Desert locust detection using Earth observation satellite data in Mauritania

Desert locust plagues have threatened food security in northern African countries for centuries. To prevent their effects, current early warning systems in arid environments need to be improved using the latest and most advanced modelling techniques and Earth observation datasets. Previous studies have analysed certain environmental predictors such as NDVI or soil moisture individually in an effort to detect suitable areas. However, we introduce new variables (Surface Temperature, LAI and Soil Moisture Root Zone) from the SMAP satellite and apply different machine learning methods in our species distribution model in order to identify desert locust presence. We obtain highly satisfactory model results (KAPPA & TSS = 0.901 and ROC = 0.986) to detect the probability of presence and, hence, likely breeding areas based on environmental factors. The most relevant variables were surface temperature, NDVI and soil moisture at root zone under different time scenarios. This study also confirms the potential of the SMAP satellite to retrieve critical temperatures due to its time pass, in addition to reinforcing the NDVI product from MODIS as a reliable environmental predictor. These results demonstrate the validity of this new approach based on machine learning methods to identify favourable breeding areas in Mauritania.