Non-stationarity of the validity of stationarity [INVITED]

The terrestrial environment is constantly changing on many time scales. Change is driven by factors that include transient extraterrestrial processes, lithospheric dynamics and interactions with the atmosphere, biological evolution, internal dynamics of the ocean-atmosphere-cryosphere system, and human activities. The effects of these processes on the environment are inherently unpredictable. Consequently, any description of the environment that is used in water planning and management is an approximation. One common approximation in water decision-making is that of stationarity: the assumption that the future will look, in essence, like the past. In practice, stationarity entails further approximations: that the observational record is sufficiently long to characterize temporal variability adequately and that the time horizon relevant to the decision is sufficiently short that the characterization will continue to apply. The usefulness of a decision based on stationarity depends on all the approximation errors and on the robustness of the decision with respect to those errors. Historically, despite errors inherent in stationarity, decisions have proven sufficiently robust that bridges and dams still stand, irrigated crops have been harvested, and ever-thirsty cities have flourished. During recent decades, however, the growing human "footprint" on the environment presumably has further weakened the validity of stationarity. From a hydrologic perspective, anthropogenic climate change anticipated during this century is unique among various sources of non-stationarity: it is both large in magnitude and globally pervasive. However, it must also be acknowledged that many of its hydrologic impacts are highly uncertain. Future water decisions might benefit from consideration of best estimates of climate-change impacts, where they are credible, and from increasing allowance for the growing uncertainties in climate.