The mid-transition in the electricity sector: impacts of growing wind and solar electricity on generation costs and natural gas generation in Alberta

In response to climate change, electricity grids are decreasing their carbon intensity with the addition of wind and solar variable renewable energy generation (VREN). This leads to a mid-transition period, where renewable energy is unable to satisfy electricity demand without contributions from other fossil sources such as natural gas, but also generates sufficiently to constrain conventional generation—changing their operating and market conditions. We use a simplified copper plate model, which scales up and down historical wind and solar generation, to examine how and when the patterns and generation costs for fossil fuel power could change by the increasing capacities of VREN on the relatively isolated Alberta electricity grid. We find that beginning at 20% VREN an increasingly diverse range and reduced hours of dispatched capacity is necessitated from the existing generation. However, even as capacity factors for fossil fuel generation decrease their costs remain reasonable and we found this to be a low-cost pathway for achieving moderate to deep emission reduction goals. A full 86% of demand could be met with VREN before generation costs exceeded 100$/MWh, allowing for an emissions reduction of 28.4–9 million tonnes yr−1 of CO2eq, on a lifecycle basis. In order to integrate the renewable generation, new and existing fossil fuel units will require market rules that incentivise flexibility and ensure they remain in place throughout the transitionary period as they are crucial to balance variable renewable generation.


Introduction
The Intergovernmental Panel on Climate Change stresses that to maintain warming to a 1.5 • C pathway, carbon emissions must reach net-zero by 2050 (Rogelj et al 2018).In 2015, world leaders signed the Paris agreement, a legally binding treaty, and agreed on firm carbon emission cuts (UNFCC 2016).As most global emissions result from the generation and consumption of energy, a key pathway for decarbonization is the transition to net-zero electricity generation and a deep electrification of our energy systems (Davis et al 2018).As a signatory to the Paris agreement, Canada committed to reducing its greenhouse gas emissions by 2030, to 30% below 2005 levels, and in 2021, Canada further strengthened their commitments increasing the target to a 40%-45% reduction (Leach et al 2021).
Many studies have explored the operation and economics of highly decarbonized electricity systems (Haas et al 2017, Jenkins et al 2018, Koltsakis and Dagoumas 2018, Cuisinier et al 2021).These studies have spanned Europe (Keatley and Hewitt 2008, Amorim et al 2014, Knorr et al 2014, Heuberger et al 2017, Pleβmann and Blechinger 2017, Schlachtberger et al 2017), Australia (Elliston et al 2014, Riesz et al 2015, Lenzen et al 2016), South America (Odeh andWatts 2019, Costa et al 2022), and North America (Mai et Radpour et al 2021, McPherson et al 2023).They vary from full net-zero electricity emission modeling e.g.(Ruhnau and Qvist 2022) to pathway modeling which includes intermediary carbon constraints e.g.(Duan et al 2022), and from simpler copper plate models e.g.(Tong et al 2020) to advanced capacity expansion and unit commitment models e.g.(De Sisternes et al 2016).These studies highlight the difficulties of reducing emissions for the final 10%-20% of electricity demand as this is where costs rise quickly to meet increasingly stringent emission targets.In their paper, Mai et al (2022) identify and discuss the advantages and disadvantages of six different pathways to achieve the final 10%: increasing variable renewable energy (VREN), seasonal storage, dispatchable low carbon, demand response, direct air capture, or fossil fuel generation with carbon capture.
Many have modeled electricity grids that include high rates of VREN generation (e.g.wind and solar), balanced with energy storage, decarbonized dispatchable power, and/or across large geographical areas (Zerrahn et al 2018, Tong et al 2020, Kan et al 2021, Clerjon and Perdue 2022, Ruhnau and Qvist 2022).Energy storage allows excess VREN electricity to be captured through chemical (batteries) or physical changes (pumped hydro, compressed air) to use later, which can fill gaps in VREN generation, while large geographic areas can also play a key role in smoothing the VREN generation variability.In their study Duan et al (2022) considered a range of VREN penetration levels between 0% and 100%.They identified the composition of the lowest cost electricity systems across 1%-100% decarbonized electricity grids, considering many firm electricity generation technologies in combination with variable generation and for a variety of wide (national level) geographical areas and demonstrate how idealized grid compositions change with current technology costs as they reduce the carbon intensities.Zerrahn et al (2018), evaluate the economic trade-offs between VREN curtailment and energy storage for several VREN penetration levels-40%, 50%, 60%, 70%, and 80%.They show that the optimal grid composition will be a balance of VREN curtailment and energy storage-avoiding curtailment can increase costs beyond the cost of oversizing and curtailing VREN.Further, Tong et al (2020) assess the impact of energy storage costs in VREN electricity systems.They show that the availability of high or low-cost energy storage will change the role that storage plays with respect to optimal storage + VREN electrical grids.Energy storage systems operate with shorter, frequent charge/discharge cycles at higher costs, filling short-term VREN gaps, and exhibit longer seasonal cycles filling long-term outages in VREN when they have lower costs.Cost reductions in energy storage can lower VREN curtailment.
Consideration has also been given to the impact of increasing levels of VREN on existing generation assets.There are several concerns that have been identified with the increasing amounts of VREN in an electricity grid.There are additional technical requirements to maintain reliability, such as fast frequency control and load following (Kroposki 2017).Next, as baseload units are required to be more flexible, they are subject to higher numbers of cold, warm, and hot starts-increasing maintenance and decreasing reliability, which shorten their lifespan (Keatley and Hewitt 2008).Then there are market concerns as VREN impacts the merit order curve, such as price cannibalization from low cost/negative cost hours (Lynch et al 2021) which may require market redesign (Hogan 2022), and finally, preferential renewable curtailment before demand is fully met with VREN, as existing assets lack flexibility (Denholm et al 2018).However, in electrical grids highly dependent on thermal natural gas power plants, their generation requirements are expected to change as VREN penetration increases.When this begins to occur, how it will impact the existing fossil fuel baseload, cogeneration, peaking requirements, and overall economics of generation throughout increasing levels of VREN generation, is poorly understood.
The point at which the new VREN resources place a constraint on the existing fossil resources but are insufficient to operate on their own has been defined as the mid-transition period (Grubert and Hastings-Simon 2022).Characterising this transition period is critical for electricity system planning and successful management of the transition.When a system reaches this period will depend on the specific properties of the grid such as the interconnections and geographic diversity available in VREN supply, which smooth gaps in VREN generation (Roques et al 2010, Odeh andWatts 2019), and the flexibility of the existing fossil fuel generation (Denholm et al 2018).
The Canadian province of Alberta represents a relatively isolated electricity grid with limited interconnection with neighboring grids in British Columbia, Montana, and Saskatchewan.A region with high shares of fossil fuel generation including industrial co-generation, the Alberta Government pledged to phaseout coal generation by 2030 (Collins 2016), and the Canadian Government followed with a national pledge in 2017 (GOC 2017).Further, the Alberta government has legislated a 30% by 2030 renewable electricity target, and the Canadian Government has set a net-zero electricity goal by 2035 as part of a 2050 net zero commitment (GOC 2022, GOA n.d.).Based on the Alberta Electric System Operator (AESO) Annual Market Statistics Reports (2016-2023) 2021, coal and natural gas generation made up 64% and 26%, respectively, of total generation in 2015 and this reversed to 17.2% and 64.4% in 2022-further, half of the coal capacity (2273 MW) was replaced through a repowering to natural gas generation capacity and the other half retired (2278 MW).The retired coal generation was substituted by a combination of new gas (406 MW installed) generation capacity, higher capacity factors in existing gas generators, and additional installed VREN capacity (3396 MW installed), all built between 2015 and 2022.The growth of VREN in Alberta's competitive electricity market has accelerated recently with 58% of the total growth in VREN capacity during this period commissioned in 2022, increasing the installed capacity from 3005 MW to 4752 MW (Authors' calculation based on AESO, 2023c).If all remaining approved VREN projects are constructed, this would bring the total VREN capacity to 9446 MW (AESO 2022b), by the end of 2024 assuming a 2 year construction timeline.Delays would lengthen this timeline and could arise as a result of issues such as supply chain availability, financing, or other factors.
Currently, the Alberta electrical grid is supplied mostly through natural gas generation, and it is often reported that decarbonization using variable energy sources will require long term energy storage (Hunter et al 2021, Ruhnau andOvist 2022).In this paper, we explore how wind and solar will impact the electricity lifecycle emissions, the levelized cost of electricity (LCOE), and the natural gas fleet operating and capacity requirements when only short-term battery energy storage systems (BESS) are available.We do this using a simplified model by scaling up or down the existing wind and solar fleet but allowing the proportion of wind to solar to vary.Then, using the historical generation and demand data we use a lowest LCOE optimization to determine how much VREN capacity and what fractions of wind and solar would be required to meet a multi-year demand profile from 1%-99% VREN penetration.We then evaluated how BESS could be used to reduce the VREN capacity using 2 h, 12 h, and a theoretical 72 h duration.Our goal is to use this simple model to evaluate how much VREN can be incorporated into the Alberta electricity grid using short-term energy storage and to look at how these increasing capacities of VREN impact the existing fossil fuel fleet in the near term to identify and characterize the mid-transition point.In doing so we explore the emissions reductions that Alberta can affordably accommodate by incorporating increasing shares of VREN, before some combination of long-term energy storage, transmission, demand side management, and dispatchable zero emission generation is necessary to further decarbonize the electricity grid.
The paper proceeds as follows, first we discuss the detailed data, optimization, and energy storage model used for the analysis.Following that, we present the VREN penetration capacity results for wind and solar with and without energy storage, the natural gas patterns of generation required to balance increasing VREN, the resulting lifecycle emissions, and the LCOE cost forecasts.We then identify the mid-transition period and highlight natural gas dispatch requirements for growing levels of VREN.Finally, we discuss the limitations of using a copper plate model with scaled historical data followed by the implications of our key findings.

Methodology
AESO maintains hourly data for electricity demand and generation (by asset type) (AESO 2015(AESO -2021)).The time series analyses in this study used 7 years of historical electrical demand and wind utilization factors, from 2015-2021.However, as solar generation was relatively negligible prior to 2021 (<100 MW of capacity), only one year of solar utilization factors were used-2021.A time series model was set up so that VREN generation would be calculated for both wind and solar in each hour using their respective installed capacity (which was considered equivalent to available capacity), and the historical hourly utilization factor.AESO defines utilization factors as 'the ratio of net-to-grid generation to net-to-grid available capacity' (AESO 2023b, p 15).
The model then incorporated a myopic energy storage algorithm.When VREN generation exceeded demand, excess electricity would be stored first-up to the energy storage power and energy capacity constraints-and curtailed second.When VREN generation for any hour falls below the required demand, the shortfall would first be made up with stored electricity, again constrained by the model's energy storage and power constraints.If there was still a gap between VREN supply (wind, solar, and storage) and demand, it was filled using natural gas electricity generation.The roundtrip storage efficiency was assumed to be 83% (Kucevic et al 2020).
In order to evaluate the effects of increasing capacities of wind and solar energy on existing assets, we used an optimization to predict the optimal wind and solar capacities, the split between the two technologies, and when and how much energy storage that would be required to meet higher and higher percentages of the total demand.The optimization decision variables were the total VREN installed capacity, the fraction of that installed capacity which is wind, and for the battery energy storage cases, the power capacity of the BESS storage in MW.The energy capacity was set by the battery duration for each of the examined scenarios (2 h, 12 h, and 72 h).Only these short-term storage durations were considered.However, as the hourly demand gap could often be less than the maximum battery power capacity for the scenario, energy storage was allowed to supply durations longer than the set scenario up until the energy capacity was exhausted.Finally, in order to obtain the lowest cost VREN generation required to meet each constraint (% of demand satisfied using VREN generation), the objective function for the optimization was the VREN LCOE including the cost of storage for scenarios 2, 3, and 4. We do not include natural gas costs or the possibility of natural gas capacity expansion as we are exploring historical demand data for which the current natural gas fleet was sufficient and evaluating the impacts to this fleet had there been more VREN generation.Optimization details and cost assumptions can be found in appendix A and a summary of the optimization scenarios considered is shown in table 1 below.
We then used the model to assess how much of Alberta's 2015-2021 demand would be met with the current installed capacities, the projects which have obtained AUC approval, and all announced solar, wind and energy storage project capacities.The current installed capacity was obtained from the Current Supply Demand Report (AESO 2023a), while the projects with AUC approval, and all announced capacity is from the AESO Long Term Adequacy Report (August 2022, p 18).These values are summarized in table 2 below.
Next, for each of the scenarios and percentage of VREN penetration in table 1, the time series requirements for the natural gas generation were analyzed.For each VREN level, we determined the maximum capacity, minimum capacity, and the overall fleet capacity factors required from the NG fleet.Then, we calculated the hourly energy generation distributional histograms required to meet the remaining (not satisfied by VREN) demand, including seasonal and diurnal comparisons, to assess trends and characterize the impact of increasing levels of VREN on the natural gas fleet over the mid-transition.Using these calculated generation profiles, both a lifecycle emission and a LCOE analysis were completed to evaluate the environmental and cost impacts.Costs were determined, including the 2025 carbon tax and credits available in Alberta (TIER 2019, GOC 2021), for the NG generation, VREN generation, and the combined generation, or the total overall cost of electricity required meet demand.Under the TIER (2019) regulation, fossil fuel power generation will pay a carbon tax where their emissions intensity is above benchmark and only on the emitted difference between their actual emissions and the benchmark level.Whereas emission free generation will be credited with the carbon tax at the benchmark emission intensity for their generation.For this analysis we used the 2025 carbon tax and benchmark values of 95$/tCO2ea and 0.3478 tCO2eq/MWh respectively (TIER 2019, GOC 2022).This is a credit of 33.04 CA$/MWh for VREN generation, a credit of 1.42 CA$/MWh for combined cycle generation as its efficiency is above benchmark and a carbon tax of 12.01 CA$/MWh for simple cycle generation.
As power plants are increasingly expected to operate more flexibly in response to increasing levels of VREN generation, they will be subjected to increasing numbers of hot, warm, and cold starts which will decrease efficiency on existing combined cycle plants and favor simple cycle generation.In order to address the issue that combine cycle power plants will face minimum generation levels and must-run constraints we used the simplifying assumption that combined cycle plants would be switched out to more flexible simple cycle plants when they are required less than half the year.Therefore, for the greenhouse gas and cost analysis, the natural gas fleet was assumed to operate as a simple cycle steam generator or peaker plant (with the associated lower efficiency of 0.4) for all natural gas capacity that is required to operate less than 50% of the time, and as a combined cycle plant with an efficiency of 0.57 if the natural gas capacity is required equal to or more than 50% of the time (Keatley andHewitt 2008, CER 2016).We performed a sensitivity analysis to test the impact of this 50% cutoff and compare our results to a capacity cutoff of 75%.We found that the emission differences between these assumptions are minimal, diverging only slightly towards 75% VREN, in addition the impact on LCOE is small with a maximum LCOE cost difference between the two cases of 4.32 CA$/MWh.The impact of this assumption on costs and emissions can be found in their respective appendices B and C. Additional cost assumptions for the LCOE calculations can be found in appendix B, while the lifecycle emissions assumptions for the greenhouse gas analysis can be found in appendix C.
This cost optimization model is highly simplified compared to a system dispatch model, but we demonstrate here that it can be useful for an analysis and characterization of the mid-transition impacts.The model design is guided by both the central question in this analysis-when and how is the existing natural gas fleet impacted by development of VREN such that it can be said to be impacted by the mid-transition dynamics in the near to mid-term, along with the observed market dynamics in Alberta where new market entrants are responsible for a significant amount of VREN build out.The relative simplicity in the modeling approach allows for more model runs, for example exploring price sensitivities, while still generating the key results needed for the practical characterization of the mid-transition targeted in the analysis.Note: a The percent of demand satisfied using VREN-either directly or by applying stored VREN.b Below 70% it was found to be more economic to add additional VREN capacity than to add energy storage, therefore in all scenarios 0%-69% follow the Scenario 1 curve.

Historical data
Alberta produces wind or solar electricity in 99% of all hours in the dataset, and for more than 80% of the year the combined wind and solar utilization factors were above 10% (50/50 split).The remaining 20% of hours where VREN generation is under 10% would require exponentially increasing capacities of VREN to meet demand in the exact hours its required.Figure 1 shows the distribution of VREN utilization factors in the dataset, for wind (6 years), solar (1 year), and a theoretical 50/50 split between the two capacities.Note that combining the factors with a 50/50 weighting decreases the interannual variability from wind and leads to a decline in the number of both high utilization factor hours and low utilization factor hours.

Wind, solar, and storage capacities
Figure 2 shows the baseline VREN capacity results with no energy storage from scenario 1 optimizations listed in table 1, as well as a comparison of scenario 1 to scenario 3 (12 h of energy storage).The current, approved and announced capacities are included on figure 2 for comparison; they correspond to an annual VREN penetration level of 12%, 29%, and almost 60% of demand, respectively.Finally, figure 2 also shows the fraction of VREN production that is curtailed under the scenarios.Initially, wind only compositions achieve the VREN renewable constraint with lower LCOEs.This is due to their higher overall capacity factors and similar Capex costs assumptions (see appendix B).However, at 51% VREN it becomes more economical to add solar rather than continuing to scale up wind during low wind daylight hours and curtail at other times.The difference between the amount of solar in the current installed capacity and the scenario optimization results is ascribed to the competitive market in Alberta where decisions about private investment in generation assets may differ from a cost optimized model due to market rules and competitive operation of existing assets, and we discuss this in more detail in the Discussion below.A sensitivity analysis in appendix B, examines how these results change if solar prices have decreased faster than predicted, as solar forecasts have historically underestimated this pace (Way et al 2022).The results show that using the Way et al (2022) 2025 prediction for solar costs (330USD/kW would indeed invert the findings such that solar would be the primary choice early on while wind capacity emerges at 27% VREN. While it is possible to scale up wind and solar installations in the model to meet demand in 99% of all hours, the required capacities are very large to meet the final 15%-20% (above 80%-85% VREN).From that point, VREN curtailment begins to grow rapidly.Curtailed VREN begins when 32% of electricity demand is met by wind energy, then at 51%, the curtailment growth changes as solar capacity is introduced to the VREN generation mix.VREN curtailment reaches 20% when 70% of demand is met by wind and solar, however, with no energy storage, it grows from 20% to over 99% to meet the more stringent 99% VREN constraint.
At current storage costs, it remains more economical to add wind and solar capacity until approximately 78% of demand is met with VREN, after which, short term BESS begins to be more economical.This includes the cost of curtailing up to 35% of the VREN for the 78% case.In the 12 h storage scenario, minimal amounts of energy storage immediately begin to reduce curtailment for the 78%-90% VREN levels, at which point curtailment again begins to grow quickly.However, at the 99% VREN level, energy storage reduces curtailment from 99.7% to 72.2%, and drastically reduces the required installed wind capacity from 651.5 GW to 54.8 GW.This is accomplished with a maximum 12 h storage power capacity of 18.2 GW for the 99% case, which results in an energy capacity of 218.8 GWh.Note that this power capacity exceeds the maximum hourly demand required in the dataset, therefore at maximum demand this storage results in more than 12 h of energy storage.Further, it provides faster charging capabilities when needed.Additionally, both the 2 h and 72 h storage scenarios resulted in a similar reduction in installed capacities.
We compared of all three BESS scenarios along with a sensitivity analysis of the energy storage costs which show how storage costs ±50% impact the results.The shorter energy storage durations were economic at lower %VREN penetrations, 78% and 79%, for the 2 h and 12 h scenarios respective, while 72 h scenarios first required energy storage at 90% VREN.The least cost scenario was the 12 h BESS which best balanced power and energy capacity requirements resulting in the lowest overall LCOEs.A 50% reduction in battery capital costs for the 2 h scenario resulted in energy storage becoming economic earlier (71% VREN) and lowered the final wind and solar capacity requirements by 18 GW replacing it with an additional 30.9GW of power capacity at 99% VREN.The resultant 99% LCOE was also 52$/MWh less than with lower cost storage (304 CA$/MWh vs 252 CA$/MWh).Further details can be found in appendix D.

Natural gas impacts
One feature of the mid-transition is the constraints the new system places on existing energy systems.Here we evaluate the impact of increasing VREN% on the operation of the natural gas fleet.Figure 3 shows the minimum and maximum dispatched natural gas for each scenario and at each %VREN.The maximum required natural gas dispatch changes very little with respect to increasing amounts of wind and solar capacity.Initially (0% VREN levels), the maximum required dispatch is 11 729 MW.This is immediately reduced to 11 689 MW at 1% VREN level with the addition of just 269 MW of wind capacity (figure 1).Additional small reductions in the maximum dispatched natural gas continue until 33% VREN levels when a maximum of 11 442 MW is reached, and that maximum remains static all the way to 99% VREN levels.The energy storage scenarios were unable to further lower the maximum until VREN levels of 90%, 97%, and 98% for 72 h, 2 h, and 12 h respectively.Even then, they had only a minimal impact reducing the maximum natural gas hourly generation required from 11 442 MW to 11 312-11 316 MW depending on the storage scenario.This result is mainly due to a period in 2019 where wind and solar production is very limited.This deficit period is shown for the 12 h storage, 95% VREN case in figure 4. Conversely, the lower bound of the NG dispatch requirement is impacted early on by increasing levels of installed capacity of wind, falling from an initial low of 6595 MW to 0 MW by the time 32% of total demand is met by VREN generation, which has an installed wind capacity of 8600 MW (figure 2).
Figure 5 shows how the hourly distribution of natural gas dispatch changes to meet the residual demand at increasing capacities of VREN with and without energy storage.At 0% VREN, the natural gas distribution directly reflects the demand distribution for the dataset, it is a tight, normal distribution.The distribution of natural gas dispatch is immediately impacted even at low levels of VREN penetration, mainly by widening the range of dispatched generation required by natural gas.However, much of the change in shape in the distribution of natural gas generation occurs between 20% and 50% VREN as the shape trends from a normal bell towards a uniform distribution Hours where VREN generation can meet 100% of demand begins to occur at 31% VREN penetration, but quickly climbs to 10% of all hours by 48%.By 50%, the flat uniform distribution is taking shape and from there as VREN capacity increases it flattens the Natural Gas Generation evenly, reducing the number of hours required at each generation level.Figure 5(b) shows the impact of adding different amounts of energy storage on these distributions for 80%-99% VREN penetration.The shorter energy storage scenario results in higher charge/discharge capacity, but lower overall energy storage (MWh), see appendix D for a comparison of the power and energy capacity results.The higher power capacity from the 2 h energy storage effectively lowers the bottom of the distribution, therefore reducing the number of low natural gas generation (<5000 MW) hours in comparison to the 12 and 72 h storage scenarios.The 12 h storage lowers the middle range (5000-8000 MW) more effectively, while the 72 h storage scenario lowers the higher range (>7000 MW) more effectively at the 95% VREN penetration, but it has much less impact on the lower generation range.As already shown in figure 3, none of the storage scenarios can eliminate the few peak natural gas hours (>11 000 MW).The percentage of hours requiring peak natural gas generation is shown in table 3. It shows that 72 h of energy storage is significantly better at reducing peak hours at 90% and 95% VREN.Seasonal and diurnal trends for natural gas dispatch, with and without energy storage, are explored in appendix F. Natural gas dispatch is required in all seasons and the distribution under 9000 MW of dispatch is similar season to season, however peak capacity (>10 000 MW) is required mainly in the winter.Diurnal trends first show a duck curve at 50% VREN which is when solar capacity first starts in the model.Seasonal energy storage histograms demonstrate the trade-offs in power and energy capacity.For example, 2 h and 12 h scenarios at 99% VREN require no NG dispatch in September, but the 72 h scenario requires approximately 15 h of low capacity (<2500 MW) dispatch.

Emissions
As VREN meets increasingly higher levels of demand the lifecycle greenhouse gas emissions of the electricity generated are impacted by several factors.First, fewer megawatt-hours of thermal generation reduces direct emissions as less natural gas is burned.However, as the fleet requirements trend towards flexibility and  simple cycle generation from a more stable constant combined cycle generation, the emission intensity of these megawatt-hours rises due to the corresponding drop in efficiency.Finally, the over installation and curtailment of VREN required to meet these targets raises the lifecycle emission intensity of VREN generated megawatt-hours.Figure 6 shows how the total annual emissions and emissions intensity changes with respect to %VREN.Initially, annual emissions drop as %VREN increases, this begins to slow around 50%VREN when curtailment and the higher lifecycle emissions of solar generation begin to have an impact.The rapid decrease beginning at 75%VREN occurs as at this point the natural gas fleet consists solely of simple cycle generation-no thermal capacity at all is required in at least half of the hours.Beginning at 91%VREN for the no storage scenario and 96%VREN for the 12 h storage scenario the lifecycle emissions attributed to the overbuild required to meet the %VREN targets begins to eliminate any gains from reducing thermal generation; the overbuild required in lieu of other technologies at these high %VREN targets is extensive.

Generation costs
One concern with respect to decreasing CF for thermal plants is the impact on the cost of electricity.Figure 7 shows the effect that decreasing generation has on the LCOE of natural gas electricity with and without capital cost recovery.For comparison, the VREN LCOE (no storage) is also shown.VREN LCOE is less expensive, mainly due to the carbon credit they receive.The natural gas LCOE increases steeply as its generation is required for fewer and fewer MWh, while the fleet capacity can not be reduced.Much of this increase is the capital cost recovery component that is necessary to recoup costs in the fewer and fewer hours for which natural gas generation is needed.As the energy storage has no ability to reduce the natural gas fleet capacity, the no storage and storage scenarios all share the same natural gas LCOE costs as VREN penetration increases.The LCOE values in figure 7 are compared to historical prices and the LCOE including capital recovery and carbon tax for natural gas generation reaches 102$/MWh (the 2021 average pool price in Alberta) when VREN generation accounts for 75% of demand, and 162$/MWh (the 2022 average pool price in Alberta) by 88%. Figure B1 in appendix B compares the component LCOE's with and without carbon taxes and credits.The carbon tax has negligible impact on natural gas LCOE's until 50% VREN as they are only charged on emissions that exceed the benchmark rate-this limits its impact until the flexibility of simple cycle generation is required.Conversely, the carbon credits significantly decrease the LCOE's of the VREN generation in Alberta, without carbon credits, new natural gas (full capital recovery LCOE) is lower than VREN LCOE until 40% VREN.The effective LCOE for all the electricity required to meet demand-a combination of both VREN and natural gas and including the 12 h storage scenario.Above 80%, even short-term battery energy storage makes a significant impact at reducing the total LCOE required to meet the demand throughout the 7 year dataset, with the 99% VREN case LCOE including capital recovery of 266$/MWh for the 12 h storage scenario.This is a drastic reduction in comparison to the baseline scenario with no energy storage, which has a LCOE of 1466$/MWh for the 99% VREN case, saving over 80% of the no storage costs.Most of this cost savings is due to the reduction in the required installed capacities and resultant curtailment of wind and solar generation.When compared to historical pool prices VREN can reach 86% of demand before exceeding the 2021 average power pool price of 102$/MWh and 92% with just 12 h of BESS. Figure B2 compares the effective LCOE with and without the carbon tax/credits, the effect of carbon taxes/credits reduces the effective LCOE starting out low and growing as VREN penetration increases.Starting at 1% VREN the difference is 1.17 CA$/MWh, then it reaches 10 CA$/MWh by 45% VREN and at 99% VREN it is 24$CA/MWh which is 3%, 19%, and 12.5% lower than the effective LCOE without carbon tax/credits.

Discussion
Figure 2 illustrates that wind is currently the lower cost option for renewable energy generation in Alberta for low levels of VREN.As VREN penetration rises the different hours of availability for solar generation vs wind leads to the addition of solar vs over installing wind and curtailing as the cost optimal solution.This is mainly because the average utilization of factors for wind is 35% in the dataset, while solar was only 20%, and the capital costs per kW installed are very similar for the two (appendix B).Currently solar energy projects are underway and expanding quickly in Alberta, even though they do not appear to have economic utility in our model until wind power can meet 50% of all demand.As the hours that solar energy is available correspond to higher power pool pricing-unlike in our utility model-there is a competitive market incentive to generate more power during these hours and increase project returns.For example, wind captured an average annual pool price at a 30% discount to the average annual power pool price in 2022, whereas solar captured a 26% premium (AESO 2023b).While the inter-annual variation in solar production is expected to be small, 2021 could also prove to be a poor solar year.A comparison of 2021 solar data to 2022 solar data is available in appendix E. The year 2022 did have a higher mean utilization factor at 22%, with much of the difference in generation occurring in July and August.
Next, we showed that it is initially significantly less expensive to over install wind and solar capacity, and curtail excess, than it is to add battery energy storage and that even short-term energy storage can drastically lower required VREN capacities at deep levels of penetration.Curtailment is necessary for lower cost grid compositions, and this is supported in other research (Zerrahn et al 2018).However, this model does not account for the economic utility that energy storage can provide a private solar or wind power plant.For example, energy storage allows renewable energy to be shifted throughout the day and to capture a higher hourly price, increasing returns and in 2022 energy storage did capture 170% premium to the average power pool price (AESO 2023b).A wind operator could capture a higher rate simply by shifting power generation from hours when there is a lot of wind generation to hours where there is little.Currently, curtailed VREN in Alberta does not receive compensation, which as VREN penetration increases and curtailment is more common, will increase the motivation for VREN + Storage installations.Furthermore, energy storage also allows for revenue stacking through dispatchable power and participation in the ancillary service markets.This can help explain why 5600 MW of energy storage have been announced and in fact 90 MW already exist in the Alberta grid (table 2).
The distribution natural gas dispatch (figure 5) shows between 20% and 50% VREN the shape changes from normal bell towards a uniform distribution and we identify this as the start of the mid-transition period.Figure 8 zooms in on this transition.Almost immediately the shape changes from bell trending to bimodal and is almost uniform by 48% VREN.Once VREN generation meets 20% of demand, more and more hours will be required in the lower range of hourly capacity generation from the existing natural gas power plants.This will increase their cold, warm, and hot starts, and begin to lower the CF of the more expensive to run plants-placing constraints on the existing fossil fuel generation and beginning the mid-transition.
The LCOE results in figure 7, highlight that a large amount of VREN capacity (meeting up to 85% of demand) can be installed on the Alberta grid before generation costs exceed 100$/MWh.This includes both the costs of VREN curtailment and the cost of idle natural gas plants-their fixed operating expenses.Safaei and Keith (2015) also found that balancing VREN with natural gas is a cost-effective pathway to reduce greenhouse gas emissions.Furthermore, as seen in other studies, relatively short storage durations of 8-24 h improve costs and facilitate the final 10%-15% VREN penetration (Zerrahn et al 2018).Unfortunately, short term storage and VREN are insufficient to meaningfully reduce the required reliable, dispatchable capacity, or the maximum natural gas generation required in all scenarios.It should be noted that while we calculate LCOEs for very high levels of VREN it is unlikely that the system would be operated in this manner.A different energy storage use strategy and longer-term storage, in some combination with increased transmission interconnection, dispatchable non-emitting forms of generation, and possibly demand response would be required to eliminate those high natural gas generation hours, when wind and solar generation are negligible for long periods of time.Additionally, the storage required exceeds the period between which renewable generation meets demand, as there is insufficient time to refill storage capacities prior to another deficit period.This was also noted by Ruhnau and Ovist (2022), where they find that these deficit periods can be flanked by low VREN generation for weeks on either side.
These natural gas capacity requirements create an environment where capacity factors for many of the conventional assets will decline meaningfully as VREN penetrates the market.As early as 31% VREN penetration or only ∼2% more than the VREN capacity than is already approved by the AUC, we find hours where VREN meets demand directly.This has implications for industrial generators and cogeneration where process or heating requirements dictate operation.VREN is likely to be curtailed sooner to accommodate these system constraints, even though carbon generating assets are still running.Further, VREN may be preferentially curtailed as it has no cycling costs (startup costs), nor a minimum run level, resulting in increased flexibility and ease of curtailment.This aligns with what has already happened in Texas and California (California ISO 2017, Denholm et al 2018).However, curtailing VREN resources while fossil fuel generators are still running detracts from net zero goals, and California is actively looking at ways to decrease renewable curtailment (California ISO n.d.).
In 2022, Alberta had a mean cogeneration load that was consumed behind the fence (directly by the industrial generators) of 28% of all demand, an implied capacity of 2769 MW (AESO 2023b).Referring to figure 8, VREN energy will begin to run up against cogeneration in Alberta at around 20% VREN penetration and approximately 5375 MW of wind capacity are installed (from figure 2).When all projects with AUC approval are installed, this is expected to occur as much as 5% of the time.This is further evidence of fossil fuel generation constraints beginning at 20% VREN penetration in Alberta.Short-term storage will enable VREN generation to shift, delaying the issue, but at the cost of round-trip efficiency losses, which also detract from decarbonization targets.
While conventional generation will see lower capacity factors as VREN generation grows, the LCOE for this natural gas generation is not impacted greatly until very high penetration levels (>80%).However, existing combined cycle natural gas plants are not designed to operate as peaker plants, they have less flexibility, take time to warm up and operate efficiently, have a mandated down time once required to shutdown, and start-up frequency has direct impacts on both cost and reliability (Keatley andHewitt 2008, AESO 2020b).These constraints must be addressed as capacity factors fall.

Limitations
This simplified analysis uses a basic storage model, scaling historical wind and solar generation data and ignoring market and transmission constraints and influences, which creates some limits on the application of this study.First, the storage model does not allow storage to be stored until all of demand is met by VRE generation.Therefore, storage cannot be considered economic until there is enough VREN capacity installed that there is sufficient, reliable oversupply with which to fill the storage often enough to be more economic than overbuilding and curtailing.Storage value is further underestimated as it is not considered in the intermediate period to reduce variability for the fossil fuel generation, reducing the number of turbine starts in natural gas generators, for example.
Wind generation, during the historical period analyzed, is located mainly in the southern corridor of Alberta.This is where the best wind resources in the province are located.However, there are locations with lesser wind resources, but with different patterns that may be more suitable for filling the VREN gaps in the historical data.Recently, wind generation in the province has diversified to include the eastern part of the province, therefore, future wind generation data would not be expected to closely match historical data.Additionally, there was limited solar generation in Alberta in 2021.The year started with only 107 MW capacity, the majority of which was a single 100 MW power plant.Capacity then jumped to 240, 290, to 336 MW throughout the year as more solar power plants came online.However, the available capacity took time to ramp up after each of these plants was installed.Therefore, the early data in the year (January-April), is very geographically limited.In comparison to 2022, 2021 data did have a lower average utilization factor of 20% to 2022's 22%, with much of difference occurring in August, and somewhat in February.A comparison is available in appendix E. The summer of 2021 was plagued by widespread high altitude forest fire smoke which may explain the drop in July and August solar generation (Dormer 2021).
Increased VREN is expected to both decrease hourly electricity prices when VREN is generating, and increase overall price volatility, because of its effect on the merit order (Macedo et al 2020).This would create an economic incentive for renewable energy generators to build and co-locate energy storage with their facilities.Further, as VREN is not paid for curtailed energy, there would be additional incentives to add energy storage earlier in the decarbonization pathway than this model predicts.Finally, as wind and solar resources are in the southern part of the province, transmission and land constraints would occur before the existing generation fleet could be scaled up to meet the provincial electricity demand.

Conclusion
The recent cost reductions in wind and solar power capital costs have led to a renewable energy boom in Alberta.As VREN generation can meet more of Alberta's electricity demand, fossil fuel generation will need to decrease-this includes units that currently operate at high capacity factors.This analysis identifies that this is expected to begin at around 20% VREN penetration.Further, if all the AUC approved VREN projects are installed, then it is expected that for 5% of all hours, the required natural gas generation will fall to levels at or below current mean cogeneration (industrial power generators).Historically, due to the ease and flexibility of VREN curtailment other jurisdictions have seen higher rates of VREN curtailment, and thus higher grid carbon intensities, than otherwise possible.Therefore, the mid-transition period as defined by Grubert and Hastings-Simon (2022) will begin in Alberta at %VREN penetrations of 20% and is likely imminent and expected to begin within the next 2 years.Further, this analysis highlighted gaps in VREN production that are insufficiently filled with up to 3 days of energy storage.Fossil fuel generation or other firm capacity will need to balance VREN generation, even at deep levels of penetration.
The cost analysis highlighted important insights into rising VREN penetration in Alberta's electrical grid.First, LCOE costs remain competitive until very high rates, >85% of renewable energy penetration.Both the overbuild and curtailed VREN LCOE and the predicted fossil fuel LCOE required to balance VREN generation, remain low throughout this mid-transition period.This leads to the second insight, that overbuilding and curtailing is a low-cost path towards decarbonizing up to 80% Alberta's electricity demand but that the early mid-transition period will require market approaches that value this low-cost pathway.Next, energy storage remains a costly addition until much deeper levels of VREN penetration are achieved.At very high levels of VREN penetration (>90% VREN) short-term energy storage can reduce LCOE (e.g. a 12 h BESS can reduce the 99% scenario from an LCOE of 1466 $/MWh to 266 $/MWh, more than an 80% reduction) but alternative approaches such as long term storage, transmission interconnection, or non-emitting forms of dispatchable generation are required to decarbonize the electricity system at reasonable cost.Finally, we find that increasing VREN in Alberta's electrical grid is a cost-effective way to lower greenhouse gas emissions from the estimated 28.4 million tonnes of CO2eq emissions with today's grid composition to 9 million tonnes of CO2ea at an 86%VREN penetration and a 100$/MWh LCOE.
These findings, which stem from the market structure, existing resource mix, and renewable resource availability are likely relevant for similar natural gas dominant northern latitude electricity grids as they pursue decarbonization.Such regions could include: Ireland (EirGrid Group 2022a), Nova Scotia (CER 2023b), New York (New York ISO 2019), New England (ISO New England 2022), and Saskatchewan (CER 2023a).As in Alberta, competitive electricity markets and high prices combined with carbon pricing can spur significant investment by lowering VREN generation costs below incumbent gas power plants.This can lead to quick adoption of variable generation which can quickly start to constrain existing assets.Understanding how quickly the residual load will change with increasing levels of VREN can inform market and policy designers in addressing these constraints.While, Ireland, New York, and New England have a competitive energy market structure.The power sectors in New York and New England have extensive interties with surrounding regions and operate under the Regional Greenhouse Gas Initiative a cap and trade system which caps emissions for the region from power generation (RGGI 2023, ISO New England n.d., New York ISO n.d.).Only Ireland has a both a carbon tax (Revenue Irish Tax and Customs 2023) and limited interties (EirGrid 2022b); it has already reached an impressive 40% of generation from VREN (EirGrid 2022a).Future work could explore the longer-term impacts of demand changes as end use electrification is expected to increase demand while demand side management could increase flexibility.

Figure 1 .
Figure 1.Historical utilization factors.Note: Distribution of the utilization factors in the dataset.Utilization factors for wind and solar were sorted from high to low for the top two graphs, and then combined with a 50/50 weighting before sorting, for the bottom graph.

Figure 2 .
Figure 2. Scenario 1 and 3 optimization results-VREN baseline, with 0 h and 12 h of energy storage.Note: (a) scenario 1 results, these are the optimal wind and solar capacities required to meet demand at each %VREN constraint from 0%-99%.The required curtailment and current and forecasted VREN capacities are also included.(b) Scenario 1 and 3 results are compared from 70%-99% VREN.

Figure 3 .
Figure 3. Hourly minimum and maximum dispatched natural gas and fleet composition.Note: (a) For each scenario, the minimum and maximum hourly dispatch of natural gas is plotted.(b) How the NG fleet changeover from Combined Cycle to Simple Cycle occurs.

Figure 5 .
Figure 5. Hourly generation histograms as a function of the %VREN satisfying demand.Note: (a) Hourly generation distributions for scenario 1 (0 h of energy storage).(b) Scenario 1, 2, 3, and 4 are compared from 80%-95% VREN.Capacity factors (CF) are shown annotated at the top of each plate and the 0 MW hours are excluded from the distribution but shown as a percentage of total hours in the separate sidebar to the left of each distribution.The axis scale is changed in row three to show the shape as CF decline significantly.

Figure 6 .
Figure 6.Annual emissions vs. %VREN penetration.Note.Annual emissions calculated using lifecycle values for electricity generation.Detailed assumptions and methodology are available in appendix C.

Figure 7 .
Figure 7. Levelized cost of electricity.Note: (a) Component LCOEs-NG with and without capital recovery, VRE with 0 h (scenario1) and 12 h (scenario 3) of energy storage.(b) Effective LCOE for the combined NG/VREN costs with and without capital recovery and with 0 h and 12 h of energy storage.The average 2021 and 2022 power pool prices are shown for (AESO 2022a, AESO 2023b).

Figure 8 .
Figure 8. NG hourly generation distribution-20%-48% VREN.Note: Figure shows the shape change from 20%-50% VREN levels.The 0 MW natural gas hours are shown in the sidebar to the left of each histogram and the fleet capacity factors are annotated.

Table 1 .
Energy storage and VREN as a percentage of demand across scenarios.

Table 2 .
Current, approved and announced VREN capacity generation in Alberta.
Note: Adapted from Current Supply Demand Report (AESO 2023a) and the Long Term Adequacy Report (AESO 2022b).

Table 3 .
Percent of hours with NG dispatch above 11 000 MW capacity.Table compares the percentage of hours that require more than 11 000 MW of Natural gas generation in each of the storage scenarios. Note: