Improved forest canopy evaporation leads to better predictions of ecohydrological processes

Canopy evaporation (Ei) is a vital process in forest ecosystems impacting hydrology and biogeochemistry through the redistribution of gross rainfall and gradual infiltration of water into the soil profile. Inaccurate representation of Ei in models may lead to flawed predictions of ecohydrological processes such as water availability, soil erosion, nutrient transport, and ecosystem productivity, thus compromising the reliability of model outputs. The Soil and Water Assessment Tool (SWAT) ecohydrological model has been widely used for various purposes worldwide. However, SWAT has shown limitations in forest ecosystems. SWAT employs a single equation to calculate canopy evaporation for crops and trees, which may not accurately account for the differences in ecophysiology and aerodynamic resistance between short and tall vegetation. In SWAT, canopy interception is calculated as a function of canopy storage and is normalized by the maximum plant leaf area index (LAI). Here we present an alternative approach to simulate forest canopy interception and evaporation with SWAT. Under our proposed approach, the LAI normalization is eliminated, and canopy storage is computed as a linear function of daily LAI and a user-defined parameter. We used remote-sensing (R-S) estimates of Ei to accurately parameterize forest canopy evaporation in the modified and default models. The Alabama-Coosa-Tallapoosa, a large (55,000 km2) and forested watershed system in the Southeast United States, is utilized as testbed. Results showed that the default SWAT largely underestimated (> 70%) forest Ei across our study domain. The modified model better matched R-S estimates of Ei, showing a mere 2% overestimation. Additionally, the modified model yielded better agreement with R-S transpiration and total evapotranspiration compared to the default model. Our alternative approach positively affected the model simulation of daily streamflow and ecologically relevant flow metrics, reducing model overestimations and leading to better agreement with observations. Also, the modified model led to reduced sediment, nitrate, and organic nitrogen loadings, with sediment and organic nitrogen being particularly affected, witnessing reductions of 13 and 11%, respectively, compared to the default model. Finally, our proposed approach resonated in better agreement between simulated net primary productivity (NPP) and R-S estimates. Although our study is in the context of SWAT, our findings can be useful to the broader modeling community since other popular process-based models are based on similar modeling assumptions. Our findings demonstrate the benefits of improved forest evapotranspiration partitioning for simulating ecological processes with SWAT.