Projecting terrestrial carbon and water flux is the basis of earth system science and relevant subjects. Current carbon and water flux models perform nonnegligible uncertainty in remote sensing applications and in predicting the future dynamic of earth system, which is mainly caused by the parameterization-structure and type-based parameters. Therefore, to improve the model performance is of great importance during extrapolation by describing plant physiological process and avoiding dividing terrestrial surface by type-based parameters. The First-Principles Theory, which considers plants intend to maximum carbon gain with least resource cost by self-optimality, provides a universal criterion for predicting its behaviour. Existing research has proved that carbon uptake by C3 plants could be estimated by a universal model based on First Principles Theory. In this research, we use this productivity model, the P model, to coupling estimate terrestrial Gross Primary Production (GPP) and evapotranspiration (ET) in China. The GPP of C4 plants is estimated by an extension of current P model. Terrestrial ET is divided into biotic transpiration and abiotic evaporation. We use the Penman-Monteith equation to estimate transpiration. The critical variable in the equation, canopy conductance, can be predicted by the P model with the environmental conditions as input. Evaporation from soil and interception is estimated on the basis of a universal empirical function. The model requires no type-based parameter to be calibrated. We carried out validation with site scale and country scale. Site-scale validation is based on the ChinaFLUX dataset. Seven sites with 54 years of observation are selected. The Yucheng site has maize flux observation for C4 sub-model. Comparison between the modelled results and observation indicates our model's robustness:R² between estimated GPP at seven sites and observation is 0.61, RMSE=2.1 gC/d, fitting slope=0.96, R² of ET estimation is 0.66, RMSE=0.85 mm/d, fitting slope=1.04. We also mapped GPP and ET in China with the calculation capacity and gridded meteorological product provided by Google Earth Engine (GPP from 2007 to 2015, ET from 2003 to 2018). We observed a reasonable spatial pattern from comparing GPP with remotely sensed Sun-Induced Chlorophyll Fluorescence (SIF) product and from comparing ET with other products. Good consistency of flux estimation with site and country scale observation proves the robustness of our proposed universal model. Moreover, a sensitivity comparison of our universal model against parameterization models indicates our universal model achieves better performance under the circumstances without accurate land cover information or an effective quantity of training samples. The robustness of theory and the reasonable result of our universal model would be helpful for earth system science and regional resource monitoring based on remote sensing.