Climate-aware decision-making: lessons for electric grid infrastructure planning and operations

Climate change poses significant risks to large-scale infrastructure systems and brings considerable uncertainties that challenge historical planning approaches. Here we focus on how climate awareness might be better incorporated into planning and decision-making in the electric power sector. To do so, we consider lessons from decision science literature where researchers have specifically focused on how to make better decisions under uncertainty. We perform a three-part review: of decision science literature on best practices for decision-making under uncertainty, of industry practice documents to understand how new uncertainties may affect the types of decisions electric utilities are making today, and of literature on electricity sector planning under climate change to assess how lessons from decision science might fit into sector-specific analyses. We discuss how characterizations of uncertainty from the decision science literature might guide approaches in the electricity sector to appropriately capture climate-related uncertainties. We also distill three key ideas from the decision science literature that can be incorporated into electricity sector planning to manage these new uncertainties: robustness, adaptive planning, and multi-stakeholder engagement. We offer example recommendations for how these key ideas might be incorporated into one essential type of planning activity, capacity expansion.


Introduction
Changing climate patterns have already begun to strain society's infrastructure systems, and projected trends in ambient and extreme conditions portend significant stresses in the coming years. Billions of people worldwide rely on large-scale infrastructure systems, such as electric power, water, and transportation, for their daily needs. Despite considerable attention on documenting the mechanisms and potential impacts of climate change on these systems, more work is needed to incorporate climate awareness into actual planning decisions, which today often rely on historical data and outdated assumptions (Gerlak et al 2018).
We focus here on electric power systems, which are affected by climate and weather patterns in myriad ways and face ongoing and acute stresses from changing conditions (Ward 2013, Chandramowli and Felder 2014, Panteli and Mancarella 2015, Craig et al 2018, Brockway and Dunn 2020, Perera et al 2020. Centralized electric grids consist of generation and delivery infrastructure that supplies electricity to enduse customers. Recent events-such as widespread service interruptions and price spikes due to cold weather in Texas in February 2021 (Busby et al 2021, Doss-Gollin et al 2021; in California, heatrelated rolling blackouts in August 2020 as well as grid-caused wildfires and associated power shutoffs intended to prevent additional fires from 2017 to 2021 (Dale et al 2018, Abatzoglou et al 2020, Wolak 2021; and extensive storm-induced outages in New Orleans in late summer 2021 (Blau et al 2021)demonstrate that system planning and operational practices have not adequately accounted for climate change. Such events jeopardize the electric sector's core goals of safety, reliability, and cost-effectiveness. We define resilience here broadly, as the ability to preserve these goals without significant disruption and/or impact on customers, and to return to safe, reliable, and cost-effective operation quickly if disruptions occur.
Building climate awareness into electricity sector planning and decision-making is far from straightforward. Current planning processes commonly focus on predicting a likely future in terms of electricity demand, weather conditions, and technology options, and identifying investments that are expected to achieve specified reliability targets under those planning assumptions. Changing climate conditions and associated uncertainties-including those related to infrastructure impacts, policy and regulatory actions, and consumer behavior-challenge this underlying deterministic premise (Grubler et al 2011, Chattopadhyay et al 2016, Moallemi and Malekpour 2018). Yet openings may now exist to rethink planning approaches, as high-profile service disruptions have focused public, regulatory, and legal scrutiny on electricity sector planning and preparedness (Lacommare et al 2017, Gundlach 2020. In response, some electric utilities have begun producing vulnerability assessments to evaluate where their systems may face specific threats (PG&E 2016, Ralff-Douglas 2016, ConEd 2019c, McMahan and Gerlak 2020.
A major barrier to change is that the uncertainties inherent in future climate trajectories and their implications require a deeper grappling with the unknown than is present in electricity sector planning today. While electricity sector planners have always had to make decisions in the context of uncertain future conditions (e.g. those related to demand growth, technological innovation, fuel prices, etc), existing practices to account for uncertainty rely extensively on historical trends and expert intuition honed in the context of previous climate conditions (Kuhn and Madanat 2005). In the face of climate uncertainty, these approaches will no longer hold. Making better decisions in the face of these uncertainties will require taking stock of the available science and integrating it into planning processes, evaluating existing baked-in assumptions and revising where needed, considering compounding impacts, and reviewing expectations about acceptable levels of risk and resilience (Linkov et al 2014, Craig et al 2018, Brockway and Dunn 2020. Separately, decision science researchers have focused on how we can make better decisions even with significant uncertainties, and concepts from decision science can help provide systematic ways of accounting for them. Here, we review their insights in the context of advancing climate-aware planning in the electricity sector.
We perform a three-part literature review with the goal of making pragmatic recommendations on how climate awareness can be better incorporated into electricity sector planning today. First, we review academic and grey literature in decision science and its applications to assess best practices for decisionmaking under climate-induced uncertainties as well as tools to manage uncertainty in planning processes. Next, we review industry practice documents from two electric utilities to ground ourselves in the types of decisions made now and consider how potential new uncertainties created by climate change may impact those decisions. Then, we review literature on electricity planning under climate change to assess how decision science insights may be implemented in this space.
We assess how climate-induced uncertainties in the electricity sector fall within decision science frameworks used to categorize uncertainties. Such frameworks provide insights about how these uncertainties may be appropriately captured in planning models, and indicate that methods designed for handling deep uncertainty should be employed. We further distill three key ideas from the decision science literature: robustness, adaptive planning, and multistakeholder engagement, and assess how these build on current practices in the electricity sector. We discuss how the electricity sector could incorporate these ideas into industry practices, and highlight existing approaches that can form a scaffolding for climateaware planning. Further, we focus in on one category of electricity sector decision-making-capacity expansion planning-and provide tangible examples of how climate awareness can be better incorporated via the key ideas identified from decision science. We suggest that embracing the concept of multiple plausible futures, setting up signposts that identify tipping points as the future evolves, and involving boundary organizations to help translate insights between climate science and electricity planning can help electricity sector decision-makers better plan for the future. In doing so, we bring together insights from three fields: decision science, electricity systems, and climate science, to advance how decision-makers in critical infrastructure sectors might rethink planning processes to better prepare for climate change.

Decision science review
Decision science refers to the process of how individuals and organizations make decisions based on available information. It is an interdisciplinary field that draws on theories from economics, operations research, forecasting, behavioral science, and statistics. Growing complexities in infrastructure planning due to climate uncertainties have spurred the development of planning approaches that borrow theories from this field (Kwakkel and Van Der Pas 2011). The new planning approaches aim to provide planners with a more holistic view and assist in solving 'wicked' problems in infrastructure planning.
We reviewed both academic and grey literature on decision-making processes related to climate adaptation and resilience policy across critical infrastructure sectors. We focus on how uncertainty can be incorporated during the modeling process, rather than on how these frameworks might be implemented within regulatory entities.
For the academic literature, we used 14 combinations of search terms in Scopus (supplementary  table 1). This yielded a total of 1018 peer-reviewed journal papers. We then removed duplicates and reviewed the abstracts using the following inclusion criteria, keeping only papers that met all of the criteria: (a) The paper considers climate as a source of uncertainty. (b) The paper is related to planning for an uncertain future or adaptation strategies in a critical infrastructure sector, or it is a review of decisionmaking frameworks in the context of critical infrastructure planning or adaptation planning. (c) The paper was published after 2000 and is presented in English.
This selection process left us with 86 papers in our full text review, for which criterion (a) was a primary limiting factor. We then identified concepts and techniques used to handle uncertainty in each paper and categorized them into overarching concepts with associated techniques.
Our focus on decision-relevant planning approaches prompted us to also review the grey literature, where innovations by practitioners working in critical infrastructure sectors may be documented. We used four combinations of search terms (supplementary table 2) in three databases: Adaptation Clearinghouse, Climate-ADAPT and US Climate Resilience Toolkit. These databases contain a rich set of public documents produced by government agencies and nonprofits with a specific focus on climate adaptation. Down-selection was performed via the filtering functions embedded in each database rather than through keyword searches.
This search yielded 109 reports across the three databases. We then removed duplicates and reviewed report summaries using the following inclusion criteria, keeping only papers that met all of the criteria: (a) The report considers climate as a source of uncertainty. (b) The report is a planning or guidance document that discusses the decision-making process to address climate change for critical infrastructure planning (reports such as vulnerability assessments or climate impact studies are not included). (c) The report was published after 2010. This selection process left us with 15 reports for full text review. We then identified concepts and techniques used to handle climate uncertainty in each report and categorized them into overarching concepts with associated techniques.

Industry practice review
To better understand the ways that climate change may impact the electric power sector, we aim to document the types of decisions that electric utilities are making today. In the United States, roughly three-quarters of electricity customers are served by investor-owned utilities (IOUs) (Energy Information Administration 2019). IOUs operate as regulated monopolies, with their investment decisions and customer rates overseen by state public utility (or public service) commissions in general rate case (GRC) proceedings. These proceedings are rich sources of public information on how electric power systems are maintained, operated, and invested in today. As these documents are extensive, we select just two utilities to compare in order to keep our scope manageable. In selecting these utilities, our criteria include: (a) geographic diversity, (b) large utilities that serve a mix of customer types (including both urban and rural representation), and (c) utilities that have been heavily impacted by extreme weather events.
We focus here on two utilities: Pacific Gas and Electric (PG&E), which serves approximately 16 million customers in urban and rural Northern California, and Consolidated Edison (ConEd), which serves over 3 million customers in the New York City area. Notably, PG&E and ConEd have already been significantly impacted by climate change: increasing wildfire activity in California and Hurricane Sandy in New York have drawn scrutiny to utility operations and focused attention on electric system safety and resilience. Here, we review the 2020 GRC filings of both utilities, focusing attention on electric power system operation and energy delivery. The approved costs and rates in these GRC filings set utility spending and investment thresholds for 3 years, from 2020 to 2022.
For PG&E, we focus on the utility's filing requesting cost recovery for 2020 spending, as approved by the California Public Utilities Commission (CPUC) (CPUC 2020b). For ConEd, we focus on the joint proposal documenting the results of a settlement agreement on utility spending from the utility and other stakeholders (ConEd 2019b), expert testimonies that provide additional context about utility planning and investments (ConEd 2019a), a report documenting anticipated capital expenditures (ConEd 2020), and the approval from the New York State Public Service Commission (NYPSC 2020). For both utilities, we seek to identify where and how they propose to invest in their systems and the decisions they are making about them. We code investment decisions by activity area (see supplementary note 1), then summarize current utility practices within each space. For each activity area, we identify emerging uncertainties that will affect electric power system planning going forward.

Electricity sector planning under climate change review
We performed a literature review of papers related to electricity sector planning that also consider climate change. Search terms used are summarized in supplementary table 3 along with the number of papers these terms returned in Scopus. From the initial returned list of papers, we removed duplicates and papers written in a language other than English, which left us with 683 total papers. Then we reviewed the abstracts using the following inclusion criteria: (a) We filtered out papers whose primary focus was not relevant to planning for electricity infrastructure and/or did not include some consideration of climate change or sustainability. (b) We removed papers with a primary focus on modeling only one area of energy systems (e.g. building energy use, energy demand, solar photovoltaic generation), or where electricity infrastructure was not a major focus of the paper. (c) We removed papers that focused on describing climate change impacts rather than planning for them or on resource characterization (e.g. solar availability). (d) We focused on electricity system planning in geographic areas with developed grid systems, though we did include some papers looking at decentralized electricity systems as an option.
Following this down-selection process, we were left with 168 papers. We then reviewed the papers to identify how the authors handle climate change and uncertainty and whether and how they incorporated the key concepts identified in the decision science review. While reviewing, we also sorted the papers into topic areas (table 5).

Results and analysis
Decision science researchers have developed approaches that can help advance how electricity sector decision-makers may consider and address climate-induced uncertainties. Here, we focus first on uncertainty characterization (section 3.1), then evaluate how different types of uncertainties may emerge in the electricity sector (section 3.2) and how they may impact electric utility practices and decisions (section 3.3). Then, we consider key concepts from decision science that may be used to manage uncertainty in planning processes (section 3.4) and evaluate the extent to which these show up in literature on electricity sector planning (section 3.5). Sections 3.1 and 3.4 are based on our review of the decision science literature (section 2.1); sections 3.3 and 3.5 are based on our review of the industry practice (section 2.2) and electricity sector planning (section 2.3) literatures, respectively. Section 3.2 draws from our review of all three areas of literature.

Characterizing climate-related uncertainties
There are many different classifications of uncertainty and debates about their characterization and importance (Kujala et al 2013), so a standard framework or taxonomy for uncertainty in the context of climate change and infrastructure planning does not exist 5 . This lack of distinction leads to confusion and impedes clear communication about uncertainty to decision-makers. Climate uncertainty is also often perceived differently by decision-makers and scientists (Berkhout et al 2014, Döll andRomero-Lankao 2017). We aim to reconcile some differences in interpretation by classifying uncertainties in a decisionrelevant context. Specifically, we approach uncertainty from a modeling perspective and distinguish the technical methods used to treat uncertainties. We expand upon work done by Hawkins and Sutton (2009) and Kwakkel et al (2010) and frame uncertainties in a grid planning context.
Uncertainty is generally defined as incomplete knowledge and/or disagreement about what is known or even knowable (Walker et al 2010, Kunreuther et al 2014. Five levels of uncertainty were proposed by Kwakkel et al (2010) (figure 1). These levels correspond to different degrees of numerical certainty, which constrain how information can be appropriately represented in models and analyses.
• Level 1 uncertainty (shallow) represents a situation when a reasonable estimate of the outcome is possible. This uncertainty may be appropriately captured with a point estimate and a range of possible deviations (e.g. tomorrow's electricity demand will peak at 5:30 p.m., +/− 5 min). • Level 2 uncertainty (shallow) refers to an uncertainty that can be reliably described through statistical terms. One can capture level 2 uncertainty through forecasting techniques (scenarios) with associated probabilities (e.g. a new generation plant is 70% likely to be operational this year, 30% likely to be delayed). Analyses may appropriately assume that historical data can be used to develop reliable future forecasts.   (Patt and Dessai 2005). One can appropriately capture level 3 uncertainty with trend-based scenarios that reflect different assumptions of the driving force (e.g. three trend-based scenarios of electric vehicle demand, based on three different assumptions about product costs). • Level 4 uncertainty (deep) are situations with known multiple alternatives but where ranking the alternatives in terms of likelihood is not possible, potentially due to inadequate data or decisionmakers' disagreement on the rankings (e.g. an optimal location for building new generation in 2030, given demand patterns, population growth and movement, community impacts, and landuse change). Analysts may struggle to specify the appropriate models, select the probability distributions to represent uncertainty about key parameters in the models, and/or to value the desirability of alternative outcomes (Lempert et al 2003). • Level 5 uncertainty (deep) represents the deepest level of recognized uncertainty. We can only acknowledge that we do not know.
These levels can be simplified to two levels: 'shallow uncertainty' and 'deep uncertainty' . Shallow uncertainty can be treated through probabilities or assigned likelihood for different future alternatives, whereas deep uncertainty refers to conditions where parties do not know and/or cannot agree on the probabilities or likelihood of different future alternatives. These levels map onto different ways of representing parameter uncertainty, which may range from probabilistic information, to bounds on a range, to trend estimation or effective ignorance (Kandlikar et al 2005). Lempert et al (2003) further define deep uncertainty as the condition where decision-makers cannot agree upon: (a) the appropriate models to describe interactions among a system's variables, (b) the probability distributions to represent uncertainty about key parameters in the models, and/or (c) how to value the desirability of alternative outcomes.
The electricity sector faces uncertainties that span each of these levels. Determining which uncertainties meaningfully impact various planning and decision contexts is critical for identifying the conceptual frameworks and analytic methods that are needed to effectively advance climate-informed practice.

Uncertainties in electricity sector planning
We can evaluate climate-induced uncertainties faced by the electricity sector through the framework discussed above. Figure 2 shows a conceptual mapping of how climate uncertainties may propagate through to impacts on the electricity system. This mapping broadly reflects the attributes of coupled natural and human systems under deep uncertainty discussed in Sharmina et al (2019). These uncertainties directly pertain to decisions made today (table 1) and should be accounted for in climate-aware planning and decision-making. To do so appropriately, we Conceptual mapping of progression of uncertainties in the electricity sector that are related to climate change. This figure is not meant to be exhaustive, but to illustrate the types of uncertainties and impacts that will emerge due to climate change. must consider the nature of the uncertainties and how they might be represented to decision-makers.

Uncertainties in climate conditions
Climate scientists have made substantial progress in developing standard approaches to work with uncertainties related to climate conditions (figure 2(a)). Uncertainty in climate prediction comes from three sources: scenario uncertainty, model uncertainty, and internal variability (Hawkins and Sutton 2009, Lehner et al 2020), which have different impacts depending on the decision time and spatial scale. In the climate science space, scenario uncertainty corresponds to uncertainty in how greenhouse gas emissions will evolve. Uncertainties in continuing emissions are represented by scenarios (often, representative concentration pathways or RCPs, see van Vuuren et al 2011) that describe how emissions may evolve over the coming decades, and constitute a range of possible outcomes. Which scenario is realized depends upon societal characteristics, behavior, and climate policy, and so is constrained by future socioeconomic scenarios. These are often considered as shared socioeconomic pathways (SSPs) (O'Neill et al 2016(O'Neill et al , 2017, each of which is constructed around a self-consistent narrative describing plausible future outcomes for population, economic growth, etc. The SSPs enable a structured exploration of how alternative societal dynamics and decisions influence future climate outcomes. The range of future emissions can be loosely bounded by such plausible socioeconomic and policy scenarios, as well as by physical constraints on fossil fuel usage. However, the SSPs and RCPs are just a handful of the infinite possible futures that lie between and around them in the scenario space. It may be possible to qualitatively assign some expected likelihood to the class of similar scenarios represented by each emissions scenario based on expert judgment regarding the underlying factors that contribute to the scenarios (e.g. Hausfather and Peters 2020). However, assigning formal probabilities remains elusive, placing emissions scenario uncertainty at Level 3 or 4 from figure 1. The climate signatures of alternative emissions scenarios tend to broadly agree through approximately mid-century, so scenario uncertainty typically matters more for longterm planning decisions such as system expansion.
Uncertainties in how climate patterns might respond to continuing emissions are captured by global circulation models (GCMs), which take emissions scenarios as inputs. Model uncertainty refers to how these different models simulate changes in climate given the same amount of emissions. Each GCM constitutes a plausible representation of how future conditions may evolve. Model-based uncertainty is deep in the sense that these models are not randomly drawn from a space of possible models and their relative likelihoods cannot therefore be formally assigned probabilities. While certain model projections can be ruled out as less credible by examining their historic performance against observational benchmarks (e.g. Brunner et al 2020, Liang et al 2020, Tokarska et al 2020), it is often impossible to narrow the range to a single 'most likely' projection or 'most credible' model. Nor is it appropriate to think of models as equally likely since some models may share development histories and therefore common biases (Knutti et al 2013). Some researchers have developed credibility-based model weighting schemes, but in practice it is necessary to evaluate many metrics to build confidence that a particular weighting approach appropriately reduces uncertainty. Another reasonable strategy is to treat each projection that passes a skill-based screen as an equally plausible yet non-probabilistic outcome (i.e. corresponding to Level 4 from figure 1). GCMs describe climate impacts at large-scale resolution, and finer geographical resolution is desirable for actual planning decisions. Downscaling methods translate GCM outputs into regional climate impacts, and these methods also contain separate model uncertainties about how these large-scale outputs may be experienced locally (Barsugli et al 2013).
Climate projections are also subject to internal variability, or the natural fluctuation of the climate system even in the absence of emissions (Deser 2020). Internal variability is stochastic in nature and in principle can be estimated probabilistically through either observation or a sufficient number of simulated climate projections contingent on a given climate model and emissions scenario. However, the rarer an extreme event (e.g. a large storm or extreme heatwave) is, the more data is required to properly characterize its probability (Tebaldi et al 2021), presenting challenges when the historic record or future projections are limited. The recent advent of large climate projection ensembles (in which the same scenario is modeled dozens to hundreds of times) has led to better statistical resolution of extreme events (Kirchmeier-Young et al 2017), but still does not address the deep uncertainty arising from the choice of climate model itself or from alternative emissions scenarios.

Uncertainties in direct climate impacts
Changing climate conditions will directly impact infrastructure, consumer behavior, and decisionmaker response (figure 2(b)). In particular, these may include how climate variables (e.g. humidity, temperature) impact power equipment performance and lifetimes, as well as the demands consumers put on that equipment (e.g. through increasing air conditioning use). Direct impacts will include those from changes in ambient conditions (i.e. trends in surrounding air temperature, precipitation, wind patterns, etc. that infrastructure is exposed to in routine operation), and in extreme conditions (i.e. increasing magnitude and frequency of events likely to cause disruptions). Researchers have estimated specific impacts from climate trends in these areas, and this ongoing work in conjunction with reasonable near-term confidence in felt local climate impacts (e.g. overall temperature trends) may make it possible to represent some uncertainties in these parameters through statistics. Yet such representations still rely on projecting the underlying climate impacts, which are nevertheless deeply uncertain (section 3.2.1). Climate conditions may also impact consumer and decision-maker responses that are more difficult to characterize empirically, such as population migration and/or political pressure.

Uncertainties in indirect climate impacts
Climate will also have additional indirect impacts on the electricity sector (figure 2(c)). These may include changing maintenance and operating practices, increasing danger for customers from heat-wave induced outages, and compounding stresses between consumer behavior and infrastructure performance. Policy actions in response to changing climate conditions may also prompt both increasing consumer actions (e.g. incentives for new technology adoption), and changing industry practices (e.g. decentralization). Adaptive responses by consumers and policymakers may pose further risks to electricity system stability and cost-effectiveness (Simpson et al 2021). These uncertainties emerge from the interrelationships among entities involved in electric power systems, infrastructure impacts, and climate conditions, and are compounded by feedback loops, including with other complex infrastructure sectors (e.g. water, transportation) (Reed et al 2022). Therefore, they are more difficult to estimate and capture through statistical approximations or modeling tools. Such uncertainties may thus be squarely considered deep, and require careful assessment that does not rely on predicting a deterministic future that may generate false confidence (Sharmina et al 2019).

Electric utility practices and emerging uncertainties
New uncertainties that arise from changing climate conditions are already disrupting the normal operation and performance of electric power systems. Accounting for these uncertainties in electricity sector planning will require evaluating how decisions are made today and how decision-making and planning processes will need to be updated for the future. From our review of the GRC filings of PG&E and ConEd (supplementary note 1), we identify six activity categories and 16 activity areas in which utilities are making decisions today that will be impacted by new uncertainties from a changing climate. We discuss each activity area in supplementary note 2 and summarize them in table 1.
These activity areas illustrate the range of practices that are exposed to new climate uncertainties. While the uncertainties discussed in section 3.2 and shown in figure 2 will impact all utility decisions, in table 1 we indicate specific climate uncertainties that may be particularly relevant to each activity area. These are meant to be illustrative rather than comprehensive: (a) Severe events refers to uncertainties in the magnitude and frequency of events with the potential to disrupt day-to-day operations (e.g. strong storms, heat waves, droughts, etc). (b) Ambient conditions refers to uncertainties in how average conditions (e.g. temperature, precipitation, humidity, etc) will change over time. (c) Speed of changes refers to uncertainties in the rate of relevant changes in both severe and ambient conditions. (d) Demand changes refers to uncertainties in changing consumer demand patterns and the resulting impacts to customers of disruptions (e.g. increasing electrification will mean greater reliance on the electricity grid for new loads, such as heating). (e) Changing regulatory and policy landscape refers to uncertainties about how decision-makers will respond, including potential new requirements for utilities with regards to risk and cost management (e.g. insurance).
In real-time operations, increasing frequencies of extreme events and changing ambient conditions may stress existing approaches to system operation, data collection and monitoring, as well as the ability of crews to effectively restore power after outages. In system maintenance, changing ambient conditions may alter the expected lifetime of system components (e.g. through corrosion or thermal stress) and necessitate more frequent maintenance and labor needs to ensure satisfactory operation. Further, while new climate patterns (e.g. drought) may place more stress on vegetation management practices, such practices will become increasingly important to preserve operation in light of extreme events and severe storms. In performance and business operations, changing regulatory and policy practices, including direct-tocustomer incentives for new technologies, may also continue to drive changes in how utilities operate. Further, new insurance risks may threaten existing accounting practices.
Here, we focus on two activity categoriesplanning and investments-in which utilities make and execute long-term decisions about their systems. Activities in these areas will need to contend with the impacts to technology of changing ambient conditions and increasing severe events, changing consumer demands for electricity, and a changing regulatory and policy landscape (effectively, all uncertainties depicted in figure 2). To illustrate how these new uncertainties may challenge current utility planning and decision-making practices, we focus here on capacity expansion planning, which is widely used in the electricity sector to evaluate the need for investments in electricity generation and delivery infrastructure to serve new and changing demand patterns (Gacitua et al 2018, Miara et al 2019). These planning activities rely heavily on capacity expansion models, which incorporate information about existing infrastructure and make technology, market, and policy assumptions to determine where and when system investments should be made to accommodate load growth at least cost (figure 3). The extent to which the results of these models represent the real world depends on how they incorporate uncertainties in climate and its direct and indirect impacts and stakeholder objectives (figure 3(a)), as well as how well the model itself represents the electricity system (i.e. model uncertainty, figure 3(c)).
Current implementations of capacity expansion models typically make assumptions about these uncertainties to create a single plausible future to plan for. In California and New York, the capacity expansion planning models that feed into and inform infrastructure investments are primarily deterministic, i.e. they seek to make the best possible predictions of a single plausible future given current information, assume that uncertainties can be treated probabilistically, and then plan for that future (Brockway and Dunn 2020). While scenario planning approaches are used, investments are still tailored to one (typically median) scenario. Further, while in some cases such models take in weather and climate data directly (e.g. using temperature to predict electricity demand), they also contain embedded assumptions that implicitly incorporate climate conditions (e.g. equipment lifetimes). Feedback loops are also necessary to consider: for example, increasing consumer adoption of air conditioning due to increasing temperatures will increase demand on the grid at the same time as hotter days reduce line capacity and generation efficiency.
The potential impact that climate uncertainties may have on the results of these models is unknown until such uncertainties are more formally and systematically explored. Certainly, not all climate uncertainties will matter equally for all decisions, but relying on historical weather data as well as historical performance neglects the impact of climate as a source of uncertainty. Moreover, relying on historical weather data means, in practice, accepting the full risk of impacts due to climate change rather than preparing appropriately. Building climate awareness into capacity expansion planning is critical, and literature that evaluates the impacts of planning with climate considerations consistently show that it is better to plan for climate impacts upfront rather than after the fact (e.g. Miara et al 2019, Peter 2019, Sridharan et al 2019, Fonseca et al 2021. By not accounting for climate uncertainty, we are planning for an inaccurate future, which may compound existing grid inadequacies.

Key concepts from decision science for managing uncertainties
Traditional decision-making approaches that use the framework of expected utility theory (Savage 1972, Weitzman 2009), such as benefit-cost analysis (Boardman et al 2005), and cost-effectiveness analysis (Garber and Phelps 1997), hinge on the ability to assign a probability density function to future events and assume all uncertainties are quantifiable. However, any single projection of the future is likely to be inaccurate (Bishop et al 2007). Moreover, planning for climate change is a case of deep uncertainty, where long-term climate uncertainties in infrastructure planning are hard to quantify and futures are hard to predict. Therefore, these traditional approaches cannot provide planners with a holistic view, adequately integrate climate uncertainty, or grapple with the long-term multiplicity of plausible climate futures (Lempert et al 2003, Ackerman et al 2009, Hallegatte 2009, Ranger 2010, Masur and Posner 2011, Dittrich et al 2016. Relatedly, oft-cited reasons for not including climate data in decision-making processes include the perceived poor reliability of climate forecasts and difficulty in assessing the credibility of climate projections (Rayner et al 2005, Barsugli et al 2013. These reasons may also stem from the need to optimize under a single future and treat climate uncertainty in a probabilistic manner. To address the limitations of the traditional decision-making approaches and climate uncertainty in policy making, a growing body of literature within the climate-aware decision science space uses concepts from decision-making under uncertainty and decision-making under deep uncertainty across different infrastructure sectors (table 2). These concepts are not new (Morgan et al 1990), particularly within the water and transit sectors, but they have not significantly factored into work by practitioners in grid infrastructure planning. A majority of this literature is targeted at water resources planning. The relatively high prevalence of these concepts in the water resources planning sector might be due to the early introduction by Matalas and Fiering in 1977(National Research Council 1977, Herman et al 2015, Dittrich et al 2016, Giuliani and Castelletti 2016. However, the similarities between water resource and grid planning (e.g. high likelihood of path-dependence, sensitivity on both demand and supply to climate impacts, long infrastructure lifetime, investment irreversibility) make these concepts transferable. As discussed in section 3.2, climate uncertainties significantly affect all decision activities (e.g. system planning, investment, maintenance) and create compounding uncertainties (e.g. policy response, consumer behavior, infrastructure performance). In particular, long-term decisions such as capacity expansion planning may need a different approach to incorporate uncertainty in the decisionmaking. Three key concepts that emerged from our review could offer an alternative approach to the traditional methods: (a) Robustness. Rather than an optimal outcome for a single future, a robust outcome performs reasonably well in a wide range of plausible future climate scenarios. (b) Adaptive planning. Uncertainties in climate conditions evolve over time, and adaptive planning refers to setting up the institutional capacity to take actions over extended time horizons. This can include identifying short-term actions, developing different long-term options, and performing continuous monitoring to assess appropriate long-term paths. (c) Multi-stakeholder engagement.
Tensions between conflicting perspectives and policy objectives may be resolved through an iterative multi-stakeholder engagement process. Table 3 shows a summary of those concepts with associated techniques and papers. The following sections provide deeper dives into each of those concepts and discuss opportunities to implement them into organizational decision-making.

Key concepts Key Techniques Example Papers
Robustness: ensuring the outcome is insensitive to errors and uncertainties in the parameter assumptions and performs reasonably well in a wide range of possible future scenarios.
• Exploratory modeling or scenario discovery: developing a set of scenarios in a systematic manner to expand the understanding of how the future might unfold and trade-offs between different prioritization of objectives. • Vulnerability assessment: Leveraging the systematic scenario discovery process to evaluate when a chosen policy performs poorly using a set of predefined metrics.
Herman et al (2015) Table 4 shows the prevalence of key concepts in the literature. Robustness is more commonly discussed in academic literature compared to adaptive planning and multi-stakeholder engagement. However that prevalence did not translate over to practice in the grey literature. The lack of prevalence in implementing robustness concept is perhaps due to the technical complexity and knowledge gap on how uncertainties may impact planning outcomes, particularly in the context of climate change.

Robustness
The concept of robustness originated in ecology during the 1970's, when it was introduced to water resources planning in the context of climate change by Matalas and Fiering (National Research Council 1977). They defined a robust design as one that may not be the optimal choice under any one scenario, but that performs reasonably well under a variety of possible climate scenarios. Robustness further refers to the insensitivity of system design to any potential errors in assumptions affecting decision performance (National Research Council 1977). Although the mathematical backbone of robustness has evolved, the definition is still broadly valid. Forgoing the optimal choice may seem counterintuitive, but decision-makers are often willing to sacrifice optimality for robustness (Clímaco 2004, Difrancesco and Tullos 2015, Herman et al 2015, Rosenhead et al 2017. This preference for robustness is at odds with traditional decisionmaking approaches, where the goal is to produce the optimal outcome under the assumption that uncertainty is well-characterized (i.e. shallow uncertainty). Given the irreducibility and depth of climate uncertainty and irreversibility of infrastructure investment, approaching planning with the robustness concept enables decision-makers to identify viable solutions that work across many possible futures without needing to predict which of those futures is most likely.
In recent years, an increasing number of decision frameworks are integrating robustness concepts into their approach, including Decision Scaling (Brown et  Here, we focus on extrapolating robustness concepts and techniques that may be relevant to the electricity sector. The exact definition of robustness is dynamic depending on the decision context and stakeholders' attitudes (Giuliani and Castelletti 2016). However, the underpinning technique to achieve a robust outcome is by exploring how different policy alternatives perform in a wide range of plausible scenarios and evaluating trade-offs among multiple performance measures (Moallemi et al 2020). There are four distinct components within a robust decision analysis: a set of decision alternatives, a range of plausible scenarios, performance measures or robustness metrics, and vulnerability assessment or robustness controls. We will provide a simplified overview of each component, refer to Herman et al (2015) for a more comprehensive discussion.
(a) A set of discrete policy alternatives. These are the different decision options that may be implemented based on future scenarios and current institutional constraints. They may be pre-specified by decision-makers or developed through a computational search. For example, a discrete policy alternative might be a set of investment choices generated by a capacity expansion model under a particular set of inputs, including a given climate scenario. This policy alternative may then be evaluated alongside other alternatives developed by running the same model with different inputs (see (b) range of plausible scenarios). In practice, the set of decision alternatives is more likely to be prespecified due to resource constraints. A multistakeholder engagement process may be effective in ensuring different objectives are incorporated into the initial policy alternatives and determining whether adaptive planning is necessary or possible. (b) A range of plausible scenarios that considers all uncertainties and translates them into parameters in the model. For example, uncertainties around the change of temperature or precipitation in the next decade would become parameters with assigned plausible ranges. The different scenarios are different combinations of those uncertainties. A multi-stakeholder engagement process may be effective to create a comprehensive list of uncertainty factors and agree on plausible ranges for uncertainty parameters. The goal is not to predict future scenarios rather to explore what could happen. Each policy alternative is evaluated in each of the scenarios using robustness metrics. (c) Performance measures or robustness metrics for each policy alternative are a list of criteria that policy alternatives are evaluated against in each scenario. Stakeholders and analysts may codevelop multi-objectives and robustness metrics. Robustness metrics are developed in a similar manner to traditional decision-making approaches, where decision-makers choose specific metrics to evaluate policy performance based on the decision context (i.e. properties of the problem space such as regulatory environment and technical constraints). Example robustness metrics are GHG emissions, system reliability, and investment cost. The performance of one policy alternative may be evaluated using a combination of several robustness criteria. Robustness is commonly assessed using two methods, least-regret or satisficing. Least-regret quantifies each policy alternative's deviation from expected performance based on robustness metrics and identifies the policy that deviates the least over a wide range of plausible futures (Lempert and Collins 2007). Satisficing evaluates each policy alternative against the robustness metrics and identifies the policy that performs reasonably well compared to the alternatives over a wide range of plausible futures (Hall et al 2012). Robustness metrics provide decision-makers with a systematic approach to evaluate and compare each policy under a wide range of climate futures. (d) Vulnerability assessment or robustness controls refer to the process of isolating the uncertain factors most responsible for system vulnerabilities and examining where policies could fail (i.e. produce unacceptable outputs with respect to the performance metrics). A common technique is scenario discovery, which simulates the performance of different policy alternatives under a wide range of plausible futures and identifies where some policy alternative may fail (Bryant and Lempert 2010). This is similar to sensitivity analysis in traditional decisionmaking approaches. The goal is to help decisionmakers target specific vulnerabilities and develop adaptive strategies to address them (Haasnoot et al 2013).
In the grey literature, robustness is discussed as an overarching concept but it is rare for planners to conduct and implement robustness concept in the standard decision-making process. Some reports acknowledge the need to move away from deterministic and definitive prediction about future climate conditions and the importance of incorporating uncertainty by planning for wide range of futures (Means et al 2010a, US Department of Energy 2016, World Association for Waterborne Transport Infrastructure 2020). However, this is often not operationalized; there are only several pilot projects in the water resources sector that leverage RDM to test its technical capability and value of running a wide range of future scenarios. One notable example is the work done by the Bureau of Reclamation in the Colorado River (

Adaptive planning
Adaptive planning is another tool used to cope with uncertainties in long-term infrastructure planning. Adaptive plans are designed to be iteratively updated with newly available information, leading to more resilience against uncertainties and more effectiveness in guiding future actions. Similar to the concept of robustness, adaptive plans also examine a wide range of uncertainties and plausible futures (Walker et al 2013a). In addition, adaptive planning refers to the flexibility of a policy alternative to change course based on the evolving environment through continuous learning (Rosenhead et al 2017, Haasnoot et al 2018. The objectives are to reduce path dependencies due to technological, institutional, and behavioral lock-ins (Fouquet 2016, Maier et al 2016 and increase tolerance to future uncertainties (Jeuken et al 2015). To achieve those objectives, adaptive planning identifies short-term actions and long-term options by systematically monitoring the environment, gathering information, and iteratively adjusting strategies to new circumstances (Yzer et al 2014). The literature presents several frameworks that encourage adaptive planning, such as Assumption Based Planning (Dewar and Wachs 1993), Adaptive Policy Making (Kwakkel et al 2010), Engineering Options Analysis (Smet 2017), and Dynamic Adaptive Policy Pathways (Haasnoot et al 2013). The fundamental idea across those frameworks is that coping with uncertainty involves a robust short-term strategy and monitoring for changes that indicate a need to develop new strategies. Here, we again focus on the common and essential components of adaptive planning.
Adaptive planning can either be static or dynamic. Static adaptive planning aims to protect a basic policy from failing through contingency actions and monitoring (Walker et al 2001, Haasnoot et al 2013. Dynamic adaptive planning goes beyond contingency planning and aims to monitor policy performance over time and develop alternative policies to switch to when certain thresholds are met . The success of any adaptive plan depends on monitoring and anticipating ongoing developments such as changing climate and social context. Both approaches consist of two crucial components: a monitoring plan and a list of signposts (Maier et al 2016). We provide a simplified overview of those two components. Refer to Herman et al (2015)  needed. An effective signpost consists of three quality criteria: salience, credibility, and legitimacy (Cash et al 2005). Salience refers to how decision-relevant a signpost is. A salient signpost will provide insights to address policy concerns and is measurable, timely, and reliable. Credibility refers to how scientifically sound are the critical values assigned to each signpost and how convincing they are at motivating potential changes needed. Legitimacy refers to the acceptability of the technical process around data gathering.
In grey literature, planners discuss the idea of adaptive plan within the context of adaptive capacity. It is often defined as the ability to adjust to change via building redundancy, resilience or recoverability (Johnson 2012, US Department of Energy 2016, World Association for Waterborne Transport Infrastructure 2020). The water sector is most advanced in incorporating ideas of monitoring plan and triggers when discussing adaptive planning.

Multi-stakeholder engagement
Multi-stakeholder engagement is an essential component for incorporating uncertainties into the planning process. Successfully developing a robust outcome and an adaptive plan relies on gathering decision-relevant information and considering perspectives from different stakeholders throughout the planning process (Babovic et al 2018). Conducting multi-stakeholder engagement is especially crucial in the context of climate planning because misrepresentations of future scenarios may exacerbate negative climate impacts on marginalized populations (Jafino et al 2021).
The literature presents many different frameworks of conducting multi-stakeholder engagement for infrastructure planning (Tompkins et al 2008, Gardner et al 2009, Herman et al 2014, Mok et al 2015, Bourne 2016, Cuppen et al 2016. A 'one size fits all' approach for multi-stakeholder engagement does not exist, since it is highly dependent on the decision context and planning stage (Rountree et al 2021). Moreover, there are discrepancies around the exact definition of a 'stakeholder' (Carney et al 2009).
Here, we use the IPCC's definition: stakeholders are individuals or groups that can influence or may be affected by the decision outcome. They might be policy-makers, scientists, communities, and/or managers in the sectors and regions most vulnerable to infrastructure failure (Rowe and Frewer 2000, Conde et al 2004. Stakeholder engagement is broadly defined as any activity such as a survey, interview, or interactive workshop with the purpose of information exchange between analysts, decision-makers, and stakeholders. There are some common frameworks infrastructure planners use to conduct stakeholder engagement. Delphi Method is commonly used to generate parameter bounds for climate scenarios (Grime and Wright 2016). Approaches such as the rapid assessment process and participatory planning are also common ways to understand different stakeholders' perspectives and generate scenarios (Meadow et al 2015). There is much room for improvement to increase input from affected stakeholders, engage with diverse values, and examine conflicting objectives during the planning process to ensure procedural equity (Cradock-Henry et al 2020, Leal Filho et al 2021. Stakeholder engagement in the energy sector can take different forms. Currently, these include intervention by non-profit and community organizations in regulatory proceedings, public comments on infrastructure siting decisions, and technical advisory groups established by regulators on specific topics (Baldwin et al 2018, Solman et al 2021. A full review of stakeholder engagement practices is out-of-scope here, but this topic deserves additional attention in the context of questions about procedural and distributional equity that have been documented by other researchers in infrastructure planning efforts (Sovacool et al 2016, Heffron and McCauley 2017, Fletcher et al 2022. Procedural equity starts with considering how to represent perspectives of affected community members by selecting a wide range of stakeholders and ensuring they are involved throughout the planning process. We offer some additional thoughts on this in section 4.3.
Although multi-stakeholder engagement takes many forms, the common objectives for engaging stakeholders in the context of climate-aware decision making are fourfold: (a) Co-produce decision-relevant metrics to measure robustness and monitoring plan performance; (b) Increase transparency to the planning process with greater input and feedback from stakeholders regarding their preferences; (c) Seek active support and build consensus from stakeholders for the decisions which are made; and (d) Incorporate diverse perspectives from different stakeholders.
Infrastructure planners cannot effectively implement the concepts of robustness and adaptive planning without multi-stakeholder engagement. Inputs from stakeholders are critical for defining policy success or failure. Many frameworks for decision-making under uncertainty include stakeholder engagement as an important step of the planning process. Ideally, analysts and decision-makers would engage different stakeholders to understand various priorities and develop decision-relevant performance criteria at the beginning of the planning stage (Ranger et al 2013, Stanton andRoelich 2021). Then analysts would translate those insights into parameters and constraints during the modeling process (Bhave et al 2016). Once modeling is complete, stakeholders are brought back for deliberation to build consensus and iteratively improve the decision outcome (Lempert et al 2003). Moreover, having a boundary spanner or boundary organization during the engagement process may help to increase effectiveness of the coproduction process. The role of boundary organization is to facilitate the engagement process by allowing scientists and decision makers to maintain their independence and objectivity while also creating some permeability of the boundary to co-produce robustness metrics and monitoring plans (Clark et al 2011).
The grey literature suggests that multistakeholder engagement is already ubiquitous in large-scale long-term infrastructure projects as a way to manage risk and obtain public buy-in. However, there is a literature gap in how to conduct stakeholder engagement to effectively incorporate uncertainties and reconcile competing priorities (Stanton and Roelich 2021).

Electricity sector planning under climate change
Researchers studying electricity systems in the context of climate change have taken a variety of approaches to this subject. To facilitate our review, we first sorted the identified papers into topic areas based on an initial survey of their titles, abstracts, and keywords (table 5). These topic areas formed clusters of papers with some commonality in their perspective, scope, approach, and/or methodology. We reviewed papers in each topic area to understand how the authors considered climate change and uncertainty, and assessed to what extent the three key concepts identified in the decision science review (robustness, adaptive planning, and stakeholder engagement) showed up in this literature. A full review of modeling methodologies used in the energy sector is out of scope here, but we direct readers to Keirstead et al ( (2021) for more discussion on the types of models used.

Climate change and uncertainty
Uncertainties, including those inherent in climate predictions as well as those stemming directly and indirectly from climate change (see section 3.2), have been considered in various ways at different stages of analyses. Researchers acknowledge the presence and potential impact of uncertainties such as technical system characteristics (e.g. generator performance, line capacity), policy goals (e.g. renewable energy targets, carbon prices), and consumer behavior (e.g. rates of technology adoption, demand growth). However, to capture these uncertainties, there is widespread reliance on tools that would only be appropriate for shallow uncertainty. These include directly estimating individual values, perhaps with some sensitivity ( Relying on value estimation and/or probabilistic estimates may further enable researchers to resolve parameter uncertainty within their analyses and report a sole solution. This raises the concern of seeking an optimized solution, which may inadvertently suggest to decision-makers that it is possible to model perfect knowledge (i.e. fully settle uncertainty) when the future is not deterministic (Chattopadhyay et al 2016). An alternative is to depict uncertainty directly in results, for example by presenting a possible range of outcomes (Willems et al 2019, Bloomfield et al 2021. Such results may more appropriately convey the continued need to consider uncertainty when presented for discussion in multi-stakeholder engagement processes. Further, few researchers acknowledge model uncertainty (see structural uncertainty in Sharmina et al 2019). Model uncertainty is present in all types of modeling exercises, including climate models (section 3.2.1) and capacity expansion models (section 3.3), but, unlike in the climate modeling literature, little attention is paid in the electricity planning literature to systematically evaluating a diversity of structural assumptions in model design.
Authors who explicitly use climate information in their analyses also take a variety of approaches. In some cases, a limited set of climate information (e.g. one emissions scenario and GCM) are used as a case study (Chen et al 2021). Others include more climate information to explicitly represent possible futures, for example by using several GCMs with one emissions scenario (Miara et al 2019, Peter 2019, or several emissions scenario with one GCM (Santos da Silva et al 2021). In some cases, climate model outputs are aggregated (for example, by averaging their results) (Miara et al 2019). Other researchers evaluate climate models as distinct plausible futures, often as a set with the goal of creating a reasonable range on the phenomena of interest alongside several emissions scenarios (Parkinson and Djilali 2015, Spalding-Fecher et al 2017, Sridharan et al 2019, Voisin et al 2020, Figueiredo et al 2021, Fonseca et al 2021. Importantly, authors in the reviewed literature commonly focus on specific types of uncertainty within a particular analysis rather than capturing the full range of uncertainties that may exist within a particular decision-making or planning process. For example, some authors use a range of GCMs and emissions scenarios while holding other assumptions constant (Li et al 2014, Miara et al 2019, Peter 2019, Voisin et al 2020, Fonseca et al 2021, while others use no or limited climate information but vary assumptions related to socioeconomic factors (Chen et al 2016). Such approaches emulate the concept of controlled experiments and may provide valuable scoping or case study information. However, they do not account for the full range of plausible scenarios that should be considered in a robust process.

Robustness
Relatively few reviewed papers focus on ensuring robustness in their decision outcomes (for an example, see Nahmmacher et al 2016). However, electricity sector researchers have developed computational techniques that capture components of the robustness concept. These techniques are designed with the acknowledgement that a single prediction of a possible future is inadequate, multiple plausible trajectories exist, and it is therefore worthwhile to consider how a policy or decision may perform across those potential futures. Techniques include: (a) Stochastic programming tools may incorporate different plausible futures and weight them by specified probabilities. The overall objective is then to minimize the expected system cost across all futures (Chattopadhyay et al 2016, Ji et al 2020b. (b) Robust optimization is frequently used to determine the best (or, least-worst) solution for a given scenario, but such tools may also consider multiple futures, with the optimal solution then defined as one that demonstrates satisfactory performance (or, avoids the worst outcomes) across alternative scenarios (Li et al 2014, Parkinson and Djilali 2015). (c) Trade-off approaches start from the development of a vast possible solution set, then iteratively screen out inferior approaches as evaluated by predefined metrics (Heinrich et al 2007). A final solution set is evaluated against all proposed futures.
Beyond these techniques, authors in the electricity planning literature selectively employ components of robustness approaches. For example, authors commonly include the development of multiple plausible scenarios and performance metrics. Some also include a vulnerability assessment or robustness controls (Chen et al 2016). It is less common in the electricity sector literature that we reviewed to explicitly incorporate multiple policy alternatives within the design of a set of scenarios (for examples, see Malekpour 2018, Bloomfield et al 2021). While few authors incorporate all elements of the robustness concept as discussed in the decision science literature, these frameworks nevertheless present scaffolding on which additional components of robustness could be integrated. It is important to note that these approaches may quickly become computationally intensive with the generation of additional scenarios and incorporation of policy alternatives.

Adaptive planning
Planning in the electricity sector is heavily influenced by decisions that have already been made; for example, existing investments will impact where new investments will need to go. Recognizing the importance of long planning timescales with multiple decision points, researchers have made use of multistage modeling tools to create the foundation for evaluating system conditions and resolving uncertainties over time (Szolgayová et al 2012, Ji et al 2017, 2020a, Peter 2019. For example, models may be run once to a selected point in time, then evaluated again based on new information about how conditions have evolved (Chattopadhyay et al 2016). Such two-stage processes are particularly useful for uncertainties that have definite dates of resolution.
Other modeling tools may prioritize maintaining flexibility throughout a given timeline, for example, by selecting scenarios that maintain flexibility towards long-term uncertainties (see stochastic programming with recourse in Heinrich et al 2007) or by incorporating feedback signals for real-time adaptability (Hung and Chang 2017). Other authors discuss adaptation options but do not explicitly incorporate them into analyses (Nierop 2014, Burillo et al 2019.
Such approaches differ from the adaptive planning methodologies proposed in the decision science literature as they do not necessarily incorporate a monitoring plan or signposts, rather relying on discrete points in time and/or metrics of flexibility. However, they may provide a computational basis for incorporating those components to improve planning resiliency.
To some extent, current electricity sector practices do incorporate elements of adaptive planning, specifically via regulatory proceedings that require utilities to evaluate and justify their investment needs every 3 years (section 2.2). However, these short timescales do not facilitate looking ahead at future climate conditions, and may thereby further path dependence and technology lock-in. This could occur, for example, by investment in conventional grid hardening at the expense of building flexibility into the electricity system through distributed assets. Moreover, planning on short timelines neglects the impact of long-term climate trends (e.g. sea level rise) that may threaten technology investments with long implementation and performance timelines (e.g. new generating facilities built near coastlines). Instead, adaptive planning principles call for proactively tracing out the different long-term options when certain trigger points are met, thereby giving signals to electricity sector participants to lay the groundwork for new approaches (e.g. legislative or regulatory actions that create markets for energy storage).

Multi-stakeholder engagement
We encountered relatively few papers that explicitly consider multi-stakeholder engagement. In some cases, authors consider assumptions about risk tolerance or other decision-maker preferences (Szolgayová et al 2012, Ji et al 2017, 2020a, or formulate solutions to be decision-relevant (Burillo et  Other authors have considered stakeholder involvement theoretically, such as by developing frameworks for engagement (Araújo and Shropshire 2021), simulating preferences via agent-based models (Hoekstra et al 2017, Teixeira et al 2018, Hanna andGross 2021), or evaluating computational approaches capable of accounting for various perspectives by solving for multiple objectives and/or including appropriate uncertainty ranges on parameters (Heinrich et al 2007). Authors have also proposed that stakeholder engagement can be used to determine where to focus within a solution space once a set of solutions are obtained, and, as each individual view is necessarily a simplification of all considerations, to provide a check on other perspectives by helping ensure that a planning exercise accounts for needed nuances (Bollinger et al 2014, Nierop 2014).

Recommendations
We have identified three key concepts from decision science-robustness, adaptive planning, and multistakeholder engagement-that can be directly incorporated into electricity system planning and decision-making. Here, we return to capacity expansion planning (see section 3.3) as a concrete example to illustrate how these key ideas could be incorporated into electricity sector activities.

Improving robustness in capacity expansion planning
Current capacity expansion models used in industry practice do not adequately account for the range of uncertainties and plausible futures created by changing climate conditions (section 3.3). The electricity planning literature (section 3.5) contains ample examples of approaches to uncertainty that would be appropriate if those uncertainties were shallow. However, deep uncertainties present in climate change impacts and responses make even stochastic approaches inadequate.
Fully embracing robustness methodologies from the decision science literature would involve running capacity expansion models with a full range of climate, policy, behavior, and technology scenarios to identify the wide range of plausible futures, then evaluating the performance of investment options to identify those that perform acceptably well across those futures. Such an effort could involve thousands of relevant scenario combinations, and extensive computational and human effort to create scenarios, track performance, and interpret outputs. This approach would undoubtedly present electricity sector decision-makers with valuable information, but may face practical barriers to implementation. A more incremental approach may help ensure buyin and enable learning (as well as minimize mistakes) while implementing modifications to current industry practices. Alongside this initial incremental approach, there is an opportunity for the research community to investigate a much broader range of scenarios and develop recommended approaches for electricity sector decision-makers to utilize richer information in decision-making.
A key step towards building robustness into capacity expansion planning is to embrace the concept of multiple plausible futures rather than planning for one possible (however well-justified) future. In practice, this could mean that electricity sector planners: (a) Develop scenarios of plausible climate change futures on the basis of projections from multiple climate models (without averaging). Consider that plausible futures may be more diverse than just low/medium/high climate change scenarios. (For example, of the ten climate projections recommended for use in California, four labeled as 'priority models' represent qualitatively different futures, including 'warm/dry' and 'cool/wet' , see CalAdapt (2021).) (b) Develop combined scenarios where different climate futures are considered alongside scenarios that account for uncertainties in demand growth, market, policy, and technology options (see also Giudici et al 2020). Risk mitigation options (e.g. greater deployment of batteries, demand response, operating reserves, and other strategies that contribute to system robustness) could also be included in combined scenarios. (c) Select several of these combined scenarios that may complement and/or form a plausible range when evaluated in conjunction with a median combined scenario. (d) Run capacity expansion models with a median and several range scenarios and evaluate investment needs for each.
(e) For each set of calculated investment needs, assess how these investments would perform under all other scenarios to identify vulnerabilities in investment choices (i.e. evaluate the counterfactual impacts or unmitigated risks of planning for a lower demand and climate impact scenario if instead a higher demand and climate impact scenario is realized). (f) Evaluate thresholds for acceptable risks to how investments will perform and incorporate those thresholds into planning decisions.
Planners may also consider identifying which uncertainties are most consequential to the outcomes of interest and using those as a feedback loop to identify dimensions to explore further with additional scenarios. Such an approach, while limited in the number of scenarios assessed, would still create measurable progress towards building climate awareness and working with irreducible uncertainties.

Incorporating adaptive planning principles into capacity expansion decisions
Ranges of plausible outcomes can be defined for a variety of uncertainties, and adaptive planning principles can be used to refine those ranges as the future evolves. To aid this effort, decision-makers should develop a monitoring plan with input from modelers and stakeholders to identify signposts that signal when we need to take certain actions or change our planning assumptions. Notably, such adaptive planning principles were considered, though not ultimately prescribed, for inclusion in a recent CPUC decision requiring California's IOUs to publish climate vulnerability assessments (CPUC 2020a).
We offer three examples of signposts that could be applied to electricity sector planning in California: (a) When California's Sierra Nevada snowpack reaches <30% of historical baseline for 3 continuous years (Siirila-Woodburn et al 2021), stop counting on late summer hydropower resources in future models and deploy battery storage instead. (b) Define planning scenarios for low, medium, and high demand growth; if air conditioning adoption reaches >5% of households annually, know that we are in the high demand growth scenario. (c) If wintertime peak electricity use due to heating electrification hits >80% of summertime peak, know that we need to start evaluating climate scenarios for their impact on cold months.
We stress that these are just example criteria. While the specific implementations of these signposts will vary, the key concept here is that setting signals now can aid which plausible futures are assessed and which assumptions are made in capacity expansion models over time. Furthermore, there may be organizational barriers to incorporate and implement adaptive planning. Decision-makers may need to formalize this concept at the beginning of the planning process and provide institutional support to reduce resistance for the potential adaptive measures. For example, the decision-making entity may establish a working group with stakeholders and experts, assign specific personnel to gather and analyze the data, and build stakeholder consensus on the specific adaptive measures.

Multi-stakeholder engagement in capacity expansion
Capacity expansion planning for regulated utilities occurs within public rate cases that provide opportunities for stakeholder involvement. However, the complexity of growing climate impacts on utility operations will require increased attention on multistakeholder engagement as a tool to identify and refine uncertainties, plausible futures, and paths forward. Decision-making entities such as utilities and regulatory bodies could consult boundary organizations to translate technical climate science information and complex planning processes between different stakeholders.
Broad stakeholder input and increased transparency is necessary at every step of the capacity expansion planning process: (a) Determining appropriate inputs (climate data, demand data, infrastructure data). (b) Identifying relevant uncertainties and appropriate numerical ranges for uncertain parameters. (c) Developing salient robustness metrics to evaluate different capacity expansion policy outcomes under a range of plausible climate futures. (d) Establishing consensus on planning goals. (e) Evaluating each policy alternative's performance for different objectives and discussing tradeoffs among objectives.
Joint deliberation on how best to implement infrastructure plans must happen throughout the process in an iterative manner. At each step, key questions about the scope and mechanisms of stakeholder engagement efforts must be considered. For example: Who are the right set of participants to include, and how should conveners ensure they equitably represent necessary perspectives? Who will decide how stakeholder input should be weighted, resolve tensions, and balance the relative influence of different voices? What if some stakeholders do not want to engage and/or consensus is not possible? Further, what is the appropriate mechanism of and limits to stakeholder engagement, to ensure deliberations are not unduly sidetracked or prolonged at each step? These questions have been more thoroughly explored in the water sector than in the electricity sector, and evidence exists that insights from this work are not necessarily transferable between resource contexts (Rountree et al 2021). More foundational work is therefore needed here. Answering these questions is beyond the scope of the present review, but they merit additional focus in the context of climate-aware electricity sector planning.

Conclusions
Current planning processes in the electricity sector are not equipped to account for the stresses or uncertainties posed by a changing climate. Recent extreme weather events and service disruptions make it clear that climate change will continue to challenge electricity sector performance and operating practices. To better address these impacts, utility decisionmaking and planning approaches must incorporate a greater awareness of climate change and associated uncertainties. Here, we perform a three-part literature review to distill lessons from decision science researchers and assess how they might be incorporated into electricity sector planning to better account for the uncertainties posed by climate change.
First, we refer to the decision science literature to characterize climate-related uncertainties as shallow and deep uncertainties. This framework provides insights on how quantitative approaches may appropriately capture uncertainty. For example, shallow uncertainties (levels 1-3), where reasonable estimates of the relative likelihoods of different outcomes are possible, may be appropriately represented through probabilistic or trend-based scenarios. Deep uncertainties (levels 4-5), where multiple alternatives cannot be ranked by likelihood, cannot be reduced to statistical treatments.
Climate-induced uncertainties that must be considered in electricity sector planning include uncertainties in climate conditions as well as their direct and indirect impacts on electricity infrastructure, consumer behavior, and decision-maker responses. Efforts to characterize these uncertainties must consider complex interrelationships and deeply uncertain climate futures. Therefore, incorporating climate awareness into electricity sector planning will require making use of decision science practices that are equipped to handle deep uncertainty.
We distill three key ideas from decision sciencerobustness, adaptive planning, and multi-stakeholder engagement-that have been developed to deal with deep uncertainty and may help incorporate climate awareness into electricity sector planning and decision-making. Robustness requires considering multiple plausible futures and identifying planning approaches that perform acceptably well across them. Adaptive planning uses signposts and monitoring plans to refine plausible futures in time, building the institutional capacity to respond as climate conditions evolve. Multi-stakeholder engagement offers the potential to iteratively consider policy objectives and possible responses to adapt to new conditions, as well as to reconcile competing objectives.
Existing literature on electricity sector planning under climate change incorporates these key ideas to varying-but often only limited-degrees. However, existing tools that have been developed for use in the electricity sector, such as robust optimization and multi-stage modeling, can provide a scaffolding for implementing relevant insights from decision science.
Further, these three key ideas can guide actionoriented steps to build climate awareness in the electricity sector. We offer recommendations for how insights from these key ideas can be implemented within capacity expansion planning, a key area of decision-making in the electricity sector that is affected by all levels of uncertainty posed by climate change. Specifically, embracing the concept of multiple plausible futures, setting up signposts that signify tipping points as the future evolves, and including boundary organizations during stakeholder engagement can help electricity sector decisionmakers better plan for the future.
Future work is needed to better connect lessons from decision science to industry practice. In particular, we suggest that social science research may provide insights on institutionalizing concepts of robustness and adaptive planning by decisionmakers, and more effectively communicating uncertainty to and involving stakeholders in planning processes.
Climate change poses significant and ongoing challenges for critical infrastructure systems. Here, we offer some initial steps that can be taken today to help decision-makers incorporate climate awareness into electricity sector planning.

Data availability statement
No new data were created or analysed in this study.

Acknowledgments
We are grateful for helpful comments from Jill Moraski, Nichole Hanus, Emma Tome, and David Anthoff. Kripa Jagannathan steered us towards relevant grey literature sources. Any remaining errors or omissions are our own. This work was supported in part by the Office of Science, Office of Biological and Environmental Research, Climate and Environmental Science Division, of the U.S. Department of Energy under Contract No. DE-AC02-05CH11231 as part of the HyperFACETS Project, 'A framework for improving analysis and modeling of Earth system and intersectoral dynamics at regional scales' (Award No. DE-SC0016605). Publication made possible in part by support from the Berkeley Research Impact Initiative (BRII) sponsored by the UC Berkeley Library.