The diurnal forcing of solar radiation is the largest signal within the Earth system and dominates the diurnal cycle of the turbulent heat fluxes and evapotranspiration ( λE ) over land. Incoming solar radiation ( R sd ) also shapes temperature, vapor pressure deficit and wind speed known as important controls on λE . Current process-based λE schemes used in remote sensing and land–surface modeling differ in how these controls on λE are represented and which input variables are required. Here, we analyze how well different surface energy balance schemes are able to reproduce the diurnal cycle and how the diurnal signals of observed input variables actually influence the resulting diurnal pattern of λE . As additional constraint for model evaluation we estimate a linear and a non-linear phase shift component of a surface variable (e.g. λE ) to incoming solar radiation. We illustrate our analysis with observations from an eddy covariance station at a temperate grassland site in Luxembourg with a focus on clear sky conditions. During the field campaign in 2015 a summer drought led to a dry-down of soil moisture which allows for studying the effect of wet and dry conditions on the diurnal cycle. We found a remarkable, almost linear relationship of λE with R sd , which exhibits a significant positive phase lag during wet periods. This phase lag in λE was compensated by a preceding phase lag of the sensible heat flux. Vapor pressure deficit ( D a , often used as input for Penman–Monteith based approaches) exhibits a strong phase lag, which is driven by air temperature reflecting large diurnal heat storage changes in the lower atmosphere. This large phase lag in D a , which is not seen in λE , explains why actual and potential evapotranspiration approaches can show systematic deviations from observations at the sub-daily time scale and highlight the need for a time-dependent non-linear compensation through the conductance parameterization. The surface to air temperature gradient used as input in energy balance residual approaches corresponds rather well with its linear response and phase lag to the observed sensible heat flux under both, wet and dry conditions. This simplifies the conductance parameterization and explains the better correlation of these models at the sub-daily time scale. We conclude that the analysis of phase lags at the sub-daily time-scale provides valuable information on the drivers of the surface–atmosphere exchange, which can be used to evaluate and improve the representation of land–atmosphere coupling in land–surface schemes.