ABSTRACT
We present an assessment of several geospatial layers proposed as models for detecting clandestine graves in Mexico. The analyses were based on adapting the classical ROC curves to geospatial data (gROC) using the fraction of the predicted area instead of the false positive rate. Grave locations were obtained for ten Mexican states that represent the most conflicting regions in Mexico, and 30 layers were computed to represent geospatial models for grave detection. The gROC analysis confirmed that the travel time from urban streets to grave locations was the most critical variable for detecting graves, followed by nighttime light brightness and population density, whereas, contrary to the rationale, a previously proposed visibility index is less correlated with grave locations. We were also able to deduce which variables are most relevant in each state and to determine optimal thresholds for the selected variables.