Hydrological modeling frameworks require an accurate representation of evaporation fluxes for appropriate quantification of, e.g., the water balance, soil moisture budget, recharge and groundwater processes. Many frameworks have used the concept of potential evaporation, often estimated for different vegetation classes by multiplying the evaporation from a reference surface ("reference evaporation") by crop-specific scaling factors ("crop factors"). Though this two-step potential evaporation approach undoubtedly has practical advantages, the empirical nature of both reference evaporation methods and crop factors limits its usability in extrapolations under non-stationary climatic conditions. In this paper, rather than simply warning about the dangers of extrapolation, we quantify the sensitivity of potential evaporation estimates for different vegetation classes using the two-step approach when calibrated using a non-stationary climate. We used the past century's time series of observed climate, containing non-stationary signals of multi-decadal atmospheric oscillations, global warming, and global dimming/brightening, to evaluate the sensitivity of potential evaporation estimates to the choice and length of the calibration period. We show that using empirical coefficients outside their calibration range may lead to systematic differences between process-based and empirical reference evaporation methods, and systematic errors in estimated potential evaporation components. Quantification of errors provides a possibility to correct potential evaporation calculations and to rate them for their suitability to model climate conditions that differ significantly from the historical record, so-called no-analog climate conditions.