Groundwater depletion in the Urucuia Aquifer System (UAS) has concerned the Brazilian water agencies since it is the principal water source in the arid length of the São Francisco river. In contrast, the irrigation fields increasing and uncertainty regarding climate variability have impaired the management of this important hydrograph region of Brazil. Therefore, the proper management of the UAS relies on the correct definition of the factors that are impacting most of the ground and surface waters, i.e., whether they are from climate or anthropogenic activities. Relying on Artificial Intelligence (AI) modeling and hydrological signatures, this study proposed a methodology to disentangle the role of climatic variability and groundwater pumping on streamflow dynamics by rebuilding natural time series in four watersheds within the UAS’s boundaries. By comparing the natural to the observed streamflow time series, the long-term 90th percentile (Q_90) discharge, from which 80% can be granted for multiple uses in the State of Bahia, decreased by up to 50% in the UAS. Yet, the watersheds’ productivity, given by the 90% specific yield, ranged from -52.3% (3.8 L/s.km2 to 1.8 L/s.km2) to -74.0% (23.4 L/s.km2 to 6.1 L/s.km2). Low flow duration was ∼ 1,650 days in natural conditions and increased to 10,354 days in the current land use and cover scenario. Changes in the maximum low flow duration length ranged from 191.7% to 1,315.7%, and in the low flow, deficits were up to 7,150%. These results highlight that groundwater pumping is the principal factor of the UAS’s dryness since climate variability could not track streamflow decreases. However, climate variability is secondary because of the intensification of the atmosphere demand. Although it is a Brazilian application, the proposed methodology can be applied in other aquifer systems to guide decision-makers' management strategies worldwide.