Climate change scenarios can provide valuable information to help better understand how the past will differ from the future. They reveal a non-stationary climate and are often the best tool available (NRC 2012a). They do, however, have limitations. For example, the spatial or temporal scales of the data might be too coarse for certain decisions (section 4.6), or other changes (e.g., changing demographics, socioeconomics, land use, and infrastructure demands (Brekke et al. 2009)) eclipse climate pressures (section 4.5). Additionally, even when climate change scenarios are being used, other investigations may add important insights (Vano et al. 2014; Lehner et al. 2017a).

For perspective, below is a brief overview of four approach categories. This is not an inclusive list, as more exist and more will likely be developed.

Climate change scenario studies: These approaches are often characterized as a chain-of-models approach where global climate model projections are downscaled and the downscaled climate change information (e.g., 30 years of daily precipitation, temperature) is then used as input to hydrology models, which generate streamflow and snowpack information, which can be used as input to reservoir operations models. This type of study is often the focus of existing guidelines because it most explicitly uses global climate model information and often requires decisions on model selection to translate global information to a local scale.

Paleoclimate studies: Paleoclimate or paleoflood information is generated using information collected from the environment which can be proxies for past climate and flood events that date back further than the instrumental record (e.g., the width of tree rings can be correlated with streamflow) (Woodhouse et al. 2006). These analogs from the past can date back thousands of years, and provide improved perspectives on natural variability, such as the length of dry periods (Woodhouse and Lukas 2006), the characteristics of past floods (Raff 2013) or how sensitive river basins are to temperature increases (Lehner et al. 2017a). Studies have also used a combination of scenario-based and paleoclimate studies to evaluate future change (Reclamation 2011a; McCabe and Wolock 2007).

Stochastic hydrology studies: stochastic precipitation and hydrology timeseries can be used to stress test a system (Rodriguez-Iturbe et al. 1987; Salas 1993; Wilks and Wilby 1999; Yates et al. 2003; Erkyihun et al. 2016). The perturbations can be informed by historical information (e.g., paleoclimate information) or by global climate model trends. These techniques aim to avoid some of the uncertainties associated with using global climate models directly, yet address risk-based issues analytically (Olsen et al. 2015). In many cases, stationarity is assumed, although there are techniques that have included non-stationary stochastic methods (Kilsby et al. 2007; Erkyihun et al. 2016). It is, however, important to recognize that these timeseries are based on statistical models that do not capture process-based understandings, which limits how these can be used to interpret future change.

Climate-informed water system vulnerability analysis: These approaches are commonly referred to as decision support modeling and include techniques such as decision scaling (Brown et al. 2012), scenario-neutral approaches (Prudhomme et al. 2010), and robust decision making (Lempert et al. 2003). Typically, the focus is first on defining the decision context and exploring sensitivities by perturbing the climate incrementally to identify system vulnerabilities to changes in temperature, precipitation, or other climate variables before considering whether and how to apply climate change information (Brown et al. 2012; Brown and Wilby 2012; Weaver et al. 2013). EPA and CWDR (2011) describe strengths and limitations of using different decision support tools.

Approaches vary in complexity (Ludwig et al. 2014) – i.e., in which processes are represented and at what spatial and temporal scales. Simpler approaches (e.g., simple perturbations, simple water balance models) can be easier to understand, but may not include processes that provide more realistic representations of climate change (NRC 2012a).

Approaches also differ in the order in which they evaluate aspects of the system. They are often referred to as top down or bottom up, reflecting either those that start with the climate change information first or those that start with the decision context first, respectively (Brown et al. 2011). In reality, this dichotomy is blurry. Climate change scenario studies should consider the decision context in model selection, and Climate-informed water system vulnerability analysis should consider realistic climate change perturbations when evaluating system performance in a changed climate.

More similarities exist than what the top down v. bottom up dichotomy suggests. Most approaches use global climate model information as part of their analysis process. All approaches have goals to better understand how the past will differ from the future and usually aim to find low-regret, robust alternatives that do well across a range of possible futures (Clark et al. 2016; Olsen et al. 2015). All approaches recognize the importance of climate variability and the importance of other changes (e.g., land cover, population changes). Often too, different approaches can complement each other (e.g., stochastic hydrology studies use perturbations based on paleoclimate information (Brekke et al. 2009). Additionally, all approaches deal with uncertainty whether it is in tree-ring reconstructions (Woodhouse et al. 2006), how climate variables are correlated (Yates et al. 2003), or how well hydrology is being simulated (Mendoza et al. 2015). And, importantly, each approach has benefits and challenges that require professional judgment to navigate.