Realistic representations and simulation of mass and energy exchanges across heterogeneous landscapes can be a challenge in land surface and dynamic vegetation models. For mixed life-form biomes such as savannas, plant function is very difficult to parameterise due to the distinct physiological characteristics of tree and grass plant functional types (PFTs) that vary dramatically across space and time. The partitioning of their fractional contributions to ecosystem gross primary production (GPP) remains to be achieved at regional scale using remote sensing. The objective of this study was to partition savanna gross primary production (GPP) into tree and grass functional components based on their distinctive phenological characteristics. Comparison of the remote sensing partitioned GPPtree and GPPgrass against field measurements from eddy covariance (EC) towers showed an overall good agreement in terms of both GPP seasonality and magnitude. We found total GPP, as well as its tree and grass components, decreased dramatically with rainfall over the North Australian Tropical Transect (NATT), from the Eucalyptus forest and woodland in the northern humid coast to the grasslands, Acacia woodlands and shrublands in the southern xeric interior. Spatially, GPPtree showed a steeper decrease with precipitation along the NATT compared to GPPgrass, thus tree/grass GPP ratios also decreased from the northern mesic region to the arid south region of the NATT. However, results also showed a second trend at the southern part of the transect, where tree-grass ratios and total GPP increased with decreasing mean annual precipitation, and this occurred in the physiognomic transition from hummock grasslands to Acacia woodland savannas. Total GPP and tree-grass GPP ratios across climate extremes were found to be primarily driven by grass layer response to rainfall dynamics. The grass-containing xeric savannas exhibited a higher hydroclimatic sensitivity, whereas GPP in the northern mesic savannas was fairly stable across years despite large variations in rainfall amount. The pronounced spatiotemporal variations in savanna vegetation productivity encountered along the NATT study area suggests that the savanna biome is particularly sensitive and vulnerable to predicted future climate change and hydroclimatic variability.