Divergence in global plant carbon use efficiency across data-driven estimates, satellite product, and process-oriented models

• High-latitude CUE patterns diverge, with the data-driven showing lower values. • CUE trends vary: MODIS up, the data-driven down, DGVMs show no clear change. • Mean annual temperature mainly drives CUE, with method-specific effects. Plant carbon use efficiency (CUE)—the proportion of net to gross primary productivity (NPP/GPP)—is a critical indicator of terrestrial carbon cycling, yet its global patterns and trends remain uncertain due to challenges in directly measuring carbon flux. Here, we compared global plant CUE estimates using three methods: (1) a data-driven estimate derived from existing GPP, ecosystem respiration and soil heterotrophic respiration datasets (hereafter referred to as data-driven CUE), (2) MODIS satellite product, and (3) dynamic global vegetation models (DGVMs). We found significant spatial and temporal divergences in CUE emerged among the three methods. Specially, the data-driven method showed lower CUE in northern high latitudes and higher CUE in sub-Saharan Africa, India, and northern Australia, while MODIS and DGVM estimates exhibited opposing patterns. Temporally, MODIS data indicated a global CUE increase during 2001–2012, data-driven method suggested a decline, and most DGVMs showed no significant trend. Machine learning-based spatial attribution showed that mean annual temperature primarily drove spatial CUE variability among the three approaches, with method-specific relationships. These discrepancies underscore the need to reconcile methodological biases to enhance the accuracy of global plant CUE mapping, thus supporting carbon sequestration strategies under a warming climate.