Using phase lags to evaluate model biases in simulating the diurnal cycle of evapotranspiration: a case study in Luxembourg

While modeling approaches of evapotranspiration (lambda E) perform reasonably well when evaluated at daily or monthly timescales, they can show systematic deviations at the sub-daily timescale, which results in potential biases in modeled lambda E to global climate change. Here we decompose the diurnal variation of heat fluxes and meteorological variables into their direct response to incoming solar radiation (R-sd) and a phase shift to R-sd. We analyze data from an eddy-covariance (EC) station at a temperate grassland site, which experienced a pronounced summer drought. We employ three structurally different modeling approaches of lambda E, which are used in remote sensing retrievals, and quantify how well these models represent the observed diurnal cycle under clear-sky conditions. We find that energy balance residual approaches, which use the surface-to-air temperature gradient as input, are able to reproduce the reduction of the phase lag from wet to dry conditions. However, approaches which use the vapor pressure deficit (D-a) as the driving gradient (Penman-Monteith) show significant deviations from the observed phase lags, which is found to depend on the parameterization of surface conductance to water vapor. This is due to the typically strong phase lag of 2-3 h of Da, while the observed phase lag of lambda E is only on the order of 15 min. In contrast, the temperature gradient shows phase differences in agreement with the sensible heat flux and represents the wet-dry difference rather well. We conclude that phase lags contain important information on the different mechanisms of diurnal heat storage and exchange and, thus, allow a process-based insight to improve the representation of land-atmosphere (L-A) interactions in models.