Performance evaluation of PML and PT-JPL models in simulating wetland evapotranspiration based on multi-stage parameter optimization

The distinct canopy structure and phenological rhythms of wetland vegetation have not been adequately take into account in the current regional-scale evapotranspiration (ET) simulation, which has led to a notable discrepancy between the simulated and measured values. To improve model applicability in the region, this study concentrated on the natural reed wetland and artificial paddy field wetland in the Liaohe Delta, utilizing multi-source observational data from 2018 to 2023. The ET simulation and local parameter optimization were conducted using Penman-Monteith-Leuning (PML) and Priestley-Taylor Jet Propulsion Laboratory (PT-JPL) models. For the reed, the four growth stages were germination, vegetative growth, reproductive growth, and mature or early yellowing; for the rice, the stages were green-up, tillering and jointing, grain filling and flowering, and ripening. Through parameter sensitivity analysis, the maximum stomatal conductance of leaves was identified as the key parameter of the PML model. The PT-JPL model’s four key parameters were the canopy extinction coefficient (PAR absorption), canopy extinction coefficient (PAR interception), evaporation regulation coefficient, and optimal growth temperature of vegetation. The five key parameters were optimized by growth stage using the Bayesian optimization algorithm and the measured data from 2018 to 2022. The validation set consisted of the 2023 flux station observation data, and the simulation performance of the models was assessed before and after optimization using the Nash-Sutcliffe Efficiency coefficient (NSE), Root Mean Square Error (RMSE), and Mean Absolute Error (MAE). Simulating wetland evapotranspiration and its components was done while assessing the optimized model’s adaptability. The results showed that both models’ simulation performance for reed and paddy field ET had improved following parameter optimization by growth stage, with the PML model outperforming the PT-JPL model. Specifically, the simulation effect of the PML model for reed and paddy field ET was as follows after parameter optimization: NSE increased by 7.6% and 51.7%, respectively, compared to before optimization, RMSE decreased by 12.3% and 45.5%, and MAE decreased by 10.8% and 55.0%, respectively. For the PT-JPL model's simulation effect for reed and paddy field ET following parameter optimization was as follows: NSE rose by 24.5% and 33.1%, respectively, in comparison to pre-optimization, RMSE dropped by 20.1% and 21.5%, and MAE dropped by 19.9% and 27.7%, respectively. Both models successfully captured the general seasonal pattern of ET, with higher values in summer and lower values in spring and autumn. However, the simulation values of the PML model in each growth stage were more consistent with the measured values. The PT-JPL model generally exhibited a systematic underestimation. In the simulation of ET components, the optimized PML model could accurately simulate the partitioning of ET components in paddy fields and reed wetlands. The simulation results of the PML model showed that ET of paddy fields and reeds was dominated by transpiration (accounting for 89.6% and 82.2%, respectively); transpiration exhibited a unimodal pattern over the phenological cycle, and soil evaporation decreased with declining canopy coverage. In contrast, the PT-JPL model substantially overestimated soil evaporation, which accounted for 48.5%-68.9% of total ET. Totally, the parameter-optimized PML and PT-JPL models have significantly improved the simulation accuracy of ET in paddy fields and reed wetlands, and of the two models, the PML model shows better applicability in ET simulation for the Liaohe Delta wetlands. This study confirms that vegetation-specific parameter optimization can effectively enhance the reliability of ET models. By integrating physiological constraints with parameter optimization, the estimation accuracy of the models for evapotranspiration components can be significantly improved. Future research could integrate multi-source observational data to further refine the parameterization schemes of the two models under complex surface conditions, to enhance the reliability of regional ET and its component simulations.