Accurate quantification of terrestrial evapotranspiration (ET) is essential to understand the Earth's energy and water budgets under climate change . However, despite water and carbon cycle coupling, there are few diagnostic global evapotranspiration models that have complete carbon constraint on water flux run at a high spatial resolution . Here we estimate 8-day global ET and gross primary production (GPP) at 500 m resolution from July 2002 to December 2017 using a coupled diagnostic biophysical model (called PML-V2) that, built using Google Earth Engine , takes MODIS data (leaf area index, albedo , and emissivity) together with GLDAS meteorological forcing data as model inputs. PML-V2 is well calibrated against 8-day measurements at 95 widely-distributed flux towers for 10 plant functional types , indicated by Root Mean Square Error (RMSE) and Bias being 0.69 mm d −1 and −1.8% for ET respectively, and being 1.99 g C m −2 d −1 and 4.2% for GPP. Compared to that performance, the cross-validation results are slightly degraded, with RMSE and Bias being 0.73 mm d −1 and −3% for ET, and 2.13 g C m −2 d −1 and 3.3% for GPP, which indicates robust model performance. The PML-V2 products are noticeably better than most GPP and ET products that have a similar spatial resolution, and suitable for assessing the influence of carbon-induced impacts on ET. Our estimates show that global ET and GPP both significantly ( p < 0.05) increased over the past 15 years. Our results demonstrate it is very promising to use the coupled PML-V2 model to improve estimates of GPP, ET and water use efficiency , and its uncertainty can be further reduced by improving model inputs, model structure and parameterisation schemes.