Uncertainty of transpiration estimates from a coupled water and carbon model in grassland ecosystems

Method This study assesses uncertainty in grassland transpiration (T) estimates using coupled water and carbon models (PML-V2 and uWUE model). Fifteen grassland eddy covariance sites from the global FLUXNET were selected. Leaf area index (LAI) at these sites were obtained from MOD15A2. Evapotranspiration (ET) and gross primary production (GPP) were used to optimise the model parameters. Meteorological observations and LAI were used to drive the model at daily scale to obtain ET, GPP and T simulations. Good model performance for ET and GPP by PML-V2 is indicated by high Nash-Sutcliffe Efficiency for ET and GPP averaged at 0.81 and 0.71 respectively. T from the uWUE model were calculated by applying an empirical relationship between GPP and ET. The estimated T from the PML-V2 model and its ratio to measured ET is then compared to that from the uWUE and literature reported experimental results.

Results A The T/ET estimated by PML-V2 model varies greatly across 15 study sites, averages at 0.53, ranges in 0.08-0.84, while uWUE model gives a mean value of 0.45, ranges in 0.25-0.58. B The PML-V2 model produces higher T/ET estimates in wetter grasslands where annual precipitation is over 800 mm/year than that of uWUE model. Similar T/ET values were found from both models in drier grasslands where annual precipitation falls in 200-700 mm/year. C Both PML-V2 and uWUE model can well capture the seasonal variation of T, with higher T/ET during growing season and lower for the rest period in a year. D The PML-V2 model shows greater variation of T/ET estimates across wetter grasslands (6 sites, precipitation > 1000 mm/year, T/ET range from 0.52 to 0.84). On the contrary, uWUE model simulates more constant T/ET values T/ET around 0.47. E Drier grasslands measurement reports T/ET range from 0.45 to 0.74, both PML-V2 and uWUE prediction shows higher variation lies in 0.08-0.41 (4 sites, precipitation < 500 mm/year).

Conclusion This study shows that high uncertainty exists in transpiration estimates across models for grassland. Larger variations of T/ET estimates were found in wetter grasslands. The higher estimates of T/ET by PML-V2 can be a result of overestimation of LAI. The lower T/ET by uWUE model may due to constant water use efficiency used to obtain T from GPP. Seasonal variations in T/ET in grasslands contributes greatly to total T/ET estimates.