Quantifying the Value of Learning for Flexible Water Infrastructure Planning

Abstract

Climate change uncertainty challenges water supply planning (Gleick, 1989(Gleick, , 2000. Infrastructure systems, which are designed to last for many decades, must perform well under highly uncertain conditions (Cosgrove & Loucks, 2015;IPCC, 2022). Traditional water supply planning approaches usually oversize water infrastructure using a safety factor to account for uncertainty and reduce the chance of system failure (Stakhiv, 2011). However, overbuilding adds costs and environmental impacts, which are especially detrimental in more resource-scarce regions. One strategy to plan under uncertainty is flexible planning. Flexible, or adaptive, planning approaches can help mitigate infrastructure over-or underbuilding by changing the design or operations of infrastructure to respond to evolving conditions over time (Bertoni et al., 2021;Culley et al., 2016;Hui et al., 2018). For example, flexible plans may respond to hydrologic shifts, like a climate trending drier (Kumar et al., 2013), or changing societal values, such as a focus toward sustainability and ecological benefits (Kermisch & Taebi, 2017). Flexible plans also have downsides. They require identifying signs of system vulnerability (Dewar et al., 1994;Haasnoot et al., 2013) to trigger adaptations (Walker et al., 2001), which assumes that we can identify a reliable signpost for when to adapt. This may not be the case in many systems (Raso et al., 2019). Flexible plans also are expensive and may require additional costs for monitoring systems or repetitive structural elements to allow for future adaptations (

Valuing Combinations of Flexible Planning, Design, and Operations in Water Supply Infrastructure

“…Future studies could investigate the interactions and relative value of combinations of flexibility in water supply for balancing different drought characteristics in conjunction with longer-term climate trends. Future work could also investigate other opportunities to learn about climate uncertainty, such as incorporating improved precipitation projections resulting from advances in climate science, and their impact on the usefulness of flexibility (Skerker et al, 2023).…”

“…Then, the Bayesian model is used to update uncertainty as new information is observed every 20 years in the infrastructure planning SDP (Fletcher, Lickley, & Strzepek, 2019). This allows us to capture the process of learning about climate change through new precipitation observations, but does not capture other opportunities to update model uncertainty (e.g., from advancements in climate science) (Skerker et al, 2023). This approach requires us to discretize the range of plausible future 20-year mean temperature and precipitation outcomes.…”

“…We develop a total of 32 discrete precipitation states ranging from 30% to 30% of the historical average to represent a range of plausible future drier and wetter climate conditions while balancing the resolution and computational costs in exploring plausible future precipitation scenarios. We model the change in long term temperature as deterministic, as GCMs project regional warming and we also find via preliminary sensitivity analysis that temperature has minimal influence on results (see Skerker et al, 2023). We consider 21 GCM projections from the fifth phase of the Coupled Model Intercomparison Project forced under the high emissions scenario (RCP8.5) to capture a wide range of uncertain future climate change scenarios, given that GCM projections are considered a lower bound on climate change uncertainty (Sanderson & Knutti, 2012).…”

“…For each climate state, the reservoir operations SDP requires us to define the probability distribution of monthly inflows into the reservoir to determine the transition probabilities P z τ+1 ⃒ ⃒ z τ , q τ s ) . To model monthly time series of inflow into the reservoir, we generate precipitation time series using a K-nn approach developed in Rajagopalan and Lall (1999) and applied in Skerker et al (2023) for each climate state. We pair this with a lumped conceptual rainfall-runoff model developed in Strzepek and Mccluskey (2010) and applied in Skerker et al (2023) to derive 200-year monthly inflow time series for each of the 21 GCMs.…”

“…To model monthly time series of inflow into the reservoir, we generate precipitation time series using a K-nn approach developed in Rajagopalan and Lall (1999) and applied in Skerker et al (2023) for each climate state. We pair this with a lumped conceptual rainfall-runoff model developed in Strzepek and Mccluskey (2010) and applied in Skerker et al (2023) to derive 200-year monthly inflow time series for each of the 21 GCMs. We find that monthly reservoir inflow distributions for each climate state can be assumed to be lognormally distributed (see Text S4 in Supporting Information S1).…”