Planning Strategies for Sustainable Electricity Systems Amidst Uncertainty: Case Study of the Yukon, Canada Authors: Chris Fitzgibbon, I. David Elder, Heather L. MacLean, I. Daniel Posen Published: 2026 ABSTRACT Electric utilities face a growing challenge in maintaining low-cost and reliable electricity amidst emerging pressures for socio-environmental sustainability and reduction of greenhouse gas (GHG) emissions. In this work, we introduce an approach for generation expansion planning (GEP) that prioritizes sustainability objectives by implementing planning strategies that constrain future generation options based on socio-environmental indicators. Capacity expansion decisions under each strategy are then simulated and evaluated under a wide range of possible futures. Using the Temoa energy system optimization model, we adopt a unique myopic foresight approach in which cost-optimal capacity expansion decisions are locked-in based on forecasts of the future, before actual values are revealed. We demonstrate this approach for a case study in the Yukon, Canada, by identifying robust strategies with respect to cost, reliability, GHG emissions, ecological sustainability, and social sustainability. In our case study, more flexible strategies, such as restricting the development of diesel generation but not other types of thermal generation, improve socio-environmental outcomes at relatively low risk to system cost. Conversely, more restrictive strategies, such as avoiding all projects with potentially negative social or ecological impacts, risk substantially increasing system costs and may even have the opposite intended effect by inducing reliance on emergency diesel generators. Introduction Global demand for electricity is rising rapidly and with it the environmental, social, and economic consequences of new electricity generators. For electric utilities responsible for maintaining affordable and reliable electricity systems, there is a growing challenge in rendering public priorities into robust decisions for long-term investments in new generation capacity. Eliciting stakeholder perspectives has become common practice and energy projects in Canada are often proposed to achieve an array of social, economic, and environmental objectives, , but there remains a lack of tools available to model trade-offs. Previous research has demonstrated the merits and cobenefits of diversifying systems away from predominantly fossil-fuelled generation, but systems comprised exclusively of low-carbon alternatives are sometimes more expensive , or could face opposition for other reasons such as aesthetics, land use changes, etc. Various forms of renewable energy, encompassing technologies such as hydroelectric, biomass, geothermal, wind, and solar energy, have gained much attention in recent years as potential solutions to satisfying diverse agendas, especially in abating emissions of greenhouse gases (GHG). , Despite lower emissions, challenges remain in fully addressing socio-environmental sustainability with renewable energy as both renewable and conventional generators alike have the potential to alter terrestrial and marine habitats and biospheres, create auditory and visual disturbances, and/or affect energy affordability, among others. , Further, variable forms of renewable energy, principally wind and solar technologies, pose challenges to servicing demand year-round and can introduce risks to system stability and reliability. Nonrenewable energy sources present their own challenges and trade-offs, thus underscoring the complexities of planning electricity systems to reliably meet demand while minimizing negative impacts to socio-environmental sustainability and energy affordability. Generation Expansion Planning (GEP) literature attests to the complex nature of generation capacity planning decisions. GEP informs the “what, when, and where” of new generation capacity to fulfill demand while addressing cost, reliability, and, increasingly, externalities associated with electrical infrastructure. Energy System Optimization Models (ESOMs) form the computational engine upon which GEP relies to produce cost-optimal deployment of new generators. Attention has focused on improving the fidelity of these models to preserve the integrity of insights amidst growing complexities arising from increased shares of variable renewable energy and energy storage technologies along with relevant uncertainties. , Bridging the gap in integrating public objectives into GEP, researchers have incorporated various sustainability metrics into capacity expansion planning. Starting with exploring trade-offs between cost and various air pollutants, , studies have since incorporated participatory approaches into decision-making, − optimized for various nonfinancial objectives, − and quantified socio-environmental impacts. ,− Such GEP studies draw insights in the form of a discrete capacity expansion plan or series of plans with varying attributes for which a decision-maker can apply preferences for implementation. These impacts can be presented to the decision-maker as irreducible trade-offs that must be prioritized or collapsed into a single metric as is the case for many Multi-Criteria Decision-Making (MCDM) techniques. Another branch of GEP research assesses the effects of various policy instruments on system outcomes with studies assessing policies such as renewable portfolio standards, − emission trading schemes, emissions intensity standards, and various levels of carbon taxation. ,, Among the GEP literature which assesses the effects of policy on sustainability outcomes, three gaps are identified, which this study aims to address. First, decarbonization is overwhelmingly the intended outcome of policy instruments within GEP studies. , However, well-intended, over-reliance on GHG emission targets or renewable portfolio standards can distract from the more fundamental objectives which policies should be addressing, with Spillias et al. arguing such focus on decarbonization can even undermine wider sustainability goals. Second, all modeling efforts are vulnerable to the impacts of uncertainty on the validity of insights, − particularly where conclusions have policy implications. This holds especially true for electricity systems where many decisions are considered deeply uncertain, facing uncertainty across a range of techno-economic, political and socio-technical conditions, and where stakeholders may have diverse competing objectives. In such cases, even the range of potential outcomes may be unknown, and forecasts often amount to little more than speculation due to the lack of underlying probability distributions. Competing objectives may then make these outcomes difficult to evaluate. While methods for handling uncertainty have been developed, such as robust optimization, stochastic linear programming, and Monte Carlo simulations, conducting these analyses requires estimates of probability distributions or upper and lower bounds of uncertain parameters thus showcasing a challenge in representing deep uncertainty in energy modeling research with these techniques. Third, many capacity expansion models solve with “perfect foresight” whereby the model has complete insight into the future, contrasting with the capabilities of real-life decision-makers who rely on forecasts to make capacity expansion decisions. In response to this, models that make sequential optimizations along their time horizons based on limited knowledge of the future can better represent realistic decision-making conditions. , Research has shown that modeling with “myopic foresight” rather than perfect foresight leads to greater dependence on established technologies, increases the cost of meeting decarbonization goals, and often prioritizes short-term advantages at the expense of long-term benefits. , This approach remains limited in capturing long-term knowledge or trends, instead making a more binary assumption that perfect information is known in the short term and no information is known in the long term. , These gaps highlight a need for an approach that assesses broader sustainability and cost implications of electricity planning decisions under conditions of deep uncertainty. In this study, we present an approach that (a) articulates explicit socio-environmental objectives as a basis for high-level electricity planning strategies, and (b) simulates planning decisions amidst uncertainty, subject to such strategies. We simulate planning decisions using a capacity expansion ESOM that solves myopically but with a rolling horizon of uncertainty resolution, as demonstrated conceptually by Cuisiner et al. and McDonald et al. In this modified form of myopic foresight, information is revealed to the decision-maker (i.e., the cost-optimizing algorithm) sequentially such that only the present is known with certainty, with long-term forecasts providing the basis for present-day decisions which are subsequently locked-in for the future. To account for uncertainty, we incorporate an approach for modeling deep uncertainty implemented by Lempert et al. whereby a Monte Carlo simulation of equally likely future conditions is layered on top of the modeling framework, running a capacity expansion model many times but with different configurations for parameters such as fuel costs or electricity demand. Finally, negative impacts to socio-environmental sustainability of capacity expansion pathways are quantified for each Monte Carlo iteration by aggregating and capacity-weighting impact indicator scores for each new generator. Outputs are thus a distribution of impacts in various sustainability, reliability, and cost metrics across many different future conditions. By directly comparing pairwise iterations from the Monte Carlo simulation (i.e., with equivalent uncertainty realizations) across different planning strategies, our results can support decision-making despite the deep uncertainties being faced. Simulating decision-making in electricity system planning with this approach allows insights to be extracted by introducing policy instruments and observing the effects on sustainability, cost, and reliability outcomes across a wide range of future conditions. As a demonstration of the approach, we apply it to assess the impacts of various simple planning strategies on electricity system expansions in the Yukon, Canada, as a case study. We investigate whether such strategies can be expected to robustly progress socio-environmental sustainability in the Yukon without substantially increasing system costs. The Yukon is a remote Territory located between the 60th and 70th parallels in northern Canada and as of 2021 had a population of approximately 46,000 people. The Yukon has a small electricity grid, with a peak load of 104.4 kW in 2021 and total energy consumption (consumer and industrial) of 572 GWh. , The Yukon grid has no interties with its neighboring jurisdictions (Alaska to the west, Northwest Territories to the east, and British Columbia to the south), emphasizing the need for local resiliency; see Supporting Information (SI) section 1.1 for additional background on the Yukon’s energy system. These attributes contribute to exposing the Yukon electricity grid to a high degree of uncertainty, making the Yukon a relevant case study. Further, it is generally understood in the Yukon that there is an urgent need to undo harm from legacy infrastructure and better serve community and environmental needs, necessitating a planning method which holistically assesses the impacts of addressing sustainability objectives. Methods Study Design For this study we constructed a cost-minimizing capacity expansion model (see section for details on the model) that optimizes electricity generation capacity for the Yukon over a planning horizon spanning 2022 to 2050 in 4-year time steps using 2021 as the base year. Given the challenges in incorporating multiple criteria (particularly those which are qualitative) directly into a single objective function, we capture socio-environmental goals by first defining a set of fundamental objectives, then derive simple strategies which would intuitively advance each objective. Such strategies, for instance banning the construction of new fossil-fueled generators to reduce GHG emissions, are imposed as constraints to the capacity expansion model (see details on strategies in section ). With this model, we run a Monte Carlo simulation of many plausible future conditions pertaining to fuel costs, energy demand, load kurtosis, and capital cost of emerging technologies, among others, which are modeled as random walks through time (see details on uncertainty in section ). Within each iteration of the Monte Carlo simulation, we assume the decision-maker forecasts conditions 8 years into the future (i.e., two 4-year model time steps) and must lock-in capacity decisions, which are then subject to a construction lead time (see details on model foresight in section ). Generation capacity is expanded with a reserve margin of 30% and the model can select from a variety of new generation options (see Figure in section ). The future is then revealed in 4-year intervals and available generators are dispatched using the simple economic-dispatch component of the capacity expansion model. In the Yukon, peak demand for electricity (typically occurring in the winter when heating loads are high) is partially met with emergency rental diesel generators. This behavior is represented in the modeling framework by assuming dispatch of emergency 1.8 MW diesel generators wherever the model detects a supply shortage. Within each iteration of the Monte Carlo simulation, the model calculates system performance as it pertains to electricity costs, GHG emissions, reliability, stability, and negative impacts to social and ecological sustainability (see details in section ). Pairwise comparisons are done for the strategies against a common set of Monte Carlo draws. Due to computational limitations, our results are limited to 100 Monte Carlo iterations, which appeared to produce sufficient spread to identify riskier versus more stable strategies. A schematic overview of the study is presented in Figure . By comparing results across strategies and assessing performance across a wide range of uncertain futures, we demonstrate how decision-makers can study the implications of explicitly addressing socio-environmental sustainability objectives through targeted planning strategies or policies. Visual representation of study design. Strategies, derived from a fundamental objective, act as constraints to the capacity expansion model. A Monte Carlo simulation of techno-economic and demand parameters produces many least-cost capacity expansion pathways, which are then processed into a distribution of interpretable outcomes such as GHG emissions and electricity generation costs. Planning Strategies We apply an objective-driven structured decision-making approach to produce strategies that an electricity system planner could conceivably employ to meet a given objective (see Figure ). Objectives represent fundamental aspirations for the electricity system. For the purposes of this study and to illustrate the approach, we focus on four high-level objectives: maximizing electricity affordability, reducing GHG emissions, improving social sustainability, and improving ecological sustainability. These are simplified, high-level objectives that are intended as representative examples to demonstrate the approach; they are not intended to capture the full diversity of needs and priorities of residents and communities in the Yukon or elsewhere. From these objectives, we derive strategies that would intuitively advance the planning objective. For instance, with the goal of reducing the GHG emission intensity of the electricity system, we adopt strategies that have been proposed elsewhere such as avoiding construction of new fossil-fueled generation , or enforcing a regional cap on GHG emissions. For the emissions cap strategy, we impose an annual emission limit which roughly corresponds to an average grid emission intensity of 65 gCO2e/kWh to align with the facility-level GHG emissions intensity standard proposed by the Clean Electricity Regulation in Canada. Likewise, for the objectives of improving social and ecological sustainability, we test strategies that either directly target projects scoring poorly against these objectives, or generator types that typically face related concerns (i.e., hydropower, which has known impacts on aquatic ecosystems , ). As there is no single objective measure of social or ecological sustainability, we instead rely on low-medium-high scoring of each prospective generation project against 32 different socio-environmental impact indicators based on Intergroup Consultants such as habitat loss for at-risk species, air pollution, and land use. We aggregate and capacity-weight the scoring to produce one quantitative indicator each for negative impacts to social sustainability and negative impacts to ecological sustainability as described in section and in SI section 2.1. Regardless of strategy, emergency rental diesel generation is permitted under emergency circumstances as a back-up to service supply gaps, in line with actual practice in the Yukon as reported by Yukon Energy, the territory’s vertically integrated electrical utility. Figure shows the planning strategies tested in this study while SI section 2.2 provides more detail on their specific implementation. Planning strategies investigated in this study are derived from various planning objectives. Structured decision-making defines specific strategies each intended to achieve a fundamental objective. Capacity Expansion Model We use the open-source Tools for Energy Modeling and Optimization Analysis (Temoa) capacity expansion ESOM to model capacity expansions for the Yukon electricity system between the years 2022 and 2050 with a base year of 2021. The model minimizes the sum of capital, fixed, and variable annual costs, discounted to 2021 dollars using a discount rate of 5.45% to align with the rate assumed by Yukon Energy. Investment and fixed costs of new generators are before taxes and based on Class 5 (+100%/–50%) cost estimates from Yukon Energy planning documents, , which include the cost of transmission connections and are inflated to 2021 dollars using the Canadian Consumer Price Index. Fuel costs in 2021 reflect actual pricing as reported by Yukon Energy. The planning horizon constitutes eight uniform four-year periods between 2022 and 2050 with hourly resolution, based on a representative day framework. Lacking specific load data for the Yukon, we proxy the hourly load profile with that of the British Columbia (BC) grid in 2019, scaling the data to the annual energy and peak load of the Yukon grid in 2021. We deem this data to be a suitable stand-in due to its shared characteristics as a winter-peaking grid with a summer lull. To permit computational feasibility, we use 12 ordered representative days at hourly resolution. To select representative days, the annual load was broken into 12 unequal segments, of which the length was manually selected based on visually similar daily dispatch patterns with the model running at full hourly resolution. The day chosen to represent each segment was selected using hierarchical clustering over total load and availability of wind energy. The exception was the segment containing the peak load day, where the peak load day was manually selected to represent that segment (see SI section 2.3 for further discussion on representative days). We configure the model with existing generators based on publicly available data on the state of the electricity system in the Yukon in 2021. The model is permitted to select among new generators which have been proposed as part of recent generation expansion planning, , with technologies including diesel, liquefied natural gas (LNG) (the Yukon has no access to piped natural gas), hydroelectric, wind, solar, geothermal, pumped hydro storage, battery storage, and demand side management (DSM) (see Figure ). We elected to also include nascent technology popular in literature in this list of new generators, namely compressed hydrogen storage and nuclear small modular reactors (SMR). Due to its remote and limited infrastructure, opportunities for new development in the Yukon are finite. We therefore assumed capacity expansions of specific generators cannot exceed the installed capacity proposed in the planning documents, with the exception being thermal generation (i.e., diesel and natural gas) and battery storage as these technologies are not inherently limited by geography or physical footprint. We assume these technologies can instead be built up to four times the proposed installed capacity. The model is linearized, meaning new generation capacity can be built up to the maximum installed capacity of that generator or scaled down by any amount (the model does not establish which new technologies are viable but rather sizes capacities of predefined technologies). As a simplifying assumption, we model the grid as a single node thus ignoring transmission losses and constraints; implications of these modeling limitations are described in section . Proposed new generation projects in the Yukon as of 2021 (note that some generators are no longer being contemplated by Yukon Energy as of 2025). Our model allows new generators to be built up to their maximum capacity (shown in parentheses) or scaled down by any amount. Temoa is technology explicit, therefore each technology archetype is represented and configured uniquely in the model. Technologies differ in which constraints are applicable, but in general all technologies endogenize generation output given exogenous inputs for constraints such as availability factor, maximum annual energy, ramp rate, storage capacity, and efficiency (see SI sections 2.4, 2.5, and 2.6). Dispatchable generators are assumed to be able to dispatch instantaneously (i.e., ramp up to full capacity in less than 1 h). The exception is LNG: Yukon Energy reported the existing units run as baseload due to ramping limitations, so we assume a 24-h ramp rate. Hydroelectric generation is modeled by endogenizing reservoir level and electrical output using daily hydrometric inflow data and conducting a mass balance on the water system to establish the net water (energy) entering the system each day. Similarly, streamflow data at daily resolution is used to model run-of-river generators. A prospective intertie with BC is modeled such that it can sell to the Yukon as much excess energy as is approximately available on the BC grid for a given hour, up to the maximum transmission capacity of the proposed line. Sales from the Yukon to BC are assumed to be negligible. Wind and solar availability are represented by a site-specific hourly capacity factor value from a single representative year (2021). , Storage technologies are modeled as seasonal storage, i.e., looped per annum. DSM resources are assumed to simultaneously reduce peak demand and overall energy demand. Refer to SI section 2.5 for detail on technology representation in the model. In line with best practices for utility-scale generation expansion planning, we assign the Planning Reserve Margin a value of 30% to align approximately with the planning margin used by Yukon Energy (see SI section 3.1 for a sensitivity analysis). Effective Load Carrying Capacity (ELCC), the fraction of a generator’s capacity that can contribute to the Planning Reserve Margin, is assigned to existing generators based on historical availability with variable renewable generation (wind and solar) receiving no ELCC. Note that for the Emission_Cap strategy, fossil-fuelled generators are given no ELCC to prevent emitting technologies from contributing to the reserve margin despite their output being constrained. Uncertainty Parameterization We model input uncertainties either as events with binary realizations (e.g., a project failing or not failing to permit; number (n) of unique events = 35) or as variables with continuous possible realizations (e.g., future capital costs or electricity demands; n = 18), many of which are highly uncertain making parametrization inherently imprecise. We prepare our Monte Carlo simulation by randomly selecting realizations for each of these variables. We use the methods described below as an example of one of many possible approaches to parametrization (see SI section 2.7 for details on parametrization). Events with binary realizations are given a ‘probability’ of a nonnull outcome, estimated considering the Yukon context. Nonnull outcomes are resolved at some random time within the future time horizon. For this study we included the following binary events, of which the ‘probabilities’ were selected based on author judgment: connection of the Yukon grid to the grid in neighboring BC (number (n) of unique events = 1, probability (p) of occurrence within time horizon = 0.3), early retirement of the Yukon’s largest hydroelectric dam (n = 1, p = 0.1), availability of federal government funding for hydro projects thus reducing the capital expenditure (n = 2, p = 0.4), availability of renewable diesel (a low-carbon drop-in alternative to fossil diesel) (n = 1, p = 0.3), rejection of proposed amendments to water licenses allowing reservoirs to be drawn lower (n = 2, p = 0.5), and rejection of a project permit for any planned project (n = 28, p = 0.2). Strictly speaking, these are realization frequencies based on the author’s subjective judgment for our Monte Carlo simulation and are used to generate a wide set of plausible future outcomes for exploration and should not be interpreted as rigorous probabilistic outcomes. In future work, it may be possible to establish more empirically grounded Bayesian probabilities using techniques such as expert elicitation, however the current values (based on author-judgment) remain sufficient to demonstrate our proposed methods and to explore potential outcomes that inform overall success of the planning strategies explored. Acknowledging these limitations we also explore how these binary events correlate to outcomes of interest (SI sections 4.3 and 5.2) along with a sensitivity analysis on the specific probability modeled for the most influential event: early retirement of Aishihik dam (SI section 5.3). Input variables with continuous outcomes are modeled as random walks and are resolved chronologically throughout the model’s rolling time horizon. For simplicity, we assume no mean reversion and interpret the consequence as allowing for broader exploration of the parameter space. Realizations for these variables start at a known value from the 2021 base year and walk forward according to a standard deviation, assuming a normal distribution with the 2021 value acting as the mean (see Tables SI.7 and SI.8 in SI section 2.7). For some uncertainties known to rise steadily over time (e.g., electricity demand), a trend is applied to the random walk according to a linear regression of known past values. We modeled the following continuous variables: capital cost of emerging or nascent technologies (n = 7), fuel costs (n = 4), hydroelectric output (n = 1), total annual energy demand (n = 1), load kurtosis (n = 1), maximum efficacy of a DSM program (n = 1), and capacity of new distributed solar generation (n = 1). Although correlation between uncertainties can influence energy model results, we do not attempt to capture such correlations and instead opt for a more general exploration of the space defined by our input uncertainties, represented as a wide set of possible worlds rather than trying to define probable worlds with such correlations. Dynamic Myopic Foresight By default, Temoa solves with perfect foresight (Figure a). For this study, we instead use Temoa’s built-in myopic modeling capabilities, optimizing a predefined look-ahead window within the planning horizon and locking-in capacity expansion decisions for future windows (Figure b). We enhance this feature by updating model parameters prior to optimizing each myopic window such that only the first model period in the window is given the actual (i.e., resolved) parameters, with subsequent periods being mere forecasts. These forecasts may be (and often are) incorrect but are not recognized as such until the optimization window moves forward and reveals the actual values for the present. This simulates uncertainties resolving chronologically with only the present ever having true certainty. We select a three-period (12-year) myopic optimization window to align (approximately) with the duration of a typical planning horizon. We assume uncertain events and input variables are forecasted naively, as if they were to remain unchanged from the known present-period value, with the exception being annual electricity demand which instead makes a prediction for the two future periods using a linear regression of all known past values. Although this approach likely underestimates the forecasting capabilities of a real-world planner, simulating forecasts in energy models in this way nonetheless improves on incumbent methods. Capacity expansion decisions are carried forward to all future myopic windows, simulating lock-in. To represent lead-times for new generators, we allow the model to build new capacity immediately but constrain the generator’s energy output to zero for all periods for which the generator would be assumed to be under construction. Although Temoa calculates dispatch decision variables for all periods, we discard dispatch decisions for all but the present period to calculate our measured outputs; in other words, outcomes are only based on the realized dispatch, and not on the forecasts. Visual representation of the modeling approach with (a) perfect foresight and (b) dynamic myopic foresight. Myopic foresight is made “dynamic” by updating future forecasts based on knowledge of the past (upper right of panel (b)). In the upper right graph of panel (b), white circles represent a single realized run, compared to the (naively) forecasted values (black circles) that the model uses in the optimization. Forecasts are often incorrect, but capacity expansion decisions made with these forecasts are nonetheless locked-in. This process is repeated 100 times to produce 100 different future realizations in the Monte Carlo simulation (each with associated imperfect forecasts). Metrics To convert raw outputs from the capacity expansion model to interpretable outcomes, we calculate six metrics which allude to the performance of the electricity system; see Table for a high-level summary of these metrics. 1Summary of Metricsmetricunitsbasic methodaverage electricity generation costCA$/kWh delivered electricitytotal system costs ÷ total electricity demand over planning periodGHG emissions intensityg-CO2e/kWh delivered electricitytotal system GHG emissions ÷ total electricity demandreliabilitydimensionless (probability)loss of load probability at peak load hourstabilitydimensionless (ratio)maximum for any given hour: wind/solar output ÷ nonwind/solar outputnegative impacts to social sustainabilitydimensionlesscapacity-weighted aggregation of social sustainability indicators (refer to SI section 2.1 for more detail)negative impacts to ecological sustainabilitydimensionlesscapacity-weighted aggregation of ecological sustainability indicators (refer to SI section 2.1 for more detail) Summary of Metrics To remain agnostic to changes in demand, we measure electricity generation costs in dollars per kWh averaged over the time horizon using Temoa’s cost outputs. Likewise, GHG emissions of fuel-consuming technologies are reported as a lifecycle emissions intensity in g-CO2e per kWh, calculated exogenously with operational emission factors sourced from national GHG inventory reporting and embodied emission factors comprised of estimates for locally relevant manufacturing and transportation emissions. , For simplicity and ease of interpretation, all outcomes of interest are all reported as averages across the 28-year planning horizon even though model decisions are made by optimizing discounted prevent value of costs; the rolling decision window also makes it less meaningful to present a single discounted cost across the study horizon. Nevertheless, we further explore the possibility to incorporate the GHG externality (the only one easily translated into a monetary equivalent) directly into the optimization formulation using a discounted social cost of carbon (results in SI section 5.1). Reliability is proxied by two simplified resource adequacy metrics. Expected Unserved Energy (EUE) is the cumulative gap between supply and demand in a given time period, quantified in energy units. We calculate EUE endogenously in our model and use this value to quantify supply gaps, which are assumed to be met with emergency rental diesel generation that incurs both fixed and variable costs. Loss of Load Probability (LOLP) is the probability of demand exceeding the available generating capacity in a given time period (usually 1 h). At each hour, LOLP takes into consideration: (a) temperature-dependent probability of forced outage for each generating unit; (b) discharge capacity of energy storage technologies; and (c) stochastic output of variable renewable energy. LOLP thus represents a probability that there can be expected to be a loss of load event at the studied hour. The industry standard for acceptable LOLP is ≤0.1, which is interpreted as one loss of load event every ten years. We calculate a simplified form of LOLP using Monte Carlo simulation of resource adequacy for only the peak demand hour, which typically occurs in the winter to support heating loads and for our load data set was February 12th at 6 pm. Despite underrepresenting interyear variability, a static peak demand hour nonetheless approximates reliability impacts with light computational effort. Valuation of generation resources at this peak hour varies per generation type. Wind, solar, and run-of-river hydro generate energy probabilistically based on historical outputs from the past 20 years. ,, We assume that the prospective intertie with BC can contribute as much excess power as is available for sale at the Yukon’s peak load hour. Thermal generators generate up to their installed capacity. As a proxy for state of charge of reservoir hydroelectric and energy storage technologies, we assume the available energy is the predetermined ELCC. Boolean availability of each generator is determined by a random draw of forced outage at probability (1 – p) where p is the temperature-dependent forced outage rate. Note that total installed capacities are broken down by individual generator units when multiplied by the Boolean availability; this avoids overestimating LOLP by assuming that forced outages affect only individual generator units rather than an entire power station. Temperature-dependent forced outage rates for generators are reported by Murphy et al. For our peak day calculation, we assume outage rates at −15 °C, the average temperature in the Yukon in February. Electricity grids with high penetration of variable renewable generation (i.e., wind, solar) can be susceptible to frequency instabilities caused by low inherent inertia. To identify any deviation in grid stability compared to the BAU case, grid stability is proxied by the ratio of variable renewable energy (VRE) (which are frequency-destabilizing in absence of supplementary inertial control) to spinning generation (which are frequency-stabilizing) to implicate the inertia present on the grid. The ratio is calculated for every hour and is the sum of all energy outputs from nonspinning technologies divided by the sum of all energy outputs from spinning technologies. Impacts to social and ecological sustainability were assessed for proposed generation projects by Intergroup Consultants as part of Yukon Energy’s generation expansion planning. Existing generators were not scored as it was assumed that retirement of an existing generator would not lessen socio-ecological impacts as (a) retirement decisions are outside of the decision context of the model, and (b) we assume infrastructure has socio-ecological impacts beyond retirement (which is especially true for hydroelectric infrastructure). Intergroup Consultants rated projects on a 3-point Likert scale (low, medium, high) for their perceived (negative) impact in 32 different impact indicators (Table ) (see section SI 2.1 for scoring against each indicator). 2Breakdown of Socio-Ecological Indicators category subcategory criterion impact indicators social sustainabilityN/Apfirst nation landsfirst nation settlement landstraditional lifestylefootprint land area impactland area loss for traditional lifestyleland quality effects on traditional lifestylecabins, camps, and structurescountry foodsheritage resourcesdensity of heritage resourcesimportance/cultural value of heritage resourcestourism, recreation, and other resources and land userecreational valuestourism valuesaestheticsnon-renewable resourcesother renewable resourcesland use and renewable resources planscultural and community well-beinginfrastructure and servicespublic safety, worker interaction, human and community healthcommunity, first nation, and personal developmentair qualityGHG emissions other air pollutantsecological sustainabilityaquatic environmentfish and fish habitatsalmon and habitatspecies at risk and habitatcommercial, recreational and aboriginal fishes species and habitatwater quantity and qualityconsumptive water userelative scale of new impoundment/floodingflow changesterrestrial environmentterrestrial species and habitatspecies and risk and habitatprotected and conservation areaswildlife key areascaribou rangesterrestrial footprint and land usefootprint terrestrial arealinear development for roads/transmissionpermafrostwetlandsaIncluded in the original report by Intergroup Consultants are indicators for “local economic benefits” and “climate change risk”, but were omitted from our analysis to limit the scope to social and ecological impacts.bOmitted from final scoring as GHG emissions are broken out as a separate output. Breakdown of Socio-Ecological Indicators Included in the original report by Intergroup Consultants are indicators for “local economic benefits” and “climate change risk”, but were omitted from our analysis to limit the scope to social and ecological impacts. Omitted from final scoring as GHG emissions are broken out as a separate output. To aggregate indicators, Intergroup Consultants averaged and rounded up the indicators for each criterion (toward high impact) to produce a single score. While it is recognized this approach implies a strong value judgment that assumes equal weight to indicators within a criterion, for the Yukon case study we simplified the categories to enhance interpretability of decision-relevant outcomes. Four prospective generators used in this study (Primrose Lake hydro, emergency rental diesel, hydrogen storage, and nuclear SMR) were not assessed by Intergroup Consultants. For Primrose Lake hydro and emergency rental diesel, scores were imputed based on similar projects, taking the higher (more negatively impactful) value where similar projects of the same technology type had different scores. For emerging technologies (hydrogen storage and nuclear SMR) which did not have similar projects to baseline against, scores were generated for each indicator category against the rubric used by Intergroup Consultants, taking a conservative approach. To generate aggregated scores in the social and ecological sustainability impact categories, the criteria score (low = 0, medium = 1, high = 2) was first multiplied by the installed capacity in kW (or annual energy in kWh for air pollution indicators), then divided by 1 × 105 to produce a number roughly between 0 and 10, then averaged to produce the categorical score. Intuitively, this capacity-weighted (or activity-weighted) score represents the proportionally larger impacts of larger projects. Although this approach is simple, it provides a first step in indicating high-level implications of policy or strategy decisions. Results and Discussion Results Figure shows the capacity expansion outputs as of 2050 for each of the 100 Monte Carlo iterations for each strategy. As the model is linearized, capacities can be built up to a technology’s specified maximum capacity (noted explicitly on the figure) or scaled down by any amount. This is visualized in Figure as many technologies were not always built to their full capacity but rather some portion thereof. As electricity demands in the Yukon are small compared to other jurisdictions and the number of discrete projects that can fulfill each generation type is limited, the effect of this linearization is that the model can to cost-optimize more so than is possible in reality; for instance the model may choose to build a small amount of a hydroelectric project to meet demand but in reality the whether to build a project is a go/no-go decision. However, we compare strategies pairwise for each Monte Carlo iteration relative to the business-as-usual (BAU) scenario to enable meaningful comparison in impacts among strategies, negating effects of under-reporting total system cost (and other metrics) stemming from linearized capacity expansions. Capacity expansion results at the end of the model’s temporal horizon (year 2050) for all 100 Monte Carlo iterations, per technology for each strategy. Being a linearized model, capacity can be expanded up to the maximum capacity of each technology (black horizontal line) or scaled down. It is assumed that emergency rental diesel generation has no maximum capacity and can expand indefinitely to meet supply gaps. Nuclear SMR technology has an assumed maximum capacity of 300 MW, but the y-axis is scaled down for clarity. Figure illustrates the impacts of imposing eight different objective-driven planning strategies to the Yukon electricity system over the 28-year planning horizon for 100 unique future conditions (see SI for unprocessed model results: section 4.1 shows the Monte Carlo outcomes in absolute terms to showcase associated ranges for each metric, section 4.2 shows economic-dispatch results, and section 4.3 illustrates the relationship between the realization of uncertainties and the metrics). Overall, most strategies are effective at advancing their intended objective, but we find that all strategies have trade-offs with cost. Strategies aimed at reducing negative impacts to social and ecological sustainability (orange and green) can abate both externalities simultaneously but at the expense of higher emissions. Among eight strategies, Emission_Cap (purple), No_Fossil (dark blue), and No_Soc_Impact (dark orange) risk unacceptably high costs (i.e., increase more than 30%) with No_Soc_Impact failing to achieve its objective entirely. In the case of No_Soc_Impact, avoiding too many generation options overconstrains the energy system and results in unintended negative consequences to social sustainability as undesirable technologies (primarily emergency rental diesel generation) are required to maintain reliability. A similar effect is observed for the Emission_Cap (purple) and No_Fossil (dark blue) strategies, where a failure to build sufficient permanent capacity resulted in an overreliance on emergency rental diesel generators, incurring high fixed costs (from renting the generators) and operational costs (from consuming diesel fuel). These “tail” risks are an important feature of optimizing with imperfect forecasts that can sometimes result in dramatically different outcomes from what was intended/expected. Pairwise comparison of system performance under various planning strategies relative to business-as-usual (BAU) planning for each of 100 Monte Carlo iterations (see Figure for strategy descriptions). Values represent percent change compared with BAU. Cost and emissions metrics are reported as averages over the 28-year planning horizon. Social and ecological impact indicators are aggregated 3-tier impact scoring of individual projects, weighted linearly by installed capacity. Both strategies aimed at reducing ecological externalities (No_Hydro and No_Eco_Impact) achieve their intended objective in 100% of futures. Avoiding only generators scoring “high” for social sustainability impacts (No_Hi-Soc_Impact) likewise appears effective, raising costs on average 1% but improves social sustainability in 91% of futures. Of the four GHG emission abating strategies, No_Diesel (mid blue) is the only strategy to show meaningful effectiveness in reducing GHG emission intensity without risking unacceptable increases in cost, with an average emissions reduction of 31% and costing on average 12% more and up to only 26% more than the BAU strategy. The implications of this analysis are that less constrained strategies, such as avoiding only the worst-performing technologies while permitting others, are more effective at efficiently achieving their target objective without disproportionately impacting other metrics. This also demonstrates that the unintended consequences of overly restrictive strategies can have severe impacts with no significant advantage over more moderate approaches. Figure shows the capacity expansion results for eight select scenarios of which the selections were based on the iteration which had the highest and lowest outcome to each of the four metrics (costs, GHG emissions, impacts to social sustainability, and impacts to ecological sustainability) to showcase the diversity of capacity expansion outcomes. With this analysis, driving factors behind various outcomes emerges: for instance, deployment of abundant pumped hydro is found in the iteration with lowest electricity costs suggesting this technology has high economic value. Given the Yukon has a highly seasonal load profile, long-term energy storage such as pumped hydro intuitively has high potential to be cost-effective in the Yukon context. Similarly, a high instance of fossil thermal use in the scenario with the highest social impacts is consistent given these technologies have high impacts to air quality and overall community well-being (see scoring in SI section 2.1). Additionally, the dramatic emergence of fossil thermal capacity late in the planning horizon suggests an unexpected uncertainty resolution resulted in rapid deployment of emergency rental diesel generators. Capacity expansion results for select Monte Carlo iterations, presented as a bar graph for existing (hashed) and new (solid) generators. Selections are based on scenarios with highest and lowest impacts for each measured output. Effects of strategies on reliability and system stability were estimated using simplified resource adequacy metrics serving as a proxy for more established reliability calculations. Shown in Figure are the effects of planning strategies on peak-hour LOLP and maximum VRE penetration. All planning strategies exhibit comparable or lower LOLP than the BAU strategy, averaging just above the industry acceptance criterion of 0.1 which represents one loss of load hour every ten years. LOLP is lower for GHG-abating strategies owing to reduced shares of diesel generation, which suffer high forced outage rates in winter temperatures. The fraction of load met with VRE generation remained similar in all cases. Estimates for system reliability are comparable among planning strategies. Loss of load probability and maximum hourly ratio of VRE generation relative to spinning generation are presented for 100 Monte Carlo iterations for each planning strategy, including the BAU strategy (red on far left). Supply shortages are detected by the model as EUE, which is where demand exceeds supply in any given hour and emergency diesel generators are dispatched to maintain supply. EUE was observed only for strategies which restrict deployment of permanent diesel generation (Emission_Cap, No_Fossil, No_Diesel, and No_Soc_Impact), with EUE appearing in 3 to 56% of simulated futures for these strategies. EUE is typically a result of running out of prospective generation options with short lead times in cases where the myopic model is surprised by changing conditions (e.g., high demand). This suggests the need for careful evaluation of resource adequacy under circumstances of constrained flexible and/or dispatchable generation. Aggregated indicators for social and ecological sustainability metrics are used to show at a high level the impact of various planning strategies on these outcomes. These two metrics are comprised of 8 impact criteria which are themselves comprised of scoring across 32 impact indicators (see Table in section ). To better understand the contribution of the 8 impact criteria to the aggregated socio-environmental impact scores, Figure shows the contribution of each criterion from each of the 100 Monte Carlo iterations, per planning strategy. Note that scores are normalized to the highest score across all strategies in each criterion (i.e., a value of 1 in Figure represents the highest observed score across all model runs). Contribution analysis for socio-environmental impacts for all 100 Monte Carlo iterations, per criterion for each strategy. All Monte Carlo iterations are stacked; color saturation/density, therefore, indicates increased alignment among simulated futures. Values are normalized for each criterion among all strategies such that 0 = no impact and 1 = highest negative impact from across all strategies. While the two ecological indicators are highly correlated, the social indicators differ substantially across scenarios. The strongest impacts to air quality and social wellbeing are observed in the No_Soc_Impacts scenario (resulting in the backfire effect observed due to the need for emergency diesel generation). In contrast, high impacts to First Nations land, heritage sites and aesthetics, tourism, and recreation, as well as high ecological impacts, are observed in the Emissions_Cap and No_Fossil scenarios due to higher reliance on hydropower facilities. Discussion This study introduces an approach that contributes to GEP. First, the approach permits holistic evaluation of the implications of policy instruments on sustainability outcomes. By explicitly defining socio-environmental objectives as a basis for forming high-level planning strategies, we introduce a structured approach for handling multicriteria decision framing. Implementing these strategies as constraints in a cost-optimizing capacity expansion model forms a tool that allows decision-makers to simultaneously assess the impacts of addressing multiple sustainability objectives. Further, as the approach produces insights about general strategies rather than specific investment schemes, it is capable of producing insights that persist as future uncertainties resolve and the specific investment path deviates from the original projection. Selecting a strategy rather than a specific investment path allows electricity system planners to adapt investment plans to new information, while still working toward a long-term coherent vision constrained by explicit objectives. In addition, given system outcomes are dependent on underlying uncertainty probabilities which may be linked to overarching policies (e.g., a government’s willingness to supply permits for new hydroelectric developments), our approach can inform policymakers how a particular policy environment impacts power system planning decisions and associated outcomes (i.e., policies which overconstrain the power system were demonstrated through our case study to increase the reliance on diesel and raise costs, strengthening the case for governments to ease restrictions on new development). Second, the dynamic myopic foresight modeling approach exposes tail effects induced by imperfect forecasts which may otherwise go unnoticed when modeling with perfect foresight or with methods that do not consider uncertainty. One such example is the 65 gCO2e/kWh emissions cap strategy, which appears as a viable and robust strategy when evaluated with perfect foresight but is exposed as high-risk when modeled myopically (see SI sections 3.2 and 3.3 for this analysis). In this case, it is plausible that the myopic decision-maker, unable to detect rapidly changing conditions, underinvests in nonemitting technologies resulting in supply shortages as energy output from existing fossil generation assets is limited by the cap. This modeling approach thus captures these strategy pitfalls which are otherwise undetected. Finally, we demonstrate the utility of modeling a wide range of electricity system uncertainties, many of which are deeply uncertain. Exploring the uncertainty space with a Monte Carlo simulation in which all futures are treated as equally likely allows for identifying robust strategies which retain their efficacy regardless of future conditions. This also highlights vulnerable strategies that have unintended consequences or produce dramatically different outcomes when faced with uncertainty, such as avoiding all new generation deemed to negatively impact communities. Moreover, this Monte Carlo approach to modeling uncertainty spurs additional assessments which can inform system vulnerabilities and robust generation technologies. Such analyses draw insights relevant to the Yukon context, illuminating the viability and competitiveness of various technologies in this remote northern setting. We note several limitations in our case study and modeling efforts, notably the use of representative days, omission of transmission constraints, linearization of capacities, use of a proxy demand profile, and a number of subjective input numerical inputs (e.g., discount rate, foresight length, binary event probabilities) for which additional sensitivity analysis would be beneficial had computational resources allowed. While a model with such simplifications may be unsuitable for use as a predictive or prescriptive tool, pairwise comparison of output distributions insulates the insights from modeling limitations to some extent, instead relying on the model to diagnose the cause and effects of system behavior rather than attempt to predict it. Similarly, while simplified metrics for socio-environmental sustainability and resource adequacy are limited in capturing underlying outcomes of interest, these high-level metrics are nonetheless useful to decision-makers as a useful starting point for integrating different objectives into decision-making. Finally, despite limitations in parametrizing uncertain binary events, we nonetheless explore the uncertainty space as intended. Overall, the approach is flexible and presents an opportunity for power system planners to (a) define and contemplate electricity system externalities and uncertainties, (b) frame present-day decisions or strategies as a structured set of alternatives, and (c) via Monte Carlo capacity expansion modeling, quantify decision alternatives to identify those which present greater or lesser long-term risks to favorable outcomes. Our Yukon case study produced results, namely the enhanced tail risks of overly constrained planning strategies, which stemmed largely from unique preconditions of the Yukon grid such as the highly limited development opportunities and highly unfavorable back-up alternatives (emergency rental diesel). Our approach also relied on existing socio-environmental scoring across a set of predefined options, which may not exist (but could be developed) elsewhere. Nevertheless, some insights are likely to be generalizable such as how (a) general planning strategies/constraints can achieve positive results for socio-environmental without requiring those goals to be embedded in the planning optimization, and (b) that overly constraining such decisions creates large risks (and potential backfire) under uncertain futures. The overall modeling approach we propose (planning strategies, coupled with Monte Carlo simulation and dynamic myopic foresight to evaluate diverse objectives) is also adaptable to other jurisdictions and case studies – though potentially with dauting data and computational requirements for larger or less isolated grids. Refinement of the approach is possible with improvements in modeling capabilities. For example, the model is limited in capturing subhourly effects and transmission constraints, stemming from inherent limitations of the ESOM. Future work could soft-link a node-based power system model such as PyPSA or apply established resource adequacy calculations to fully assess implications of strategies on reliability. We also see potential in leveraging the differences in decisions made by a model with perfect information and its myopic counterpart to identify “signposts” where decisions consistently diverge. This could open the possibility of developing a proactive planning approach whereby a myopic planner can pre-emptively identify the need for new capacity investments. We anticipate validation of such a proactive plan would require a mixed integer capacity expansion model rather than a linearized model as to more clearly observe differing decisions. Finally, valuable extensions of this method could establish how decisions can be adapted over time as uncertainties are realized; such an approach would require a deeper analysis of decision timelines and of which decisions could realistically be altered or otherwise conditioned on these future uncertainties. These decision adaptations could then be layered into the Monte Carlo modeling backbone. We introduce an approach to evaluate the implications of objective-driven decision-making on electricity generation expansion planning. Motivated by the prospect of improving the sustainability outlook of the electricity system in the Yukon, Canada, we demonstrate our approach by simulating strategies to maximize electricity affordability, reduce GHG emissions, improve social sustainability, and improve ecological sustainability while adding new capacity to the Yukon electricity grid. We demonstrate a series of strategies which are effective at achieving their target objectives but each induces trade-offs with cost. Among sustainability outcomes, improving social and ecological sustainability can be achieved simultaneously, but trade-offs exist between GHG emissions and these other sustainability objectives. Strategies which take a flexible approach, such as avoiding only new diesel generation rather than all new fossil-fueled generation, can be effective at meeting their stated objectives while balancing trade-offs in cost. Conversely, more restrictive and inflexible strategies such as a 65 gCO2e/kWh emissions cap or avoiding all generators which may negatively impact social sustainability have a dramatically higher risk of elevated costs or electricity supply shortages. This study makes progress toward addressing an important gap in GEP literature by demonstrating how simulations of capacity expansion decision-making can identify effective planning strategies which perform well in advancing socio-environmental and affordability objectives amidst uncertainty. The approach has implications in empowering electricity system planners to better address objectives garnered through public engagement, providing both specific insights for the Yukon electricity system and a generalized approach to supplement GEP methods. Supplementary Material