Making energy simulation easier for future climate – Synthesizing typical and extreme weather data sets out of regional climate models (RCMs)

Applied Energy 177 (2016) 204–226

Contents lists available at ScienceDirect

Applied Energy journal homepage: www.elsevier.com/locate/apenergy

Making energy simulation easier for future climate – Synthesizing typical and extreme weather data sets out of regional climate models (RCMs) Vahid M. Nik Division of Building Physics, Department of Building and Environmental Technology, Lund University, Lund, Sweden

h i g h l i g h t s

g r a p h i c a l a b s t r a c t


A method for the impact assessment

of climate change on the energy performance of buildings.
Synthesizing one typical and two extreme weather data sets out of regional climate models.
Evaluating the method for eight climate scenarios and two energy modeling tools.
Simulating one office building in Geneva and the building stock in Stockholm.
Number of simulations decreases extensively while results are accurate enough.

a r t i c l e

i n f o

Article history: Received 12 March 2016 Received in revised form 9 May 2016 Accepted 18 May 2016 Available online 24 May 2016 Keywords: Climate change Weather data Energy simulation Regional climate models Big data Building

a b s t r a c t Higher availability of future climate data sets, generated by regional climate models (RCMs) with fine temporal and spatial resolutions, improves and facilitates the impact assessment of climate change. Due to significant uncertainties in climate modeling, several climate scenarios should be considered in the impact assessment. This increases the number of simulations and size of data sets, complicating the assessment and decision making. This article suggests an easy-to-use method to decrease the number of simulations for the impact assessment of climate change in energy and building studies. The method is based on synthesizing three sets of weather data out of one or more RCMs: one typical and two extremes. The method aims at decreasing the number of weather data sets without losing the quality and details of the original future climate scenarios. The application of the method is assessed for an office building in Geneva and the residential building stock in Stockholm. Results show that using the synthesized data sets provides an accurate estimation of future conditions. Variations and uncertainties of future climate are represented by the synthesized data. In the case of synthesizing weather data using multiple climate scenarios, the number of simulations and the size of data sets are decreased enormously. Combining the typical and extreme data sets enables to have better probability distributions of future conditions, very similar to the original RCM data. Ó 2016 Elsevier Ltd. All rights reserved.

E-mail addresses: [email protected], [email protected] http://dx.doi.org/10.1016/j.apenergy.2016.05.107 0306-2619/Ó 2016 Elsevier Ltd. All rights reserved.

V.M. Nik / Applied Energy 177 (2016) 204–226

1. Introduction By realizing the importance of climate change adaptation in different aspects of human life and following the advances in computing future climatic conditions, which has resulted in higher availability of future climate data sets, assessing the probable impacts of climate change has turned to be an interesting research topic for different fields of science such as energy and buildings. Planning for climate change adaptation is complicated since it is difficult to predict the expected degree of warming and pace [1]. Impact assessment of climate change is usually performed by means of the climate data generated by global climate models (GCMs) – also known as the general circulation models. GCMs contain atmospheric model, ocean model, land surface scheme and the sea ice model and simulate climatic conditions under different initial and boundary conditions such as emissions scenarios. GCMs simulate future climatic conditions for the spatial resolution of 100–300 km2 [2] which cannot be considered as weather and is coarse for the purpose of impact assessment. The impacts of a changing climate and the consequent adaptation strategies to deal with occur on the regional and national scales, where regional climate downscaling (RCD) provides projections with much greater detail and more accurate representation of localized extreme events [3]. More than having coarse spatial resolutions, direct use of the GCM output in impact assessment is not recommended due to recognized biases [4,5]. Therefore the GCM data should be downscaled by means of statistical or dynamic downscaling techniques. One well-known statistical technique is morphing [6] which combines present-day observed weather data with the GCM results. Climate projections show changes in both average conditions and variability, including changes in the frequency and magnitude of extreme events. Morphing technique however reflects only changes in the average weather conditions and neglects changes in future weather sequences. For example it is not possible to see changes in extreme climatic conditions for the morphed data, though extremes will be more often and stronger in the future [7]. Future regional and distributional shifts are diverse and dependent to the time period (e.g. the considered season), regions and considered phenomena [8,9]. Dynamic downscaling of GCMs by means of regional climate models (RCMs) has the advantage of generating physically consistent data sets across different variables [10,11]. RCMs provide weather data with suitable temporal (down to 15 min) and spatial resolutions (down to 2.5 km2) for direct use in building and energy simulations (e.g. [12]). The morphed weather data are likely to underestimate the impacts of climate change relative to RCM, though having similar emissions scenario and timeframe [13]. RCMs generate a state of the atmosphere for each time step as well as in long integrations over a century. The consequent simulated state is exactly what can be termed ‘‘weather” at each time step. However, it is not possible to rely on short time spans when dealing with future climate scenarios and the recommendation is considering time spans of 20–30 years. Moreover, there are different uncertainties which affect the simulated climate data, such as the selected GCM, RCM, emissions scenario and the spatial resolution [9]. Hence a valid impact assessment should consider several scenarios and it is not possible to rely on few climate scenarios [14–16]. This creates the challenge of dealing with large data sets and uncertainties, which has been discussed thoroughly in some previous works of the author (e.g. [9,17]). Estimating the probable future conditions is important for energy and infrastructure planning and several researchers have investigated the effects of climate change on energy use and generations. A large amount of work exists on assessing the impacts

205

of climate change on buildings as the main energy users in cities and urban areas, focusing on different categories of buildings such as office, residential, hospital and supermarket (e.g. [18–24]), thermal comfort and Indoor environmental quality (e.g. [25–30]), building characteristics as well as construction and retrofitting strategies (e.g. [31–35]) and assessing the energy saving potentials and the performance of building components ([20,33,36–38]). Uncertainties due to climate change have been considered in several works (e.g. [17,33,39–43]), some combined with other important uncertainty factors. For example Zhou et al. [44] studied uncertainties due to climate change, policy, population and GDP growth in modeling the effects of climate change on the U.S. state-level building energy demands. They showed that effects of climate change (and the associated uncertainties) are not limited to the energy performance of buildings and are extendable to energy systems and infrastructure. Impacts of climate change on the energy network/system have been investigated by some other researchers as well; Mirasgedis et al. [43] assessed the sensitivity of electricity demand in the Greek interconnected power system to climate change and socio-economy, using a multiple regression model. They used RCM climate data for the monthly time step and investigated the consequences of the extreme conditions in that scale for 30-year periods. According to their work, what is more troubling by climate change is the increase in seasonal variability, both in percentage and in absolute terms, which may reach to 34% of average annual values. Ahmed et al. [45] used a regression based model to analyze climate induced future electricity demand in New South Wales of Australia, estimating that per capita demand (associated only to climate change) in summer and spring for year 2100 may rise by 6.14% and 11.3% respectively. More than changing future energy demand, climate change can alter the energy market in other ways, for example by affecting energy generation. Most of the studies about effects of climate change on the energy generation are focused on renewables, especially wind, hydropower and solar energy. There are few studies on future changes in solar resource, mostly affected by uncertainties in cloud cover estimations by GCMs, with not that much agreement between historical observations and GCM output [8]. Pryor et al. [46] studied changes in near surface wind speeds due to global climate change by downscaling GCMs and generating probability distributions of wind speeds at sites in northern Europe for historical periods (1961–1990 and 1982–2000) and two future periods (2046–2065 and 2081–2100). They did not find any evidence of substantial evolution in projections for the 21st century relative to the end of the 20th century, suggesting that mean and 90th percentile wind speeds will decrease slightly by 2100. They recognized uncertainties due to downscaling different GCMs and indicated the increased divergence of results toward the end of 21st century. Seljom et al. [47] identified the effects of climate change on the Norwegian energy system toward 2050 using five different global models and six emission scenarios. They found that effects of climate change on the wind power potential are very limited. Fant et al. [8] investigated impacts of climate change on wind and solar resources in southern Africa, considering climate uncertainties. They indicated that GCMs impart substantial uncertainties in resolving wind and solar variables, although changes in wind and solar potential by 2050 are expected to be small. Sailor et al. [48] considered uncertainties in climate data for estimating of wind turbine energy generation by statistically downscaling four GCMs, forced by two different emissions scenarios from the Special Report on Emissions Scenarios (SRES A1B and A2) [49]. Kao et al. [50] studied impacts of climate change on annual hydropower generation in the United States, developing future climate scenarios using a GCM, a nested regional climate model and a hydrological model. In their results, the model uncertainty range was ±9 TW h for the hydropower energy generation per year.

206

V.M. Nik / Applied Energy 177 (2016) 204–226

Despite of the focus and subject in the impact assessment of climate change, e.g. buildings, energy systems or renewables, a critical part of the assessment is always the weather data sets which are used in the assessment. Synthesizing weather data sets for energy simulations has a long history and several techniques have been developed, which some have been inspiring for creating typical future weather data sets (e.g. [6,51]). The advantage of using typical/representative weather years is reducing the computational and data handling efforts by using one year, representing all the years in a long period (usually 30 years). Besides, a consistent form of weather data is ensured so that results from different studies can be compared [52]. Several techniques are available to create typical or reference weather files for energy simulations which Chan et al. [53] have provided a review of some of the most important ones. Creating typical meteorological year (TMY) was introduced by Hall et al. [54], which is based on selecting typical meteorological month (TMM) for each month and concatenating them to create the weather file for one year. American Society of Heating, Refrigerating and Air-Conditioning Engineers (ASHRAE) has created different typical or reference weather data sets, such as typical reference year (TRY), Weather Year for Energy Calculations (WYEC) and International Weather Year for Energy Calculation (IWEC) [55–58]. National Solar Radiation Data Base (NSRDB) released their own TMY which is known as TMY2 [59]. Though typical or representative meteorological years facilitate energy calculations, they can have some drawbacks. For example Hong et al. [60] studied multi-decade simulations with Actual Meteorological Year (AMY) and compared with TMY3 to assess the impact of weather on the long-term performance of buildings. They showed that energy savings and peak demand reduction can be significantly underestimated or overestimated in the case of using TMY3. Kershaw et al. [61] found that TRY in general does produce representative average data for a building in a given location. They also concluded that the design summer year (DSY) consistently underestimates the levels of human discomfort as well as overheating risk. Most of the efforts for creating typical future climate files are based on extending the available approaches and applying them on the statistically downscaled GCM data. In this way, those variations and anomalies which induce more extreme conditions in future will be neglected. One approach for creating weather files is integrating the monthly mean changes from GCMs into the existing typical meteorological year by mean of morphing technique (e.g. [62]). Yang et al. [52] developed typical principal component years (TPCYs) and extended the technique to develop twelve typical principal component months (TPCMs), using only monthly data out of GCMs for five cities in China during the 21st century. They notice the difficulties of working with future climate and mention that developing a typical weather year for future climate from a multi-year database is time-consuming or even not practicable due to the need for downloading many years of hourly/daily predictions from GCMs as well as the need for statistical analysis of the long-term frequency distributions of the meteorological variables. Huang and Hwang [63] used morphing technique to produce future hourly weather years for building simulations based on the predicted values provided by a GCM and created TMY data sets to assess human comfort. Their results have the limitations of using morphed data and only one GCM. Jentsch et al. [13,64] developed a tool to generate future weather series under the A2 emissions scenario with TMY2 and EPW formats. As they mention, one of the major limitations of the synthesized weather data is considering only one emissions scenario. Zhu et al. [65] proposed a time series forecasting method based on climate periodicity analysis and applied that to future monthly temperature prediction of Shanghai (adopting morphing technique). They established a Dual-Periodic Time Series Model (TSM) which has shown more

accurate results for the recent two years than their considered GCM under RCP4.5. Arima et al. [66] constructed a prototype of the near-future design weather data of the 2030 s. They used RCM data and selected a representative weather data for the average conditions during 10-year periods among the results of downscaled weather data. Afterwards the representative data were corrected with observations to reduce bias of RCM and GCM. Wang and Chen [22] compared the projected TMY3 data using HadCM3 global model with the actual TMY3 data and applied morphing technique to create hourly weather data. Jylhä et al. [67] performed the impact assessment of climate change by creating hourly weather data sets, using test reference years (TRYs) and considering multiple climate scenarios. In several works for assessing impact of climate change, combined with socio-economic factors, cooling and heating degree days (CCD and HDD) were calculated and applied in regression analysis (e.g. [43,45]). Christenson et al. [68] developed heating degree days (HDD) and cooling degree days (CDD) out of monthly temperature data using several climate scenarios for different locations. They neglected the associated loss in accuracy due to using monthly temperature data. They were interested in assessing long term changes and trends than in precisely predicting individual monthly HDD or CDD values. The usual assumption in using degree days is that the building energy demand has a significant linear correlation with degree days and therefore suits mostly for buildings with a relatively constant internal temperature, thermal gains and building properties [65]. A method was developed for using UK Climate Projections (UKCP09) in energy simulations under PROMETHEUS project by Eames et al. [69], which is based on creating future probabilistic reference years. Through generating 100 samples of 30 years on a daily time series and then using a disaggregation procedure, they produce an hourly time series for a given decade, location and emission scenario. Although each set of stochastically produced 30 years includes natural variability, climate change signal is stationary within it. The generated weather data sets have been used in several building simulations [70] and compared against the morphed data [71]. Kershaw et al. [72] showed the application of the probabilistic reference years in risk assessment for buildings and occupants. They modeled the internal conditions and energy use of a building with all 3000 example years produced by the UKCP09 weather generator. By comparing results out of the probabilistic reference years [69], they showed that almost the full range of values predicted by the 3000 files can be estimated using the probabilistic reference years. Most of the above mentioned approaches are based on statistical downscaling of GCMs and in the case of dynamic downscaling, based on the daily or monthly temporal resolution of RCMs. Many of these methods neglect the probable climatic variations in different time scales, e.g. from seasonal to hourly. In this way, it is not possible to estimate the probable extreme conditions. Kershaw et al. [72] pointed to the fact that not considering all the possible future scenarios results in losing critical information about the potential range of the response of buildings and of the risk occupants might be subject to. By higher availability of the hourly data out of RCMs, there will be more impact assessment using the hourly weather data extracted directly from climate models. As it was mentioned, because of the existence of several climate models/uncertainties and long-time data sets, an authentic impact assessment will be a laborious task due to generating huge data sets. When it comes to buildings and energy systems, the number of possible scenarios and uncertainties to be considered grow enormously which make the impact assessment more difficult. This article suggests an approach for the impact assessment of climate change on buildings and their energy performance, based on creating three sets of weather data out of RCMs: (1) typical downscaled year (TDY), (2) extreme cold year (ECY) and (3)

V.M. Nik / Applied Energy 177 (2016) 204–226

extreme warm year (EWY). The weather data sets are created using one or several RCMs. This work investigates if the synthesized weather data sets represent the original RCM data and its variations in different time scales. Moreover it is assessed if the climate uncertainties and differences between scenarios are recognizable between the synthesized data sets when they are based on one climate scenario. The main motivation for creating such weather data sets is decreasing the calculation load while keeping a high accuracy in estimating the variations in hourly time scale. This work aims at synthesizing limited number of weather data sets out of RCMs with the hourly temporal resolution, in a way that the synthesized data represent the average conditions without underestimating the climate uncertainties, extremes and variations in different time scales. Another intention is facilitating the procedure of creating representative weather data sets out of RCMS, without weighting weather parameters in time series. To enhance the impact assessment and decision making procedure, the suggested method evolves in a way to cover climate uncertainties, still keeping the number simulations limited. Application of the suggested method is tried for six climate scenarios in Geneva and two in Stockholm, investigating the energy performance of an office building in Geneva and the residential building stock in Stockholm. This article is divided to the following sections: Section 2 explains about the background of the current work, the method for synthesizing typical and extreme weather data sets (TDY, ECY and EWY), the building models and the climate data sets which are used in this work. Results are thoroughly assessed in Section 3, followed by conclusions about the suggested approach in Section 4. 2. Methodology This section starts with a short description about the background which has resulted in the current work. Afterwards, the main idea of synthesizing typical and extreme weather data sets is discussed. In the following the building models and climate data sets which are used in this work to evaluate the suggested method are presented briefly.

207

in cities [23,73]. The method was dividing the time frame of simulations into several 20- or 30-year simulations and simulating each building only once during each period. Probabilistic distribution of the results was very similar to the cases where all the buildings were simulated during all the years and it was possible to assess the average and long term performances with high accuracy (for more details see [73]). When it comes to designing future air conditioning and energy systems for buildings in urban areas, having accurate estimations of the hourly variations, peak loads and extreme conditions is necessary. This can be underestimated when the number of simulations are decreased as it is visualized by Fig. 1. In this figure the black graph represents the hourly heating demand of the building stock in Stockholm. The building stock is statically represented by 153 buildings and the heating demand is calculated using the method suggested in [73] for a 30-year period, which is simulation buildings over a semi-random distribution of years. The light gray graph in Fig. 1 shows the hourly heating demand of the building stock when all the buildings have been simulated for all the 30 years (4590 years of simulations). The magenta graph shows the same result for only one arbitrary year of simulation (153 years of simulations). Though the probabilistic distribution of the black and gray graphs are very similar (see [73] for details), hourly variations of the black line are smaller than the magenta line (smaller amplitudes for the black graph). This occurs because not simulating all the buildings for every year dampens the hourly effect of climate and underestimates its hourly variations. Although this kind of (semi-) random distribution of years and buildings can estimate the average and long term performances with high accuracy [73], results conceal the short term variations and underestimate the extreme conditions. Therefore it is necessary to use other techniques for simulating future conditions and assessing extreme conditions. The common methods, like using TMY, enables having short time series, representing longer periods. However they underestimate extreme conditions as it was discussed in the introduction. Moreover, due to the increased number of data sets and uncertainties, it is complicated to use the common methods for synthesizing weather data sets out of hourly RCM data. 2.2. Synthesizing weather data sets: TDY, ECY and EWY

2.1. Background of the work The method for synthesizing weather data sets out of RCM data is a response to the challenges that the author has faced for impact assessment and working with future climate data sets. Several techniques and methods have been developed and/or used by the author for building simulations as well as data assessment (e.g. [9,17,23,73,74]), confirming the importance of climate uncertainties and its dependence on the considered phenomena and time scale (e.g. [33,73,75]). Using the hourly RCM data provided a better platform for assessing extreme climatic conditions, which is usually neglected in the case of using common (statistical) approaches for synthesizing weather data sets. However a big challenge has been always dealing with large data sets due to the existence of several climate scenarios (and uncertainties) for long periods; or example even more than 100 years on the hourly time scale. More than that, there exist several scenarios for buildings and energy systems; for example assessing several retrofitting conditions of buildings [74] connected to the energy system in an urban area. This makes the assessment (e.g. data analysis, storage, transfer and visualization) and decision making difficult and costly. Among the statistical methods which were used or developed (e.g. [9,17,73,74]), some aimed at decreasing the number of simulations [73,74] and some at assessing large data sets in different time scales [9,23,74]. A method was specifically developed to decrease the number of simulations for energy modeling of building stock

The backbone of the proposed method is synthesizing three sets of weather data for a 30-year period: (1) typical downscaled year (TDY): representing the typical conditions during the considered period, (2) extreme cold year (ECY): representing the coldest conditions, and (3) extreme warm year (EWY): representing the warmest conditions. Assuming a 30 year period (e.g. 1961–1990), TDY is synthesized in a similar way as TMY by Hall et al. [54], which is based on selecting twelve typical meteorological months (TMMs) and concatenating them to create a weather file for one year. In this work, creating TDY is based on the hourly values of the outdoor air temperature and (unlike TMY) the other climate parameters are not weighted (reasons are explained at the end of this section). The hourly temperature of the 30-year RCM weather data is a 30  8760 matrix, which will be divided into 12 matrices corresponding to 12 months in a year. For each month, temperature distribution is found by calculating its quantiles for each year separately and for all the 30 years together, which the latter is considered as the reference. For each month, the year with the most similar distribution to the reference will be selected as the year with the typical meteorological month (i.e. the year which its quantiles have the least absolute difference from the quantiles of the reference during the considered month). This is similar to comparing the cumulative distribution function (CDF) of the single and reference (or longterm) data sets and finding the one closest to the long-term

208

V.M. Nik / Applied Energy 177 (2016) 204–226

Fig. 1. Hourly profiles of heating demand for 4590 years of simulations (light gray) and semi-random distribution of years and buildings. The considered case is the building stock in Stockholm with RCA3-CNRM-A1B3 climate scenario during 1961–1990.

distribution (which is referred as Finkelstein–Schafer (FS) statistics [54]). For this work, quantiles were calculated using ‘‘quantile” command in Matlab and the cumulative probability interval between 0 and 1 was divided into 100 evenly spaced probabilities to get a finer and more accurate distribution. After repeating the procedure for all the months, 12 years (among 30 years) which represent the typical temperature of each month are recognized. A similar procedure is used to create ECY and EWY data sets, however instead of looking for the least absolute difference, the years with the maximum (for ECY) and minimum (for EWY) difference (a real number instead of a positive real number for TDY) are selected as the years representing the extreme temperatures for each month. Three sets of data which are created in this way, TDY, ECY and EWY, are referred to as the ‘‘synthesized” data sets hereafter in this article. Fig. 2 compares the distribution of the outdoor temperature between the synthesized and original data sets for RCA3-CNRMA1B3 climate scenario during 1961–1990 in Stockholm. Apparently the distribution of temperature for TDY is very similar to the original RCM data for 30 years, unlike ECY and EWY. What is mentioned above is about synthesizing weather data sets based on single climate scenario. The same technique can be applied to create synthesized weather data based on several climate scenarios. Assuming two climate scenarios of SC1 and SC2 for the period of 1961–1990, two sets of 30-year weather data exist (60  8760 matrix), which TDY and extremes are created based on them. In this way, the three sets of synthesized weather data represent both of the climate scenarios during considered period. For

Fig. 2. Nonparametric comparison of the hourly temperature for 30 years of the original RCM data and 1-year synthesized weather data sets of TDY, ECY and EWY. The considered climate scenario is RCA3-CNRM-A1B3 during 1961–1990 in Stockholm.

this work such weather data sets were synthesized out of six and two climate scenarios for Geneva and Stockholm respectively. These are referred to as ‘‘synthesized weather based on multiple climate scenarios” in this work. Such synthesized data sets do not carry climate uncertainties anymore, however they provide a valid and representative image for the considered period. This is discussed with more details in Section 3. The common approach for making representative years (for example making TMY) is using some major weather indices – e.g. dry bulb temperature, dew point temperature, wind speed and solar radiation – and calculating a weighted sum for the indices and recognizing the month with the lowest weighted sum as the typical month. In this work, the suggested technique is based on using only one weather parameter/index out of RCMs which is the dry bulb temperature. Some important reasons for recognizing the typical and extreme months only based on air temperature are as follows:
Climate change does not equally affect all the climate parameters and its signals are not visible or do not have the same strength for all the climatic parameters. For example, climate change signals are strong in temperature and rain precipitation while they are weak for wind and solar radiation (as it is discussed in Section 1). This affects weighting the climatic parameters and complicates the process of finding the representative month.
Since the aim is creating typical and extreme weather data sets, similar indices/parameters should be used to recognize the typical and extreme data sets, which is difficult in the case using weighting factors and several parameters.
Difficulty in weighting the climatic parameters gets more serious when more than one climate scenario is considered (i.e. synthesizing weather data based on multiple climate scenarios). Uncertainties of future climate data sets affect each parameter separately. For example two scenarios may show 30% difference for 30-year mean values of temperature, but 10% difference for wind data. For the users of RCM data, it is almost impossible to decode or trace back these differences and set the correct weighting factors for each parameter.
Climate data out of GCMs and RCMs reflect the interactions of several components of the climate system, similar to the real climate system. This means that each parameter which is calculated by a climate model has been affected by many other parameters and it is not a very accurate approach to weight that for finding a representative month/year.
This work aims at simplifying the process of synthesizing weather data sets for future conditions and decreasing the calculations loads for the impact assessment. Calculating the

209

V.M. Nik / Applied Energy 177 (2016) 204–226 Table 1 Some characteristic of the BESTEST building in Geneva. Material

Thickness [mm]

Total U-value [W/m2/K]

Wall

Gypsum Insulation Gypsum Wood

25 200 9 25

0.23

Roof

Light insulation Wood Gypsum

365 22 13

0.10

Floor

Floor coating Light weight concrete Insulation Concrete

5 50 50 250

0.60

Characteristic

Value [unit]

Solar heat gain coeff. (SHGC) Solar transmittance Visible transmittance Glazing U-value Emissivity

0.76 0.70 0.80 2.90 0.84

Window

daily indices results in increasing the calculation load (considering many climate scenarios) as well as losing the track of hourly variations. 2.3. Building models Two building models are used in this work to evaluate the suggested method: (1) one single building in Geneva, and (2) residential building stock in Stockholm. Since the main intention of the work is assessing the usefulness and accuracy of the synthesized climate files for hourly energy simulations, the assessment technique is based on the comparative analysis of the energy simulation results. This implies using exactly the same building model for all the weather data sets and time periods. The first building model, which is used for the climatic conditions of Geneva, is a BESTEST [76] room with the dimensions of 8 m  6 m  2.7 m (L  W  H), having a 2 m  3 m (W  H) window in the middle of the southern wall with the dimensions of 6 m  2.7 m (W  H). To enable tracking differences between climate scenarios, no shading, blind or curtain is assumed for the window. Otherwise, setting a shading strategy as a function of solar radiation would influence the comparison and diminish the sensitivity of the simulation results to climate uncertainties. Airtightness of the building is 0.5 ACH at 50 Pa and its heating and cooling set points are 21 °C and 25 °C respectively. The maximum power of both the heating and cooling systems are set to 5000 Watt to enable tracking extreme conditions in the energy calculations. It is assumed that two people work from 09:00 to 18:00 during weekdays (Monday–Friday) and there is no one inside the building during weekends. Some characteristics of the building envelope are available in Table 1. The BESTEST building was modeled as one zone and simulated in IDA Indoor Climate and Energy (IDA ICE) simulation tool which is a validated multi-zone simulation application [77]. The second building model in this work has been used previously to simulate and assess future conditions for the residential building stock in Stockholm, considering several climate scenarios and uncertainties [73]. The building stock of Stockholm is statistically represented by 153 sample buildings from the BETSI investigation by the Swedish National Board of Housing, Building and Planning (Boverket) in year 2009 [78], which is the major source of information for the energy performance of residential buildings in Sweden and has been used previously in several works (e.g. [23,35,73,74,79]). According to the previous investigation [73],

[–] [–] [–] [W/m2/K] [–]

heating demand of the building stock in Stockholm will decrease in the future; e.g. during 2081–2100 it will be 25–30% less than the demands before 2011. However climate uncertainties play an important role in the assessment. For example, in the case of having different GCMs, there can be differences up to 30 kW h/m2 (relatively around 30%) in the 20-year mean values. Moreover, variations of the heating demand (hourly standard deviations) can reach to values more than 50% of the average heating demand with 25–30% uncertainties due to different GCMs. Uncertainties increase for cooling demand up to 500%; however the calculated cooling demand for future is still low in Stockholm. Among all the uncertainty factors of the climate data, different GCMs introduce the largest uncertainties in the calculations. For more details about modeling and assessing the future energy performance of the building stock in Stockholm the reader is referred to [23,73]. 2.4. Future climate scenarios The weather data sets which are used in this work were generated by two versions of the Rossby Centre regional climate model: RCA3 [11] and RCA4 [80]. Weather data for Stockholm were dynamically downscaled by RCA3 with the spatial resolution of 50 km and for Geneva by RCA4 with 12.5 km resolution. Each RCM downscaled different GCMs with different forcing conditions; for Stockholm, two GCMs were downscaled by RCA3: (1) CNRM: the third version of the ocean-atmosphere model initially developed at CERFACS. (2) ECHAM5: a coupled atmosphere-ocean GCM developed at the Max-Planck Institute for Meteorology in Germany. A1B scenario of the Special Report on Emissions Scenarios (SRES) [49] is considered for Stockholm. Approximate carbon dioxide equivalent concentrations, corresponding to the computed radiative forcing due to anthropogenic greenhouse gases and aerosols in 2100, for the SRES A1B scenario is about 850 ppm [81] (for more details see [15,17]). For Geneva, RCA4 downscaled four GCMs: (1) CNRM-CM5 general circulation model which has been developed jointly by CNRM–GAME (Centre National de Recherches Météorologiques—Groupe d’études de l’Atmosphère Météorologique) and Cerfacs (Centre Européen de

210

V.M. Nik / Applied Energy 177 (2016) 204–226

Fig. 3. Cumulative heating (top) and cooling (bottom) demand for simulations using 30 years of climate data (light gray) and one year of TDY (black), ECY (blue) and EWY (red) climate data. The considered case is BESTEST building for Geneva with RCA4-MPI-ESM-LR-rcp85 climate scenario. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Table 2 Comparison of the annual heating and cooling demand during 2070–2099 for six climate scenarios: RCA4- (1) CNRM-CM5-rcp45, (2) CNRM-CM5-rcp85, (3) ICHEC-EC-EARTHrcp45, (4) ICHEC-EC-EARTH-rcp85, (5) IPSL-CM5A-MR-rcp85, (6) MPI-ESM-LR-rcp85. Values for the original 30-years weather data (30Y), TDY, ECY, EWY and the average of the last three (Triple) as well as the relative differences of TDY and Triple compared to 30Y and the first climate scenario are shown. Climate scenario

Heating 30Y

Cooling TDY

ECY

EWY

Triple

30Y

TDY

ECY

EWY

Triple

6787.0 5620.9 6137.0 5250.7 5156.7 5045.2

3316.9 2767.7 3020.0 2662.7 2575.5 2873.3

4983.4 4210.7 4508.1 3950.6 3834.4 3938.3

1495.0 2231.6 1437.9 2298.7 3464.4 3031.4

1465.9 2190.5 1298.0 2326.3 3298.2 3070.5

356.7 929.1 463.6 1103.2 1808.6 1532.7

2693.2 3750.5 2672.6 3437.0 5305.5 4492.1

1505.3 2290.0 1478.0 2288.8 3470.8 3031.8

13.2 11.4 6.5 12.8 6.1 13.5

0.0 0.0 0.0 0.0 0.0 0.0

1.9 1.8 9.7 1.2 4.8 1.3

0.0 49.3 3.8 53.8 131.7 102.8

0.0 49.4 11.5 58.7 125.0 109.5

1 2 3 4 5 6

Annual demand [kW h] 4403.8 4846.3 3779.4 4243.5 4233.1 4367.4 3502.0 3938.4 3614.2 3770.9 3469.1 3896.5

1 2 3 4 5 6

Relative difference from 30Y [%] 0.0 10.0 0.0 12.3 0.0 3.2 0.0 12.5 0.0 4.3 0.0 12.3

1 2 3 4 5 6

Relative difference from the first climate scenario [%] – Climate uncertainties 0.0 0.0 0.0 0.0 0.0 14.2 12.4 – – 15.5 3.9 9.9 – – 9.5 20.5 18.7 – – 20.7 17.9 22.2 – – 23.1 21.2 19.6 – – 21.0

– – – – – –

– – – – – –

Recherche et de Formation Avancée). Horizontal resolution of the model varies from 2.8° to 1.4° in the atmosphere and from 2° to 1° in the ocean [82]. (2) ICHEC-EC-EARTH which are results of the collaboration between ICHEC (Irish Centre for High-End Computing) and Met Éireann [83]. EC-EARTH is an Earth system model,

– – – – – –

– – – – – –

0.0 – – – – –

0.0 – – – – –

0.7 2.6 2.8 0.4 0.2 0.0 0.0 52.1 1.8 52.1 130.6 101.4

developed by a number of European national weather services. A prime advantage of EC-EARTH is the operational infrastructure which allows an enormous amount of observations to be assimilated into the model and its behavior can be verified against observations from the daily and seasonal to decadal timescales [84].

211

V.M. Nik / Applied Energy 177 (2016) 204–226

Fig. 4. Hourly profiles of the outdoor temperature for 30 years of simulations (light gray) and (from top to bottom): (1) hourly average of 30 years (1=30 ECY and (4) EWY. The considered climate scenario is RCA4-MPI-ESM-LR-rcp85.

(3) IPSL-CM5A-MR is the mid resolution (1.25°  2.5°) version of the IPSL-CM5A Earth system model, developed by IPSL (Institut Pierre Simon Laplace). IPSL-CM5A includes 5 component models representing the Earth System climate and its carbon cycle, including; atmosphere, ocean, oceanic biogeochemistry, sea-ice, continental surfaces and vegetation as well as atmospheric chemistry [85]. (4) MPI-ESM-LR is the coupled Max Planck Institute Earth System Model at base resolution. MPI-ESM is the improved ECHAM5/MPIOM climate model which its adopted coupled carbon cycle allows studying feedbacks of climate change on the carbon cycle itself. The representation of the middle

P30

y¼1 T y;h ),

(2) TDY, (3)

atmosphere as well as the land surface with interactive vegetation dynamics are also incorporated into the design of the MPI-ESM [86–88]. The first two GCMs for Geneva are forced by two Representative Concentration Pathways (RCPs), RCP4.5 and RCP8.5, and the other GCMs are forced only by RCP8.5. This gives six different climate scenarios for Geneva. RCPs are four greenhouse gas concentration trajectories adopted by the IPCC for its fifth Assessment Report (AR5) in 2014 [89–91] (RCPs supersedes SRES). The global warming increase for RCP4.5 during 2046–2065 is 1.4 °C on average with the likely range of 0.9–2.0 °C, compared to early 21st century (1986–

212

V.M. Nik / Applied Energy 177 (2016) 204–226

Fig. 5. Hourly profiles of heating demand for 30 years of simulations (light gray) and (from top to bottom): (1) hourly average of 30 years (1=30 (4) EWY. The considered climate scenario is RCA4-MPI-ESM-LR-rcp85.

2005 average). The increase for 2081–2100 is 1.8 °C (likely range: 1.1–2.6 °C). These value for RCP8.5 are 2.0 °C (likely range: 1.4–2.6 °C) during 2046–2065 and 3.7 °C (likely range: 2.6–4.8 °C) during 2081–2100 [86,92]. All the RCM weather data for Stockholm and Geneva were synthesized in Matlab before being used in energy simulations. For example climate parameters were synchronized and shortwave components of the solar radiation were calculated based on a method developed by Taesler and Andersson [93]. A more detailed description about preparing the climate data for building simulations is given by Nik [17]. 3. Results and application of the synthesized weather data sets In this section, results of the energy simulation are assessed and compared for the cases with the original RCM weather data (longer

P30

y¼1 Q y;h ),

(2) TDY, (3) ECY and

simulations) and the synthesized one. In the following, the word ‘‘Triple” refers to the occasions when the TDY, ECY and EWY cases are combined and studied together.

3.1. BESTEST building in Geneva The energy performance of the BESTEST building is simulated for future climatic conditions of Geneva in Switzerland. At first, results are analyzed when the synthesized weather data sets are made out of one RCM scenario; in Section 3.1.1 results for one climate scenario are discussed thoroughly and six climate scenarios are compared separately in tables, focusing on seasonal changes and climate uncertainties. In Section 3.1.2 results are investigated when the synthesized weather data sets are made of six climate scenarios.

V.M. Nik / Applied Energy 177 (2016) 204–226

213

P Fig. 6. Hourly profiles of cooling demand for 30 years of simulations (light gray) and (from top to bottom): (1) hourly average of 30 years (1=30 30 y¼1 Q y;h ), (2) TDY, (3) ECY and (4) EWY. The considered climate scenario is RCA4-MPI-ESM-LR-rcp85.

3.1.1. Synthesized weather data based on single climate scenario The cumulative distributions of the hourly heating and cooling demand for four sets of weather data (30 years, TDY, ECY and EWY) as well as 30 single years of the original RCM weather data are compared in Fig. 3 for three 30-year periods. Apparently, the distribution of TDY (black line) is very similar to the case when the building is simulated for all the 30 years (dark gray line). Difference between 30-year and TDY increases by time for heating demand; the annual heating demand for TDY is around 4, 6 and 12 percent more than the average of 30 years, respectively for 2010–2039, 2040–2069 and 2070–2099. This is not the same for the annual cooling demand which TDY shows around 2, 6 and +1 percent differences compared to the 30-year case. ECY weather data results in large heating demand and small cooling demand while the opposite happens for the EWY weather data. In other

words, the two extremes define the (pessimistic) borders of the distributions, as it is visible in Fig. 3. The annual heating and cooling demand of the BESTEST building in Geneva for six (separate) climate scenarios during 2070– 2099 are compared in Table 2 for five cases: ‘‘30Y” which are the averages of the annual demand over 30 years for the original weather data from RCMs, ‘‘TDY”, ‘‘ECY” and ‘‘EWY” – which are the simulation results using, respectively, TDY, ECY and EWY synthesized weather data sets – and finally ‘‘Triple” – which is the average of the calculated demand for TDY, ECY and EWY. To have a better understanding of the differences between the synthesized and original weather data, the relative differences of TDY and Triple compared to 30Y are shown in Table 2 under ‘‘Relative difference from 30Y [%]”. For an ideal case with no difference between the synthesized and the original 30-year data, values

214

V.M. Nik / Applied Energy 177 (2016) 204–226

Fig. 7. Boxplots of the outdoor temperature for 30 years (30Y), TDY, ECY, EWY and Triple (combination of TDY, ECY and EWY) for RCA4-MPI-ESM-LR-rcp85 climate scenario.

Fig. 8. Boxplots of the relative humidity for 30 years (30Y), TDY, ECY, EWY and Triple (combination of TDY, ECY and EWY) for RCA4-MPI-ESM-LR-rcp85 climate scenario.

under TDY column should be equal to zero. However this is not the case and there are differences, varying depending on the climate scenario. It does not make sense to compare the extreme weather files and their columns are left empty. On average (considering all

the climate scenarios), the relative difference for heating demand is around 10% (TDY: +9% and Triple: +10.6%) and for cooling demand is around 1% (TDY: ±2.5% and Triple: +1%). The relative difference decreases for the Triple data set since the need for cooling in Geneva is more dependent on the extreme warm conditions, therefore adding them in the comparison decreases the differences from the original 30Y case. On the other hand, adding extremes for the annual heating demand mostly increases the relative differences. Another comparison in Table 2 is done by checking the differences due to climate uncertainties for the original 30-year data (30Y values are based on the average of 30 years), TDY and Triple (under ‘‘Relative difference from the first climate scenario [%] – Climate uncertainties”). The first climate scenario is taken as the reference in this comparison, which is the reason of having zero in all the cells of the row for climate scenario 1. In an ideal case, each row should show one unique value, dealing with the fact that the relative differences between climate scenarios are exactly the same for all the weather data sets. By checking values for 30Y and the synthesized data sets (TDY and Triple), it is found that the synthesized data represent the climate uncertainties, though with some differences from the original 30Y case. Similar to the previous comparison, considering extremes (Triple) increases the differences from 30Y for the heating demand and decreases for

Fig. 9. Boxplots of the wind speed (left) and direction (right) for 30 years (30Y), TDY, ECY, EWY and Triple (combination of TDY, ECY and EWY) for RCA4-MPI-ESM-LR-rcp85 climate scenario.

215

V.M. Nik / Applied Energy 177 (2016) 204–226

Fig. 10. Boxplots of the components of the direct normal (left) and diffusive horizontal/right) solar radiation for 30 years (30Y), TDY, ECY, EWY and Triple (combination of TDY, ECY and EWY) for RCA4-MPI-ESM-LR-rcp85 climate scenario.

Fig. 11. Boxplots for the hourly heating (left) and cooling demand for 180 years (180Y), three single years of TDY, ECY and EWY and their combination (Triple) for RCA4-MPI-ESM-LR-rcp85 climate scenario.

Table 3 Annual average and standard deviation of the outdoor temperature [°C], heating power [Watt] and cooling power [Watt] for the period of 2070–2099 and six climate scenarios: RCA4- (1) CNRM-CM5-rcp45, (2) CNRM-CM5-rcp85, (3) ICHEC-EARTH-rcp45, (4) ICHEC-EARTH-rcp85, (5) IPSL-CM5A-MR-rcp85, (6) MPI-ESM-LR-rcp85. Annual average Climate scenario Outdoor temperature 1 2 3 4 5 6

30Y

Standard deviation

TDY

ECY

EWY

Triple

30Y

TDY

ECY

EWY

Triple

9.91 11.58 10.18 12.16 13.54 13.02

9.88 11.59 10.21 12.24 13.60 13.03

5.38 7.98 6.60 8.55 10.16 9.64

13.59 15.31 14.01 15.64 17.93 16.51

9.61 11.63 10.27 12.14 13.90 13.06

7.73 7.78 7.84 8.38 8.81 8.24

7.62 7.74 7.73 8.34 8.63 8.18

8.05 7.12 7.39 7.72 7.48 7.42

8.09 7.95 8.27 8.37 10.01 8.67

8.61 8.18 8.37 8.65 9.32 8.58

Heating power 1 2 3 4 5 6

502.39 431.17 482.92 399.52 412.31 395.76

553.16 484.36 498.50 449.54 430.41 444.75

774.69 641.58 700.49 599.33 588.60 575.87

378.60 315.91 344.71 303.93 293.97 327.96

568.82 480.62 514.57 450.93 437.66 449.53

544.61 500.65 534.92 493.02 496.26 487.36

560.55 525.87 545.41 515.04 501.90 503.92

677.62 577.55 625.89 586.92 560.43 557.07

471.64 420.48 452.99 416.76 424.75 431.06

598.49 529.12 565.10 525.05 513.10 510.16

Cooling power 1 2 3 4 5 6

170.55 254.59 164.04 262.24 395.23 345.82

167.32 250.03 148.15 265.53 376.47 350.47

40.71 106.05 52.92 125.92 206.44 174.95

307.41 428.09 305.05 392.31 605.58 512.74

171.82 261.39 168.71 261.25 396.16 346.05

402.42 514.83 374.27 489.35 663.68 617.82

391.94 494.88 338.77 486.80 607.97 617.79

169.33 322.96 218.81 353.84 458.19 437.67

519.98 630.74 488.77 575.21 772.44 703.57

403.41 516.09 380.32 492.78 647.14 612.43

216

V.M. Nik / Applied Energy 177 (2016) 204–226

Table 4 Relative difference [%] of the values for TDY and Triple in Table 3 from the corresponding values for 30Y. Annual average

Standard deviation

Climate scenario

TDY

Triple

TDY

Triple

Outdoor temperature 1 2 3 4 5 6

0.3 0.1 0.3 0.7 0.4 0.1

3.0 0.4 0.9 0.2 2.7 0.3

1.4 0.5 1.4 0.5 2.0 0.7

11.4 5.1 6.8 3.2 5.8 4.1

Heating power 1 2 3 4 5 6

10.1 12.3 3.2 12.5 4.4 12.4

13.2 11.5 6.6 12.9 6.1 13.6

2.9 5.0 2.0 4.5 1.1 3.4

9.9 5.7 5.6 6.5 3.4 4.7

Cooling power 1 2 3 4 5 6

1.9 1.8 9.7 1.3 4.7 1.3

0.7 2.7 2.8 0.4 0.2 0.1

2.6 3.9 9.5 0.5 8.4 0.0

0.2 0.2 1.6 0.7 2.5 0.9

the cooling demand (on average for the six climate scenarios). In general, according to Table 2 and Fig. 3, simulation results using TDY and extreme weather data sets can represent the 30-year data sets with acceptable accuracy, showing the maximum difference of around 10% for the calculated annual demand. Moreover TDY and its combination with extremes are reflecting climate uncertainties. More than the annual scale, it is interesting to see how the synthesized weather data perform on the hourly scale. Hourly profiles of the outdoor temperature for RCA4-MPI-ESM-LR-rcp85 climate scenario in Geneva during three 30-year periods of 2010–2039, 2040–2069 and 2070–2099 are illustrated in Fig. 4. The hourly profile of all the 30 years of the original RCM data are plotted for all

the periods (light gray lines) as the reference. These values are compared to their hourly average values on the graphs on top. Apparently the hourly variations are dampened too much, though the trend is similar for both of the data sets. The TDY data set, on the second graphs from top, reflects the natural variations of the outdoor temperature. Compared to the ECY and EWY in the last two graphs in Fig. 4, TDY covers an area with the highest probability, while ECY is shifted to the lower bound of 30 years graph and EWY to the upper bound. Energy simulation results for different weather data sets are illustrated in Figs. 5 and 6, which show respectively the hourly profiles of the heating and cooling demand (in kW h which for the hourly time scale is equal to power in Watt). Similar to the temperature profile, the hourly average of the energy demand profiles does not represent the natural hourly variations (check dark gray lines in the top graphs of Figs. 5 and 6), while TDY represents the natural hourly variations and cover an area with the highest probability. ECY weather data results in large heating demand (Fig. 5) and small cooling demand (Fig. 6), while the opposite occurs for EWY weather data. In the figures which show the hourly profiles of the outdoor temperature, heating and cooling demand (Figs. 4–6 respectively), considering TDY, ECY and EWY together results in covering the span which the original 30-year data covers, meanwhile keeping the number of simulations down to three (10 times less). In the following, data sets are further analyzed using boxplots. Distribution of the data sets is investigated thoroughly in Figs. 7–11, using boxplots. For each 30-year time period, five boxplots are plotted; comparing data sets of 30-years (30Y), TDY, ECY, EWY and the combination of the last three (Triple). The horizontal green dot-lines in Fig. 7 are drawn in line with the median, lower and upper quartiles and whiskers of the Triple case (which whiskers are set to extend up to 1.5 times of the interquartile range). It is not expected to have similar distributions for 30Y, ECY and EWY, since the last two represent the extreme conditions. The box (or the interquartile range) for 30Y and TDY have very similar ranges as it is expected, as well as the Triple does. However, whiskers and outliers are underestimated for TDY compared to 30Y for

Fig. 12. Cumulative heating (top) and cooling (bottom) demand for simulations using 180 years of climate data (light gray) and one year of TDY (black), ECY (blue) and EWY (red) weather data when six climate scenarios are considered. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

217

V.M. Nik / Applied Energy 177 (2016) 204–226

Fig. 13. Hourly profiles of the outdoor temperature for 180 years of simulations (light gray lines) and (from top to bottom): (1) hourly average of 180 years (1=180 (2) TDY, (3) ECY and (4) EWY. The considered case is BESTEST building for Geneva with six climate scenario.

all the three time periods in Fig. 7. These values with lower probabilities are much better estimated by the Triple case. It is interesting to check the distributions for other climate parameters. In general, it should not be expected to see the same pattern as Fig. 7 for other climate parameters, since the synthesized weather data sets have been created only based on the temperature distribution. For the relative humidity in Fig. 8, values for TDY are very similar to the original RCM data sets. Adding extremes to the typical data results in getting distributions with higher RH as it is visible for the interquartile range of the Triple case. Differences for the wind speed in Fig. 9 are quite small, however Triple shows the most similar distribution to 30Y. In the same figure, differences for the wind direction are larger and TDY distribution is the most similar with 30Y. For the components of the solar radiation in Fig. 10, distributions of the data sets are very sim-

P180

y¼1 T y;h ),

ilar; during some periods TDY is the best match with 30Y and during the other periods, it is the Triple case. A very important part of the assessment is to see how the synthesized weather data sets affect the distribution of the energy simulation results. This is assessed in Fig. 11: similar to the outdoor temperature, considering extremes improves the distribution of results. The most similar distribution to 30Y belongs to Triple. In other words, the combination of TDY, ECY and EWY provides a distribution very similar to the original distribution, especially for the extreme conditions or those with less probabilities (whiskers and outliers). In Geneva, the need for heating is higher than cooling, both in amplitude (higher power) and time span (longer cold season). The annual average of the hourly outdoor temperature, heating and cooling demand (or power) and their variations are shown in

218

V.M. Nik / Applied Energy 177 (2016) 204–226

Fig. 14. Hourly profiles of heating demand for 180 years of simulations (light gray lines) and (from top to bottom): (1) hourly average of 180 years (1=180 (3) ECY and (4) EWY. The considered case is BESTEST building for Geneva with six climate scenario.

Table 3 for six climate scenarios. Considering the original case with 30 years of simulations (30Y) as the reference, the outdoor temperature and heating demand for TDY show the best agreement, both for the annual averages and their hourly variations (standard deviation). For cooling demand, which is less often and more dependent on extreme conditions, considering extremes in the Triple data set improves the results, showing smaller differences from the 30Y case. By checking the values for 30Y, TDY and Triple in Table 3, it is obvious that climate uncertainties are reflected for TDY and Triple in the hourly scale. Differences are calculated in Table 4 for TDY and Triple. The average values for the outdoor temperature are very close to the original 30-year values: less than 1% and 3% difference for the TDY and Triple data respectively. Even for the outdoor temperature, taking the extreme cases into account (Triple) does not increase the differences from the 30Y case considerably (compared

P180

y¼1 Q y;h ),

(2) TDY,

with TDY). For standard deviation, differences for TDY are less than 2% and for Triple data it varies between 3% and 11%. Naturally, standard deviations increase when two extreme data sets are introduced. For the heating power, both the TDY and Triple data overestimate around 3–13%. Standard deviations are smaller, varying between 1% and 9% depending on the scenario, with larger differences for Triple data. For cooling power, the calculated values for Triple data are in a better agreement with 30-year data, showing less than 3% difference, while for TDY it varies between 1% and 10%, both for the averages and standard deviations. In general, TDY underestimates the cooling power compared to the 30-year data. 3.1.2. Synthesized weather data based on multiple climate scenarios The weather data sets which are used in this section are synthesized out of six climate scenarios for Geneva, which are introduced in Section 2.4. The synthesized weather data sets (TDY and

219

V.M. Nik / Applied Energy 177 (2016) 204–226

Fig. 15. Hourly profiles of cooling demand for 180 years of simulations (light gray lines) and (from top to bottom): (1) hourly average of 180 years (1=180 (3) ECY and (4) EWY weather data. The considered case is BESTEST building for Geneva with six climate scenario.

extremes) represent all the scenarios, thereby climate uncertainties are fading away. The energy simulation results of the BESTEST building in Geneva are investigated in a similar way as the previous section, without discussing climate uncertainties. An overall view of the energy simulation results is provided by the cumulative distribution of the heating and cooling demand for three time periods in Fig. 12. Light gray lines in all the figures represent single years of the original RCM data set, which is equal to 180 years during each 30-year period (six scenarios, each 30 years). Dark gray lines represent the average of 180 years and the black line represents TDY (one year), which shows a very similar distribution to the dark gray line. ECY (blue lines) and EWY (red lines) define the boundaries when the extremes occur during the whole year, which is the least probable scenario in reality. The

P180

y¼1 Q y;h ),

(2) TDY,

hourly profiles of temperature, heating and cooling demand are respectively plotted in Figs. 13–15 . The color coding is similar what was mentioned above for Figs. 4–6. Similar to the case of considering only one climate scenario, results of the synthesized weather data sets cover the appropriate range of data with natural hourly variations. For example, the heating demand calculated using TDY covers a range between 180 single years, while for ECY the range is shifted to the largest hourly values unlike the EWY with small hourly heating demands (see Fig. 14). Distribution of the hourly heating and cooling demands are compared using boxplots in Fig. 16 in a similar way as Fig. 11. The only difference is that the original case contains 180 simulations (180Y) in Fig. 16. Similar to Fig. 11, the Triple case shows the best agreement with the original distribution (180Y).

220

V.M. Nik / Applied Energy 177 (2016) 204–226

Fig. 16. Boxplots for the hourly heating (left) and cooling demand for 180 years (180Y), three single years of TDY, ECY and EWY and their combination (Triple) when six climate scenarios are considered.

Table 5 Seasonal averages of the heating power and their variations [Watt] during three periods when the synthesized weather data is based on six climate scenarios. Relative difference [%] of the cases with synthesized data (TDY and Triple) from the original case (180Y) are shown in the second section of the table. The third section of the table compares interperiodical changes [%] between seasons for three cases and the last section shows differences from the original case [%]. 2010–2039 Spring

Summer

2040–2069 Autumn

Winter

Spring

2070–2099

Summer

Autumn

Winter

Spring

Summer

Autumn

Winter

Seasonal average of the heating power (mean) and hourly variations of it (sd) 180Y mean 558.3 158.1 429.5 979.6 520.5 sd 506.7 238.0 480.5 591.2 487.1 TDY mean 505.2 157.5 468.9 1053.3 500.9 sd 552.4 239.5 505.9 586.2 478.1 Triple mean 585.2 184.0 493.5 1154.5 519.5 sd 564.3 288.0 562.8 755.1 538.0

[Watt] 126.7 211.9 109.4 194.9 140.6 246.0

377.7 447.6 463.8 481.5 440.1 495.9

927.1 571.2 1020.1 568.1 1069.1 645.6

455.4 453.5 459.7 459.3 482.1 512.1

98.9 184.4 102.4 194.8 138.0 251.3

349.1 428.7 390.6 439.6 417.9 498.5

852.7 552.2 993.9 551.3 1016.0 643.6

Relative difference from the original case (180 years of weather data) [%] 180Y mean 0.0 0.0 0.0 0.0 0.0 sd 0.0 0.0 0.0 0.0 0.0 TDY mean 9.5 0.4 9.2 7.5 3.8 sd 9.0 0.6 5.3 0.8 1.8 Triple mean 4.8 16.4 14.9 17.9 0.2 sd 11.4 21.0 17.1 27.7 10.4

0.0 0.0 13.7 8.0 11.0 16.1

0.0 0.0 22.8 7.6 16.5 10.8

0.0 0.0 10.0 0.5 15.3 13.0

0.0 0.0 0.9 1.3 5.9 12.9

0.0 0.0 3.5 5.6 39.5 36.3

0.0 0.0 11.9 2.5 19.7 16.3

0.0 0.0 16.6 0.2 19.2 16.6

Inter-periodical variations for the seasonal average of the heating demand and its hourly variations [%] 180Y mean – – – – 6.8 19.9 12.1 5.4 sd – – – – 3.9 11.0 6.8 3.4 TDY mean – – – – 0.9 30.5 1.1 3.2 sd – – – – 13.5 18.6 4.8 3.1 Triple mean – – – – 11.2 23.6 10.8 7.4 sd – – – – 4.7 14.6 11.9 14.5

12.5 6.9 8.2 3.9 7.2 4.8

21.9 13.0 6.4 0.1 1.8 2.2

7.6 4.2 15.8 8.7 5.0 0.5

8.0 3.3 2.6 3.0 5.0 0.3

Relative difference from the original case (180 years of weather data) [%] 180Y mean – – – – 0.0 sd – – – – 0.0 TDY mean – – – – 87.4 sd – – – – 247.7 Triple mean – – – – 65.8 sd – – – – 20.5

0.0 0.0 34.2 43.0 42.4 30.2

0.0 0.0 70.8 99.6 91.6 116.6

0.0 0.0 108.4 106.1 33.4 112.4

0.0 0.0 68.0 11.1 38.1 90.7

0.0 0.0 53.8 69.8 18.8 33.0

Differences between Triple and 180Y are larger for the outliers of cooling demand compared to Fig. 11, however the number (or probability) of outliers are much less than the other values. Seasonal averages of the heating power and their variations for TDY, its combination with extremes (Triple) and the original RCM data (180Y) are compared in Table 5 for the case of synthesizing weather data sets out of six climate scenarios. In the second section of Table 5 (‘‘Relative difference from the original case (180 years of weather data) [%]”), the relative differences of the values for TDY and Triple in the first section of the table are shown in percentage (compared to 180Y). According to the last column, which shows the averages over all the periods and seasons, TDY deviates less from 180Y values.

0.0 0.0 91.0 29.6 10.3 73.6

0.0 0.0 41.2 8.7 38.0 328.7

Average 0.0 0.0 4.6 1.7 15.1 17.5

Average 0.0 0.0 28.8 29.0 11.6 13.2

Climate change can also affect the inter-periodical variations among the considered periods. For example if the average temperature in springs of 2040–2069 increases 2 °C more than 2010–2039, the increment for the next period is not the same and springs of 2070–2099 might be e.g. 1 °C warmer than 2040– 2069. These inter-periodical variations will be reflected in building simulations as it is shown in some previous works of the author (e.g. [9,17]). The third section of Table 5 shows the interperiodical variations for the seasonal averages of heating demand and its hourly variations in percentage. Values under ‘‘2040– 2069” and ‘‘2070–2099” columns are the relative differences from ‘‘2010–2039” and ‘‘2040–2069” periods respectively. To check how much the values out of TDY and Triple data sets are different from

V.M. Nik / Applied Energy 177 (2016) 204–226

221

Fig. 17. Hourly profiles of heating demand for 4590 years of simulations (light gray lines) and (from top to bottom): (1) semi-random distribution of buildings and years (153 years), (2) TDY, (3) ECY and (4) EWY weather data. The considered case is the building stock in Stockholm with RCA3-CNRM-A1B3 climate scenario.

the original case (180Y), the relative differences from the original case are shown in the fourth section of Table 5. The largest deviations for the Triple data set, specifically for standard deviations (sd), occur during summer and winter which extremes are more often. For example, in springs of 2040–2069 the inter-periodical difference (compared to 2010–2039) for the average heating power out of TDY is 87.4% less than 180Y. In other words, the interperiodical changes are underestimated by TDY; springs of 2040– 2069 have 6.8% less heating power than 2010–2039 for 180Y, while for TDY this value is 0.9% (check the third section of Table 5). It is important to check the values in the third section of Table 5 when reading the values in the fourth section, since the large values in the fourth section might be misleading. The interesting outcome of the fourth section is in the last column of Table 5 which shows that the average of the relative differences over all seasons are less

for the Triple case. This means that considering the extremes in calculating the inter-periodical variations of the seasonal averages results in having values closer to the original data sets. 3.2. Building stock of Stockholm After evaluating the application of the synthesized weather data sets in energy simulation of a single building, this section evaluates their application in the impact assessment of climate change on the energy performance of a group of buildings with different properties. In this section the building stock of Stockholm, which is statistically represented by 153 buildings, is considered. Similar to the Geneva case, the first set of synthesized weather data is created using one future climate scenario which its results are discussed in Section 3.2.1. Afterwards, in Section 3.2.2 the energy

222

V.M. Nik / Applied Energy 177 (2016) 204–226

Fig. 18. Boxplots for the hourly heating (left) and cooling (right) demand of 153 buildings for 30 years (All years – equal to 4590 year-simulation), one semi-random year for each building (1 year – equal to 153 year-simulation), TDY for all buildings (153 year-simulation), ECY for all buildings (153 year-simulation), EWY for all buildings (153 yearsimulation) and Triple (combination of TDY, ECY and EWY).

Fig. 19. Cumulative heating (top) and cooling (bottom) demand for simulations using 9180 years of weather data (light gray), hourly average of them (dark gray) as well as one year of synthesized TDY (black), ECY (blue) and EWY (red) weather data when two climate scenarios are considered. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

simulation results are assessed when several climate scenarios are used for synthesizing weather data. Unlike the Geneva case, climate uncertainties as well as the annual and seasonal variations are not discussed in this section since climate uncertainties and variations by time are better illustrated when the focus is only one building and results are not influenced by changes in building characteristics. 3.2.1. Synthesized weather data based on one climate scenario TMY, ECY and EWY weather data sets from RCA3-CNRM-A1B3 climate scenario for three 30-year periods of 1961–1990, 2021– 2050 and 2071–2100 were used to run energy simulations. The

hourly profiles for the heating demand of the building stock are compared in Fig. 17; the original case is plotted on every figure which is the case when all buildings are simulated for all the 30 years during each period, resulting in 4590 (153 [#buildings]  30 [#years]) simulation years. The graphs on top in Fig. 17 compare the semi-random simulations (dark gray lines – discussed in Section 2 and [73]) with the original case (light gray lines). The semi-random simulation technique does not represent the hourly variations accurately and similar to taking the hourly averages (as it was shown for Geneva case), the hourly variations are dampened. Unlike that, TDY and the two extreme weather files represent the natural hourly variations, while TDY covers the most

V.M. Nik / Applied Energy 177 (2016) 204–226

223

Fig. 20. Hourly profiles of heating demand for 9180 years of simulations (light gray lines) and (from top to bottom): (1) TDY, (2) ECY and (3) EWY. The considered case is the building stock in Stockholm with two climate scenarios (RCA3-CNRM-A1B3 and RCA3-ECHAM5-A1B3).

probable area and extremes estimate the hourly heating demand in the case of having extreme cold or warm climatic conditions. Nonparametric distribution of the heating and cooling demand are compared in Fig. 18. According to the boxplots for heating demand, semi-random simulation of the building stock (shown as ‘‘1 year” in Fig. 18) underestimates the extreme conditions in all the periods which is obvious by comparing the whiskers for ‘‘All years” and ‘‘1 year” in Fig. 18. Semi-random simulations also underestimate average conditions as it is visible for the hourly heating demand during 2071–2100 in Fig. 18: the box (interquartile range) for ‘‘1 year” is shifted downwards compared to ‘‘All years”. Using TDY weather data set gives a better distribution of the values, though with underestimating the hours with very large heating power (compare the upper whiskers and outliers for ‘‘All years” and ‘‘TDY” in Fig. 18). Including extreme weather data sets in the assessment improves the estimation; ‘‘All years” and ‘‘Triple” have very similar distributions during all the periods and Triple data set estimates the extreme heating power better than TDY. As it was investigated previously [73], cooling demand for Stockholm is very small even for future climate. The hourly profile of cooling demand does not reveal that much information and it is enough to look at its nonparametric distribution using boxplots. In Fig. 18 there is no box in the plots for the hourly cooling demand (unlike the Geneva case), since there are very few cooling hours compared to the total number of hours in the considered time spans. Not surprisingly the EWY data set and consequently the Triple case are the most similar to the original case (All years).

3.2.2. Synthesized weather data based on multiple climate scenarios Energy simulation results for the building stock in Stockholm are examined here when the synthesized weather data sets are made out of two different climate scenario: RCA3-CNRM-A1B3 and RCA3-ECHAM5-A1B3. In the case of simulating the building stock using the original weather data sets, 153 buildings were simulated for 30 years for two climate scenarios, resulting in 9180 simulation years per period. Results are shown as light gray lines in Figs. 19 and 20, which the first figure shows the cumulative heating and cooling demand and Fig. 20 shows the hourly profiles for heating demand. Hourly averages of the light gray lines are shown as dark gray lines. Using TDY, ECY and EWY weather data sets decreases the number of simulations to 459 per period (20 times less than the original case). In both Figs. 19 and 20, TDY covers the most probable area of results (compared to the original case), while ECY results in having the highest heating demand and the lowest cooling demand, in contrary to the results out of using EWY weather data. It is interesting to see how much cooling demand can increase by time in Fig. 19 in the case of using EWY data. However, as it was discussed previously, the two extremes define the pessimistic boundaries and the probability of having cumulative distributions is very low since the worst conditions do not happen continuously for one year. Having an overall picture about the probability of extremes, such as Fig. 21, helps in resilient design of buildings and energy systems. Boxplots in Fig. 21 compare the distribution of the heating and cooling power for five different cases. ‘‘All years” corresponds to the original case which

224

V.M. Nik / Applied Energy 177 (2016) 204–226

Fig. 21. Boxplots for the hourly heating (left) and cooling (right) demand of 153 buildings for 30 years and two climate scenarios (All years – 9180 year-simulation), TDY for all buildings and all scenarios (153 year-simulation), ECY for all buildings and all scenarios (153 year-simulation), EWY for all buildings and all scenarios (153 yearsimulation) and Triple (combination of TDY, ECY and EWY).

‘‘TDY” and ‘‘Triple” have quite similar distributions to the original case, however Triple shows a better distribution and whiskers and outliers are more similar to the original case. It is interesting to see that it is possible to reach to very good distributions with such details by simulating only for three years instead of 30 years per building. 4. Conclusions This work suggested a method for performing the impact assessment of climate change on the energy performance of buildings using weather data sets out of regional climate models (RCMs) with the hourly temporal resolution. The method is based on synthesizing three weather data sets for each 30-year period: (1) typical downscaled year (TDY), (2) extreme cold year (ECY) and (3) extreme warm year (EWY). Each weather data set is created based on comparing the cumulative distribution of the outdoor (drybulb) temperature and finding the typical and extreme months. The weather data sets were synthesized using single or multiple climate scenario(s). In the case of using multiple scenarios, climate uncertainties (among the considered scenarios) fades away and the synthesized data represents all the considered scenarios. Several synthesized weather data sets were created in this work, considering two RCMs (RCA3 for Stockholm and RCA4 for Geneva), six GCMs (four for Geneva and two for Stockholm) and three emissions scenarios/pathways (RCP4.5 and RCP8.5 for Geneva and SRES A1B for Stockholm). Six different climate scenarios were considered for Geneva, which seven sets of synthesized weather data (TDY and extremes) were created; six sets when climate scenarios were considered separately (single climate scenario) and one set based on all the six scenarios (multiple climate scenarios). For Stockholm two climate scenarios were considered and two sets of weather data were synthesized; one based on a single scenario and one based on two scenarios. The synthesized data sets were used in the energy simulation of an office building in Geneva and the residential building stock in Stockholm. Based on the results, it is possible to use the synthesized weather data sets in energy simulations and produce reliable results, representing the cumulative energy distributions as well as the hourly variations. The cumulative distribution of the heating and cooling demand using TDY are very similar to the original weather data set. The hourly profiles of temperature, heating and cooling demand by TDY represent the most probable conditions while ECY and EWY define the extreme conditions. Distributions

of the hourly heating and cooling demand were studied for different data sets, using boxplots. According to the results, considering TDY, ECY and EWY together (which is called Triple) enables to have estimations very similar to the cases where the original weather data sets are used. Assessing climate uncertainties and seasonal variations revealed the fact that the synthesized data sets (TDY and Triple) reflect climate uncertainties (when being created based on single climate scenario) as well as climate variations in different time scales (e.g. hourly, periodical and seasonal). For example, it was shown that considering extremes in calculating the interperiodical variations of the seasonal averages results in having values closer to the original data sets. Applying the presented method and using the synthesized weather data sets has the advantage of decreasing the number of simulations extensively without neglecting the extremes and variations of the original RCM data. By considering multiple scenarios in synthesizing the weather data sets, it is possible to have a scientifically valid assessment while keeping the number of simulations much lower than working with the original RCM data sets. In the case of need, it is possible to synthesize weather data sets from several climate scenarios in a structured way. For example, in the case of assessing the importance of emissions scenario or RCPs for future energy conditions, data sets from several RCMs and/or GCMs with a similar RCP can be used to synthesize TDY and extreme data sets. Acknowledgment The author thanks Dr. Erik Kjellström and Dr. Gustav Strandberg at the Rossby Centre, SMHI for their support and help with the climate data. This research was financed by the Swedish Research Council (Formas), which is gratefully acknowledged. References [1] The Global Risks Report 2016. World Economic Forum, Geneva, Switzerland, 80116; 2016. [2] Meehl GA, Stocker TF, Collins W, Friedlingstein P, Gaye A, Gregory J, et al. Global climate projections climate change 2007: the physical science basis. In: Qin M, Manning Z, Chen M, Marquis K, Averyt M, Tignor HL, et al., editors. Contribution of working Group I to the fourth assessment report of the intergovernmental panel on climate change. Camb. Univ. Press; 2007. p. 747– 845. [3] ‘CORDEX’. Available: . [4] Fowler HJ, Blenkinsop S, Tebaldi C. Linking climate change modelling to impacts studies: recent advances in downscaling techniques for hydrological modelling. Int J Climatol 2007;27(12):1547–78.

V.M. Nik / Applied Energy 177 (2016) 204–226 [5] Prudhomme C, Wilby RL, Crooks S, Kay AL, Reynard NS. Scenario-neutral approach to climate change impact studies: application to flood risk. J Hydrol 2010;390(3–4):198–209. [6] Belcher S, Hacker J, Powell D. Constructing design weather data for future climates. Build Serv Eng Res Technol 2005;26(1):49–61. [7] Field CB, Barros V, Stocker TF, Dokken DJ, Ebi KL, Mastrandrea M, et al. Summary for policymakers. In: Managing the risks of extreme events and disasters to advance climate change adaptation. Cambridge, UK, and New York, NY, USA: Cambridge University Press; 2012. [8] Fant C, Adam Schlosser C, Strzepek K. The impact of climate change on wind and solar resources in southern Africa. Appl Energy 2016;161:556–64. January. [9] Nik VM, Sasic Kalagasidis A, Kjellström E. Statistical methods for assessing and analysing the building performance in respect to the future climate. Build Environ 2012;53:107–18. July. [10] Giorgi F. Regional climate modeling: status and perspectives. J Phys IV – Proc 2006;139:18. [11] Samuelsson P, Jones CG, Willen U, Ullerstig A, Gollvik S, Hansson U, et al. The Rossby centre regional climate model RCA3: model description and performance. Tellus A 2011;63(1):4–23. [12] Rummukainen M. State-of-the-art with regional climate models. Wiley Interdiscip Rev Clim Change 2010;1(1):82–96. [13] Jentsch MF, James PAB, Bourikas L, Bahaj AS. Transforming existing weather data for worldwide locations to enable energy and building performance simulation under future climates. Renew Energy 2013;55:514–24. July. [14] IPCC. Climate change 2007: synthesis report. In: Contribution of working Groups I, II and III to the fourth assessment report of the intergovernmental panel on climate change. Geneva, Switzerland: IPCC; 2007. [15] Kjellström E, Nikulin G, Hansson U, Strandberg G, Ullerstig A. 21st century changes in the European climate: uncertainties derived from an ensemble of regional climate model simulations. Tellus A 2011;63(1):24–40. [16] Christensen J, Kjellström E, Giorgi F, Lenderink G, Rummukainen M. Weight assignment in regional climate models. Clim Res 2010;44:179–94. December. [17] Nik VM. Climate simulation of an attic using future weather data sets – statistical methods for data processing and analysis. Licentiate thesis. Sweden: Chalmers University of Technology; 2010. Available online: http:// publications.lib.chalmers.se/records/fulltext/114053/114053.pdf. [18] de Wilde P, Coley D. The implications of a changing climate for buildings. Build Environ 2012;55:1–7. [19] Kalvelage K, Passe U, Rabideau S, Takle ES. Changing climate: the effects on energy demand and human comfort. Energy Build 2014;76:373–80. June. [20] Shibuya T, Croxford B. The effect of climate change on office building energy consumption in Japan. Energy Build 2016;117:149–59. April. [21] Braun MR, Beck SBM, Walton P, Mayfield M. Estimating the impact of climate change and local operational procedures on the energy use in several supermarkets throughout Great Britain. Energy Build 2016;111:109–19. January. [22] Wang H, Chen Q. Impact of climate change heating and cooling energy use in buildings in the United States. Energy Build 2014;82:428–36. October. [23] Nik VM. Hygrothermal simulations of buildings concerning uncertainties of the future climate PhD thesis. Gothenburg, Sweden: Chalmers University of Technology; 2012. Available online: http://publications.lib.chalmers.se/ records/fulltext/159222.pdf. [24] Lomas KJ, Giridharan R. Thermal comfort standards, measured internal temperatures and thermal resilience to climate change of free-running buildings: a case-study of hospital wards. Build Environ 2012;55:57–72. September. [25] Alves CA, Duarte DHS, Gonçalves FLT. Residential buildings’ thermal performance and comfort for the elderly under climate changes context in the city of São Paulo, Brazil. Energy Build 2016;114:62–71. February. [26] Barbosa R, Vicente R, Santos R. Climate change and thermal comfort in Southern Europe housing: a case study from Lisbon. Build Environ 2015;92:440–51. October. [27] Sailor DJ. Risks of summertime extreme thermal conditions in buildings as a result of climate change and exacerbation of urban heat islands. Build Environ 2014;78:81–8. August. [28] Fisk WJ. Review of some effects of climate change on indoor environmental quality and health and associated no-regrets mitigation measures. Build Environ 2015;86:70–80. April. [29] Coley D, Kershaw T. Changes in internal temperatures within the built environment as a response to a changing climate. Build Environ 2010;45 (1):89–93. [30] Wan KKW, Li DHW, Pan W, Lam JC. Impact of climate change on building energy use in different climate zones and mitigation and adaptation implications. Appl Energy 2012;97:274–82. September. [31] Chow DHC, Li Z, Darkwa J. The effectiveness of retrofitting existing public buildings in face of future climate change in the hot summer cold winter region of China. Energy Build 2013;57:176–86. February. [32] Karimpour M, Belusko M, Xing K, Boland J, Bruno F. Impact of climate change on the design of energy efficient residential building envelopes. Energy Build 2015;87:142–54. January. [33] Nik VM, Sasic Kalagasidis A, Kjellström E. Assessment of hygrothermal performance and mould growth risk in ventilated attics in respect to possible climate changes in Sweden. Build Environ 2012;55:96–109. September.

225

[34] Nik VM, Mata E, Kalagasidis AS. Assessing the efficiency and robustness of the retrofitted building envelope against climate change. Energy Procedia 2015;78:955–60. November. [35] Nik VM, Mata E, Sasic Kalagasidis A, Scartezzini J-L. Effective and robust energy retrofitting measures for future climatic conditions—reduced heating demand of Swedish households. Energy Build 2016;121:176–87. June. [36] Sehizadeh A, Ge H. Impact of future climates on the durability of typical residential wall assemblies retrofitted to the PassiveHaus for the Eastern Canada region. Build Environ 2016;97:111–25. February. [37] Nik V, Sasic Kalagasidis A. Wind driven rain and climate change: a simple approach for the impact assessment and uncertainty analysis. In: Presented at the NSB2014, Lund, Sweden, 2014, vol. Challenges of the future. [38] Pakkala TA, Köliö A, Lahdensivu J, Kiviste M. Durability demands related to frost attack for Finnish concrete buildings in changing climate. Build Environ 2014;82:27–41. December. [39] de Wilde P, Rafiq Y, Beck M. Uncertainties in predicting the impact of climate change on thermal performance of domestic buildings in the UK. Build Serv Eng Res Technol 2008;29(1):7–26. [40] Wang X, Chen D, Ren Z. Assessment of climate change impact on residential building heating and cooling energy requirement in Australia. Build Environ 2010;45(7):1663–82. [41] Tian W, de Wilde P. Uncertainty and sensitivity analysis of building performance using probabilistic climate projections: a UK case study. Autom Constr 2011;20(8):1096–109. [42] Gaterell MR, McEvoy ME. The impact of climate change uncertainties on the performance of energy efficiency measures applied to dwellings. Energy Build 2005;37(9):982–95. [43] Mirasgedis S, Sarafidis Y, Georgopoulou E, Kotroni V, Lagouvardos K, Lalas DP. Modeling framework for estimating impacts of climate change on electricity demand at regional level: case of Greece. Energy Convers Manage 2007;48 (5):1737–50. May. [44] Zhou Y, Clarke L, Eom J, Kyle P, Patel P, Kim SH, et al. Modeling the effect of climate change on U.S. state-level buildings energy demands in an integrated assessment framework. Appl Energy 2014;113:1077–88. January. [45] Ahmed T, Muttaqi KM, Agalgaonkar AP. Climate change impacts on electricity demand in the State of New South Wales, Australia. Appl Energy 2012;98:376–83. October. [46] Pryor SC, Schoof JT, Barthelmie RJ. Winds of change?: projections of nearsurface winds under climate change scenarios. Geophys Res Lett 2006;33(11): L11702. [47] Seljom P, Rosenberg E, Fidje A, Haugen JE, Meir M, Rekstad J, et al. Modelling the effects of climate change on the energy system—a case study of Norway. Energy Policy 2011;39(11):7310–21. [48] Sailor DJ, Smith M, Hart M. Climate change implications for wind power resources in the Northwest United States. Renew Energy 2008;33 (11):2393–406. [49] Nakicenovic N, Alcamo J, Davis G, de Vries B, Fenhann J, Gaffin S, et al. Special report on emissions scenarios: a special report of working Group III of the intergovernmental panel on climate change; 2000. [50] Kao S-C, Sale MJ, Ashfaq M, Uria Martinez R, Kaiser DP, Wei Y, et al. Projecting changes in annual hydropower generation using regional runoff data: an assessment of the United States federal hydropower plants. Energy 2015;80:239–50. February. [51] Guan L. Preparation of future weather data to study the impact of climate change on buildings. Build Environ 2009;44(4):793–800. April. [52] Yang L, Wan KKW, Li DHW, Lam JC. A new method to develop typical weather years in different climates for building energy use studies. Energy 2011;36 (10):6121–9. October. [53] Chan ALS, Chow TT, Fong SKF, Lin JZ. Generation of a typical meteorological year for Hong Kong. Energy Convers Manage 2006;47(1):87–96. [54] Hall IJ, Prairie RR, Anderson HE, Boes EC. Generation of a typical meteorological year. Sandia Labs., Albuquerque, NM (USA), SAND-78-1096C; CONF-780639-1; January 1978. [55] AHSRAE. International weather for energy calculations (IWEC Weather Files) User’s Manual. ASHRAE; 2002. [56] ASHRAE. Weather year for energy calculations. American Society of Heating, Refrigerating and Air-Conditioning Engineers; 1985. [57] Crow L. Summary description of typical year weather data. Chicago midway airport: American Society of Heating, Refrigerating and Air-Conditioning Engineers; 1970. [58] Crawley DB. Which weather data should you use for energy simulations of commercial buildings? ASHRAE Trans 1998;104(2):498–515. [59] Marion W, Urban K. Users manual for radiation data base TMY2s derived from the 1961–1990 national solar@. Natl Renew Energy Lab 1995. [60] Hong T, Chang W-K, Lin H-W. A fresh look at weather impact on peak electricity demand and energy use of buildings using 30-year actual weather data. Appl Energy 2013;111:333–50. November. [61] Kershaw T, Eames M, Coley D. Comparison of multi-year and reference year building simulations. Build Serv Eng Res Technol 2010;31(4):357–69. [62] Chan ALS. Developing future hourly weather files for studying the impact of climate change on building energy performance in Hong Kong. Energy Build 2011;43(10):2860–8. [63] Huang K-T, Hwang R-L. Future trends of residential building cooling energy and passive adaptation measures to counteract climate change: the case of Taiwan. Appl Energy.

226

V.M. Nik / Applied Energy 177 (2016) 204–226

[64] Jentsch MF, Bahaj AS, James PAB. Climate change future proofing of buildings– Generation and assessment of building simulation weather files. Energy Build 2008;40(12):2148–68. [65] Zhu M, Pan Y, Huang Z, Xu P. An alternative method to predict future weather data for building energy demand simulation under global climate change. Energy Build 2016;113:74–86. February. [66] Arima Y, Ooka R, Kikumoto H, Yamanaka T. Effect of climate change on building cooling loads in Tokyo in the summers of the 2030s using dynamically downscaled GCM data. Energy Build 2016;114:123–9. February. [67] Jylhä K, Jokisalo J, Ruosteenoja K, Pilli-Sihvola K, Kalamees T, Seitola T, et al. Energy demand for the heating and cooling of residential houses in Finland in a changing climate. Energy Build 2015;99:104–16. July. [68] Christenson M, Manz H, Gyalistras D. Climate warming impact on degree-days and building energy demand in Switzerland. Energy Convers Manage 2006;47 (6):671–86. [69] Eames M, Kershaw T, Coley D. On the creation of future probabilistic design weather years from UKCP09. Build Serv Eng Res Technol 2011;32 (2):127–42. [70] Mylona A. The use of UKCP09 to produce weather files for building simulation. Build Serv Eng Res Technol 2012;33(1):51–62. [71] Eames M, Kershaw T, Coley D. A comparison of future weather created from morphed observed weather and created by a weather generator. Build Environ 2012;56:252–64. October. [72] Kershaw T, Eames M, Coley D. Assessing the risk of climate change for buildings: a comparison between multi-year and probabilistic reference year simulations. Build Environ 2011;46(6):1303–8. [73] Nik VM, Sasic Kalagasidis A. Impact study of the climate change on the energy performance of the building stock in Stockholm considering four climate uncertainties. Build Environ 2013;60:291–304. February. [74] Nik VM, Mata É, Sasic Kalagasidis A. A statistical method for assessing retrofitting measures of buildings and ranking their robustness against climate change. Energy Build 2015;88:262–75. February. [75] Nik VM, Mundt-Petersen SO, Kalagasidis AS, De Wilde P. Future moisture loads for building facades in Sweden: climate change and wind-driven rain. Build Environ 2015;93(2):362–75. [76] Judkoff R, Neymark J. International Energy Agency Building Energy Simulation Test (BESTEST) and diagnostic method. Colorado, USA: National Renewable Energy Laboratory (NREL), NREL/TP-472-6231; February 1995. [77] EQUA. Validation of IDA indoor climate and energy 4.0 build 4 with respect to ANSI/ASHRAE Standard 140-2004. Equa Simulation AB, Solna, Sweden, Technical report; April 2010. [78] BETSI. Description of the existing buildings: technical characteristics, indoor environment and energy consumption. (Bebyggelsens energianvändning, tekniska status och innemiljö). Karlskrona, Sweden: Boverket – the National Board of Housing, Building and Planning; 2009.

[79] Mata É, Sasic Kalagasidis A, Johnsson F. Energy usage and technical potential for energy saving measures in the Swedish residential building stock. Energy Policy 2013;55:404–14. April. [80] Samuelsson P, Gollvik S, Jansson C, Kupiainen M, Kourzeneva E, van de Berg WJ. The surface processes of the Rossby Centre regional atmospheric climate model (RCA4). Swedish Meteorological and Hydrological Institute (SMHI), Meteorologi 157; 2015. [81] Solomon S. Climate change 2007: the physical science basis: contribution of working Group I to the fourth assessment report of the intergovernmental panel on climate change. Cambridge Univ Pr; 2007. [82] Voldoire A, Sanchez-Gomez E, y Mélia DS, Decharme B, Cassou C, Sénési S, et al. The CNRM-CM5.1 global climate model: description and basic evaluation. Clim Dyn 2012;40(9–10):2091–121. [83] ‘Met Éireann | ICHEC’. Available: [accessed 14.01.16]. [84] Hazeleger W, Bintanja R. Accomplishments, current status and future plans of EC-EARTH: a European earth system model. EC-Earth Steering Group; May 2014. [85] ‘IPSL Climate Modelling Centre (ICMC)’. Available: [accessed 14.01.16]. [86] Giorgetta MA, Jungclaus J, Reick CH, Legutke S, Bader J, Böttinger M, et al. Climate and carbon cycle changes from 1850 to 2100 in MPI-ESM simulations for the Coupled Model Intercomparison Project phase 5. J Adv Model Earth Syst 2013;5(3):572–97. [87] Block K, Mauritsen T. Forcing and feedback in the MPI-ESM-LR coupled model under abruptly quadrupled CO2. J Adv Model Earth Syst 2013;5(4):676–91. [88] ‘New Earth system model of Max Planck Institute for Meteorology’. Available: [accessed 14.01.16]. [89] IPCC. Climate change 2013: the physical science basis. In: Contribution of working Group I to the fifth assessment report of the intergovernmental panel on climate change. Cambridge, United Kingdom and New York, NY, USA: Cambridge University Press. [90] van Vuuren DP, Edmonds J, Kainuma M, Riahi K, Thomson A, Hibbard K, et al. The representative concentration pathways: an overview. Clim Change 2011;109(1–2):5–31. [91] Moss RH, Edmonds JA, Hibbard KA, Manning MR, Rose SK, van Vuuren DP, et al. The next generation of scenarios for climate change research and assessment. Nature 2010;463(7282):747–56. [92] Stocker TF, Qin D, Plattner G-K, Tignor M, Allen SK, Boschung J, et al. Summary for policymakers. In: Climate change 2013: the physical science basis. Contribution of working Group I to the fifth assessment report of the intergovernmental panel on climate change. Cambridge, United Kingdom and New York, NY, USA: Cambridge University Press. [93] Taesler R, Andersson C. A method for solar radiation computations using routine meteorological observation. Energy Build 1984;7:341–52.