Evaluation indicator of regional energy vulnerability to climate change - spatial and temporal analysis

25 Climate change might affect energy production and therefore the energy security of a country or region. This 26 vulnerability situation may affect Renewable Energy Sources (RES) such as hydroelectric and has consequences 27 on effective energy transition. Since the transition to RES is a key for decarbonizing the economy in line with 28 the Paris Agreement this situation is critical for many countries in which their energy systems are linked to 29 resources strongly affected by climate. The aim of this study is to purpose a vulnerability indicator (VI) to 30 evaluate the electric energy vulnerability of an on-grid system to climate change at national and regional level 31 taking as base the case of study of Colombia, a country with a system based on 70% RES. VI is computed with 32 different variables that may be related to climate change, the energy matrix, and vulnerability. Principal Component Analysis (PCA) was used to select the variables involved in the VI calculation. The VI was 34 calculated for the whole country and the 32 departments (states) showing that the regions with the larger 35 vulnerability correspond to the more energy demanding regions. These vulnerable regions to climate change are 36 more than 50% of the maximum possible vulnerability, meanwhile, the vulnerability of the whole country was 37 estimated as 43%. The analysis was developed for the current situation of Colombia in which there are two 38 regions: interconnected (SIN) and not-interconnected (ZNI) areas.

Climate change might affect energy production and therefore the energy security of a country or region. This 26 vulnerability situation may affect Renewable Energy Sources (RES) such as hydroelectric and has consequences 27 on effective energy transition. Since the transition to RES is a key for decarbonizing the economy in line with 28 the Paris Agreement this situation is critical for many countries in which their energy systems are linked to 29 resources strongly affected by climate. The aim of this study is to purpose a vulnerability indicator (VI) to 30 evaluate the electric energy vulnerability of an on-grid system to climate change at national and regional level 31 taking as base the case of study of Colombia, a country with a system based on 70% RES. VI is computed with 32 different variables that may be related to climate change, the energy matrix, and vulnerability. Principal 33 Component Analysis (PCA) was used to select the variables involved in the VI calculation. The VI was 34 calculated for the whole country and the 32 departments (states) showing that the regions with the larger 35 vulnerability correspond to the more energy demanding regions. These vulnerable regions to climate change are 36 more than 50% of the maximum possible vulnerability, meanwhile, the vulnerability of the whole country was 37 estimated as 43%. The analysis was developed for the current situation of Colombia in which there are two 38 regions: interconnected (SIN) and not-interconnected (ZNI) areas. Energy transition has turned imminent. Since the human appearance on the globe, the temperature has been 56 stable until the 19th century with the beginning of the industrial revolution (Houghton and Woodwell 1989;57 Wallace and Hobbs 2006). Between 1850 and 2012, the near-surface air temperature increased by 0.8 °C mainly 58 due to anthropogenic emissions produced during the energy production demanded by transport, industry, 59 residential, among other sectors (González 2007 Colombia is a developing country with a population of more than 50 million inhabitants in 2020. It is an Andean 66 country that produces around 75.5 TWh/yr of electric energy(UPME 2020). Colombia is currently energetically 67 autonomous and has an energy mix that produces around 70% of its electric energy based on renewables, mostly 68 hydropower production facilities, only a few small RES-based solar and wind projects take place currently in 69 the country, and they do not represent more than 2% in the near future (UPME 2020). The current energy 70 situation makes the country sensitive to climate change. Large scale blackouts have been registered in Colombia 71 in 1992, and more recently in 2016 during the most recent El Niño-Southern Oscillation (ENSO) period due to 72 the decrease in the dam and reservoir levels (Mateus 2016). The aforementioned situation is evidence that during 73 strong dry seasons, mainly characterized by the ENSO, the country faces a lack of energy supply leading to 74 blackouts (Cuadros et al. 2019). Additionally, the electric energy demand is trending to increase due to the 75 development processes, the energy transition, migration from the countryside to urban areas, and important 76 population growth due to the massive immigration of refugees from Venezuela in recent years (Berg et al. 2020;77 Betts 2019). These two issues: the hydropower based energy matrix and the increasing electric energy demand, 78 will drive the decision-making process regarding the energy strategies of the country in the next years. 79 80 To assess the vulnerability of regions and countries to climate change several studies have been published. The 81 RES vulnerability has been assessed in 2012 considering several factors such as endowment, infrastructure, 82 distribution, and transmission of energy based on many studies (Schaeffer et al. 2012). Energy is identified as 83 one of the areas with more impact on climate change (Mideksa and Kallbekken 2010), authors also mentioned 84 that energy supply is impacted as a result of climate alteration since variables such as wind speed, river flow, 85 evaporation rates, and solar radiation are changing. Additionally, it has been found that climate change is 86 affecting the performance of non-RES (mainly thermal and nuclear power plants) in Russia (Klimenko et al. 87 2018). 88 89 The main aim of this study is to propose a vulnerability indicator (VI) to evaluate the electric energy vulnerability 90 of an on-grid system to climate change at national and regional level taking as base the case of study of Colombia. 91 The proposed indicator is based on the different variables that may be related to climate change, the energy 92 matrix, and therefore the vulnerability. To identify the variables that must be used for the indicator calculation 93 the Principal Component Analysis (PCA) method was used since this approach allows identifying the 94 relationships between the different variables. The data used was retrieved for the last 20 years at the country 95 level and the last 7 years at the regional level. 96 97 The VI was calculated for the whole country and the 32 departments (states). The spatial comparison between 98 the different geographical locations was evaluated using different regional distributions of the VI. The method 99 based on the indicator proposed is designed to be easily adapted and used in other countries or regions 100 worldwide. 101

Current Situation of energy in Colombia 103
104 The current situation of Colombia is described in terms of the climate and energy variables. Climate and energy 105 data were retrieved for the last 20 years nationwide and last seven years to the departmental level. Climate data 106 was downloaded from the ERA-5 database from ECMWF (European Centre for Medium-Range Weather 107 Forecasts)(ECMWF 2020). 108 109 The energy demand and production were obtained from the historic records from UPME (Unidad de Planeación 110 Minero Energética). Figure 1a shows the electric energy demand (ED) per department. Figure Figure 3a shows the electric energy production (EP) and the ED of Colombia between 2000 and 2020. 127 Differences between demand and production are also shown to identify the strong deficit of energy production 128 since 2015. Between the years 2003 and 2015, no deficit of energy supply was presented, even so in the early 129 2000s small deficits took place in the country. Additionally, Figure 3a also shows the increasing trend of ED 130 and energy production with time. The data used for Figure 3a were retrieved from the UPME website (UPME 131 2021). This data is composed of ED and production records per month for the whole country. 132 133 Figure 3b shows the composition of the energy mix based on the energy generation sources per month between 134 19 august 2019 and 20 August 2020, data was retrieved from SIEL (UPME 2020), the current electric energy 135 mix is characterized by hydropower as the main production source with 70.4%, followed by natural gas and coal 136 with a 13.7% and 12.7 % respectively.

Vulnerability indicator (VI) 151
The indicator utilizes several variables for which the time profiles must be retrieved as the first step in the 152 calculation of the indicator. The variables must correspond to the same spatial area of study, i.e. the same region, 153 city, or country, and during the same period of time. Profiles of each used variable should be based on the same 154 time basis i.e. daily, weekly, monthly or yearly data. These variables might be related to climate change and 155 between them in order to be significant for the indicator calculation of the region of study. Variables can be 156 linked with different types of fields such as economic, climatic, and electric. Since the index calculation is 157 focussed on the variables strongly linked to climate change, to identify these variables a statistical method such 158 as PCA (Principal Component Analysis) (Abdi and Williams 2010) can be used. In addition, for some clusters 159 with many variables, it is necessary to use a correlation matrix to choose the most representative variable of the 160 cluster (Bartholomew 2010). 161 162 The Vulnerability Indicator of the region ( ) can be computed using the equation (1). In this equation, is 163 the weighting factor of the variable , and is the score factor of the time with increasing vulnerability for 164 variable .
The scoring factor is calculated according to equation (2), in which is the time unit (e.g. months, years, etc.), 170 and can take the value of 0 or 1 in each timestep depending if the vulnerability in terms of the variable 171 decreases or increases respectively. 172 173 The weighting factor is calculated according to equation (3), when is the standard deviation of variables 179 identified as the suitable ones for the vulnerability consideration, in this expression the indexes and denote 180 the aforementioned variables. Additionally, the condition of equation (4) must be complied with in any case. 181 182 With a good indicator, it might be possible to compare against cases (regions, cities, or countries), and against 186 the minimum and maximum possible values. The formulation of presented in this study allows this type 187 of analysis. To perform this analysis the calculation of the maximum possible can be carried out taking the 188 maximum possible value as = . This will lead to equation (5). Differently, the minimum possible 189 will always take the value of zero, see equation (6). for the region, city or country will mean r has been strongly affected by climate change, and a low % will mean 199 The variables selection in the vulnerability index of energy to climate change strongly depends on the study area. 221 For some cases the difference between energy and environmental policies makes necessary the use of more 222 specific data. In any case can be seen the use of satellite-based data is possible and a good source of information 223 for the vulnerability the analyses such as the presented in this article. In this research data used was retrieved from different sources of information e.g. local agencies reports, public 232 databases and satellite-based data. We apply the PCA approach to select the variables used in the calculation of 233 the VI, the weighting of variables was performed based on the standard deviation of the independent variable-234 time-series. This study is performed at country scale. 235 236 For clusters 1, 2, and 3 one single variable was selected as the representative of the set of variables within each 257

Results and discussion
cluster. The criterion used was the strength of the correlation in terms of the arrow length, i.e. the variable with 258 the larger arrow (the arrow closer to the circle) was selected. The cluster 4 has several variables, and the criterion 259 used for the cluster 1, 2, and 3 is not applicable, for this reason, to select the representative variable of cluster 4, 260 a correlation matrix analysis was performed for the variables within the cluster 4 ( Figure 4b). For cluster 4 the 261 ED was selected as the representative variable since according to the correlation matrix it has a higher correlation 262 coefficient than the other variables within the same cluster. 263 264 According to the analysis performed the selected variables for the vulnerability index calculation were ER, EE, 265 EI, and ED since they are representative of the identified clusters 1, 2, 3, and 4 respectively.  with RV, but at the same time, EE decreases with time. The increase of EI and RV is explained by the installation 281 of new dams between January 2000 and January 2020, it is estimated that the capacity of the reservoirs increased 282 by 27.7%, rising from 18 to 23 dams in the whole country (XM S.A. E.S.P. 2021b). As a consequence of the 283 energy supply deficit after 2015, the EE presents the fast reduction seen in Figure 5a. 284 285 There is an inverse relationship of RV with P (clusters 1 and 3) (Figure 4a), e.i. The decrease in precipitation 286 leads to a water level decline in the dams of the energy matrix, this is evidence of the relationship between the 287 energy system of the country and climate change through climate variables such as P. This is also evidence of 288 the valid clusterization performed based on the PCA method. 289 290 Figure 4a shows an inverse relationship between clusters 2 and 4, ED, EP, and Pop are the variables with major 291 influence in this relationship. Cluster 4 vectors show how when ED increases with Pop, this needs a major EP, 292 and this affects negatively the ER. Analysis of T and CE is important since they are variables related to climate 293 change. T and CE variables also conform to Cluster 4, with increasing but slightly weaker trends. 294 295 The CE trends to increase in developing countries such as Colombia, this due to the link between GHG emissions 296 and activities linked to industry, transport, agriculture, etc. In the case of Colombia T has the capability to 297 influence the energy structure of the country since the biggest share of the energy mix is dependent on water, a 298 resource-sensitive to T. 299 300 An interesting relationship between clusters 1 and 3 is observed, the P has a positive relationship with EE, 301 meaning that when rains are presented a super-habit of energy takes place, and therefore EE increases during 302 these periods. P is also inversely related to RV, for the same reason when no rains have presented the level of 303 water in the dams decreases leading to fewer reserves. This observation is coherent with the trend in Figure 5. 304 These observations evidence the high vulnerability of the Colombian energy system to climate variations since 305 this energy system is rich in hydropower (70% hydropower); this is, therefore, evidence of hydropower-RES-306 based-systems vulnerability to climate change. 307 308 Niña" phenomenon took place in South America. Figure 5c shows a net decrease of 4 TWh for ER between 2000 329 and 2020. 330

National vulnerability 332
To analyze the vulnerability at the national level, the VI was computed using the method proposed in this study. 333 The variables selected based on the PCA and correlation matrix were used for the computation of the VI as 334 suggested in section 2.2. 335 336 A VI of 8.64 was estimated for Colombia based on the data from 2000 to 2019. This index value corresponds to 337 43% of the maximum possible vulnerability (VImax=20), e.i. The country has 43% of its maximum possible 338 energy vulnerability to climate change. This percentage represents the vulnerability of SIN regions only in this 339 study. 340

Regional vulnerability 341
To analyze the distribution of the vulnerability in the country the VI was calculated for all the departments of 342 Colombia. This distribution allows analyzing the vulnerability in spatial terms. Figure 6 shows the percentage 343 of vulnerability in the departments of Colombia, calculated based on the VI proposed in this study for the period 344 between 2000 and 2019. ZNI regions are not included in the map. 345 To analyze the spatial distribution of the vulnerability, the average percentage of vulnerability was computed 350 considering the values grouped in 3 ways: 1-north/south, 2-east/west, and 3-per natural region (i.e. Amazonia, 351 Andina, Caribe, Orinoquia, and Pacifica A small difference is observed between the east and west averages (0.4%), the east average is slightly larger than 359 the average for the west. This is due to the distribution of the surface covered by the interconnected areas (SIN) since the SIN areas are balanced in east/west, differently to the difference between north and south which is 361 greater, being the north more vulnerable than the south with an average percentage of vulnerability 1.7% larger. 362 363 The region Andina has the largest average percentage of a vulnerability index that might be linked with the large 364 electricity demand in this region. differently, the Amazonia region has the lowest percentage of vulnerability, a 365 result in agreement with the lower electricity demand of this region. Pacific and Orinoquia regions have similar 366 average percentages of vulnerability, with values 0.6% and 0.7% lower than the Andina region. 367

Conclusions 368
This study presents a method to evaluate the energy vulnerability to climate change based on a vulnerability 369 index (VI), for this calculation a numerical analysis performed with the PCA method and correlation matrix was 370 performed. The study presents the method to calculate the VI and its implementation using Colombia as a case 371 of study. 372 373 Relationships between the variables linked to the energy system and climate change were identified. Evidence 374 linking climate change and energy vulnerability was observed in the relationships of the analyzed variables 375 (Figure 4), e.g. P is inverse to EI, and T inverse to ER. Additionally, the trends of the studied variables with 376 time show the impact of ENSO and "La Niña" phenomena on the system during the period of time between 2000 377 and 2019 (Figura 3a and Figura 5). This demonstrates that an important risk of energy supply lack in Colombia 378 is linked to hot dry years, and therefore to conditions induced by climate change in the future. This effect also 379 contributes to the estimated vulnerability, e.i. A link between climate change and the increasing energy 380 vulnerability in Colombia was evidenced based on the analysis performed with the variables involved. 381 382 The PCA analysis performed on the variables of the Colombian system allowed us to identify relationships 383 between the variables in clusters 1 and 3, leading to identify a high vulnerability of hydropower-RES-based-384 systems to climate variation, and therefore to climate change. 385 386 The vulnerability of the energy system of Colombia to climate change was quantified as 43% of its maximum 387 possible at the national SIN scale. The analysis per region was based in the department borderlines of the country 388 SIN areas showing that the higher vulnerability is located in the regions with higher demand (e.g. departments 389 of Andina region), even if they have an important EP infrastructure and are part of SIN. Some regions such as 390 Chocó located at the west of the country on the pacific coast, have a large vulnerability, this can be explained by 391 the lack of EP and transmission infrastructure due to geography and security issues due to the presence of illegal 392 groups. 393