Mon, Aug 15, 2022: 5:00 PM-6:30 PM
ESA Exhibit Hall
Background/Question/MethodsSoil water content (SWC) is a major factor governing plant and ecosystem functioning, especially in water-limited systems. Climate change is altering the hydrological cycle, leading to SWC deficits in semiarid regions that impact vegetation and the carbon cycle. Thus, reliable SWC data is central to evaluating overall dryland ecosystem water balance, understanding how SWC varies over time and space, and evaluating its influence on critical plant and ecosystem functions. However, sensors that record SWC at high frequencies often malfunction, leading to incomplete timeseries. Here, we developed a novel analytical approach to impute missing SWC data and tested the approach at six AmeriFlux sites along an elevation gradient in the southwestern US. We assume missing data can be imputed as a mixture of (1) simple linear interpolation of missing SWC between the observed endpoints of the missing data gap and (2) SWC simulated by an ecosystem water balance model (SOILWAT2). We implemented the mixture model in a Bayesian framework for multiple depths and allowed the relative utility (mixture weight) of each mixture component (linearly interpolated vs SOILWAT2) to vary by depth, site, and data gap characteristics.
Results/ConclusionsWe evaluated two model variants: mixture weights that (1) are constant over time versus (2) vary over time as a function of cumulative precipitation, antecedent temperature, and gap length. Across all sites and depths, both models estimated missing SWC data well (R2 = 0.81-0.98 for “fixed” weights; R2 = 0.86-0.99 for “dynamic” weights). Results from the “fixed” weights model indicate that linearly interpolated values are generally of greater utility than SOILWAT2, but the utility of SOILWAT2 increases with the increasing influence of small, ephemeral precipitation events at shallower depths and more arid sites (desert grass- or shrubland). Linearly interpolated values were most useful for data gaps that were short and characterized by little/no precipitation and/or cooler temperatures. In some sites, the utility of linear interpolation was governed by a gap’s precipitation characteristics for shallow layers but by temperature for deeper layers, indicating that SOILWAT2 is needed to reliably impute missing SWC during gaps when precipitation occurred or when evapotranspiration modified SWC. In summary, the mixture model reliably imputes missing SWC, while lending insight into the processes governing SWC dynamics. The approach could accommodate more than two estimators of the missing data and be extended to other timeseries.
Results/ConclusionsWe evaluated two model variants: mixture weights that (1) are constant over time versus (2) vary over time as a function of cumulative precipitation, antecedent temperature, and gap length. Across all sites and depths, both models estimated missing SWC data well (R2 = 0.81-0.98 for “fixed” weights; R2 = 0.86-0.99 for “dynamic” weights). Results from the “fixed” weights model indicate that linearly interpolated values are generally of greater utility than SOILWAT2, but the utility of SOILWAT2 increases with the increasing influence of small, ephemeral precipitation events at shallower depths and more arid sites (desert grass- or shrubland). Linearly interpolated values were most useful for data gaps that were short and characterized by little/no precipitation and/or cooler temperatures. In some sites, the utility of linear interpolation was governed by a gap’s precipitation characteristics for shallow layers but by temperature for deeper layers, indicating that SOILWAT2 is needed to reliably impute missing SWC during gaps when precipitation occurred or when evapotranspiration modified SWC. In summary, the mixture model reliably imputes missing SWC, while lending insight into the processes governing SWC dynamics. The approach could accommodate more than two estimators of the missing data and be extended to other timeseries.