Policy Research Working Paper 10858 Assessing Power System Disruptions and Associated Economic Impacts from Increasing Extreme Heat Events in Southeast Europe Using an Idealized Design Methodology David Farnham Ross Eisenberg Luc Bonnafous Urban, Disaster Risk Management, Resilience and Land Global Practice July 2024 Policy Research Working Paper 10858 Abstract As the world endeavors to decarbonize and shift toward analysis reveals that between 2021 and 2070, urban cen- sustainable energy sources, power systems will become ters in Southeast Europe may be at risk of an estimated increasingly dependent on weather conditions. This depen- four to nine power system disruptions per decade due to dence creates the challenge of managing fluctuations in increasing trends in extreme heat events. These disruptions both power supply and demand (particularly for cooling), have the potential to incur annual economic costs of up which can jeopardize system reliability, particularly during to tens of millions of dollars in some cities. The projected extreme weather events. To what extent will the increases in disruptions highlight the challenges of adapting power sys- peak cooling demands manifest more frequent power system tems to climate change, even with idealized regular redesign disruptions, posing risks to human health and economic and maintenance efforts. To mitigate power system fragility activity? This paper focuses on urban centers in Southeast during heat waves, the paper recommends implementing Europe and utilizes state-of-the-art climate simulations to measures such as securing reserve power capacity, promot- estimate changes in the magnitude of extreme heat events. ing urban cooling through greening initiatives, adopting It also estimates the frequency of potential associated power demand-side management with smart-grid infrastructure, system disruptions and their ensuing impacts on economic and increasing the deployment of solar power, which typi- activity by assuming an idealized design methodology. The cally has high generation potential during heat waves. This paper is a product of the Urban, Disaster Risk Management, Resilience and Land Global Practice. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at david.f@climate.ai, reisenberg1@worldbank.org, and lbonnafous@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Assessing Power System Disruptions and Associated Economic Impacts from Increasing Extreme Heat Events in Southeast Europe Using an Idealized Design Methodology David Farnham, Ross Eisenberg, Luc Bonnafous Keywords: Power system disruptions, Extreme heat events, Cooling power demands JEL Codes: R10, P48 Table of Contents LIST OF FIGURES ....................................................................................................................................................................3 LIST OF TABLES ......................................................................................................................................................................4 INTRODUCTION .......................................................................................................................................................................5 URBAN CENTERS OF SOUTHEAST EUROPE........................................................................................................................................... 6 METHODOLOGY ......................................................................................................................................................................7 DATA ................................................................................................................................................................................................................ 7 QUANTILE DELTA MAPPING POST-PROCESSING FOR CMIP6 SIMULATIONS ............................................................................. 8 ESTIMATING COOLING DEMANDS ............................................................................................................................................................ 9 SIMULATING THE ELECTRICITY SYSTEM CAPACITY DESIGN PROCESS ......................................................................................... 9 SIMULATING SYSTEM RELIABILITY BY SIMULATING THE NUMBER OF DEMAND EVENTS THAT ECLIPSE THE DESIGN CAPACITY OVER THE SYSTEM LIFETIME ..............................................................................................................................................11 ESTIMATING THE COST OF DISRUPTIONS .............................................................................................................................................12 RESULTS ................................................................................................................................................................................... 14 PROJECTIONS OF UNDER-SUPPLY DISRUPTION FREQUENCY..........................................................................................................14 PROJECTIONS OF ECONOMIC COSTS FROM UNDER-SUPPLY DISRUPTIONS .................................................................................16 DEPENDENCE ON DESIGN PARAMETERS ..............................................................................................................................................20 Dependence on design record length ........................................................................................................................................... 20 Dependence on design year (2021 to 2090) ............................................................................................................................... 21 MITIGATING THE COSTS OF OUTAGES AND NAVIGATING THE UNCERTAINTY OF PROJECTIONS ........................................................................................................................................................................ 21 CONCLUSIONS ....................................................................................................................................................................... 23 REFERENCES ......................................................................................................................................................................... 25 SUPPLEMENTARY MATERIAL ...................................................................................................................................... 28 QUANTILE DELTA MAPPING DETAILS .................................................................................................................................................28 THE IMPACT OF A 15% RESERVE MARGIN ..........................................................................................................................................29 THE IMPACT OF CONSIDERING BOTH HEATING A COOLING DEMAND .........................................................................................29 THE IMPACT OF DIFFERING DESIGN METHODOLOGIES ....................................................................................................................31 DEPENDENCE ON DESIGN METHODOLOGY RESULTS .......................................................................................................................32 IDEALIZED DESIGN PROCESS AND SOURCES OF UNCERTAINTY ....................................................................................................33 APPENDIX 1: CHANGES IN THE EXTREME PEAK HEATING AND COOLING DEMANDS OVER THE FIRST HALF OF THE 21ST CENTURY FOR GLOBAL CITIES ........................................................................... 36 2 List of Figures FIGURE 1: (A) GLOBAL MAP WITH THE SOUTHEAST EUROPE REGION HIGHLIGHTED WITH A GREEN RECTANGLE; (B) THE URBAN CENTERS INCLUDED IN THE SOUTHEAST EUROPE REGIONAL ANALYSIS. ............................................................................................................................................................................................ 7 FIGURE 2: OVERVIEW OF DESIGN AND EVALUATION PROCESS. ...............................................................................12 FIGURE 3: THE MEDIAN NUMBER OF POWER SYSTEM DISRUPTIONS DURING THE PERIOD OF 2021 TO 2070 DUE TO COOLING DEMANDS EXCEEDING SUPPLY ARE ESTIMATED TO BE BETWEEN ABOUT 4 TO 9 EVENTS PER DECADE FOR THE MAJOR URBAN CENTERS OF SOUTHEAST EUROPE. ...............................................................................................................................................................................................14 FIGURE 4: THERE IS SUBSTANTIAL UNCERTAINTY IN THE ESTIMATES OF MEAN NUMBER OF POWER SYSTEM DISRUPTIONS DURING THE PERIOD OF 2021 TO 2070 DUE TO COOLING DEMANDS EXCEEDING SUPPLY DUE TO A) CLIMATE MODEL UNCERTAINTY (BOXPLOT SPREAD) AND B) SOCIOECONOMIC PATHWAYS (THREE COLUMNS OF PANELS). THE HORIZONTAL LINES IN THE MIDDLE OF THE FILLED BOXES REPRESENT THE MEDIAN, THE FILLED BOXES SPAN THE 25TH TO 75TH PERCENTILES, AND THE WHISKERS EXTEND TO THE 5TH AND 95TH PERCENTILES. THE SPREAD SHOWN VIA THE BOXPLOTS REPRESENTS THE INTER-MODEL SPREAD; I.E., EACH POINT THAT DETERMINES THE BOXPLOTS IS A SINGLE MODEL ESTIMATE FOR THE MEAN NUMBER OF UNDER-SUPPLY EVENTS PER DECADE OVER THE PERIOD OF 2021 TO 2070. ................................................................................................................................................15 FIGURE 5: THE MEAN NUMBER OF POWER SYSTEM DISRUPTIONS GENERALLY INCREASES AS THE DESIGN RECORD LENGTH INCREASES FOR THE HIGH EMISSIONS SCENARIO. THE EMPIRICAL DESIGN METHOD (HISTORICAL 90TH%) GENERALLY RESULTS IN MORE POWER SYSTEM DISRUPTIONS THAN EITHER OF THE STATISTICAL GEV-FIT DESIGN METHODS FOR THE HIGH EMISSIONS SCENARIO................................................................................................................................................................20 FIGURE 6: THE MEAN NUMBER OF POWER SYSTEM DISRUPTIONS GENERALLY INCREASES AS YOU MOVE FURTHER INTO THE 21ST CENTURY FOR THE HIGH EMISSION SCENARIO, DECREASES AS YOU MOVE FURTHER INTO THE 21ST CENTURY FOR THE LOW EMISSION SCENARIO, AND MODESTLY DECREASES AS YOU MOVE FURTHER INTO THE 21ST CENTURY FOR THE MIDDLE OF THE ROAD EMISSIONS SCENARIO. .............................................................................................................................21 FIGURE A1 1: ESTIMATES (USING THE MEAN OF ALL CLIMATE MODELS) OF THE PERCENT CHANGE (BETWEEN THE PERIODS 1990-2019 AND 2030-2059) IN THE 1-IN-10-YEAR PEAK COOLING (TOP) AND HEATING (BOTTOM) DAILY DEMANDS FOR THE GLOBAL SET OF LARGE URBAN CENTERS ACROSS EACH THREE OF THE FUTURE SOCIOECONOMIC PATHWAYS, NAMELY “LOW EMISSIONS” (SSP126; LEFT), “MIDDLE OF THE ROAD EMISSIONS” (SSP245; CENTER), AND “HIGH EMISSIONS” (SSP585; RIGHT). WE TRUNCATE THE INCREASES TO 100% FOR VISUALIZATION PURPOSES. ...................................................................................................................................................38 FIGURE A1 2: SAME AS FIGURE A1 1 BUT FOR THE PERIODS OF 1990-2019 AND 2070-2099. .............................38 FIGURE A1 3: THE REGIONAL BOUNDARIES OF THE CLIMATE REGIONS USED FOR GROUPING RESULTS. ....39 FIGURE A1 4: ESTIMATES (USING THE MEAN OF ALL CLIMATE MODELS) OF THE CHANGE (BETWEEN THE PERIODS 1990-2019 AND 2030-2059) IN THE 1-IN-10 YEAR EXTREME PEAK DAILY COOLING AND HEATING DEMANDS GROUPED BY CLIMATE REGION. THE PERCENT 3 INCREASE THE HORIZONTAL LINES IN THE MIDDLE OF THE FILLED BOXES REPRESENT THE MEDIAN, THE FILLED BOXES SPAN THE 25TH TO 75TH PERCENTILES, AND THE WHISKERS EXTEND TO THE 5TH AND 95TH PERCENTILES. THE SPREAD SHOWN VIA THE BOXPLOTS REPRESENTS THE SPREAD OF THE MEAN MODEL ESTIMATES FOR EACH OF THE URBAN CENTERS IN EACH REGION. NOTE THAT WE HAVE ALSO ADDED A REGION CALLED “CAR” TO INCLUDE THE CARIBBEAN CITIES OF PORT-AU-PRINCE (HAITI), SANTO DOMINGO (DOMINICAN REPUBLIC), AND HAVANA (CUBA). ....................................................................................................41 FIGURE A1 5: SAME AS FIGURE A1 4 BUT FOR THE PERIODS OF 1990-2019 AND 2070-2099. .............................42 FIGURE A1 6: GRAY POINTS SHOW THE MEAN MODEL ESTIMATES FOR THE 1-IN-10-YEAR PEAK HEATING AND COOLING DEMAND FOR EACH URBAN CENTER OVER THE PERIOD OF 1990- 2019, PURPLE POINTS SHOW THE MEAN MODEL ESTIMATES FOR THE 1-IN-10-YEAR PEAK HEATING AND COOLING DEMAND FOR EACH URBAN CENTER OVER THE PERIOD OF 2030- 2059, AND THE ARROWS CONNECT EACH URBAN CENTER AND ILLUSTRATE THE UPWARD TRENDS IN PEAK COOLING DEMANDS (Y-AXIS) AND THE DOWNWARD TRENDS IN PEAK HEATING DEMANDS (X-AXIS). ..............................................................................................................................................42 FIGURE A1 7: SAME AS FIGURE A1 6 BUT FOR THE PERIODS OF 1990-2019 AND 2070-2099 ..........................43 List of Tables TABLE 1: THE MEDIAN, 5TH, AND 95TH PERCENTILE ESTIMATES FOR THE NUMBER OF POWER SYSTEM DISRUPTIONS DURING THE PERIOD OF 2021 TO 2070 DUE TO COOLING DEMANDS EXCEEDING SUPPLY. ...................................................................................................................................................................16 TABLE 2: THE MEDIAN, 5TH, AND 95TH PERCENTILE ESTIMATES FOR THE PERCENT OF ANNUAL GDP LOST DUE TO UNDER-SUPPLY EVENTS DURING THE PERIOD OF 2021 TO 2070 DUE TO COOLING DEMANDS EXCEEDING SUPPLY USING THE HIGH ESTIMATE OF 0.35% PER DISRUPTION EVENT. ....................................................................................................................................................................18 TABLE 3: SAME AS TABLE 2 BUT FOR 2015 GDP (IN 2015 US $MILLION)..................................................................18 TABLE 4: THE MEDIAN, 5TH, AND 95TH PERCENTILE ESTIMATES FOR THE PERCENT OF ANNUAL GDP LOST DUE TO UNDER-SUPPLY EVENTS DURING THE PERIOD OF 2021 TO 2070 DUE TO COOLING DEMANDS EXCEEDING SUPPLY USING THE LOW ESTIMATE OF 0.06% PER DISRUPTION EVENT. ....................................................................................................................................................................19 TABLE 5: SAME AS TABLE 4 BUT FOR 2015 GDP (IN 2015 US $MILLION)..................................................................19 TABLE A1 1: THE FULL NAMES OF THE CLIMATE REGIONS DEFINED IN FIGURE A1 3. ............................................40 4 Introduction As efforts to decarbonize progress, it is vital to ensure the reliability of power systems within the context of the challenges of phasing out fossil fuels, aging transmission and distribution infrastructure, and increasing weather-induced fluctuations on energy supply (i.e., wind, solar, and hydro-power) and demand (i.e., heating and cooling demand). Even without these increasing challenges, the economic costs of unreliable critical infrastructure are already costly, particularly in less developed countries. A 2019 World Bank report estimated that the cost of unreliable infrastructure in 137 low- and middle-income countries (a cohort of countries that represents about 80 percent of the GDP of all low- and middle-income countries) was nearly $300 billion a year (about 1.5% of GDP), with over half of that cost associated with electricity outages and the subsequent losses to sales and the costs of self-generated electricity (Rentschler et al. 2019). A 2019 survey found that power outages in a typical month in Kosovo, North Macedonia, Albania, and Serbia were estimated to occur about 8, 5, 4, and 2 times each month, respectively (Dalloshi et al. n.d.). Extreme demand for cooling is among the many reasons why power systems can fail to maintain reliable supply. High temperatures heat the land surface and infrastructure, trigger the usage of air conditioning where available, and cause some of the highest electricity demand spikes that systems experience (Franco et al. 2007, Waite et al. 2017). As such, periods of elevated electricity demand driven by hot temperatures are critical times for electrical grids and can lead to soaring energy prices and, most critically, power outages or rolling blackouts (Hatvani-Kovacs et al. 2016). Examples include the blackouts triggered in California in July 2018 when spikes in electricity demand during a heat wave could not be met by the power system (Khan 2018). The July heat wave of 1995 was an even more dire example, where power failures were partially to blame in the deaths of over 800 people in the United States (Changnon et al. 1996). Furthermore, anthropogenic warming is expected to exacerbate these power system reliability issues (Allen et al. 2016, Li et al. 2012, Franco et al. 2007, Miller et al. 2008) via rising temperatures, and the anticipated increased adoption of air conditioning will lead to more energy systems being stressed during hot times (Waite et al. 2017). In addition to climate change’s effects of exacerbating the magnitude of extreme heat, the trend of increasing urbanization can also increase the magnitude of extreme cooling demand events for two 5 reasons. First is the well-known phenomenon of urban heat islands, whereby urban environments become substantially hotter than the neighboring lands due to the increased absorption of energy (in the form of solar radiation) and reduced cooling through evapotranspiration. Second, higher population density centralizes the demand for cooling during extreme heat events in locations where air conditioning is available, thereby eliminating the natural process of smoothing demand spikes via more spatially distributed demand nodes that are unlikely to all peak in cooling demands simultaneously. Having established that the costs of power system blackouts are considerable, and that anthropogenic warming is increasing the magnitude and frequency of heat waves, the goal of this paper is to use the latest generation of climate projections to estimate the propensity of urban power systems to eclipse their design capacity as a result of extreme cooling demands into the 21st century in major urban centers globally, with a special focus on the urban centers of Southeast Europe (Figure 1). Secondarily, this paper provides estimates of the economic costs associated with these urban power system disruptions. Lastly, this paper discusses the implications of increasing extreme heat events on critical power (and energy) system infrastructure and what adaptive responses can help ensure system reliability into the coming decades. Urban Centers of Southeast Europe Power system disruptions can be particularly salient in urban areas because of the high population density, the potential for cascading failures due to interconnected and interdependent systems, and the concentration of economic and industrial activities in and around urban centers. Higher population densities mean that more humans are impacted by power system disruptions. The interconnected and interdependent nature of many critical systems in urban settings, such as transportation and water systems, mean that power system interruptions can spark cascading failures and disruptions to multiple systems. Lastly, due to the high concentration of businesses and industries that rely on a stable power supply to operate in and around urban centers, urban power system disruptions can result in substantial lost revenue, reduced productivity, and increased costs for businesses. 6 Figure 1: (a) Global map with the Southeast Europe region highlighted with a green rectangle; (b) the urban centers included in the Southeast Europe regional analysis. Power system disruptions can be especially difficult to prevent and respond to in cities in Southeast Europe due primarily to aging infrastructure that has resulted from a lack of investment. Many cities in Southeast Europe have not seen significant investment in their power infrastructure in recent years (Vasquez et al. 2018), which has led to aging power infrastructure (e.g., power lines and transformers) that is more prone to failure and more difficult to maintain, which in turn increases the likelihood and severity of power system disruptions. Methodology The goal of this paper is to estimate the trends in the occurrence of potential power system disruptions due to power demand exceeding power supply during extreme heat waves. To accomplish this, we estimate the power demands and power supply capacities using a variety of datasets and methods outlined below. We provide a set of rough estimates of the economic impacts of these projected power system disruptions using a variety of datasets and methods. Data We use the following data sources for the analysis underlying this paper. • Historical temperature reanalysis data: Near-surface temperature (2 meters above the surface) at a spatial resolution of 0.25 deg by 0.25 deg and a temporal resolution of hourly 7 is acquired from the ERA-5 reanalysis dataset (Hersbach 2018). We derive estimates for the maximum and minimum temperatures at the daily level by taking the maximum and minimum of the hourly values for each day. • Future temperature projections from climate model simulation data: Maximum and minimum hourly temperatures over the previous 24 hours are acquired from Earth system model simulations from the Coupled Model Intercomparison Project, phase 6 (CMIP6). We consider three shared socioeconomic pathways, namely SSP1 (Sustainability, taking the green road), SSP2 (middle of the road), and SSP5 (fossil-fueled development). SSP2 currently appears to be the most likely scenario from an emissions perspective (Pielke Jr et al. 2022). And while SSP1 and SSP5 currently appear to be unlikely, understanding climate projections under these relatively extreme scenarios provides us with a view of the full range of possibilities. Moving forward, we refer to these scenarios as “low emissions” (SSP1), “middle of the road emissions” (SSP2), and “high emissions” (SSP5). We consider projections from the following global climate models: AWI-CM-1-1-MR, EC-Earth3-Veg- LR, FGOALS-g3, GFDL-ESM4, INM-CM4-8, INM-CM5-0, IPSL-CM6A-LR, KACE-1- 0-G, MIROC6, MPI-ESM1-2-HR, MPI-ESM1-2-LR, MRI-ESM2-0, and NESM3 since each of these models has less than 2% missing daily values, over the full simulation period of 1979 to 2100, for each of the three socioeconomic scenarios considered here. • Economic data: Gross domestic product (GDP) estimates, based on purchasing power parity, at a resolution of 30 seconds in both latitude and longitude (i.e., a spatial resolution of just under 1 km in the region of Southeast Europe) from Kummu et al. (2018). We use estimates from the year 2015 since these are the most recent available estimates. Quantile Delta Mapping Post-Processing for CMIP6 Simulations We post-process the Earth system model simulations using a quantile mapping method, which corrects systematic distributional biases in climate model output data by removing historical biases relative to observations. Specifically, we apply the Quantile Delta Mapping (QDM) method (Cannon et al. 2015) in order to preserve the trends in the climate model simulations in such a way as to ensure that the climate sensitivity (i.e., the temperature response to greenhouse gas forcing) of the model is not affected by bias correction. QDM adjusts the distribution of simulated climate data to more closely match the distribution of observed (or reanalysis) climate data, while 8 preserving the magnitude of the trends in each of the quantiles. Quantile Delta Mapping is described in detail, including presentation of the underlying equations, in the Supplementary Material. Estimating Cooling Demands We defined maximum inferred cooling demand during day (CD()) using the daily time-series of maximum temperature (i.e., the hottest hour of the day) as defined in eq. 1. CD() = max(max () − 18 °C, 0) eq. 1 Where max () is the maximum temperature for day . We note that the results shown in this paper are not sensitive to modest adjustments of the threshold at which we presume power demand for cooling begins to occur (i.e., 18 °C in this case). To complement this main text of the paper, we also present an analysis of the projected changes in extreme cooling and heating demands events over the 21st century for all urban centers of the globe with populations over 1 million (Appendix 1). The associated dataset will allow practitioners to quickly identify which urban centers are likely to be most or least in need of investments to bolster their power systems in order to keep pace with climate changes. Simulating the Electricity System Capacity Design Process We simulate the design process that a system designer would undertake to determine the appropriate size (or capacity) of a power system’s generation fleet during the summer months in order to achieve the ultimate goal of understanding the propensity of a system to become unreliable in the face of rising extreme temperature events. We assume that the largest loads on the electricity system during the summer occur during events of extreme temperature, and that other loads are relatively constant in nature. We focus on hourly extreme events and associated loads, though this same analysis can be carried out for longer duration events. We simulate the capacity design of a power system by estimating the peak daily demand that an electricity system should be designed to support, assuming that we are concerned with designing 9 a summertime system to withstand the burden of a 1-in-10 year 1 extreme event. The method entails taking the event size from the historical record with a 1-in-10 year return period to be the design event. Since this paper is focused on the impact of climate change on the reliability of the power grid, we do not explicitly consider changing population dynamics, major changes in electricity demand patterns, the effect of energy efficiency investments in buildings that make cooling more efficient and increases in air conditioning system penetration (linked to an increase in income and demand for comfort). Neglecting population changes should not have an impact on the results of this analysis because changes in population would be reflected on both the demand and supply side and thus would not impact the difference between demand and supply, which is the most critical quantity. Increasing efficiency in buildings to lower the power demand for cooling and increases in the penetration of air conditioning will have opposing impacts on this analysis, though the magnitude of their impacts is yet unknown and will likely be heterogenous across regions and urban centers. As such, we hold these values fixed for the time being but note that a follow-on analysis should investigate the extent to which these impacts themselves can be estimated. We repeat the system sizing simulation for each climate model (to estimate the impact of climate model uncertainty in our projections and subsequent design and system performance simulations), for each shared socioeconomic pathway (to estimate the dependence of our simulations on the future emissions scenario), and for each urban center separately. We have also included two alternative methods for estimating the system capacity to illustrate the uncertainty (or lack thereof) that is inherited from the design methodology (see Supplementary Material). Lastly, we investigate the impact of varying two design parameters to understand the extent to which power system disruptions due to under-supply can be expected to accelerate into the future and the extent to which issues of potential under-design are related to the historical record length used in the capacity analysis. For all simulations, the system redesign is conducted every ten years. • Design year (): defined by the year in which the updated design and system updates are assumed to have occurred. 1This is a common design standard in power system capacity adequacy planning (https://www.esig.energy/five-principles-of-resource-adequacy-for-modern-power- systems/). 10 • Record length (): defined by the length of the record used to estimate the design event. Simulating System Reliability by Simulating the Number of Demand Events That Eclipse the Design Capacity over the System Lifetime Now that we have simulated the design process to establish the capacity of the system, the next step is to evaluate the performance of the hypothetical system to support the projected demands over the system lifetime (assumed to be 10 years in this case) via eq. 2. Dec 31st of +10 #of cooling-demand-related-outages = =Jan 1st of �() > ()� eq. 2 where is the design year and () is the design capacity associated with design year . The full process is illustrated in Figure 2. We also illustrate the impacts of a 15% safety margin applied to this process in the supplemental section (Figure S1). Not surprisingly, the implementation of a 15% safety margin reduces the number of estimated disruption events considerably (by about an order of magnitude for many urban centers.) We also illustrate the impacts of considering both peak heating and cooling demand simultaneously (assuming complete electrification of heating; Figure S2). Notably, under this assumed scenario, peak annual demand is dominated by heating for the next many decades still and the total numbers of disruptions are reduced by about 2x to 4x, depending on the urban center of interest. There are several caveats associated with this analysis. The first is that this paper considers undersupply at the bulk system level (i.e., when the installed generation is not sufficient to meet demand) and does not explicitly consider outages due to stress on specific channels along the transmission and distribution systems during times of peak demand. Second, this framework does not explicitly estimate the duration nor size of an outage event (i.e., how long does the disruption last and what fraction of the city is impacted). Instead, this framework simply allows us to estimate the frequency of events when undersupply during heat waves is likely to stress the system to the point of failure. Lastly, the reality is that power systems are consistently undergoing slow gradual changes, rather than the stepwise system updates that we model here. We have simplified this into distinct, regular redesigns to isolate the effects of fluctuating capacity needs (required to meet peak demands during heat waves) over the course of decades. This approach aligns with the reality that power system capacity evaluations do occur at specific, discrete times. 11 Figure 2: Overview of design and evaluation process. Estimating the Cost of Disruptions Electricity availability is a key ingredient in enabling economic activity in any modern economy. Still, it is difficult to estimate the fraction of economic activity for each of the urban centers that depends on the availability of reliable electricity. In fact, to our knowledge, there are no good estimates for the fraction of economic activity that is dependent on electricity availability for any economies of the world. Additionally, the duration of power outages associated with an overburdened system can vary depending on whether proactive steps are taken by the system 12 operator, the level of maintenance of the power system, etc. As such, it is difficult to derive a precise estimate for the impact of power disruptions on an economy tailored to specific locations. Because of this, we use some relatively recent power outages that have been well documented and studied to derive a range of possible impacts on the economy of an urban center due to an under- supply event. Specifically, we use the February 2021 blackout in Texas and the 2019 rotating blackouts in South Africa. While these two events represent fundamentally different types of events, one where rotating blackouts were necessary throughout the year for a total of about 530 hours and one where a discrete event caused disruptions for many days, they both represent events that were brought on by insufficient generating capacity. We use economic damage estimates from these two events to provide a range of cost estimates. In the case of Texas, the estimated cost of the power system disruption caused by insufficient supply was estimated at around $130 billion (Busby et al. 2021), out of an estimated GDP for Texas for that year of about $2.05 trillion (Economic Analysis 2022). In other words, the event cost about 6% of annual GDP. “Given the fact that the February 2021 event had associated power outages that spanned 17 days, we compute the average loss per day in GDP, which is about 0.35% per day.” In the case of South Africa, the rotating blackouts throughout the year lasted a total duration of about 530 hours (or about 22 days) (Trace 2020) and cost between $3.5 billion and $7 billion (Trace 2020), out of an estimated annual GDP of about $400 billion. 2 Taking a central cost estimate of about $5.25 billion, the cost of the rotating blackout were about 0.06% per day. These estimates are simplistic but provide a crude range (spanning nearly an order of magnitude) with which to approximate the economic damage associated with blackouts or rotating blackouts due to insufficient generating capacity. We therefore use a low estimate of 0.06% of annual GDP per disruption and a high estimate of 0.35% of annual GDP per disruption. 2 https://data.worldbank.org/indicator/NY.GDP.MKTP.CD?locations=ZA 13 Results Projections of Under-Supply Disruption Frequency Figure 3: The median number of power system disruptions during the period of 2021 to 2070 due to cooling demands exceeding supply are estimated to be between about 4 to 9 events per decade for the major urban centers of Southeast Europe. Our simulations indicate that many urban centers in Southeast Europe can be expected to face issues of under-supply for the purpose of space cooling during heat waves, with a frequency of about 4 to 9 days per decade, under an idealized power system design and upgrade methodology with assumed effective execution (Figure 3). There are differences in these estimates depending on which urban center is of interest, with Skopje, North Macedonia, standing out as being particularly vulnerable to disruptive events as a result of increasing extreme temperatures and associated space cooling demands. In addition to the differences by specific urban center, there is substantial uncertainty in our estimates that results from the variation in climate model simulations and the assumed socioeconomic pathway (Figure 4). For example, the median estimate for Belgrade for the middle of the road vs. the high emissions scenario is 6 vs. 12 events per decade, while the 90 percent confidence interval for Tirana for the middle of the road emissions scenario spans 4 to 19 events per decade. 14 Figure 4: There is substantial uncertainty in the estimates of mean number of power system disruptions during the period of 2021 to 2070 due to cooling demands exceeding supply due to a) climate model uncertainty (boxplot spread) and b) socioeconomic pathways (three columns of panels). The horizontal lines in the middle of the filled boxes represent the median, the filled boxes span the 25th to 75th percentiles, and the whiskers extend to the 5th and 95th percentiles. The spread shown via the boxplots represents the inter-model spread; i.e., each point that determines the boxplots is a single model estimate for the mean number of under-supply events per decade over the period of 2021 to 2070. Our simulations of urban power system disruptions resulting from under-supply indicate that the simulated number of disruptions is consistently greater than the expected number of disruptions (i.e., one per decade when the design level was set for the 1-in-10 year event), regardless of socioeconomic pathway. While this is not surprising given that the assumed design methodology is backward looking in its derivation of the design capacity and there is a positive trend in extreme temperatures over the time periods considered here for the middle and high emissions scenarios, the magnitude of the projected number of disruptions above 1 per decade is marked (median across models of 4 to 9 events per decade for the middle of the road emissions scenario; Figure 3). 15 Table 1: The median, 5th, and 95th percentile estimates for the number of power system disruptions during the period of 2021 to 2070 due to cooling demands exceeding supply. Results for two potential alternative design methodologies are discussed in the Supplementary Material. Notably, using different assumed design methodologies does not have a marked impact on the primary disruption and cost estimates presented above (Figure S3). Projections of Economic Costs from Under-Supply Disruptions We estimate the annual economic cost associated with heat wave related urban power system disruptions for urban centers in Southeast Europe (between 2021 and 2070) for the middle of the road emissions scenario to be anywhere from about 0.03% to 0.33% of GDP or about $2 million to $120 million, depending on which urban center is the focus, based on our simulations and economic assumptions outlined in the methods section (Table 2-Table 5). There are differences 16 in these estimates depending on which urban center is specifically of interest due to differences between regions in our simulated frequency of urban power system disruptions and because of the disparate sizes of the economies of urban centers. As a reminder, our estimates for GDP are based on 2015 GDP figures and no assumptions regarding economic growth are made for the economies modeled here. Because the estimated frequency of urban power system disruptions is a primary driver of these cost estimates, there are again substantial uncertainties in our estimates that result from the variation in climate model simulations and the assumed socioeconomic pathway (Table 3 and Table 5). In fact, in some cases, these uncertainties result in our 90 percent confidence interval spanning about half an order of magnitude, as is the case for Tirana. Despite the large ranges of the estimated cost of disruptions due to extreme cooling demands, increases in the extreme cooling demands are unlikely to be a primary driver of lost economic activity in many parts of Southeast Europe given the fragile state of the electricity system that currently exists. For example, rolling blackouts have recently been introduced in Kosovo (including in the capital of Pristina) due to lack of supply, which has been blamed on lack of investment. 3 In the near-term, the power supply of many countries in Southeast Europe may be more threatened by aging coal infrastructure (resulting from a lack of investments in maintenance and rehabilitation) and drought due to large dependence on coal and hydropower generation. For example, as of 2020, it was estimated that Kosovo depended on coal for over 94% of electricity generation, while it was estimated that Albania, Montenegro, Bosnia and Herzegovina, Serbia, and North Macedonia relied on hydropower for over 99%, over 42%, over 27%, over 25%, and over 23% of electricity production, respectively. 4 3 https://www.bloomberg.com/news/articles/2022-08-26/europe-energy-crisis-kosovo- learns-to-live-with-rolling-power-blackouts- again?utm_source=website&utm_medium=share&utm_campaign=copy 4https://www.iea.org/countries/kosovo, https://www.iea.org/countries/albania, https://www.iea.org/countries/montenegro, https://www.iea.org/countries/bosnia-and- herzegovina, https://www.iea.org/countries/serbia, https://www.iea.org/countries/north-macedonia 17 Table 2: The median, 5th, and 95th percentile estimates for the percent of annual GDP lost due to under-supply events during the period of 2021 to 2070 due to cooling demands exceeding supply using the high estimate of 0.35% per disruption event. Table 3: Same as Table 2 but for 2015 GDP (in 2015 US $million) 18 Table 4: The median, 5th, and 95th percentile estimates for the percent of annual GDP lost due to under-supply events during the period of 2021 to 2070 due to cooling demands exceeding supply using the low estimate of 0.06% per disruption event. Table 5: Same as Table 4 but for 2015 GDP (in 2015 US $million) 19 Dependence on Design Parameters Dependence on design record length Our simulations of urban power system disruptions resulting from under-supply indicate that the number of disruptions is generally greater when longer historical record lengths are used in the design process (Figure 5). This trend is particularly prominent for the high emissions scenario where the positive trend in the underlying peak cooling demand events is most strong, in turn resulting in the past record being a poor predictor of future events. Figure 5: The mean number of power system disruptions generally increases as the design record length increases for the high emissions scenario. The empirical design method (historical 90th%) generally results in more power system disruptions than either of the statistical GEV-fit design methods for the high emissions scenario. There is non-monotonic behavior in the number of under-supply events for the empirical estimates as we increase design record length for all emissions scenarios. This dynamic is the result of the fact that short design record lengths do not sample extreme events sufficiently to establish the 90th percentile, while longer design record lengths include more and more data that is not reflective of the upcoming years due to the increasing trend in the data. While there is clearly a relationship between the projected number of disruptions as the design record length is varied, it is important to acknowledge that the assumed emissions scenario exerts more influence over the projected number of failures, particularly when comparing the middle of the road and high emissions scenarios. 20 Dependence on design year (2021 to 2090) Our simulations of power system disruptions resulting from under-supply during times of high heat indicate that trends in the number of simulated disruptions over time depend on the assumed socioeconomic pathway, with the high emissions scenario leading to increasing frequencies of disruptions over time and the low emissions scenario leading to decreasing frequencies of disruptions over time (Figure 6). This is a manifestation of the fact that the power system design methodology uses past data to estimate the 1-in-10 year maximum event, and as such, accelerating trends in the annual maximum events will lead to increases in the number of simulated disruptions while decelerating trends in the annual maximum events will lead to decreases in the number of simulated disruptions over time. Figure 6: The mean number of power system disruptions generally increases as you move further into the 21st century for the high emission scenario, decreases as you move further into the 21st century for the low emission scenario, and modestly decreases as you move further into the 21st century for the middle of the road emissions scenario. Mitigating the Costs of Outages and Navigating the Uncertainty of Projections We have estimated the economic cost of projected power system failures due to under-supply during extreme heat events over the next 50 years and shown that these failures may occur with substantially greater frequency than would be expected based on the engineering design of the system (i.e., about once every ten years). Reducing these costs, in the absence of dramatic globally collective action to reduce emissions and thus greatly slow or halt global warming, can be achieved through several methods of mitigating and adapting to these increasing hazards. We discuss several options below, namely using explicitly forward-looking climate simulations, conducting system 21 assessments often, increasing the extent to which generation and demand co-vary, and reducing the impact of urban heat island effects through infrastructural and land cover interventions, respectively. a. Using explicitly forward-looking methods for power system capacity design, which will ultimately entail securing greater reserve capacity in order to keep up with climate change and the increasing frequency and magnitude of extreme heat events. One way of accomplishing this is to use forward-looking temperature projections rather than exclusively relying on the historical record. This should be undertaken with care, however, as many climate models have some deficiencies with respect to their ability to simulate extreme weather events (Ehret et al. 2012, Boberg & Christensen 2012). b. Increasing the frequency with which capacity assessments are conducted will allow system analysts to update the system to withstand current climates and will limit the extent to which they must rely on extrapolations from climate models. c. Increasing the resilience of the energy system to support increasing space cooling demand spikes during heat waves by increasing the potential power supply during heat waves. One such approach is the increased deployment of solar power. This is because the fluctuations in solar power are generally well correlated with cooling demand at the hourly timescales (their daily curves are generally strongly correlated), multi-day timescales (extreme hot surface conditions during heat waves are associated with increased solar radiation at the surface resulting from clear skies), and seasonal timescales (both peak in summer) (Richardson & Harvey 2015, Solomon et al. 2016, Barnham et al. 2013, De Jong et al. 2013). This same solar radiation can be transformed into power via solar photovoltaics or other solar collection methods, and that power can be used to cool indoor spaces to give humans a place to stay cool during periods of high heat. d. Reducing the impact of the urban heat island effect by increasing the local reflectivity of the urban landscape to absorb less solar radiation and thus lower surface and near-surface temperatures. This can be achieved through the application of lighter paint colors or other engineered reflective coatings on roofs and roads, as has been shown to reduce urban heat in many cases (Sen and Khazanovich, 2021). e. Increasing the ability of the land surface to cool through evapotranspiration by increasing the density of trees and other vegetation within urban centers. 22 f. Scaling up energy efficiency in buildings to achieve the same level of cooling with less energy. g. Adopting more efficient cooling systems (e.g., inverter AC units). h. Adopting demand-side response solutions (e.g., the power utility can remotely lower consumption of AC units to avoid an outage when conditions require). While the most effective and efficient approach for implementing these design and infrastructural interventions will depend on the specific conditions of each urban environment, reducing demands during heat waves (e.g., reducing the urban heat island effect) is generally likely to be more economically efficient than simply attempting to extrapolate increases in extreme demands and building out more generation capacity. In other words, reducing the amount of power generation capacity that a system must maintain (e.g., through the implementation of energy efficiency measures) is generally preferable to increasing the size of the system to meet increased demands. As was alluded to above, climate projections carry with them uncertainty, rooted in scenario uncertainty (i.e., to what extent will we collectively alter our greenhouse gas emissions into the coming decades), model uncertainty (the ability of global climate models to simulate the earth system’s response to greenhouse gas inputs), and natural internal variability. As such, it is not recommended that practitioners rely on exact projections 30 or 50 years into the future. Instead, power system capacity assessments should be updated at regular cadences (at a higher frequency than the ten years used here). As more observations are collected over time, the reality with respect to emission levels and how they manifest in near-surface extreme temperatures will become increasingly well understood and from a practical standpoint it is more important to attempt to keep pace with these changes rather than projecting them perfectly. Therefore, while this study is useful for understanding the approximate potential magnitude of disruptions, it should not be viewed as a precise blueprint of what to expect over the next several decades. We include a comprehensive discussion of the sources of uncertainty inherent in this study in the Supplementary Material. Conclusions Designing, maintaining and operating power system infrastructure that can keep up with a changing climate is a significant challenge. This challenge is expected to increase in the coming 23 decades. Power system failures caused by insufficient power supply during extreme temperature events can be costly and can pose risks to human health and safety. In the Southeast Europe region, we have projected that the frequency and cost of power system failure events due to under-supply may be tens of millions of $US per year in some urban centers if we apply cost assumptions based on the 2021 February blackout in Texas. Mitigating the economic damage from power outage events entails improved system planning and design (e.g., securing enough reserve capacity during months of the year when peak events are possible), improving the way in which power grids are operated during times of high demands to ensure that demands do not exceed supplies (e.g., controlled rotating blackouts), and/or decreasing the impact of the urban heat island effect by increasing the reflectivity and increasing the ability of the land surface to cool through evapotranspiration by increasing the density of trees and other vegetation within urban centers. From an investment perspective, investing in types of energy generation that align with demand peaks is one strategic approach to mitigate the risks associated with power outages and disruptions. By investing in energy sources that have high availability during times of peak demand, such as solar during midday summer peak demands driven by cooling demand or wind power during winter peak demands driven by heating demand, investors can help to ensure a reliable and sustainable power supply. These types of energy sources are becoming increasingly cost- competitive and are substantially less carbon intensive than fossil fuel generation. In conclusion, this analysis highlights the increasing challenges that power systems are facing with a changing climate, specifically the overburdening of grids during extreme temperature events, that can lead to power outages and disruptions to economic activity. The study illustrates the importance of timely investment in resilient and sustainable power systems that can help to mitigate these risks and ensure a reliable power supply by keeping up with the increasing demands stimulated by a warming climate. 24 References Allen, M. R., Fernandez, S. J., Fu, J. S. & Olama, M. M. (2016), ‘Impacts of climate change on sub-regional electricity demand and distribution in the southern United States’, Nature Energy 1(8), 1–9. Barnham, K., Knorr, K. & Mazzer, M. (2013), ‘Benefits of photovoltaic power in supplying national electricity demand’, Energy Policy 54, 385–390. URL: http://dx.doi.org/10.1016/j.enpol.2012.10.077 Boberg, F. & Christensen, J. H. (2012), ‘Overestimation of Mediterranean summer temperature projections due to model deficiencies’, Nature Climate Change 2(6), 433–436. URL: http://www.nature.com/articles/nclimate1454 Busby, J. W., Baker, K., Bazilian, M. D., Gilbert, A. Q., Grubert, E., Rai, V., Rhodes, J. D., Shidore, S., Smith, C. A. & Webber, M. E. (2021), ‘Cascading risks: Understanding the 2021 winter blackout in Texas’, Energy Research & Social Science 77, 102106. URL: https://linkinghub.elsevier.com/retrieve/pii/S2214629621001997 Cannon, A. J., Sobie, S. R. & Murdock, T. Q. (2015), ‘Bias Correction of GCM Precipitation by Quantile Mapping: How Well Do Methods Preserve Changes in Quantiles and Extremes?’, Journal of Climate 28(17), 6938–6959. URL: http://journals.ametsoc.org/doi/10.1175/JCLI- D-14-00754.1 Changnon, S. A., Kunkel, K. E. & Reinke, B. C. (1996), ‘Impacts and responses to the 1995 heat wave: A call to action’, Bulletin of the American Meteorological Society 77(7), 1497–1506. Dalloshi, P., Misini, S. & Hoti, A. (n.d.), ‘Analyzing business environment- Power outages as an obstacle for firms’, p. 19. De Jong, P., Sanchez, A. S., Esquerre, K., Kalid, R. A. & Torres, E. A. (2013), ‘Solar and wind energy production in relation to the electricity load curve and hydroelectricity in the northeast region of Brazil’, Renewable and Sustainable Energy Reviews 23, 526–535. URL: http://dx.doi.org/10.1016/j.rser.2013.01.050 Ehret, U., Zehe, E., Wulfmeyer, V., Warrach-Sagi, K. & Liebert, J. (2012), ‘HESS Opinions “Should we apply bias correction to global and regional climate model data?”’, Hydrology and Earth System Sciences 16(9), 3391–3404. URL: https://hess.copernicus.org/articles/16/3391/2012/ Flores, Nina M., Heather McBrien, Vivian Do, Mathew V. Kiang, Jeffrey Schlegelmilch, and Joan A. Casey. “The 2021 Texas Power Crisis: Distribution, Duration, and Disparities.” 25 Journal of Exposure Science & Environmental Epidemiology 33, no. 1 (January 2023): 21– 31. https://doi.org/10.1038/s41370-022-00462-5. Franco, G., Sanstad, A. H. & Low (2007), ‘Climate change and electricity demand in California’, Climatic Change 87(August 2006), 1–13. Hatvani-Kovacs, G., Belusko, M., Pockett, J. & Boland, J. (2016), ‘Assessment of Heatwave Impacts’, Procedia Engineering 169, 316–323. URL: http://dx.doi.org/10.1016/j.proeng.2016.10.039 Hersbach, H., B. B. B. P. B. G. H. A. M. S. J. N. J. P. C. R. R. R. I. S. D. S. A. S. C. D. D. T. J.- N. (2018), ‘ERA5 hourly data on single levels from 1959 to present. Copernicus Climate Change Service (C3S) Climate Data Store (CDS).’. URL: 10.24381/cds.adbb2d47 Jasiūnas, J., Lund, P. D., Mikkola, J. & Koskela, L. (2021), ‘Linking socio-economic aspects to power system disruption models’, Energy 222, 119928. URL: https://linkinghub.elsevier.com/retrieve/pii/S0360544221001778 Khan, A. (2018), ‘Heat wave sparks major power outages around Los Angeles, officials say’. URL: http://www.latimes.com/local/lanow/la-me-ln-ladwp-heat-outage- 20180707- story.html Kummu, M., Taka, M. & Guillaume, J. H. A. (2018), ‘Gridded global datasets for Gross Domestic Product and Human Development Index over 1990–2015’, Scientific Data 5(1), 180004. URL: https://www.nature.com/articles/sdata20184 Li, D. H., Yang, L. & Lam, J. C. (2012), ‘Impact of climate change on energy use in the built environment in di↵erent climate zones - A review’, Energy 42(1), 103–112. URL: http://dx.doi.org/10.1016/j.energy.2012.03.044 Miller, N. L., Hayhoe, K., Jin, J. & Auffhammer, M. (2008), ‘Climate, extreme heat, and electricity demand in California’, Journal of Applied Meteorology and Climatology 47(6), 1834–1844. of Economic Analysis, U. B. (2022), SAGDP1 State annual gross domestic product (GDP) summary, Technical report. URL: https://apps.bea.gov/itable/?ReqID=70step=1acrdn=1eyJhcHBpZCI6NzAsInN0ZXBzIjpbM SwyNCwyOSwyNSwzMSwyNiwyNywzMF0sImRhdGEiOltbIlRhYmxlSWQiLCI1MzEiXS xbIkNsYXNzaWZpY2F0aW9uIiwiTm9uLU Panteli, M. & Mancarella, P. (2015), ‘Influence of extreme weather and climate change on the resilience of power systems: Impacts and possible mitigation strategies’, Electric Power Systems Research 127, 259–270. URL: https://linkinghub.elsevier.com/retrieve/pii/S037877961500187X 26 Pielke Jr, R., Burgess, M. G. & Ritchie, J. (2022), ‘Plausible 2005–2050 emissions scenarios project between 2 °C and 3 °C of warming by 2100’, Environmental Research Letters 17(2), 024027. URL: https://iopscience.iop.org/article/10.1088/1748-9326/ac4ebf Rentschler, J., Kornejew, M., Hallegatte, S., Braese, J. & Obolensky, M. (2019), Underutilized potential: The business costs of unreliable infrastructure in developing countries, Technical report. Richardson, D. B. & Harvey, L. D. (2015), ‘Strategies for correlating solar PV array production with electricity demand’, Renewable Energy 76, 432–440. URL: http://dx.doi.org/10.1016/j.renene.2014.11.053 Solomon, A. A., Kammen, D. M. & Callaway, D. (2016), ‘Investigating the impact of wind-solar complementarities on energy storage requirement and the corresponding supply reliability criteria’, Applied Energy 168, 130–145. URL: http://dx.doi.org/10.1016/j.apenergy.2016.01.070 Sen, Sushobhan, and Khazanovich, L. “Limited Application of Reflective Surfaces Can Mitigate Urban Heat Pollution.” Nature Communications 12, no. 1 (June 9, 2021): 3491. https://doi.org/10.1038/s41467-021-23634-7. Sivak, Michael. “Air Conditioning versus Heating: Climate Control Is More Energy Demanding in Minneapolis than in Miami.” Environmental Research Letters 8, no. 1 (March 1, 2013): 014050. https://doi.org/10.1088/1748-9326/8/1/014050. Trace, S. (2020), ‘South Africa’s crippling electricity problem’. Vasquez, Claudia, Begolli, Rhedon, Van Gelder, Linda & Shukla, Sameer (2018), Western Balkans: Directions for the Energy Sector, Technical report. Waite, M., Cohen, E., Torbey, H., Piccirilli, M., Tian, Y. & Modi, V. (2017), ‘Global trends in urban electricity demands for cooling and heating’, Energy 127, 786– 802. URL: http://dx.doi.org/10.1016/j.energy.2017.03.095 27 Supplementary Material Quantile Delta Mapping details We start by defining the time-dependent cumulative distribution function (CDF) of the climate model projected time-series for a time window (we use a time window of 30 years) around time (eq. 3). () = CDF � ()� eq. 3 where is the non-exceedance probability associated with the value of the projected time series, , at time t, and CDF is the cumulative distribution function of the projected time series. The value associated with the modeled quantile in the historical period (i.e., the period over which we have both climate model simulated time-series and observations), as well as the observed value associated with the quantile can be found by the entering this value into the historical modeled inverse CDF (CDF−1 −1 ℎ ) and observed inverse CDF (CDF ), respectively. ℎ () = CDF−1 ℎ � ()� eq. 4 � () = CDF−1 � ()� eq. 5 The absolute change in quantiles between the historical period and the projected period at time is given by eq. 6. () = CDF−1 −1 � ()� − CDFℎ � ()� = () − ℎ () eq. 6 Lastly, the bias-corrected future projection at time t can be acquired by applying the absolute change additively to the historical bias-corrected value (eq. 7). � () = () + CDF−1 � () � ()� = () + eq. 7 In this case, we apply the absolute change to conserve the climate sensitivity of each model. Having said that, the bias-corrected future projection at time can also be acquired by applying the relative change multiplicatively to the historical bias-corrected value, which is recommended for variables such as precipitation. 28 The impact of a 15% reserve margin The addition of a 15% safety margin reduces the number of estimated disruptions by about an order of magnitude, though this does vary substantially depending on which urban center is of interest (Figure S1). Figure S1: Same as Figure 4 but with the additional of a 15% safety margin applied during the design process. The impact of considering both heating a cooling demand We now present an alternative analysis that considers both the impact of heating and cooling demand together, under the following assumptions: 1) Maximum heating demand is defined similarly to maximum cooling demand, except that it is proportional to the difference between 18C and the minimum temperature that occurred on each day. 2) We assume that the power required to meet one degree of heating demand is four times higher than the power required to meet one degree of cooling demand (Sivak, 2013). Because peak heating demand is generally much larger than peak cooling demand in this part of the world, primarily because of the climate and because of the higher power demand to support 29 heating demand (assumption 2 above), the simulation become primarily determined by trends in peak wintertime heating demands. There are two primary takeaways from the results of this analysis (Figure S2): 1) The number of estimated disruptions for the middle of the road is decreased in this analysis (by about 2x to 4x in terms of median estimates depending on the urban center of interest) because extreme cold events are, on average, decreasing in magnitude. 2) The number of projected disruptions in this scenario is lower for the high emission scenario because that scenario is associated with the most warming, and warming serves to decrease the magnitude of cold extreme events and thus the number of simulated disruptions in a system where peak demand is dominated by heating demand. These results indicate that future disruptions due to undersupply are not likely to be primarily driven by increasing extreme heat for the next many decades, especially if all heating is met by electricity (as is implicitly assumed here), and if the generation available to meet winter peaks is assumed to be available during summer. Having said that, extreme cold events that are more extreme than the 90th percentile event from the previous 20 years are projected to continue to occur, despite the projections that extreme cold events are decreasing in magnitude. Figure S2: Same as Figure 4 but considering both heating and cooling peak demands simultaneously. 30 The impact of differing design methodologies We simulate the capacity design of a power system by estimating the peak daily demand that an electricity system should be designed to support, assuming that we are concerned with designing a system to withstand the burden of a 1-in-10 year event, in the following three ways. Note that option (a.) is equivalent to what is presented in the main text. a. Empirical estimate: The method entails taking the event size from the historical record with a 1-in-10 year return period to be your design event. This is the most simplistic way to estimate a 1-in-10 year event. b. Stationary generalized extreme value (GEV) distribution: The 1-in-10 year event as estimated via statistical fitting procedure on the historical record assuming a GEV distribution. Fitting of a GEV distribution, rather than solely depending on an empirical estimate, is an attempt to mitigate the issues of relatively small sample sizes and variability in the annual maximum time-series data that can create instability in empirical estimates of extremes. c. Nonstationary generalized extreme value (GEV) distribution: The 1-in-10 years event as estimated via statistical fitting procedure on the historical record assuming a GEV distribution where the GEV distribution’s location parameter is assumed to be nonstationary in time (i.e., a function of time). This methodology is capable of incorporating the impact of underlying secular trends in the data when estimating the 1-in- 10 years events into the future. For each urban center, we repeat all three of these system sizing simulations for each combination of climate model and shared socioeconomic pathway. We do this to estimate the impact of climate model uncertainty in our projections and subsequent design and system performance simulations, and to estimate the dependence of our simulations on the future emissions scenario for each urban center separately. 31 Dependence on design methodology results Relatively speaking, our projected estimates for disruptions are not highly dependent on the design methodology (Figure S3). Having said that, the nonstationary GEV design methodology generally resulted in the lowest median estimates for the number of under-supply events, but the largest amount of inter-model uncertainty. There is also some seemingly counter-intuitive behavior in the form of increased number of under- supply events for longer records for the nonstationary method that is intended to account for a trend in the underlying data. This dynamic is a clear indication that simply using a nonstationary GEV and allowing the location parameter to depend on time does not guarantee that estimates will properly, or fully, capture the salient trends in the data. 32 Figure S3: Same as Figure 4 but for each of the three assumed design methods (three rows of panels). Idealized design process and sources of uncertainty The assumed power system design methodology and recurring redesign and upgrades are in some respects best case scenarios due to two assumptions. • System redesigns are undertaken on regular intervals (every ten years in the scenario highlighted here), and 33 • System upgrades are assumed to be undertaken immediately after the redesign process. On the other hand, the assumed design methodology and grid operation are in other respects not best case scenarios. • The statistical models used here to estimate the power system design capacity are not optimal in the presence of warming trends because they are not forward-looking; the 1-in- 10 years event that is being estimated is implicitly assumed to be valid over the system lifetime (10 years in this case). Forward-looking extrapolation is only possible for the nonstationary GEV method, but we have not considered that to maintain consistency across design methods. • A related factor is the fact that we simulated the system re-design to happen only every ten years. A best case approach would be to re-evaluate the system at least once a year. There are several sources of uncertainty in both the extreme temperature projections and the estimated frequency and economic impacts of electricity system disruptions for urban centers due to insufficient supply during extreme heat days. In some cases, these uncertainties are explicitly incorporated into this work. The sources of uncertainty explicitly considered in this work include: • Socioeconomic scenario uncertainty, i.e., the dependence on which of the three emissions scenarios is assumed. • Climate model uncertainty, i.e., the dependence on the climate models that were considered and listed in the data section. • Region/location, i.e., the variation of urban centers considered. There remains a great deal of uncertainty regarding the range of economic damage cost estimates. Swift and effective grid operator interventions to mitigate power system disruptions through rolling or partial blackouts are not explicitly modeled here and are only implicitly considered through the use of the range of economic damage cost estimates. We know that grid operators can take action in many cases to limit the extent of power system failures and these actions should ultimately reduce the economic costs associated with power system disruptions, however, modeling such actions would require a more granular modeling endeavor. In addition to the sources of uncertainty outlined above, this work does not consider the risks of power system failures resulting from extreme weather apart from power under-supply for space 34 cooling during times of extreme temperatures. These include a) physical damage to infrastructure, such as power lines, that may fail above the temperatures for which they were designed, b) under- supply due to extreme cold temperatures and the increased use of electric heating, c) lightning strikes of power lines that can cause equipment-damaging voltage surges, d) extreme winds that can damage overhead lines, e) heavy rain and floods that can damage substation equipment . For the reasons laid out above, it is appropriate to view these projections for weather- and climate- related power system interruptions as a) simplified scenarios that solely consider the impacts of trends in extreme hot temperatures and b) do not consider the mitigating actions that can be taken by grid operators to manage partial blackouts. 35 Appendix 1: Changes in the extreme peak heating and cooling demands over the first half of the 21st century for global cities In the main paper, we focus on the urban centers of Southeast Europe and estimate the potential costs of power system disruptions due to under-supply of power during heat waves. In the process, we illustrate how this is likely to increase in the coming decades, even under the idealized power system design and upgrade methodology outlined in the main text. One way to get a sense for how these changes may manifest in any city of the globe is to look at the evolution of extreme cooling and heating demands events over the 21st century. All else being equal, the extent to which the extreme events (e.g., 1-in-10-year peak cooling demands during heat waves and 1-in-10-year peak heating demands during winter storms) are increasing is a central predictor for the risk of power systems disruptions due to power under-supply. Given this, we compute climate projections for all urban centers globally with populations of over 1 million people and briefly present the results and data here (Figure A1 1 - Figure A1 7). This dataset will allow practitioners to quickly identify which urban centers are likely to be in most or least need of investments to bolster their power systems in order to keep pace with climate changes. We show the percent changes in the 1-in-10-year peak daily cooling and heating demands between the periods of both 1990-2019 and 2030-2059 as well as 1990-2019 and 2070-2099 by urban center (Figure A1 1 and Figure A1 2). We show the changes in the 1-in-10-year peak daily cooling and heating demands between the periods of both 1990-2019 and 2030-2059 as well as 1990-2019 and 2070-2099 in units of cooling degree days and heating degree days and by individual city for each of three shared socioeconomic pathways scenarios (Figure A1 1, Figure A1 2, Figure A1 6, and Figure A1 7) and grouped by climate regions defined by the IPCC (Figure A1 3, Table A1 1, Figure A1 4, and Figure A1 5). The dataset is rich with insights, and we highlight three below: 1. The rate of change in extreme hot (increasing) vs. cold (decreasing) events that necessitate power are highly variable across different cities of the globe when viewed as percent changes (Figure A1 1, Figure A1 2) or in absolute terms of cooling degree days or heating degree days (i.e., the difference in degrees between the daily temperature and the level that is considered a comfort temperature as outlined in the main text; Figure A1 4 and Figure A1 5). 36 2. There is a clear asymmetry between changes in extreme peak cooling and heating demands over the coming decades that varies by specific urban center and region. That is, the relationship between the increase in cooling degree days and the decrease in heating degree days for the 1-in- 10-year peak events is not consistent across the globe. Consider the top three regions in terms of the median increase in cooling degree days, namely Central Europe, East North America, and South Europe/Mediterranean for the case comparing 1990-2019 to 2030-2059 (Figure A1 4). The projected decreases in extreme peak heating demands for Central Europe and East North America are generally much larger (in absolute terms) than the projected increases in extreme peak cooling demands, while the projected decreases in extreme peak heating demands for the South Europe/Mediterranean region are generally approximately equal to the projected increases in extreme peak cooling demands. The implications of this for practitioners are not straight forward, given that cooling and heating peaks happen during different seasons and the historic magnitude of these events varies widely by location. The strong variation between regions and lack of symmetry, however, does demonstrate how the increases in extreme heat and decreases in extreme cold are not equivalent across climate regions. While the specific numeric results are different for the case of changes between 1990-2019 and 2070-2099, the asymmetry noted above is still clearly present (Figure A1 5). 3. Projected trends in extreme peak cooling demands and extreme peak heating demands between the periods of 1990-2019 and 2030-2059 are relatively similar regardless of which of the three socioeconomic (and emission) scenarios we consider (Figure A1 6). That is, these projections indicate that the changes in extreme events are likely to be similar over the next several decades, regardless of how much mitigation we as humans undertake over the coming years. Having said that, when we look out to end of the century (i.e. projected trends between the periods of 1990- 2019 and 2070-2099; Figure A1 7), difference between the emission scenarios emerge more clearly. The full dataset is attached and available for exploration. 37 Figure A1 1: Estimates (using the mean of all climate models) of the percent change (between the periods 1990-2019 and 2030-2059) in the 1-in-10-year peak cooling (top) and heating (bottom) daily demands for the global set of large urban centers across each three of the future socioeconomic pathways, namely “low emissions” (SSP126; left), “middle of the road emissions” (SSP245; center), and “high emissions” (SSP585; right). We truncate the increases and decreases at 75% and -75% for visualization purposes. Figure A1 2: Same as Figure A1 1 but for the periods of 1990-2019 and 2070-2099. 38 Figure A1 3: The regional boundaries of the climate regions used for grouping results. 39 Table A1 1: The full names of the climate regions defined in Figure A1 3. 40 Figure A1 4: Estimates (using the mean of all climate models) of the change (between the periods 1990-2019 and 2030-2059) in the 1-in-10 year extreme peak daily cooling and heating demands grouped by climate region. The percent increase The horizontal lines in the middle of the filled boxes represent the median, the filled boxes span the 25th to 75th percentiles, and the whiskers extend to the 5th and 95th percentiles. The spread shown via the boxplots represents the spread of the mean model estimates for each of the urban centers in each region. Note that we have also added a region called “CAR” to include the Caribbean cities of Port-au-Prince (Haiti), Santo Domingo (Dominican Republic), and Havana (Cuba). 41 Figure A1 5: Same as Figure A1 4 but for the periods of 1990-2019 and 2070-2099. Figure A1 6: Gray points show the mean model estimates for the 1-in-10-year peak heating and cooling demand for each urban center over the period of 1990-2019, purple points show the mean model estimates for the 1-in-10-year peak heating and cooling demand for each urban center over the period of 2030-2059, and the arrows connect each urban center and illustrate the upward trends in peak cooling demands (y-axis) and the downward trends in peak heating demands (x-axis). 42 Figure A1 7: Same as Figure A1 6 but for the periods of 1990-2019 and 2070-2099 43