(1) According to the signs of eddy-covariance fluxes of sensible heat (H) and latent heat (LE), observational data over each surface are organized into four heat-flux regimes: I (H > 0; LE > 0), II (H < 0; LE > 0), III (H > 0; LE < 0), and IV (H < 0; LE < 0). Relevant constraints on the Bowen ratio (i.e. β = H / LE) are then incorporated into the data screening, so that supersaturation in the air (e.g. fog and mist) is forbidden over saturated surfaces (i.e. water, snow, and ice herein). Following examinations of Bowen-ratio similarity', we introduce auxiliary measures for data screening in regime II, given the original constraint of β < 0 can be naturally satisfied and is practically ineffective. Quality-controlled measurements suggest that regimes I and II are dominant over inland water, and regimes II and IV are dominant over snow and ice.
(2) Over inland water, α differs significantly between regimes I (β > 0) and II (β < 0, indicative of advective effects mostly in the afternoon). For instance, linear regressions of LE versus the equilibrium evaporation' result in α values of 1.13 and 1.33 for regimes I and II, respectively. Deviations from the commonly used constant value of 1.26 imply that it is indeed inappropriate to assume α as a fixed parameter. Over snow and ice, we also find fundamentally different ranges of α between regimes II (β < 0) and IV (β > 0) both representing stable boundary layers over snow and ice: α in regime II (absolute values) varies wildly between 0 and infinitive, while α in regime IV falls into a much narrower range of 0 to 1. Despite the wide variations as observed, a simple function of the Bowen ratio is found to reproduce the behavior of α fairly well.
(3) To assess the significance of taking account of the Bowen ratio-dependence of α, we compare estimates of the latent heat flux from the PriestleyTaylor formulae (using different approaches of α) against the eddy-covariance measurements (LE). Over inland water, the original α value of 1.26 leads to overestimated and underestimated latent heat fluxes for regimes I and II, respectively, though the relative differences are within 10% overall. In comparison, when α is taken as a function of the Bowen ratio, the estimated latent heat fluxes have a perfect match with the measurements. Furthermore, over snow and ice, assuming α as locally determined constant values (e.g. -3.09 and -1.19) leads to significant errors in the flux estimates for regime II (sublimation over the glacier), often with incorrect directions of water vapor transfer. To remedy this deficiency, the Bowen ratio-dependence of α is incorporated, leading to notably improved flux estimates. Similarly, for regime IV (deposition), α has to be expressed as a function of the Bowen ratio, so as to derive accurate flux estimates from the PriestleyTaylor formulae. To sum, the original Priestley-Taylor model is extended in this work to a wider variety of heat-flux regimes irrespective of the presence of advective effects, direction of water vapor transfer, and boundary-layer stability.