SWH双源蒸散模型模拟效果验证及不确定性分析

SWH模型是在经典Shuttleworth-Wallace双源蒸散模型的基础上发展起来的蒸散模型。过去的研究结果表明在站点尺度上SWH模型表现出较高模拟精度,但有关模型对主要参数及驱动变量的敏感性以及模型模拟的不确定性来源等缺乏深入理解与认识。本文通过与51个陆地生态系统站点多年的蒸散观测数据对比,在季尺度、年尺度上验证了全国范围内SWH模型的模拟效果,并分析了关键参数和驱动变量对模型不确定性的贡献大小。结果表明：SWH模型在区域尺度上取得了较好的模拟效果,模拟蒸散与实测值R²均在0.75以上。模型各参数中,冠层导度估算涉及的两个参数对蒸散模拟不确定性影响较大;驱动数据中,归一化植被指数对蒸散模拟不确定性影响较大。尽管部分数据（如降水）因插补存在较大的误差,但总体上气候驱动数据对蒸散模拟的不确定性的贡献仍低于NDVI。

Evaluation and uncertainty analysis of a two-source evapotranspiration model

Evapotranspiration (ET) is one of the core processes of water cycle in ecosystem and ET modeling is a hotspot and frontier in the field of the global climate changes. It is therefore important to provide spatiotemporal information of ET across diverse ecosystems in order to predict the response of ecosystem carbon and water cycles to changes in global climate and land use. The SWH model incorporates the Ball-Berry stomatal conductance model and a light use efficiency-based gross primary productivity (GPP) model into the Shuttleworth-Wallace model, which can simulate both ET and GPP. The newly developed SWH model presents a satisfactory prediction ability of simulating ET in a forest and a grassland ecosystem, respectively. However, the SWH model still lacks comprehensive evaluation and uncertainty analysis at regional scale. In this study, we (1) tested the model's performances on estimating ET and GPP at seasonal and annual time scales; (2) quantified the uncertainties of the model parameters and driving variables, including Normalized Difference Vegetation Index, NDVI and meteorological data; (3) quantified the sensitivity of model outputs to the parameters and driving variables; (4) quantified and separated the uncertainties of ET simulation from the parameters and driving variables. Results showed that the SWH model performed well for ET simulation at regional scale as indicated by high coefficient of determination (R² = 0.75) of linear regression of modeled against measured ET. Among the key parameters in the SWH model, two parameters related to estimating canopy stomatal conductance (g₀ and a₁) make great contribution to the model uncertainty. Among the forcing variables, NDVI is most critical in estimating GPP, which contributes much to uncertainty in ET simulation. In comparison, the climatic forcing variables contributes less to uncertainty in ET simulation owing to the high accuracy of the climate data (such as radiation and air temperature) or model’s low sensitivities to some variables (such as precipitation).