Societal drought vulnerability and the Syrian climate-conflict nexus are better explained by agriculture than meteorology

The study examines whether the 2007–2009 Middle East drought produced an agricultural collapse in Syria that contributed to migration and the 2011 conflict. Prior work often relies on meteorological proxies to infer agricultural impacts, leading to debate about the drought–migration–conflict nexus in Syria. The authors argue for moving beyond weather data to directly observe agricultural activity. Using satellite-based time series on cropland dynamics from 2000–2016, they pose one overarching research question: How did the 2007–2009 drought affect the Syrian agricultural system right before the conflict outbreak? Three subquestions guide the analysis: (1) What were the effects of climate variability on agriculture, approximated via cropland extent changes? (2) How did vulnerability of Syrian agriculture to drought and seasonal dryness change between 2000 and 2016? (3) Do observed cropland dynamics meet criteria of an agricultural collapse? The work is motivated by the need to clarify the mechanisms linking drought, agricultural productivity, migration, and conflict in a context of limited subnational agricultural data.

Research has linked weather extremes to conflict risk, often proposing agricultural production shocks and migration as mediators. However, many studies rely solely on meteorological indicators and assume widespread crop failure. The Syrian case has been widely cited as a climate-induced conflict where severe 2007–2009 drought allegedly caused agricultural collapse, mass rural–urban migration, and contributed to the 2011 uprising. Yet evidence is mixed, often anecdotal, and some studies suggest unrest was concentrated in less drought-affected areas and migrants were not primary protestors. Pre-drought structural stressors in Syria included groundwater depletion, salinization, subsidy removals (fuel 2008, fertilizer 2009), and a 2010 yellow rust outbreak, potentially increasing vulnerability. The concept of agricultural collapse is clarified via social–ecological regime shift criteria: abrupt change, substantial loss, persistence, and structural change. The literature gap concerns direct observation of agricultural impacts rather than inference from meteorology alone.

Cropland dynamics: Annual land use/land cover mapping for Iraq and Syria (28.35–37.84°N, 35.29–49.14°E) from 2000–2016 was produced using MODIS Terra 250 m surface reflectance (MOD09Q1) to derive 8-day NDVI time series, smoothed with a 30-day moving average. Phenological separability underpins classification. Training data: 1,573 visually interpreted training points across 2003, 2007, 2013, and 2015 using Landsat composites and MODIS phenology. Classification: Two-stage Random Forest. Stage 1: 500 trees with 66 inputs (46 NDVI time steps + 20 spectral-temporal metrics), rank variable importance. Stage 2: select top 15 features to train final model and predict classes annually (2000–2016), distinguishing single- and double-cropped areas and other land covers. Fallowness: Define a pixel’s “normal cropland extent” as the modal class over 2000–2016. For each year, fallowness is the area within normal cropland that is not actively cropped (bare/fallow) that year. Accuracy assessment: Independent validation sampled 1,700 points in stable cropland and 1,003 points in change-prone areas across years, with visual interpretation using MODIS phenology and Google Earth. Average overall accuracy ≈90% (SD 6%) across annual maps; class-wise user’s/producer’s accuracies generally high; 2014 single-year assessment showed overall 80% (single-crop producer 95%, user 66%; double-crop producer 74%, user 92%). Confidence intervals on area estimates computed following best practices. Statistical analysis of dryness–fallowness: Construct a 0.5°×0.5° grid (65 cells with cropland in Syria), annual panel 2000–2016 (1,105 observations). Dryness measure: SPEI-6 (October–March), panel-standardized, from CRU via PRIO-GRID; captures combined precipitation–PET effects. Model: spatial Durbin unit fixed-effects (with and without year fixed effects), maximum likelihood estimation, including spatial lag of dependent variable and spatial lag of SPEI to account for spatial dependence. Test for changing sensitivity across periods (2000–2007 vs 2008–2016; conflict years 2011–2016, 2013–2016), differences by irrigation reliance (AEI-EARTHSTAT_IR; share equipped for irrigation, square-root transformed) and northeastern governorates (Deir ez-Zor, Aleppo, Raqqa, al-Hasakah), and effects of local violence. Sensitivity analyses included detrending, excluding cells with minimal cropland, and adding lagged dependent variables. Livestock: National FAOSTAT livestock numbers assessed qualitatively in SI to compare patterns with cropland activity.

- Cropland activity varied widely; median annual fallowness was 21% (2000–2016). Years 2000 and 2008 had about 50% fallowness, while 2010 and 2013 had low fallowness (~10%). Alleged drought years 2006–2007 showed below-median fallowness (15%–16%). - The 2007–2009 drought peak (2008) coincided with high fallowness (~50%), followed by partial recovery in 2009 (~25% fallowness) and near-full recovery in 2010 (~90% active cropland). - Spatially, 2008 fallowness was widespread, particularly in Al-Hasakeh. Recovery prioritized high-intensity farming areas; marginal/less frequent cropping areas in the northeast remained more fallow in 2009. - Permanent abandonment was minimal: only 113 km² (0.5% of normally active cropland) remained fallow for five consecutive years after 2008 (2008–2012). - Livestock numbers declined during the drought but recovered thereafter (national FAOSTAT), mirroring cropland dynamics. - Dryness effect: For 2000–2016, a 1 SD decrease in SPEI-6 increased fallowness by ~11–12%. Pre-2008 (2000–2007): 1 SD decrease increased fallowness by ~7–8%. Post-2008 (2008–2016): ~13–18%. Conflict years 2013–2016: ~17% (fixed effects) to ~22% (two-way fixed effects) increase per 1 SD decrease, though not always statistically significant. - Irrigation: No strong evidence that irrigated areas became more vulnerable after subsidy cuts; areas with below-median irrigation showed slightly larger sensitivity increases. - Northeastern governorates resembled national patterns. Only weak indications that local violence increased drought sensitivity. - Collapse assessment: Abruptness (criterion i) observed in 2008; substantial loss (ii) relative to past events is questionable given similar fallowness in 2000 and recovery by 2002; persistence (iii) not met due to rapid recovery by 2010; structural change (iv) partially indicated by increased drought vulnerability post-2008, likely due to subsidy cuts, unsustainable practices, and conflict-related factors rather than the drought alone. - Migration: Patterns are more consistent with temporary, adaptive migration during drought with return by 2010 rather than widespread permanent rural–urban migration leading to long-term abandonment.

The findings show that the 2007–2009 meteorological drought caused a sharp but temporary shock to Syrian agriculture rather than a sustained collapse. Rapid cropland recovery by 2010, minimal long-term abandonment, and livestock rebound contradict the narrative of a pre-conflict agricultural collapse driving mass permanent migration. While sensitivity of agriculture to dryness increased after 2008, particularly during conflict years, attribution points more to structural and policy factors (fuel and fertilizer subsidy removals), pre-existing resource degradation, and conflict-related disruptions than to the drought itself. This underscores the need to directly measure agricultural impacts rather than infer them from meteorological proxies when evaluating climate–conflict linkages. The results refine the drought–migration–conflict narrative by highlighting agricultural system resilience, the priority of high-intensity cropland in recovery, and limited evidence for violence amplifying drought sensitivity at the local scale.

The study demonstrates that Syria’s agriculture experienced a drought-induced shock in 2008 but not a system-wide collapse; croplands largely recovered by 2010 and long-term abandonment was minimal. Agricultural vulnerability to dryness increased after 2008, likely reflecting policy changes, unsustainable practices, and conflict impacts. These insights challenge simplified drought–migration–conflict narratives and emphasize the importance of observing land-use dynamics to understand climate-related risks. The authors advocate remote sensing approaches to track agricultural activity at high temporal and spatial resolution, enabling more robust links between meteorological droughts, agricultural impacts, and societal outcomes. Future research should extend dense time-series mapping to earlier decades, integrate socioeconomic and migration data, and focus on resilience-building strategies for drought-prone rural communities.

- Cropland dynamics may not capture the full agricultural system (e.g., livestock and grazing), and fine-scale livestock data were unavailable; livestock analyses were national and qualitative. - Migration data are scant and uncertain, limiting direct testing of drought–migration pathways. - Classification uncertainties remain despite high average accuracy (~90%); single-year 2014 accuracy was 80%, and early-year estimates (e.g., 2001) had wider confidence intervals. - Coarse 0.5° grid for meteorological linkage may mask finer-scale heterogeneity. - Irrigation data reflect areas equipped circa 1990, potentially misrepresenting conditions in the 2000s. - Statistical attribution of increased vulnerability to local violence is weak; multiple contemporaneous shocks (subsidy cuts, pests, conflict) complicate causal inference. - Mild temporal trends required sensitivity checks; models without lagged dependent variables avoid Nickell bias but leave potential dynamics unmodeled.