Here, we systematically document the magnitudes, spatial patterns, and causes of the response of Penman-Monteith (i.e. physical) potential evapotranspiration to strong future greenhouse warming in at least sixteen new CMIP5-generation climate models, using 3-hourly surface output. We find that in every model, nearly all land gridpoints see increases in this quantity on the order of tens of percent, and to zeroth order this is due to the increase in atmospheric moisture deficit that follows from warming at constant relative humidity, especially if that relative humidity is low [deficit = esat*(1-RH)]. However, in each model there are numerous first-order effects that modify this response, namely changes in: relative humidity, surface net radiation, surface wind, and Clausius-Clapeyron slope. The magnitudes and structures of many of these "other" changes vary widely from model to model.
We also assess the skill of empirical fits to climatological temperature (such as the Thornthwaite method) in predicting this response, compare the response to that of net radiation alone, and compare the fidelities of both the physical and empirical methods to the modeled actual evaporation in wet climates. Finally, we quantify the consequences of these changes for ecological aridity, comparing several different prominently-employed aridity and drought indices.