Event Navigation
Title: Hydrometeorological Controls on Transpiration in a Semiarid, Snow-Dominated Watershed
Abstract: Evapotranspiration (ET) is a major, and largely unobserved, component of the terrestrial water budget. Transpiration, the component of ET that represents water loss from terrestrial vegetation, impacts soil moisture through plant water uptake, which in turn influences groundwater recharge, watershed storage, and ultimately streamflow. Transpiration is controlled by several different hydrometeorological variables including air temperature (AT), vapor pressure deficit (VPD), soil moisture (SM), precipitation (P), and available energy (AE). The Penman-Monteith equation provides a theoretical framework relating ET and these hydrometeorological controls. High temporal resolution observations of transpiration from representative vegetation types at a study site provide an invaluable dataset for understanding the hydrometeorological controls on ET. Improved understanding of site-specific hydrometeorological drivers of ET can facilitate better interpretation of remote sensing and model-based estimates of ET, and provide important context for local hydrologic processes.
This research aims to improve the understanding of the direction and magnitude of influence of each hydrometeorological variable in controlling transpiration in a semiarid, snow-dominated mountain watershed on seasonal time scales. We installed sap flow sensors in two mature Ponderosa pine trees in Dry Creek Experimental Watershed that were operated continuously for 14 months. Over this period, we obtained hourly observations of sap flow, AT, relative humidity (RH), SM, P, and longwave and shortwave solar radiation for calculating AE from a nearby weather station. Four machine learning (ML) algorithms were used to create predictive models of transpiration as a function of observed hydrometeorological variables. [Predicted transpiration was evaluated using root mean squared error (RMSE) for these four ML approaches varied between 0.001 and 0.92 L/hr. Predictive model performance was also assessed via other metrics such as mean absolute error (MAE), coefficient of determination (R2), and Nash-Sutcliffe Efficiency (NSE). These ML models were then used as input to a Shapley analysis (SHAP), an explainer method based on game theory that facilitates the assessment of the relative importance and direction of different predictors.
The SHAP analysis showed that AE, VPD, and AT were positively associated with transpiration, meaning that higher values of these variables were associated with higher transpiration. Meanwhile, SM and P were negatively associated with transpiration. Seasonal transpiration peaks occurred in late spring and summer as AE, VPD, and AT increased, and are at its lowest in the fall, winter, and early spring as P and SM increased. The machine learning models and SHAP predicted that AE, VPD, and AT are the most important factors driving transpiration, and P and SM are the least important overall. This study illustrates that ML and SHAP approaches can be used conjunctively to understand complex relationships between transpiration and hydrometeorology. In this study, the combined use of ML and SHAP revealed that the relative importance of AE and VPD in driving transpiration is seasonally dependent. This approach could be applied more broadly in space and time, which could facilitate an improved understanding of hydrometeorological controls on transpiration at landscape or watershed scales. This improved understanding could give forest and resource managers improved insight into watershed and forest health.
Zoom Link: https://boisestate.zoom.us/j/94644700403
Advisor: Alejandro Flores
Committee: Anna Bergstrom, David Huber, Qifei Niu