Water-limited ecosystems play a fundamental role in the global carbon cycle, yet they have been increasingly affected by pressures from aridification and land-use and land-cover change. The Caatinga, the largest seasonally dry tropical forest in South America, exhibits strong ecohydrological coupling resulting from centuries of adaptation, in which vegetation productivity is closely linked to precipitation pulses. However, distinguishing the driving vectors of degradation remains a challenge. This study aimed to model the net primary productivity (NPP) of the Caatinga from 2004 to 2024, analyse its trends and spatially decouple the dominant drivers. The Carnegie-Ames-Stanford approach (CASA) model was implemented on the Google Earth Engine platform, integrating Theil-Sen and Mann-Kendall tests with a hierarchical attribution analysis. Although the overall trend was statistically neutral, the spatial analysis revealed contrasting processes of degradation and restoration. Driver decoupling showed the spatial dominance of precipitation across the biome (56.4%); however, thermal stress emerged as a limiting factor with significant impact over large portions of the territory (26.1%). Land-use and land-cover change represented acute vectors of landscape alteration (17.4%). In conclusion, the Caatinga currently faces a dual threat: chronic thermal stress driven by climate change and anthropogenic fragmentation of the biome. Conservation strategies should focus on restoring vegetation to re-establish the ecohydrological resilience of the Caatinga in the face of aridification and desertification.