The End of Reliability

The first choice that seems to dominate current discussions and research funding is the attempt to reduce the uncertainty that nonstationarity implies while giving little regard to the design of the system. That is, can we develop better methods to improve our ability to estimate the probabilities associated with hydrologic risk and update our designs while retaining the static design paradigm? Although popular, this approach is presently unsatisfying because it inevitably leads to attempts to ascertain future design variables and reliability from deterministic models of coupled physical-social systems we don’t understand well enough. The dominant example involves attempting to develop design or planning guidance from the output from a GCM. Unfortunately, the typical wide range of GCM projections, the coarse spatial resolution of GCM output, the employment of downscaling techniques that rely on their own assumptions of stationarity, the lack of skill that GCMs have in reproducing variability and the use of emissions scenarios that do not have probability assigned to them severely limit the utility of GCM output for the purposes of water resources planning and management. For example, the IPCC states that climate change projections are most skillful at the continental scale, hardly applicable to the problem of design of a particular water resource system (Solomon et al. 2007). Downscaling increases the resolution of this output but cannot reduce the range of GCM projections or produce agreement among the GCM input if they disagree. Furthermore, it is inadvisable to attempt to select a single GCM that performs best for a particular analysis (Gleckler et al. 2008). The use of paleoclimatology is useful to better estimate the range of historical variability but its usefulness for future planning is again dependent on a stationarity assumption.

These issues apply only to the climate change issue. Human alteration of the hydrologic cycle is another important source of nonstationarity. If the nascent field of hydromorphology is supported and flourishes, in the future we may be able to predict the direction and implications of human action on the hydrologic cycle; but at this point it is largely unknown. All of these issues make it unlikely that we can produce better estimates of reliability in the near term. This topic was entertainingly explored by Klemes (2002), and Fiering and Matalas (1990) visited it as well, stating, “we can spend megabucks on climate research ... and still not answer the questions” regarding water planning and management under climate change.

An alternative to the dominant approach that is advocated here is accepting the end of reliability and the static design paradigm for water resource systems. Our current understanding of change in the statistics of hydrologic variables in the anthropocene is so limited that it is fruitless to attempt to reduce the uncertainty of reliability estimates in the practice of water resources design. Note that I’m not arguing against attempts in the research community to reduce this uncertainty—clearly improvement in GCM and advances in hydromorphology are needed; but they are either not imminent or likely to produce marginal improvements. What is advocated is a new research direction that focuses on the design of the water system to create innovations in water resources planning and management that enable water systems to respond dynamically to a changing world.

There is good reason to rethink the static design paradigm for water systems even without the death of stationarity. The traditional mindset is to design a system to “not fail” up to some reliability and then give little regard to what happens when an exceeding event does occur. However, when an engineer designs a system to be reliable up to a 1 in 100 event, she is leaving the job 99% done. Given enough time and space, inevitably we’ll be inundated in 1 in 100 events and the associated negative consequences—that is, failure—and we are! That may be acceptable in some fields, but when that 1% event entails the human suffering of Katrina or the mayhem in Atlanta during the drought of 2007, it cannot be deemed acceptable in this field. Instead, there is a need to replace the binary straitjacket of reliability (fail or not fail) with a continuous vision. We need to design for all events, not just a design event (Green et al. 2000; Barros and Evans 1997).

Some would argue that water resources engineers have always dealt with uncertainty and the historical paradigm continues to be adequate. Certainly there is much to be learned from the manner in which water engineers have addressed uncertainty in the past (Burges 1979). Many of these lessons are still relevant. However, the nature of the current problem is different. We currently recognize not just uncertainty due to short records, but also temporal structure in climate variability and change that contributes to the uncertainty (such as interannual and multidecadal variability, as well as secular trends due to anthropogenic climate change). Much else has changed since the development of the static design paradigm. From a planning standpoint, we’ve moved from an era where optimality and economic efficiency were accepted objectives to a time where common objectives are difficult to specify amid multiple and competing interests and the new objective is to find “politically feasible solutions.” In addition, the realization of the importance of flows for the sustenance of ecosystems has reduced the operational space for meeting other objectives. On the positive side, technological advances in remote sensing, information and communication technology, and computation provide us with a cornucopia of information about the environment that enables real time monitoring and in some cases skillful forecasts. The management of water resources requires updating not just to address nonstationarity but also to better reflect our values, and perhaps most important, to realize the potential benefit of the observational and information technology resources that are now available. We need a water resources management paradigm that is commensurate with these innovations.

What does the end of reliability mean for water resources planning and management? A useful thought exercise is to consider what one would do if there were no historical hydrologic data and therefore it was impossible to estimate the probabilities or return periods of hydrologic events of interest. I offer it as a koan to my colleagues in the profession, a question to reflect on as an exercise toward enlightenment. My answer has three parts.

These are components of a water resources paradigm that promotes dynamic response to changing conditions at all timescales. The death of stationarity implies that the future will hold surprises, whether in short-term supply expectations or long-term demand projections, and our water resources systems must be prepared to react to them. A summary of these new principles might be called the three “F’s”: forecasts, flexibility, and fields (as in, multidisciplinary). Forecasts (daily, weekly, seasonal, longer) of both supply and demand represent the best estimate of the future that we have, inaccuracies and uncertainties and all. This includes using unconditional climatology when there isn’t a strong climate signal. Where they have skill they allow us to enact anticipatory actions that mitigate the effects of hydrologic extremes. Their utility presupposes that anticipatory actions are possible and that we have back-up plans for when an anticipatory action might be mistaken (i.e., when a forecast is “wrong,” the issue of whether a probabilistic forecast can be wrong notwithstanding). At present it is often difficult to employ forecasts beneficially in systems that were designed to function without them. Flexibility enables the system to adapt to changing and unexpected conditions and to call upon resources that might be needed only rarely and would be prohibitive to provide continuously. Examples would include economic mechanisms such as temporary water transfers and controlled flooding of designated areas to relieve flood risk (Characklis et al. 2006; Kundzewicz and Takeuchi 1999). Multidisciplinary fields allows a focus on water resources management that is not limited by the single field and its prevailing assumptions that may have dominated the subject in the past; or that allows the field to evolve beyond the traditional borders of departments and disciplines.

Many of the ideas cited here are not new. However, they will remain as good ideas with only anecdotal support while the static paradigm persists. The end of reliability presents an opportunity for a scientific initiative to develop the new paradigm for water resources planning and management. For that to happen, the subject of water resources research calls for an approach that is consistent with the concept of sustainability science, a new field and section of the Proceedings of the National Academy of Sciences which “transcends the concerns of its foundational disciplines and focuses instead on understanding the complex dynamics that arise from interactions between human and environmental systems” (Clark 2007). Particularly relevant to water resources research is the idea that sustainability science “serves the quest for advancing both useful knowledge and informed action by creating a dynamic bridge between the two” (Clark 2007). In our field, there is a need for strong linkage between practitioners and theorists to ensure science benefits society and a perhaps stronger need for practitioners to advocate for the science that will yield innovations in practice.

This essay has called for a new paradigm for water resources planning and management and presented a list of possible principles that reflect the challenges and opportunities of the current era. Unfortunately, the resources for testing and expanding these principles are limited due to a lack of research funding for water resource management initiatives within the field of water resources research (Lettenmaier 2008). Certainly water is a fundamental aspect of sustainability and may even be considered a bellweather of sustainability. Now we need support for the science that will produce the innovations in water resources planning and management that are sorely needed to replace the static design paradigm in this era of nonstationarity.