The performance of Mediterranean low-income housing in scenarios involving climate change

Energy and Buildings Volume 202, 1 November 2019, 109374 The performance of Mediterranean low-income 1 housing in scenarios involving climate change. 2 Samuel Domínguez-Amarillo1, Jesica Fernández-Agüera1*, Juan José Sendra1, Sue Roaf2 3 1Instituto Universitario de Arquitectura y Ciencias de la Construcción, Escuela Técnica 4 Superior de Arquitectura, Universidad de Sevilla, Spain. [email protected], [email protected] 5 2Heriot-Watt University, Edinburgh. [email protected] 6 *jfernande[email protected] 7 Abstract 8 Social housing dating from the period between the Second World War and the end of the oil 9 crisis is one of the major stores of residential stock of European cities. This housing stock is a 10 major target for retrofitting given its characteristic poor thermal performance and inefficient 11 control of energy consumption. This article proposes a method for analysing the dynamic 12 capacity of thermal enclosures on moderate energy flows in building stock in climate change 13 scenarios, estimating the potential for adaptation and strengths and weaknesses of several 14 building categories exposed to different present and future climate scenarios. A pilot study 15 applying the procedure is carried out in the city of Seville, one of the largest in southern Europe, 16 with a representative northern Mediterranean climate. The approach designed is equally 17 applicable to other urban centres in southern Europe. Although indoor comfort in cold weather 18 must be addressed even in the least favourable future scenarios, the predominant concern for 19 this stock is controlling heat gain. This study shows how, regardless of individual situations, 20 thermal insulation alone does not guarantee an optimal response for the stock as a whole. 21 Different categories can be identified within a given stock, where some buildings display 22 significant resilience and potential for adaptation to new scenarios, while others have less scope 23 for improvement. These conclusions can provide guidelines for the design of future intervention 24 policies in southern Europe. 25 Keywords: social housing; climatic change; southern Europe; energy demand; monitoring; 26 simulation. 27 1. Introduction 28 Social housing, in its collective building form, is central to the configuration of current cities in 29 southern Europe. The considerable volume of housing built in Europe after World War II [1], [2] 30 and in Spain after the Civil War greatly affects the current energy behaviour of urban housing, 31 and must be taken into account when assessing the energy performance expected from these 32 cities. In Seville over 48 % of multi-family residential buildings - the most common type - were 33 built between 1939 and 1979 [3], [4]. Together with the buildings constructed in the early 34 twentieth century, this figure adds up to over 51 % of present housing stock. Consequently, over 35 half of the city’s homes are to some extent obsolete. Of this 51%, 60% qualifies as ‘social housing’ 36 Energy and Buildings Volume 202, 1 November 2019, 109374 and accounts for over 30 % of Seville’s total housing stock, which is at serious risk of 37 underperformance. As a result, social housing and its capacity to continue to accommodate a 38 large part of the population when faced with future changes is called into question, given the 39 effects of climate change on consumption and indoor comfort conditions. 40 Social housing is usually occupied by medium to low-income families with limited resources to 41 invest in the control of the indoor environment. These conditions lead to situations affecting the 42 health, comfort, and quality of life of residents and should not be approached from the 43 standpoint of energy consumption alone. These factors must therefore all be taken into 44 consideration both when designing global policies to improve the performance of building stock 45 in general and when planning specific interventions. 46 Given current energy and emissions requirements [5] [6] and the sub-standard habitability 47 deriving from shortcomings in the building stock [7], suitable solutions are needed to lower 48 energy demands - and in turn energy consumption - in order to substantially improve indoor 49 environmental conditions. The most pressing concerns are the envelopes, particularly the 50 façades of multi-family buildings [8], given their crucial importance in ensuring the quality of 51 the indoor environment (thermal, acoustic control and air quality), and building aesthetics in 52 terms of the image of the city [9]. 53 This research aims primarily to present a wide-ranging study on the energy performance of 54 social housing in the city of Seville, one of the largest in Spain, with a characteristic climate highly 55 representative of southern Europe [10]–[12]. In cities with mild winters and warm summers 56 (even extreme ones), indoor thermal conditions, particularly in social housing, are usually 57 conditioned by economic considerations as well as a widespread lack of cooling and heating 58 systems in homes and poor building performance [7]. This analysis aims to establish probable 59 bands for the potential modification of the energy behaviour of building stock, rather than to 60 establish specific values for buildings which should be the subject of specific studies. 61 One of the main innovations of this study, in the context of urban buildings, is that local weather 62 is constantly changing, both through its natural variability and the effect of anthropogenic 63 factors such as the different processes of climate change (CC). The climate which affects 64 buildings has undergone – and will continue to undergo – changes which will have a direct effect 65 on buildings’ energy performance [13], [14] and comfort [15], [16]. The characteristics of this 66 evolution and the main interactions with these buildings are analysed, generating a future 67 evaluation scenario in order to establish a correlation between current and potential future 68 scenarios in terms of factors driving building energy performance and energy use in indoor 69 climate control. 70 Although this analysis focuses on the evolution of housing stock in the city of Seville, the work 71 methodology and classification procedures followed are equally applicable to other cities and 72 urban areas in southern Europe. 73 As mentioned above, current environmental conditions, coupled with the effects of climate 74 change, affect energy balances. This is especially noticeable among social groups of lower 75 economic status, often affected by increased demands on energy supply systems for the 76 Energy and Buildings Volume 202, 1 November 2019, 109374 improvement of indoor conditions [17]. As the energy demand for the thermal control of the 77 buildings is directly related to climatic conditions, modifying these will lead to new energy 78 scenarios for cities and urban areas [14], [17]–[19]. 79 Retrofitting actions to reduce CO2 emissions from residential buildings and enhance energy 80 savings are usually evaluated considering the conditions of the current climate (and on occasion 81 historical data). This is particularly useful when establishing the performance of present-day 82 buildings in future climate conditions, as well as the potential performance of the energy 83 improvement measures currently under development for residential buildings, and their 84 effectiveness in a future iteration of the current climate. This issue is examined by Hooff [20], 85 Gupta [21] and Roders [22] although focusing fundamentally on colder areas in Europe. 86 A comprehensive review of the literature on the impact of climate change on building 87 performance can be found in [23]–[25] for the fundamental concepts and methods and a global 88 review in [26]–[29], covering the most recent trends. In general, it is emphasized that in 89 predominantly warm conditions or in areas with a balance between heating and cooling needs, 90 the potential for reduction of energy (or emissions) is strongly altered by the effects of climate 91 change [23] [30]. In the Mediterranean area, the increase in the cooling needs of homes, 92 especially those of lower constructive quality, is especially significant, as in [31] for Greece, 93 where a significant increase in cooling degree days and maximum indoor temperatures 94 (naturally ventilated homes) is forecast. 95 2. Methodology 96 This paper proposes a method for the analysis at housing stock-scale of the dynamic and 97 evolutive capacity of the thermal envelopes trading energy flows within climate change 98 scenarios. A probability assessment has been developed to establish the performance expected 99 for city housing stock by 2050. This short to medium timeline will provide the key points for 100 effective and economic actions in cities, calculating the potential for adaptation as well as the 101 strengths and weaknesses of various building categories when exposed to different present and 102 future climate scenarios. 103 In order to evaluate the potential impact of a climate change scenario on a multi-family social 104 housing block in southern Spain, this study simulates a representative sample for housing units 105 from the current social housing stock in 6 different scenarios. For the first scenario complying 106 with the current demand stipulations of Spanish CTE-DB-HE1 regulation [32] an alternative 107 scenario is introduced to provide a realistic assessment of the current energy use in the 108 dwellings, and the same model is subsequently used to evaluate energy variables for climate 109 conditions predicted for the year 2050. 110 This study is made up of the phases of: 111  characterisation of social housing 112  selection of case studies 113  monitoring 114 Energy and Buildings Volume 202, 1 November 2019, 109374  climate model development 115  simulations under six scenarios. 116 2.1. Characterisation of social housing 117 Given the lack of a database for social housing in Seville, this study required an exhaustive data 118 collection process to identify the developments built between 1939 and 1979. A thorough 119 review was carried out of original documents in national, regional and local historic and 120 government archives, as well as in the records of engineering firms. Documentary analysis was 121 followed by the on-site inspection of several buildings for the comparison of as-built and 122 planned or design data, identifying any changes made since their construction. 123 The information and documentation compiled was entered into a database with historic, 124 geometric and technical data in numerical and graph formats. This information was 125 supplemented with new drawings and the inspection of construction systems. The data 126 compiled and the specific characterisation is developed in [9] for performance and construction 127 characteristics of envelopes and in [33] where the cataloguing process and data set is discussed. 128 This database features information from 99,437 social dwellings built in the period under study. 129 2.2. Selection of case studies 130 The buildings studied were selected after a comprehensive process to faithfully represent the 131 building stock. The first phase consisted of the identification and characterisation of the social 132 housing developments built within the city limits in the period studied (1940-1979), 250 133 developments totalling 99,437 dwellings of an expected total of 100,510 were identified, and 134 those which were especially small or far too unique to have representation in the stock were 135 discarded [4]: in other words 98.93 %, which for all practical purposes can be considered as 136 the entire population. Exhaustive data collection was carried out, identifying affordable and social-137 type housing units in the city, including location, typological characterisation and the cataloguing of 138 units. National, regional and local archives from historical, governmental and technical organisations 139 were consulted in the data collection process, while information and documentation from press, 140 technical journals from the period studied and a scientific literature review were also analysed. Further 141 fieldwork was carried out to inspect buildings on site, collecting data for the comparison of construction 142 plans and existing buildings, allowing further evolution and transformations experienced since their 143 construction to be identified. This procedure has resulted in an extensive database of historical and 144 descriptive data: geometric, typological and constructive parameters and other technical aspects, both 145 numerical and graphical. Data set and catalogue details are included in [33] 146 Analysis led to the identification of the characteristics common to each time frame (walls, roofs 147 and other constructive element types, size of dwellings, windows and wall areas, etc.), with 148 decades selected as sub-periods to guarantee an improved practical approach. A sample group 149 of the stock (covering 83 developments and 46,476 units or 47 % of the population) was used 150 to establish the essential morpho-constructive features of these developments [9], represented 151 by the buildings selected (Tables 1 and 2, Figure 1).This group was selected for Stratified 152 sampling and used to analyse energy demand performance in the different present and future 153 climate scenarios. This group includes a representation (typical buildings) of each decade based 154 Energy and Buildings Volume 202, 1 November 2019, 109374 on the differences revealed between time periods in earlier analysis, and mostly resulting from 155 the implementation of housing construction programmes and ordinances [9]. Sampling 156 precision was improved through stratification with variable strata sizes [34]. The sample size of 157 each stratum was adapted depending on standard deviation to ensure the minimum variance in 158 the mean of the sample [35]. This resulted in a sample covering 14 developments (13,898 159 housing units in total) with a sampling fraction (fh) of 0.056. The buildings were defined using 160 non-probabilistic, directed selection, making up a modelling sample for exploratory research 161 design built using the ‘typical cases’ method [36]. The data sourced from an earlier study (matrix) 162 included quotas to ensure that all the usual types present in each sub-period were represented 163 and that clear in-depth information was provided on the performance of these characteristic 164 types. The development component was used as it was considered the minimum grouping for 165 multi-family buildings within the housing stock. When the number of units is used as a 166 parameter it does not allow suitable discrimination, given that it is a discrete variable (units are 167 grouped in buildings forming part of developments). Nonetheless, for the fit resulting from 168 applying an fh factor = 0.138, the approximation is suitable and compatible with the 169 developments selected and assignment error is therefore minimised. The major geometrical 170 envelope parameters are given in Table 1, while a comprehensive data analysis of the stock can 171 be found in [33]. 172 Table 1: Sample selected for energy modelling: main quantitative parameters 173 Model Building Development Year Decade ND SF Sw SR ND SF SW SR A 1952 50 20 878 171 273 1,180 51,802 10,061 16,107 B 1955 50 60 5,672 1,387 689 300 28,358 6,933 3,445 C 1959 50 8 248 44 89 1,611 49,975 8,848 17,922 D 1961 60 8 309 66 115 1,013 39,107 8,302 14,562 E 1963 60 20 1,009 157 324 554 27,940 4,347 8,975 F 1964 60 8 322 65 138 1,680 67,620 13,650 28,980 G 1964 60 30 2,071 252 358 300 20,712 2,520 3,580 H 1966 60 20 822 290 230 840 34,524 12,180 9,660 I 1971 70 24 1,520 242 470 2,768 178,164 28,365 55,090 J 1974 70 45 2,480 511 330 540 29,765 6,132 3,960 K 1976 70 40 5,302 866 710 640 84,832 13,856 11,354 L 1977 70 8 329 56 149 1,048 42,954 7,311 19,453 M 1979 70 32 2,603 304 359 800 65,075 7,600 8,975 N 1979 70 16 1,106 162 378 624 43,134 6,318 14,723 (dwellings) (sqm) (dwellings) (sqm) This analysis aims to establish distributions and patterns to define and classify the actions based 174 on: 175  Patterns and aggregation by time period of developments. 176  Aggregation by basic magnitude: construction type and general dimensions. 177 Energy and Buildings Volume 202, 1 November 2019, 109374  Morphological parameters related to energy performances of the building 178 envelope: wall or roof surfaces, wall to window ratio or wall to indoor area, 179 compactness and other parameters. 180  Construction systems and temporary distribution. 181 The energy and morphological parameters for the models are listed in Table 2. Developments 182 with the lowest % of openings on the façade (approximately 12 %) are models M and G. The 183 models with the highest percentage of openings on the façade are B and H (24% and 35 % 184 respectively). Models A, D and E are composed of single-brick façades. The housing units of 185 models M and N are representative of developments with thermal insulation in the envelope, 186 both on the roof and the façade. The 1970s saw an increase in the number of sloping roofs in 187 multi-family housing (models I, M and N). 188 Table 2. Energy and morphological parameters 189 Model Morphological parameters (m2) Energy parameters (W/m2K) Model Year ND SF SW SR UF UR UG A 1952 20 878 171 273 1.83 2.40 2.45 B 1955 60 5 672 1 387 689 1.53 1.23 2.25 C 1959 8 248 44 89 1.28 1.23 1.66 D 1961 8 309 66 115 1.83 1.23 2.07 E 1963 20 1 009 157 324 1.97 1.57 2.17 F 1964 8 322 65 138 1.28 1.23 1.81 G 1964 30 2 071 252 358 1.68 1.57 2.04 H 1966 20 822 290 230 1.53 1.23 1.91 I 1971 24 1 520 242 470 2.17 2.40 2.49 J 1974 45 2 480 511 330 1.53 1.84 2.20 K 1976 40 5 302 866 710 1.53 1.23 2.06 L 1977 8 329 56 149 1.53 1.54 1.97 M 1979 32 2 603 304 359 0.72 0.67 1.18 N 1979 16 1 106 162 378 0.75 0.67 1.22 190 Where: 191  SF Area of opaque façade (m2) 192  SW Window area (m2) 193  SR Roof area (m2) 194  ND No. of dwelling 195  UF Thermal transmittance, opaque façade enclosure (W/m2K) 196  UR Thermal transmittance, roof (W/m2K) 197  UG Thermal transmittance, building as a whole (W/m2K) 198 In Seville, the smaller housing units associated with social programmes for population with 199 limited means, have mostly been built in medium-height or tall buildings (models A, C, D, E, F, 200 H, I ,L,N). The housing units in taller tower blocks, approximately 27 % of the total (models B, G, 201 Energy and Buildings Volume 202, 1 November 2019, 109374 J, K), have larger surfaces than medium-height buildings. There is also a correlation between the 202 size of the housing unit and the block type, as housing units tend to be smaller in linear than in 203 H-type blocks [9]. 204 205 Energy and Buildings Volume 202, 1 November 2019, 109374 1940 - 1959 A B C 1960 - 1969 D E F G H 1970 - 1979 I J K L M N Figure 1: Building energy modelling by main time period (DesignBuilder). 206 Energy and Buildings Volume 202, 1 November 2019, 109374 2.3. Actual performance data gathering 207 One of the main aims of this study is to identify the difference between real use patterns and 208 those proposed by the National Standards [10] (Table 3) . The actual intensity of use and 209 conditions within the buildings tend to differ, leading to a distortion when evaluating demand. 210 This is reflected mostly in the variations in energy flow due to the different indoor temperatures. 211 In some cases, a monitoring process was carried out to identify the most frequent indoor 212 temperatures, and in turn, to establish new comparison scenarios. Indoor environmental 213 parameters (temperature, relative humidity and CO2 levels) were monitored continuously for a 214 full year in the selected housing units using a Wöhler CDL 210 multi-parametric monitoring 215 system (one control -housing unit in each development). Outdoor humidity, temperature, and 216 wind velocity were provided by the Spanish meteorological agency (AEMET). These 217 measurements were used to establish a pattern of use closer to that normally expected in this 218 type of housing unit rather than the standard patterns defined by national regulation for energy-219 demand compliance simulations. The starting conditions for the construction of the models are 220 based on the operational patterns established by the Spanish Standards for energy in buildings: 221 the National Energy Labelling procedure [37] and the national requirement for energy 222 conservation (CTE DB-HE) [38], regulations which implement the European Energy Performance 223 Building Directive (EPBD) at national level [39]. (Table 3): 224 Table 3. Heating/AC temperature set-point schedule as in Spanish national Standards [37] [38]. 225 Target temperature ( ˚C) 1:00 - 7:00 8:00 9:00 - 15:00 16:00 - 23:00 24:00 January to May (lowest) 17 20 20 20 17 June to September (highest) 27 free running free running 25 27 October to December (lowest) 17 20 20 20 17 *free running = mechanical thermal control off 226 This operational definition can be defined as the Normative Scenario. It should be noted that this 227 pattern assumes an almost continuous use of heating in winter, which is not the usual situation 228 in social housing stock [40], [41]. For the purposes of comparison assessment, an alternative and 229 complementary scenario was proposed as part of the discussion of the results from the 230 environmental variables and the analysis of user surveys monitored (Table 4), and is developed 231 in [10]. 232 This scenario is introduced to provide a more consistent model for the actual energy-use of 233 housing (especially in social housing), occupational profile, and heating and air conditioning 234 operation. The statistical development for the definition of the schedule and the analysis of the 235 indoor environmental data are covered in [10]. 236 237 Energy and Buildings Volume 202, 1 November 2019, 109374 The set temperature described in Table 2 was used. 393 b) Alternative 394 The set temperature described in Table 3 was used. 395 2.6.3. Climate 396 a) Present 397 This scenario, based on the present weather situation, was represented by the standard year 398 defined in the document Spanish Weather for Energy Calculations (EPW formatSWEC). 399 b) Future assuming climate change 400 A weather profile for the year 2050 was created to assess the behaviour of the building stock 401 exposed to climate change, based on HadCM3 and A2 emission conditions. 402 2.6.4. Combined scenarios 403 The comparison of different energy models is especially interesting, based on those derived 404 from the monitoring of protocols proposed in national standards, seeking alternatives 405 resembling this behaviour more closely and proposing alternatives for aspects where the biggest 406 differences have been detected (for example, use of heating). Occupant actions affecting 407 envelope performance were modelled following the National Energy Labelling procedure 408 [38]:The use of blinds and solar devices in summer was emulated in the models, considering that 409 in warm periods the aperture level of windows is reduced by 33% as a result of outer blinds - 410 the most frequent - [69], [70], in keeping with the findings of research on solar shading carried 411 out in the area [71].The impact of window aperture was standardised according to this 412 procedure, assuming that windows remain closed during winter, with very short and barely 413 noticeable operation, and in warm periods during the hours in the middle of the day, also with 414 a very short ventilation period. During late evening and night-time hours complete window 415 aperture is expected. This natural ventilation action, with a mean of 4 ACH [69], is within the 416 range identified for the area in the literature [72]–[74]. Although these values can vary greatly 417 and depend on the climate conditions at each point, this study aims to represent the common 418 values of the housing stock in order to ensure the suitable comparison of the complex based on 419 individual behaviour. 420 The modelling incorporates the effect of the presence of neighbouring buildings and its impact 421 on the solar horizon of the model for the different orientations and façades and roofs surfaces. 422 As these buildings are within an urban layout this aspect is crucial to ascertaining the real 423 performance of the building and its enclosures as well as its correlation to solar radiation. The 424 main effect occurs in winter (given the lower solar trajectory), where obstructions prevent solar 425 gains from entering through windows or being stored in walls, while in the summer greater 426 protection is provided to the roofs (in the case of lower buildings) and façades, especially those 427 with SW-NW and NE-SE orientation. This allows the model to closely simulate the real conditions 428 of use and to assess the different urban layouts. 429 Energy and Buildings Volume 202, 1 November 2019, 109374 3. Results and discussion 430 This section discusses and analyses the findings for the energy models of the individual scenarios 431 defined. The data were normalised for inter-model comparison, using the parameter of the 432 building’s environmentally controlled gross floor area (routinely applied in residential energy 433 labelling and standard compliance). The energy performance indicators (EnPIs) defined were 434 total thermal energy demand (TD) and cooling (CD)/heating demand (HD) per year. 435 3.1. Energy demand per model 436 The behaviour of the different models was studied through the comparison and evolution of the 437 different variables —heating, cooling and yearly total demand for individual building models— 438 in each of the scenarios introduced in order to analyse the degree of response and variation of 439 each of these models. Each building model was represented based on the average energy 440 demand of all the housing units within, comparing the mean value for each scenario (intra-441 scenario analysis) and between scenarios (inter-scenario analysis), as well as the use of lineal 442 and multiple regression analyses to identify the influence of parameters when needed. 443 Multivariate visualisation is used for pattern recognition. 444 Z-score was used to apply standardised demand result values in order to compare the 445 behaviours of each of the models included below each set of conditions (with very different 446 demand values in each case). , The difference between the result and the sample mean of the 447 set analysed was established and expressed in standard separation deviations (σ). Thanks to this 448 adjustment the models with the most extreme behaviours in each of the parameters analysed 449 (demands) are identified. In general, no models with extreme behaviours (outliers >3σ) were 450 identified and the different models are within the range of +/- 1.96σ in most situations. These 451 values have been represented with a multivariate Star Plot [75] (Figure 4), which is most useful 452 when the scales are comparable. Each ray represents individual study variables (in this case 453 upper vertex: annual demand; lower left vertex: cooling demand; lower right vertex: heating 454 demand). 455 In the original scenario (OA) (Figure 4a) annual energy forecasting performance varies by up to 456 1.5 times within the group. There is some relation between figures and age, with higher demand 457 in older buildings and lower demand in more modern ones (an R-square value of 62.3915% and 458 a correlation factor of 0.789883 show a moderately strong relationship between the variables 459 —p value: 0.0008—). Nonetheless, some of the older buildings have demand figures close to the 460 sample mean (D type), while some modern examples show higher results (model L). It should be 461 noted that the M and N building types date from the final period when insulation was introduced 462 into the construction and, despite the minimal energy-demand sample-minimum values, total 463 figures are very similar to those in non-insulated buildings of a similar age. The building with the 464 highest annual energy demand (A) had mass single-wythe construction, while the non-insulated 465 building with the lowest demand (J) had ceramic brick cavity-walls. However, this does not 466 appear to be a determining factor, since demand in the same sample buildings with ceramic 467 brick single-layer construction shows figures around central values. 468 Energy and Buildings Volume 202, 1 November 2019, 109374 In this initial scenario for the heating needs the correlation between the most demanding and 469 least demanding models is over double the energy (excluding insulated models), where model 470 E (building with single-wythe enclosures) displays the highest demand compared to L and F (both 471 also have single-wythe enclosures). Given that in this instance all three types had single-wythe 472 façades, the difference is due mostly to a combination of morphological and boundary factors 473 rather than to the specific construction system alone. In this case the different behaviours are 474 probably the result of the joint intervention of additional factors, rather than of the sole 475 influence of the constructive system. In this instance the most influential factors are the 476 aperture of the dwellings and solar obstruction, with the lowest demands found in the buildings 477 with a higher solar capture (orientation, aperture degree and no obstructions).. In heating 478 demand insulated buildings M and N display the highest difference in relation to the sample 479 group. Excluding models with thermal insulation which alters behaviour in some way, regression 480 model analysis establishes that the best explanation for the sample corresponds to the variables 481 linked to the envelope, especially global transmittance and the ratio of envelope per square 482 metre (with the lower Mallows Cp of 3.9692 and a r-square: 58.5182). However, variability is 483 very high as most variables show major correlations and most importantly, this is in keeping with 484 the high intensity of use and prolonged heating periods for this scenario. 485 Cooling behaviour in this scenario was almost a mirror image of heating performance, as the 486 models with the lowest heating demand exhibit the highest cooling demand. The same occurs 487 with variability, with the maximum value almost doubling the minimum. The best annual overall 488 performance was found for model I - with mean heating values and low cooling demand figures 489 for a fabric of single layer concrete-block walls and a rather high wall U-value (2.17 W/m2K)- 490 which appeared to strike the best balance. Figures for this building type are extremely low, with 491 a z-score of -2 σ. This performance can be associated with the presence of continuous balconies 492 across the entire façade, providing horizontal solar protection thus regulating solar capture in 493 winter and preventing it in summer, as a result of orientation and morphology rather than 494 specific wall solutions. Attention should be drawn to the relative high cooling demand in 495 buildings with originally insulated façades (models M and N) compared to non-insulated 496 buildings with a similar configuration (J and K). Insulation during warm periods has limited effect 497 when the morphology is not optimal. 498 In the present weather alternative low-energy intensity scenario (OB) (Figure 4c) types with 499 insulated façades (M and N) show greater differences compared to the sample as opposed to 500 scenario OA, with much lower yearly energy demands. Without this specific type of buildings, 501 demand differentials display similar relative values to those of the OA scenario (around one and 502 a half times higher), although absolute values are lower (50% less). Building performance 503 distribution reflects that from OA, albeit with some differences. In scenario OB the highest total 504 demands are again found in models A and E (close to model D), while models F to H and L 505 represent central values, and the minimum is for I to K types —excluding fabric-insulated M and 506 N. Parameters such as U global and U wall, connected with envelope thermal resistance, are less 507 noticeable in this scenario (linear regression models have no significance over annual energy-508 demand with p-values over 0.05 in both cases). The seasonal patterns show similar relative-509 profiles although heating requirements are around 30% lower and 40% for cooling. 510 Energy and Buildings Volume 202, 1 November 2019, 109374 From the above it is deduced that in an alternative energy-use laxer profile the energy demand 511 profile is moderate in buildings with dense, single-layer fabrics in which the effects of other 512 strategies such as solar control and thermal storage and buffering carry greater weight. The 513 buildings with cavity walls and lower envelope thermal mass were less sensitive to this change 514 of scenario (in relative values). In general there is less scattering of extreme values in 515 distribution. This suggests an important correlation with intermittent use - change in use 516 patterns - which is detrimental to buildings with lower thermal accumulation capacity and a 517 greater exposed surface, provoking the opposite effect in more compact cases with higher 518 thermal masses. 519 Following retrofitting (scenario RA) (Figure 4b), as expected, total demand declined significantly 520 with a roughly homogeneous façade thermal resistance for all types (thermal resistance 521 converges between 0.50 and 0.57 W/m2K due to the addition of insulation), with a substantial 522 reduction in the effect of heating and an increase in the relative weight of cooling in the annual 523 figures. Demand distribution also varied, with a change in the clustering identified in the two 524 preceding scenarios. Somewhat extreme figures were found in type C for annual demand (2.1σ) 525 and in L-Type for cooling (2.2σ). The lesser impact of façades heightened that of other 526 parameters such as the roof or WWR. Under an alternative indoor control pattern (RB) (Figure 527 4d) demand is lowered substantially, particularly for cooling, following much the same pattern 528 as that observed in scenarios OA and OB. As in scenario RA, the highest total and heating 529 demands were associated with model C, although the absolute values are 20% lower. The 530 combination of low (but not the lowest) fabric thermal resistance, low WWR and small housing 531 unit size makes this type of building highly sensitive to envelope losses. 532 Scenarios RA and RB (Figures 4b and 4d) show similar distribution of means for the models with 533 attenuated values related to OA and OB. However, as they are clearly differentiated sample sets, 534 specific distributions show major changes in the behaviour of the scenario (Kolmogorov-Smirnov 535 tests show a statistically significant difference between both distributions with a level of trust of 536 95.0% and a DN-value of 0.7857 and 0.7142 and p-value of 0.000352 and 0.0015 for OA-RA and 537 OB-RB respectively). The differences between maximum and minimum were mostly attenuated 538 in RB with respect to RA, as is the case in the original scenarios (OA and OB). 539 When climate change was assumed in the scenarios, heating accounted for far less of the total 540 demand (23% for the mean values) than cooling. 541 Comparing both climate situations, it is worth highlighting the significant reduction in heating 542 and increase in cooling for future forecast. There is also a reduction in efficiency of insulation 543 measures between the original envelope and the improved one in scenario FOB/FRB. The B 544 scenario was selected for its closer representation of the operation of actual buildings, and 545 consequently greater capacity for evaluating the potential for energy change among the 546 different scenarios. 547 In the future scenario with buildings in original conditions the effect of heating is far less 548 noticeable in the overall requirements (23% of total mean values), whereas behaviour in cooling 549 conditions is far more significant. The lowest total demand is observed in the insulated buildings 550 (models M and N), albeit with very similar values to those of non-insulated buildings I and J, 551 Energy and Buildings Volume 202, 1 November 2019, 109374 which show the lowest demand in the group without insulation. Models M and N present 552 medium cooling demand, with low total values due to the reduced heating demand resulting 553 from the façade insulation. However, this does not appear to be particularly effective for heating 554 demand control. In contrast, the demand values for the most balanced non-insulated models 555 within the sample (I and J) are slightly higher (close to the mean), in keeping with an envelope 556 without insulation. Nevertheless, their morphology benefits cooling control. Model E displays 557 the highest heating demand, with behaviour in keeping with prior analyses. However, models A, 558 C and D also present high heating demand values, which can be linked to a lower thermal 559 resistance of these enclosures compared to the rest of the group. Although heating demand 560 values are relatively low compared to the current scenarios (with a 35% reduction between 561 scenarios), maximum and minimum values vary greatly, exceeding double the value without 562 taking into account freestanding buildings. There is a high incidence of the parameter associated 563 to the thermal resistance of façades (U-value) in the distribution of heating demand values, 564 albeit with great variations(linear correlation r-square: 46.2407% indicates a moderately strong 565 relationship between variables with a standard deviation of the residuals of 2.4105 with p-value 566 of 0.0075). Variability is reduced in the case of cooling, with an approximate minimum-maximum 567 ratio of 1.6, despite the much higher absolute values and the significant increase when 568 compared to the current situation, doubling the mean cooling demand of scenario OA, although 569 in this case there are no predominant parameters in the distribution process and B-type and I-570 Type are found in extreme positions, with z-scores over +/-2 (2.1; -2.1) 571 For the situation of future climate (2050) and improved envelope, - and ahead of M and N, which 572 could be considered to have excessive insulation - models I and J display the lowest total energy 573 demand and jointly the lowest cooling demand. Heating demand is also reduced in both cases. 574 Models C and F display the highest total energy demand, with the highest cooling demand also 575 observed in model F. The features noted above are confirmed as the poor behaviour of model 576 C cannot be linked to the thermal resistance of its enclosures —wall R-value— (in this case with 577 equal values throughout the sample) or Global transmittance value, with no actual correlations 578 as R-square only reaches 12.2233% for FOB and a very low 0.8720 % for FRB both with p-value 579 over 0.005 (0.22 and 0.7508). This model also shows the highest heating demand, followed by 580 model E. Although the maximum overall value of model F is also due to its high cooling demand, 581 it presents one of the lowest heating demands in the group. This can be attributed to high solar 582 radiation capture throughout the year (high ratio of openings to surface area and effect of the 583 roof), which allows control of the need for heating but is especially problematic in the warm 584 period, despite the presence of insulation. 585 When observing the symmetry and scope of the behaviour in relation to the group mean (Figure 586 4) in general the models displaying the most balanced behaviour and lowest demands are J and 587 N, whereas models C, F and L show the least balance and the highest demands. 588 Energy and Buildings Volume 202, 1 November 2019, 109374 589 Figure 4: Mean total, heating and cooling demands for each model in the sample and star plot 590 multivariate visualisation of model energy demands under six scenarios (bar graph: 591 blue=cooling demand; red=heating demand; green=total demand; Chambers graph: top 592 vertex=yearly demand; lower left vertex=cooling demand; lower right vertex=heating demand) 593 Energy and Buildings Volume 202, 1 November 2019, 109374 594 3.2. General energy demands for the sample studied: inter-595 scenario comparison 596 Partial and total energy demands are shown in Figure 5, comparing the results for the six 597 scenarios: present, retrofitted and for the year 2050 assuming climate change, each with and 598 without retrofitting. The average values for all the scenarios and the differentials between the 599 non-retrofitted and retrofitted versions in each group (expressed in relative and absolute values) 600 are listed in Figure 6 for heating and in Figure 7 for cooling demand. 601 602 Figure 5: Mean HVAC energy demand (kWh/m2) in six scenarios. 603 A B FB A B FB O 15.9 -21.38% 12.5 -34.40% 8.2 O 15.9 -3.4 12.5 -4.3 8.2 -45.91% -51.20% -54.88% -7.3 -6.4 -4.5 R 8.6 -29.07% 6.1 -39.34% 3.7 R 8.6 -2.5 6.1 -2.4 3.7 Figure 6 : Comparison of average heating demand (kWh/m 2 ) for all models by scenario (left: relative variation (%) with and without retrofitting; right: absolute variation (kWh/m2) with and without retrofitting) Energy and Buildings Volume 202, 1 November 2019, 109374 A B FB A B FB O 21.6 -36.57% 13.7 101.46% 27.6 O 21.6 -7.9 13.7 13.9 27.6 -18.52% -16.79% -14.13% -4 -2.3 -3.9 R 17.6 -35.23% 11.4 107.89% 23.7 R 17.6 -6.2 11.4 12.3 23.7 Figure 7: Comparison of average cooling demand (kWh/m2) for all models by scenario (left: relative variation (%) with and without retrofitting; right: absolute variation in kWh/m2) with and without retrofitting). Where:  A Spanish standard set temperature  B alternative set temperature  O original  R retrofitted  F Future climate.  HD Heating demand (kWh/m2)  CD Cooling demand (kWh/m2)  TD Total demand (kWh/m2) A comparison of the type O (original) scenarios showed the significant effect of the intensity of 604 use of HVAC systems and the adoption of different set-points. The implementation of an 605 alternative schedule, closer to common practice, and the strict application of the analysis to the 606 areas of housing units that are currently conditioned (OB) lowered yearly demand to 607 approximately 70 % of the initial value (OA). 608 The distribution of seasonal demand also varied. In scenario B heating and cooling tended to be 609 more balanced (heating: 47.7% / cooling: 52.3%), whereas cooling carried greater weight in 610 scenario A (heating: 42.4% / cooling: 57.6%). A seasonal analysis showed that modifying the 611 indoor conditions greatly affected summer values, as the heating demand from OA to OB 612 decreased by 21.4%, compared to a 36.6% reduction in cooling. 613 Type R scenarios reflected the effect of improving the vertical opaque envelope (façade 614 enclosures) through energy retrofitting. Improving thermal insulation lowered yearly demand in 615 both cases (RA, RB), although the reduction was more noticeable in the lower intensity scenario 616 than in the higher one: RB=33.2%; RA=29.9%. Lower demand was observed primarily in winter, 617 with a greater decrease in heating (RA: 45.9%; RB: 51.2%) than in cooling demand (RA: 18.5%; 618 RB: 16.8%) as a result of façade insulation. As stated, façade improvements greatly impacted 619 heating, reducing the energy needs to around half the initial requirement. Although demand for 620 cooling was also reduced, this decrease was less than one-fifth of the initial value. 621 The climate change (CC) scenarios assumed an intensity use of type B or lower, considered to 622 best represent the predominant conditions in this housing stock. The assessment for the year 623 2050 indicated a significant change in energy performance, even under moderate use. In the 624 future scenario, climate change with higher mean temperatures and longer summers made the 625 winters much less severe, thereby raising the weight of summer time demand in the total. The 626 Energy and Buildings Volume 202, 1 November 2019, 109374 net result was a considerable reduction in heating demand, even with the original envelopes, 627 with an equally considerable rise in cooling demand, even under less strict summertime 628 temperature targets (compared to current standards for premises with mechanical HVAC 629 systems). 630 In the future scenario, façade improvements (FRB) improved the control of yearly demand, 631 reduced by 23.3%, attenuating the effects of CC and delivering yearly values similar to those 632 recorded for the original situation (OB). However, this effect was not balanced as the relative 633 weights of seasonal performance were highly impacted and both scenarios (OB and FRB) were 634 rendered unsuitable for comparison. In FRB, cooling (86%) clearly prevailed over heating 635 (13.4%). 636 In this scenario, heating demand was marginal, dipping to values below half those of the 637 scenario without façade improvements (FOB) and to less than one-third of the current values 638 assuming low intensity use (OB). These observations reinforce the idea that thermal resistance 639 of the enclosure is a primary factor in preventing energy loss. In contrast, the difference in 640 present and future cooling demand assuming CC is under 15%. 641 While the relative values would appear to indicate much more significant reductions in heating 642 than in cooling demand, the absolute values revealed a more balanced situation. Insulating the 643 façade lowered heating demand by 54.9% and cooling demand by just 14.1%, whereas the 644 overall reduction was actually 4.5 kWh/m2 for heating and 3.9 kWh/m2 for cooling. This can be 645 explained by the relatively low heating and high cooling demand in this scenario. The 646 differentials in absolute terms given in Table 7 show that façade insulation had a greater effect 647 on heating; the greater the indoor-outdoor temperature difference, the higher the impact. The 648 same pattern was observed in connection with the overall reduction in cooling demand between 649 FOB and FRB, compared to the more moderate findings for OB and RB. 650 3.3. Application to general stock models 651 After analysing the variability and dispersion of energy demand for each regime under the 652 different study scenarios (Annex 1) together with the density traces, behavioural models can be 653 established to represent the population, providing an image of possible evolution under the 654 different scenarios of the set of residential buildings. 655 Therefore, the proposed models should be applied to the information of the entire housing 656 stock, returning to previous analyses for the study of specific cases, or to behaviour groups to 657 avoid possible deviation of the data when modifying the scale. 658 Distribution models were selected for the best fit. A Kolmogorov-Smirnov non-parametric test 659 was applied to verify the fit to the proposed distributions, with the best fit within 70.11% of the 660 population (95% significance) and non-rejection of the null-hypothesis (K-S p-value>0.05). The 661 results and complementary tests are shown in Annex 2. 662 Analysis of the data obtained allows representative probabilistic distributions to be incorporated 663 in order to forecast a performance model to be exported to the general case set, providing a 664 Energy and Buildings Volume 202, 1 November 2019, 109374 general prediction model based on probability (assuming the approximation). Normal 665 distribution or related types of distribution (i.e. inverse Gaussian) were selected to ensure better 666 applicability and commonality. The prediction models of the annual energy demand for thermal 667 conditioning of the OB, RA, R and FOB scenarios can be adjusted to a normal probabilistic 668 distribution. At the same time, the OA and FRB scenarios are better represented by an inverse 669 Gaussian distribution (Figure 8). Table 7 shows the defining statistical parameters of the 670 different distributions. 671 672 Figure 8: Probability distribution for overall energy demand per unit of gross floor area (kWh/m2) 673 in six scenarios. OA: Red; OB: Blue; RA: dot-Orange; RB: dot-Cyan; FOB: Green; FRB: dot-green 674 Table 7. Statistical parameters for the distributions in Figure 8. 675 Scenario Mode Scale Tolerance intervals Upper limit (kWh/m2) Lower limit (kWh/m2) OA 34.67 4.92 49.01 29.00 FRB 27.48 41.2 24.88 12.52 27.4643 41.0988 34.8 21.2 Scenario Mean Standard deviation Tolerance intervals Upper limit (kWh/m2) Lower limit (kWh/m2) OB 26.22 4.7 47.8693 4.6020 RA 26.25 4.93 49.01 3.42 RB 17.48 3.57 33.92 1.03 FOB 35.88 4.42 33.17 18.98 676 The results obtained are in keeping with other research, which establishes that total energy 677 needs in mild and warm zones will increase despite the significant reduction in the influence of 678 the heating. This is the case of the USA [30], mild Australian climate zones [76], and especially 679 Energy and Buildings Volume 202, 1 November 2019, 109374 [65] L. Pérez-Lombarda, J. Ortizb, R. González, and I. Maestre, “A review of benchmarking, 914 rating and labelling concepts within the framework of building energy certification 915 schemes,” Energy Build., vol. 41, no. 3, pp. 272–278. 916 [66] European Organisation for Technical Approvals, External Thermal Insulation Composite 917 Systems ( Etics ) With Rendering, no. March 2000. 2013. 918 [67] F. Kurtz, M. Monzón, and B. López-Mesa, “Obsolescencia de la envolvente térmica y 919 acústica de la vivienda social de la postguerra española en áreas urbanas vulnerables. El 920 caso de Zaragoza,” Inf. la Construcción, 2015. 921 [68] T. H. (ed.), “Sustainable refurbishment of exterior walls and building facades Final report, 922 Part A – Methods and recommendationso Title,” Espoo (Finland), 2012. 923 [69] IDAE, Condiciones de aceptación de Procedimientos alternativos a LIDER y CALENER. 924 Madrid: Instituto para la Diversificación y Ahorro de la Energía. 2009. 925 [70] E. y T. Ministerio de Industria, Ministerio, and D. Fomento, “PROCEDIMIENTO PARA EL 926 RECONOCIMIENTO CONJUNTO POR LOS MINISTERIOS DE INDUSTRIA, ENERGÍA Y 927 TURISMO Y DE FOMENTO DE LOS DOCUMENTOS RECONOCIDOS DE CERTIFICACIÓN 928 ENERGÉTICA DE EDIFICIOS,” 2015. 929 [71] A. L. León, S. Domínguez, M. A. Campano, and C. Ramírez-Balas, “Reducing the energy 930 demand of multi-dwelling units in a mediterranean climate using solar protection 931 elements,” Energies, vol. 5, no. 9, 2012. 932 [72] E. Spentzou, M. J. Cook, and S. Emmitt, “Modelling natural ventilation for summer 933 thermal comfort in Mediterranean dwellings,” International Journal of Ventilation, 2017. 934 [73] K. Imessad, L. Derradji, N. A. Messaoudene, F. Mokhtari, A. Chenak, and R. Kharchi, 935 “Impact of passive cooling techniques on energy demand for residential buildings in a 936 Mediterranean climate,” Renew. Energy, vol. 71, pp. 589–597, Nov. 2014. 937 [74] G. A. Faggianelli, A. Brun, E. Wurtz, and M. Muselli, “Natural cross ventilation in buildings 938 on Mediterranean coastal zones,” Energy Build., 2014. 939 [75] J. M. Chambers, W. S. Cleveland, B. Kleiner, and P. A. Tukey, Graphical Methods for Data 940 Analysis. Chapman and Hall/CRC, 1983. 941 [76] X. Wang, D. Chen, and Z. Ren, “Assessment of climate change impact on residential 942 building heating and cooling energy requirement in Australia,” Build. Environ., vol. 45, 943 no. 7, pp. 1663–1682, Jul. 2010. 944 [77] M. A. Triana, R. Lamberts, and P. Sassi, “Should we consider climate change for Brazilian 945 social housing? Assessment of energy efficiency adaptation measures,” Energy Build., 946 2018. 947 [78] R. Barbosa, R. Vicente, and R. Santos, “Climate change and thermal comfort in Southern 948 Europe housing: A case study from Lisbon,” Build. Environ., 2015. 949 [79] C. Cartalis, “Climatic change in the built environment in temperate climates with 950 emphasis on the Mediterranean area,” in Energy Performance of Buildings: Energy 951 Efficiency and Built Environment in Temperate Climates, 2015. 952 953 Energy and Buildings Volume 202, 1 November 2019, 109374 954 Energy and Buildings Volume 202, 1 November 2019, 109374 Annex 1 955 Table A1.1. Statistical summary of heating energy demands in six scenarios 956 Scenario Mean Standard deviation CV Minimum Maximum Range Standard bias Standardised kurtosis OA 15.91 6.16 38.7% 6.72 25.45 18.73 0.097 -0.849 OB 12.53 4.44 35.4% 5.2 19.7 14.5 -0.081 -0.722 RA 8.62 4.33 50.2% 3.62 15.67 12.05 0.717 -1.014 RB 6.08 3.65 60.1% 2.08 12.45 10.37 0.902 -0.811 FOB 8.24 3.15 38.3% 3.06 13.18 10.12 -0.006 -0.760 FRB 3.74 2.13 57.0% 1.61 7.86 6.25 1.356 -0.436 Total 9.19 5.71 62.2% 1.61 25.45 23.84 3.011 0.278 957 Table A1.2: Statistical summary of cooling energy demands in six scenarios 958 Scenario Mean Standard deviation CV Minimum Maximum Range Standard bias Standardised kurtosis OA 21.60 4.29 19.8% 13.17 29.22 16.05 -0.575 -0.120 OB 13.68 2.43 17.7% 8.53 17.76 9.23 -0.766 0.257 RA 17.63 4.03 22.8% 11.15 26.11 14.96 0.448 0.155 RB 11.39 2.81 24.7% 6.91 16.06 9.15 0.015 -0.182 FOB 27.64 3.27 11.8% 20.98 34.68 13.70 0.256 0.938 FRB 23.74 4.26 17.9% 17.77 32.60 14.83 1.204 0.147 Total 19.28 6.66 34.5% 6.91 34.68 27.77 0.667 -1.617 959 960 Table A1.3: Statistical summary of total energy demands in six scenarios 961 Scenario Mean Standard deviation CV Minimum Maximum Range Standard bias Standardised kurtosis OA 37.52 6.27 16.7% 29.02 49.06 20.04 0.680 -0.701 OB 26.22 4.70 17.9% 18.98 33.17 14.19 -0.034 -0.823 RA 26.26 4.96 18.9% 19.67 37.32 17.65 0.962 0.232 RB 17.47 3.56 20.4% 12.52 24.88 12.36 0.686 -0.242 FOB 35.88 4.42 12.3% 29.66 44.37 14.71 0.078 -0.376 FRB 27.48 4.41 16.0% 21.23 34.84 13.61 0.396 -0.679 Total 28.47 8.18 28.7% 12.52 49.06 36.54 0.752 -0.753 962 963 Energy and Buildings Volume 202, 1 November 2019, 109374 Annex 2 964 The distribution characteristics for each case were defined based on their parameters. The Test 965 battery panel performs different approaches designed to determine if the data could reasonably 966 come from the selected distribution or not (most on the case of normality). For each test the 967 hypotheses are: 968 • Null hypothesis: the data are independent samples of a normal distribution 969 • Hypothesis Alt .: the data are not independent samples of a normal distribution 970 Since the smallest P-value of all the tests performed is greater than or equal to 0.05, the selected 971 distribution cannot be rejected with 95% confidence. 972 The tolerance interval for each distribution was provided, with 95% confidence, and the 973 certainty that at least 70.11% of the population is included (Table 7). 974 Table A2.1. Statistical summary of models in six scenarios 975 Distribution Mean STD Scale OA Inverse Gaussian 37.528 39.1797 OB Normal 26.2357 4.700 RA Normal 26.2571 4.96072 RB Normal 17.4786 3.5732 FOB Normal 35.8929 4.43577 FRB Inverse Gaussian 27.4643 41.0988 Chi-square Chi-square G.1. p-value >0.05 (95%) OA 2.3710 2 0.3055 OB 0.77965 2 0.6775 RA 0.8450 1 0.3579 RB 1.5786 1 0.2089 FOB 2.49434 1 0.1142 FRB 2.23574 2 0.3269 KolmogorovSmirnov D+ DDN p-value >0.05 (95%) OA 0.18366 0.09111 0.18366 0.7324 OB 0.104798 0.135352 0.135352 0.9596 RA 0.101252 0.0931152 0.101252 0.9987 RB 0.16268 0.1231 0.16268 0.8525 FOB 0.16485 0.16148 0.16485 0.8412 FRB 0.08639 0.10684 0.10684 0.9972 Anderson-Darling A^2 Mod. form p-value >0.05 (95%) OA 0.338383 0.338383 >=0.10 OB 0.264951 0.282186 0.637368 RA 0.2188 0.23312 0.7978 RB 0.3314 0.3529 0.4655 FOB 0.452514 0.4819 0.2310 FRB 0.20117 0.20117 >=0.10 976