Quantifying the contribution of biophysical and environmental factors in uncertainty of modeling canopy conductance

Abstract

Canopy conductance (Gc) is a crucial variable in accurately estimating latent heat flux and evapotranspiration (ET) over vegetated surfaces. Gc is highly dependent on the plant species and the surrounding environment; consequently, it is difficult to accurately simulate Gc using a simplified universal model. A number of empirical parameters and functional forms have been introduced to appropriately account for various plant species. This study examines the efficacy of a widely adopted Gc model and Penman-Monteith (PM) approach over a selection of vegetated sites in arid and semi-arid regions of China to investigate the factors that influence the accuracy of the model. Comprehensive micrometeorological and land surface monitoring datasets from eight eddy covariance (EC) stations in the Hei and Hai river basins were used for our analysis. ET calculated from a widely used surface conductance and PM method was compared with EC measurements, and the residuals of the comparison were analyzed with factors potentially influencing the residuals using their correlations and regression coefficients to identify factors accounting for the errors caused via Gc. The results showed that the vapor pressure deficit and the maximum stomatal conductance (gsm) are most significantly correlated with uncertainties in Gc for all plant species. The Gc error vs. vapor pressure deficit exhibited distinctive sensitivities among the C3 crop/grass, C4 crop and C3 tree groups. The maximum stomatal conductance showed variation with the leaf age, reaching the maximum level in the mid-growth stage for most types of plant species except for coniferous trees. The residuals increased with radiation levels for sites covered with trees, which appeared to be caused by the decreasing trend of the stomatal conductance vs. radiation at high radiation levels for trees. Finally, we propose potential methods of improving canopy conductance models based on our error analysis.