Vulnerability Index for croplands of Greece, with a twist

Agricultural systems in Mediterranean regions face increasing pressures from climate variability, particularly recurring droughts, extreme rainfall, and heat stress. This case study presents an integrated, fully open-source, and globally scalable methodology for assessing climate vulnerability in agricultural landscapes, using Greece as an illustrative example. Following the framework of the GIZ Vulnerability Sourcebook, vulnerability is conceptualized as the interaction of exposure, sensitivity, and adaptive capacity, allowing a holistic evaluation of how climatic hazards impact cropland systems. Exposure was quantified using openly accessible, global remote sensing indicators, including the Standardized Precipitation Evapotranspiration Index (SPEI) for drought intensity, CHIRPS precipitation for extreme rainfall detection, and MODIS Land Surface Temperature for thermal stress. Sensitivity was characterized using NDVI, SMAP soil moisture, SRTM terrain data, and JRC Water Occurrence, capturing variations in vegetation health, soil water availability, topography, and flood-prone areas. Adaptive capacity was approximated through WorldPop population density and VIIRS night-time lights, representing socio-economic resources and infrastructural robustness. All datasets used in this analysis are free, globally consistent, and regularly updated—ensuring that the approach remains transparent, accessible, and directly applicable to agricultural regions worldwide.

The workflow was implemented entirely within the Google Earth Engine (GEE) cloud environment, enabling efficient processing of multi-temporal, high-volume datasets. Each indicator was normalized and weighted using the Analytical Hierarchy Process (AHP), informed by expert judgments from the departmenet of Physics of the  National and Kapodistrian University of Athens. This produced spatially explicit Drought Vulnerability Index (DVI) and Flood Vulnerability Index (FVI) maps, revealing moderate to high vulnerability patterns across Greece (DVI: 0.14–0.84; FVI: 0.22–0.81). Combining these into a Composite Vulnerability Index (CVI) highlighted areas where drought and flood hazards overlap and intensify risks, especially in low-lying, intensively cultivated zones with limited adaptive capacity. To strengthen agricultural system characterization, the case study incorporated Google’s Satellite Embeddings, an open, globally available dataset offering 64-dimensional feature representations at 10 m resolution. These embeddings were paired with the Copernicus Crop Map (2021) to train a Random Forest classifier across 19 crop categories in the Larisa region. Using 3,774 samples, the model achieved an internal accuracy of 0.66 and a 0.90 agreement with Copernicus reference data (κ = 0.89), demonstrating strong performance for major crops such as wheat, maize, and olives. The results showcase the advantages of embedding-based feature spaces for scalable, transferable crop mapping across diverse agro-ecological settings.