Agriculture is a cornerstone of the Moroccan economy: it accounts for about 15% of the country’s GDP and employs over 40% of the population. Despite its importance, the sector is increasingly threatened by rising temperatures, erratic rainfall, and increasingly severe droughts. Smallholder farmers, who make up the majority of rural households, often struggle to recover from these climate shocks due to structural constraints such as limited access to irrigation facilities and formal climate information services. While traditional vulnerability analyses typically focus on objective climatic or economic indicators, they often overlook the subjective dimensions of sensitivity and adaptability. This study argues that a more holistic approach is needed—one that combines large-scale environmental monitoring with farmers’ “lived experiences” and local knowledge to develop more effective adaptation strategies.

Integrating Earth Observation and Local Knowledge for Climate Adaptation

The study focuses on the most productive regions in northern and central Morocco, which are crucial for national food security but are highly sensitive to seasonal climate changes. To address this complexity, the researchers combined Earth observation (EO) indicators from 2010 to 2024—including precipitation, temperature, and vegetation cover (NDVI)—with georeferenced survey data from 3,591 smallholder farmers. These datasets were analyzed using advanced machine learning algorithms, specifically Random Forest and XGBoost, to predict perceived vulnerability levels. A key component of the methodology was the use of spatial block cross-validation, which accounts for geographic clustering to provide more realistic and policy-relevant performance estimates than standard random validation. Additionally, the study employed SHAP (SHapley Additive Explanations) to translate these complex “black-box” models into results that policymakers can interpret.

The Dynamic Nature of Agricultural Risk and Model Performance

The modeling results show that vulnerability is an extremely dynamic process, with summer models achieving the best predictive performance (F1 scores > 0.70). A key finding is the high explanatory power of perception-based data; models relying exclusively on survey data consistently outperformed those using only satellite indicators. However, integrating both data sources yielded additional benefits, particularly when assessing vulnerability in winter. Spatial analysis revealed that while precipitation variability is highest in the dry southern zones, NDVI variability is most concentrated in the northern croplands, where cropping cycles and land use are more dynamic. These patterns provided a solid foundation for identifying the specific socio-ecological factors that influence how farmers assess their own risk.

From Climate Exposure to Adaptive Capacity: A Seasonal Perspective

The study reveals a clear seasonal difference in the factors influencing perceived risk. Vulnerability in winter is primarily linked to environmental factors such as hydroclimatic fluctuations and soil moisture, reflecting the dependence of rain-fed crops like wheat and barley on seasonal precipitation. In contrast, vulnerability in summer is more strongly determined by adaptability and management decisions. During the dry season, farmers’ access to groundwater, irrigation practices, and knowledge of climate-smart agriculture become critical factors in risk mitigation. Furthermore, the study identified a “digital divide” regarding climate information services: While younger, better-educated farmers prefer to receive alerts via mobile apps, those with lower levels of education continue to rely on traditional channels such as radio, mosques, or local advisors. This suggests that national adaptation strategies must employ multi-channel communication to be truly inclusive.

The study shows that assessing climate vulnerability in semi-arid systems requires a multidimensional approach that goes beyond biophysical exposure. The results confirm that Earth observation data complement local knowledge but do not replace it. By linking participatory data with remote sensing data, the proposed framework enables a more nuanced interpretation of climate risks that reflects the actual challenges smallholder farmers face throughout the agricultural annual cycle. Ultimately, building agricultural resilience in Morocco—and similar semi-arid regions—requires policy measures that promote crop diversification, improve soil and water management, and prioritize the participation of farmers most affected by climate change.