Locusts are grasshopper species that exhibit phase polyphenism resulting in the expression of gregarious behaviors that favor the development of large devastating bands and swarms. Desert locust preventative management aims to prevent crop damage by controlling populations before they can reach high densities and form mass migrating swarms. The areas of potential gregarization for Desert locust are large and need to be physically assessed by survey teams for efficient preventative management. An ongoing challenge is to be able to guide where prospection surveys should occur depending on local meteorological and vegetation conditions. In this study, we analyzed the relationship between historical prospection data of Desert locust observations from 2005 to 2009 and spatio-temporal statistics of a vegetation index gathered by remote-sensing with the help of multiple models of logistic regression. The vegetation index was a composite Normalized Difference Vegetation Index (NDVI) given every 16 days and at 250 m spatial resolution (MOD13Q1 from MODIS satellite). The statistics extracted from this index were: (1) spatial means at different scales around the prospection point, (2) relative differences of NDVI variation through time before the prospection, and (3) large-scale summary of vegetation quantity. The multi-model framework showed that vegetation development a month and a half before the survey was amongst the best predictors of locust presence. Also, the local vegetation quantity was not enough to predict locust presence. Vegetation quantity on a scale of a few kilometers was a better predictor but varied non-linearly, reflecting specific biotope types that support Desert locust development. Using one of the best logistic regression models and NDVI data, we were able to derive a predictive model of probability of finding locusts in specific areas. This methodology should help in more efficiently focusing survey efforts on specific parts of the gregarization areas