes scarce resources and can actively mislead donor allocation.
Resolution floors and what they mean for field-level analysis
At Sentinel-2's 10 m resolution, a square field of roughly 0.1 ha (about 32 m per side) occupies only around ten pixels. Mixed-pixel effects at field boundaries mean the practical minimum for reliable NDVI time-series classification is closer to 0.5 ha, depending on field geometry and surrounding land cover. Most agricultural plots in the Middle East and Central Asia exceed this threshold, but smallholder plots in parts of sub-Saharan Africa or terraced hillside agriculture in Yemen can fall below it.
MODIS at 250 m is a different instrument for a different question. A single MODIS pixel covers 6.25 ha. It is appropriate for regional anomaly detection and for constructing the long baselines needed to separate inter-annual climate variability from conflict-driven change, but it cannot attribute abandonment to individual farms or villages. Landsat-9 at 30 m sits between the two: adequate for field-level work in landscapes with medium to large plots, and carrying an archive that predates the Syria and Yemen conflicts by decades.
Cloud cover is the operational constraint that does not appear in sensor specification sheets. In the Sahel or the Levant, optical gaps of two to four weeks are manageable with 5-day Sentinel-2 revisit. In the Ethiopian highlands or parts of South Asia, monsoon cloud can obscure an entire growing season. SAR becomes the primary sensor in those cases, with optical used only for cloud-free windows.
Building a defensible baseline in an active conflict zone
Anomaly detection is only as credible as the baseline it compares against. The standard approach uses five to ten years of pre-conflict growing-season composites to define the expected NDVI or backscatter range for each pixel, accounting for inter-annual rainfall variability. For Syria, pre-2011 Landsat and MODIS data provide a reasonable baseline. For Ukraine, Sentinel-2 data from 2017 onwards gives a cloud-free, high-resolution reference for the pre-2022 agricultural calendar.
The baseline must be climatically normalised. A dry pre-conflict year used as the reference will make abandonment appear smaller than it is; a wet year will exaggerate it. Published methods from the FAO and academic groups working on Ukraine use the z-score of growing-degree-day-adjusted NDVI to account for this. The University of Maryland GLAD laboratory has published related methodology for cropland monitoring that is directly applicable.
Access for ground-truth validation is, by definition, limited in active conflict zones. Satellite-derived abandonment maps should carry explicit confidence tiers: high confidence where SAR and optical agree and climate anomaly is ruled out, medium confidence where only one sensor is available or cloud gaps are significant, and low confidence where the signal is ambiguous. Presenting a single unqualified map is analytically dishonest.
From abandonment maps to food-system impact estimates
Mapping which fields are uncultivated is the first step. Translating that into food-system impact requires coupling the abandonment area with crop-type maps, yield models, and population exposure data. The FAO-GIEWS and WFP VAM units have published frameworks for this conversion, and the methodology is well-documented in the public domain.
Satellize applies time-series anomaly detection on open constellations for sovereign clients and humanitarian partners. The Tonga crop-estimation programme demonstrated the core pipeline: multi-year Sentinel-2 NDVI baselines, per-field phenological modelling, and area-weighted yield proxies delivered as GIS layers. The same pipeline, adapted for conflict-zone constraints including cloud fallback to SAR and explicit confidence tiering, is the operational basis for conflict agricultural monitoring.
One honest limit: satellite data measures cultivated area, not yield. A field that appears green may be under-fertilised, poorly irrigated, or planted with a lower-value crop than the baseline year. Area-based food-security estimates therefore carry an additional uncertainty layer that remote sensing alone cannot resolve. Coupling with market price data and household survey results, where available, substantially tightens the estimate.
Typical figures
| Primary optical resolution | 10 m (Sentinel-2 MSI, red/NIR bands used for NDVI) |
| SAR resolution | 10 m range x 10 m azimuth (Sentinel-1 IW mode, ground range detected) |
| Regional baseline sensor | 250 m (MODIS MOD13Q1/MYD13Q1 NDVI composites) |
| Optical revisit | 5 days at equator (Sentinel-2A + 2B combined); 16 days (Landsat-9) |
| SAR revisit | 6 days (Sentinel-1 IW, single satellite) |
| Minimum detectable field (Sentinel-2) | Practical threshold approximately 0.5 ha for reliable NDVI time-series classification; theoretical minimum approximately 0.1 ha (10 pixels) |
| Minimum detectable field (MODIS) | Approximately 6 ha (one 250 m pixel); mixed-pixel effects significant below 25 ha |
| Archive depth | Sentinel-2 from 2015; Landsat continuous from 1972; MODIS from 2000 |
| Cloud constraint | Optical unusable under cloud; SAR C-band cloud-penetrating but sensitive to surface moisture changes that can mimic crop signals |
| Baseline period recommended | 5 to 10 pre-conflict growing seasons, climatically normalised |
Analytics Satellize can run
| Seasonal abandonment anomaly map | Per-pixel z-score of growing-season NDVI against multi-year baseline, thresholded and spatially filtered | GeoTIFF and vector polygon layer with confidence tier (high/medium/low), updated each growing season |
| SAR tillage and emergence absence detection | Sentinel-1 C-band backscatter time-series change detection; expected tillage-season increase compared against observed values | Binary change layer per agricultural calendar window, cloud-independent, delivered as GIS layer |
| Cause-attribution report | Convergence of NDVI anomaly, SAR backscatter, GPM/CHIRPS rainfall anomaly, and infrastructure coherence loss to separate drought, displacement, and deliberate destruction | Structured PDF report with per-zone attribution confidence and supporting evidence layers |
| Abandoned area time series | Annual growing-season composites from Sentinel-2 and Landsat-9, classified against crop mask; area statistics extracted per administrative unit | CSV time series and dashboard feed, updated annually or per growing season |
| Food-production shortfall proxy | Abandoned area multiplied by crop-type-specific yield coefficients from FAO-GAEZ or national statistics; uncertainty range reported explicitly | Tabular estimate of caloric production loss by district, with confidence intervals, for WFP/FAO integration |
| Recovery monitoring layer | Year-on-year comparison of NDVI peak timing and magnitude to detect partial or full return to cultivation in post-conflict or ceasefire periods | Annual update GIS layer showing recovery status per field polygon, suitable for reconstruction planning |
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.