Geo-dependent heat demand model of the Swiss building stock: method, results and example of application

(1)

Report

Reference

Geo-dependent heat demand model of the Swiss building stock:

method, results and example of application

SCHNEIDER, Stefan, et al. & SCCER Future Energy Efficient Buildings & Districts

Abstract

This paper presents a statistical regression bottom up model for the spatial characterisation of the final energy for space heating and domestic water production. For each building of the Swiss national building and dwelling register, the model estimates the heated surface and the floor specific final energy demand for space heating and domestic hot water production, by way of average statistical indicators being derived from two large calibration sets. For each pixel of territory a bootstrap algorithm allows to estimate the confidence interval around the average value given by the model. For a portion of territory with known demand, we checked that the confidence interval is in accordance with the selected confidence level, comforting that the bootstrap algorithm is adequate. Furthermore, the error decreases inversely proportional to the square root of the number of buildings, as predicted by the central limit theorem. At national level, the total aggregated final energy demand (93.7 TWh/year) is in concordance with existing statistics. Finally, we estimate that if the entire Swiss building stock would undergo deep [...]

SCHNEIDER, Stefan, et al. & SCCER Future Energy Efficient Buildings & Districts.

Geo-dependent heat demand model of the Swiss building stock: method, results and example of application. Genève : SCCER Future Energy Efficient Buildings & Districts, 2018

Available at:

http://archive-ouverte.unige.ch/unige:103112

Disclaimer: layout of this document may differ from the published version.

(2)

Swiss Competence Center for Energy Research

Report

Geo-dependent heat demand model of the Swiss building stock: method, results and example of application

SCCER Future Energy Efficient Buildings & Districts

Work package: 3, task 3.1.1: Geo-dependent Energy demand (heat, cooling &

electricity)

Stefan Schneider, Jad Khoury, Bernard Lachal and Pierre Hollmuller

March 2018

if the journal requires them.

Subject classification codes: include these here if the journal requires

(3)

Nomenclature

Abbreviations

DHW domestic hot water SF single-family SH space heating MF multi-family Databases

GEAK building energy certification database, Switzerland GWR national building and dwelling database, Switzerland IDC building heat consumption index database, Geneva Latin symbols

AE heated floor surface (m²)

AE relative error on heated surface (m²/m²) Adwell dwelling surface of a building (m²) Agross gross surface of a building (m²)

Ath surface of the building thermal envelope (m²) DD heating degree days (K.day)

DDref heating degree days, reference weather station (K.day)

fdwell ratio between heated floor surface and dwelling surface (m²/m²) fgross ratio between heated floor surface and gross surface (m²/m²)

e final energy for heat production (DHW and SH), floor specific (MJ/m²/year) Ebld final energy for heat production (DHW and SH), at building level (MJ/year) Epixel final energy for heat production (DHW and SH), at pixel level (MJ/year) Nfloors number of floors of a building

qdhw heat demand DHW (MJ/m²/year)

qdhw,SIA heat demand DHW, standard value (MJ/m²/year) qsh heat demand SH (MJ/m²/year)

qsh,avg heat demand for SH, reference weather station, average over building category and age (MJ/m²/year)

qsh,norm normed heat demand for SH after retrofit (MJ/m²/year) qsh,ref heat demand SH, reference weather station (MJ/m²/year)

(4)

qsh,net realistic (or net) saving potential after retrofit (MJ/m²/year)

qsh,norm normed (or gross) saving potential after retrofit (MJ/m²/year)

Qsh,net total realistic (or net) saving potential after retrofit (TWh/year)

Qsh,norm total normed (or gross) saving potential after retrofit (TWh/year) Greek symbols

 efficiency of heat production (-)

(5)

1 Introduction

In 2013, the Swiss building stock consumed around 92 TWh for space heating and domestic hot water production, accounting for 40% of the total final energy demand (Kemmler et al., 2014). The Swiss Energy Strategy 2050 targets to reduce the demand of the Swiss building stock by 63%, while increasing the share of renewable energy (Kirchner, Bredow, & Ess, 2012). As put forward by Mavromatidis, Orehounig, Richner and Carmeliet (2016), strategies for drastic reduction of the associated CO2 emissions imply reduction of the heat demand per square meter and simultaneous increase of the share of renewables. The way to progress most effectively along these two axis requires answering preliminary questions, as for example: Where and when do we consume heat in buildings? What buildings should be retrofitted in priority? How large are the

expected savings? Where is renewable heat available? In relation with demand, what zones are adequate for the deployment of district heating? In this regard, a general finding is that spatial and temporal characterisation of energy demand is a valuable input for drawing roadmaps leading to the decarbonisation of our energy mix.

Increasing computing power and data storage capacities combined with big data analysis techniques contribute to enlarge the scope of problems which are tackled by way of geographic information systems (GIS). GIS databases to assess energy demand and renewable resources are relevant at several territorial scales. Studies at regional scale use building level demand as input for their models (Saner, Vadenbo, Steubing, &

Hellweg, 2014; Orehounig, Mavromatidis, Evins, Dorer, & Carmeliet, 2014). The potential of combining thermal and PV solar panels is assessed with a GIS analysis in an urban and a suburban district (Quiquerez, Faessler, Lachal, Mermoud, & Hollmuller, 2015). At Swiss national scale Eicher, Pauli, Sres and Nussbaumer (2014) use GIS heat demand models aggregated at hectare raster. At a much bigger scale, as the European union, Connolly et al. (2013) or Persson, Möller and Werner (2014) use such a model with a one square kilometre raster grid.

Swan and Ugursal (2009) outline a clear classification of available models to estimate the heat demand of a building stock. For a GIS characterisation, bottom-up models are an adequate choice because in contrary to top down approaches they estimate the energy demand of small consumer aggregates, as for example buildings, which can be linked to geographical coordinates. Bottom-up models are split into engineering and statistical models. Engineering models require detailed information on the building geometries as well as on the envelope components and associated U and g values. This turns out prohibitive at national level, as well in terms of access to such data, as in terms of computational time. To overcome this difficulty, default values for unknown parameter may be used (Perez, Kämpf, Wilke, Papadopoulou, & Robinson, 2011; Perez, 2014). A simplified approach can be used at larger scale, when buildings have relatively little descriptive information (Kämpf & Robinson, 2007). So as to overcome the lack of information on detailed building characteristics, statistical models use a smaller set of explanatory parameters. Regression models estimate these

(6)

parameters using calibration samples of buildings with known energy consumption. An advantage of basing the model on measured energy demand is that it will include the influence of the users (inhabitants and facility manager), which often leads to a difference between calculated and real consumption, even in cases where the characteristics of the building are known in detail (De Wilde, 2014). Significant

differences between projected and actual demand are also observed on large samples of residential buildings (Majcen, Itard, & Visscher, 2013). This difference, often called

“performance gap”, is also observed in the case of retrofit operations (Khoury, Hollmuller, & Lachal, 2016).

As a response to preceding state of the art, this paper presents a statistical regression bottom up model for the spatial characterisation of the final energy demand for space heating and domestic water production, for the Swiss building stock. It is based on measured heat demand and heated surface values of a representative set of around 27’000 buildings, which is used for estimation of the heat demand of each building of the Swiss national building register.

As an improvement to previous work such as Eicher, Pauli, Erb, & Gutzwiller (2011), our approach bears following advantages: i) the relation between heated floor surface and gross surface and/or dwelling surface is analysed in more details; ii) space heating demand is differentiated by category and construction period; iii) within each category, the observed demand variability is used for determination of confidence intervals around the estimated values.

Section 2 presents the methodology for estimating the floor specific heat demand of each building as a function of its category, age and location, and its heated surface as a function of the category, gross surface and/or dwelling surface. Several subsections treat the following aspects: datasets used to calibrate the model;

classification of buildings into categories; estimation of heated floor surface and of final energy at building and pixel level; quantification of the uncertainty inherent to the use of average demand values per age and category in relation with the level of spatial aggregation. The resulting GIS database is used to produce heat demand map for pixels of several sizes and is shared by way of a web-service.

In section 3 we compare the aggregated final energy demand at national level with other existing statistics. We also perform a statistical analysis of the model errors and a numerical validation of the bootstrap confidence intervals. Finally, we use the resulting database to estimate the energy saving potential which could be achieved if the entire Swiss building stock would undergo deep energy retrofit.

2 Geo-dependent heat estimation model

The general structure of the model is shown in Figure 1. The combination of basic building information from the Swiss national building and dwelling register (GWR) with average statistic indicators from two calibration sets (GEAK, IDC) allows, for each building, to estimate the heated surface AE (m²) and the floor specific final energy

(7)

demand for heat production e (MJ/m²/year), from which we derive the building specific energy demand for heat production Ebld (MJ/year). For each pixel of territory, the estimated energy demand for heat production is summed over all buildings, and a bootstrap algorithm allows to estimate the confidence interval around the average value given by the model.

Figure 1: Model overview

2.1 Datasets

The model uses the following datasets:

(i) GWR (edition 2013); the Swiss national building and dwelling register of the Swiss Federal Statistical Office (OFS, 2014), which contains basic building information such as geographical coordinates, category, age, ground and dwelling surface, as well as the main energy carrier, for around 2 million of buildings. This database is not exhaustive for buildings without residential purpose. A comparison with the Swiss building footprint data layer (Swiss topo, 2015a) shows that around 400’000 buildings are missing. A unique federal building identification number (EGID) identifies each building, allowing to link this information with other databases.

(ii) GEAK (edition 2014); the inter-cantonal building energy certification database, which contains actual final energy demand for heat production (DHW and SH), averaged over three years. Note that this certification scheme is non-compulsory. The database contains 11’500 buildings (about 0.5 % of the building stock), spread over the whole Swiss territory.

(iii) IDC (edition 2014); the heating index database of the Canton of Geneva, which contains the measured final energy consumption as well as the heated floor surface AE for multi-family residential buildings and buildings used for non-residential purposes, as calculated in accordance with the standard SIA 416/1 (SIA, 2007). The declaration of these values by the building owners is mandatory, and is updated each year by a network of trained agents. The database (SITG, 2014) contains around 16’000 buildings, on a total of around 48’000 buildings for the Canton.

(8)

2.2 Building categories

For the sake of our study, the buildings of the GWR database are divided in 4 main categories (multi-family residential, single-family residential, non-residential, and mixed-use). The multi-family buildings are further divided in 2 sub-categories (residential only, residential with annex use) and the non-residential and mixed-use buildings are divided in 3 sub-categories (office & commercial, agricultural & industrial, others). Table 1 gives an overview of the number of buildings of each category.

Category Sub category Buildings

MF residential only residential 286'083 14.8%

with annex use 73'906 3.8%

SF residential only res. or with annex use 1'255'909 65.0%

Non residential office & commercial (*) 23'433 1.2%

agricultural & industrial (**) 67'153 3.5%

others 149'860 7.8%

Mixed use office & commercial (*) 13'301 0.7%

agricultural & industrial (**) 5'060 0.3%

others 56'888 2.9%

Total 1'931'818 100.0%

(*) inc. hotels, schools, hospitals (**) inc. depositories, garages, museums

Table 1: Building categories and number of buildings

2.3 Estimation of heated surface

For each building of the GWR database, the heated floor surface AE is estimated by way of the gross surface (number of floors x ground surface), or the dwelling surface:

𝐴_𝐸 = 𝑓_{𝑔𝑟𝑜𝑠𝑠}∙ 𝐴_{𝑔𝑟𝑜𝑠𝑠} (1)

𝐴_𝐸 = 𝑓_{𝑑𝑤𝑒𝑙𝑙}∙ 𝐴_{𝑑𝑤𝑒𝑙𝑙} (2)

The fgross and fdwell coefficients, which depend on the building sub-category, are the slopes of linear regressions on the data of the IDC dataset. As an example, Figure 2 shows the two linear regressions for the MF residential buildings. The entire set of regression coefficients are given in Table 2. Note that: (i) in the GEAK database the building address is not provided, so that AE could not be linked to the gross and dwelling surface of the GWR, reason why it is not used in this part of the process; (ii) the fdwell coefficient can obviously not be derived for the non-residential buildings, and has very low significance (low R² value) in the case of mixed-use buildings; (iii) when estimating AE for a given building of the GWR database, we use in priority fgross and only use fdwell if the gross surface of the building is not available.

(9)

Category Sub category fgross R² N fdwell R² N

MF residential only residential 0.890 0.973 5103 1.275 0.952 4591 with annex use 0.845 0.962 2639 1.406 0.889 2350 SF residential only res. or with annex use 0.732 0.936 462 1.398 0.963 313 Non residential office & commercial (*) 0.787 0.917 282

Agricultural & industrial (**) 0.681 0.920 134

others 0.702 0.921 29

Mixed use office & commercial (*) 0.636 0.894 313 1.847 0.189 161 agricultural & industrial (**) 0.664 0.906 31 12.428 0.553 27

others 0.631 0.889 57 2.337 0.232 42

*) inc. hotels, schools, hospitals **) inc. depositories, garages, museums R²: coefficient of determination of the linear regression N: number of buildings

Table 2: Relation coefficients between heated floor surface and gross surface (fgross) and between heated floor surface and dwelling surface (fdwell), as observed on the IDC database

Figure 2: Heated floor area AE as a function of gross surface Agross and Adwel for multifamily buildings

Finally, for each building of the IDC database and for both regressions, we compute the relative error AE (%) between the effective and the estimated heated surface:

∆𝐴_𝐸 =^𝐴^𝐸_𝑓^−𝑓^{𝑔𝑟𝑜𝑠𝑠}^∙𝐴^{𝑔𝑟𝑜𝑠𝑠}

𝑔𝑟𝑜𝑠𝑠∙𝐴_{𝑔𝑟𝑜𝑠𝑠} (3)

∆𝐴_𝐸 =^𝐴^𝐸_𝑓^−𝑓^{𝑑𝑤𝑒𝑙𝑙}^∙𝐴^{𝑑𝑤𝑒𝑙𝑙}

𝑑𝑤𝑒𝑙𝑙∙𝐴_{𝑑𝑤𝑒𝑙𝑙} (4)

These values, which are stored per building sub-category, will be used further down for the calculation of confidence intervals.

(10)

2.4 Estimation of final energy demand for heat production

In this section, the floor specific final energy for heat production Ebld is estimated for each building of the GWR database, by way of its age and category, on hand of average values computed on hand of the IDC and GEAK datasets, which are normalized on a common reference.

2.4.1 Normalization of calibration data to a common reference

Since the GEAK and IDC calibration data concerns a variety of energy carriers and locations, it has first to be normalized on a common reference. Therefore, for each building of these datasets, the measured final energy demand e (MJ/m²/year) is converted into useful heat demand for space heating:

𝑞_𝑠ℎ = 𝜂𝑒 − 𝑞_𝑑ℎ𝑤 (5)

For each of the energy carriers, the efficiency of the heat production 𝜂 is estimated by way of standard values, as used in a previous study (Khoury, 2014), which are summarized in Table 3.

Energy carrier 

Billed district heat 1.00

Direct electric 1.00

Distributed heat network 1.00 Electric driven heat pump 2.50

Firewood 0.85

Heating oil 0.83

Natural gas 0.90

Other 0.90

Unknown 0.90

Wood chips 0.85

Wood pellet 0.85

Table 3. Estimated efficiencies of the heat production system. Source (Khoury, 2014)

The useful heat demand for DHW production qdhw is estimated by way of the standard SIA 380/1 (SIA, 2009), which provides typical average values qdhw,SIA per building category.

For buildings with thermal solar panels, qdhw only accounts for the fraction of the DHW demand which is produced by the main heat production system. In such a case, using qdhw,SIA in equation (5) would overestimate qdhw and underestimate qsh. To avoid unrealistic qsh values we add the following constraint, which implies that at least one third of total produced heat is dedicated to space heating.

𝑞_𝑑ℎ𝑤 = min(𝑞_{𝑑ℎ𝑤,𝑆𝐼𝐴}, 2/3 𝜂𝑒) (6)

(11)

This thumb rule sets an upper limit for the ratio between qdhw and qsh that corresponds to very efficient buildings. Note that this choice will have limited influence on the final result since it only impacts the fraction qsh of e that is subject to the climatic correction used below. So that the various qsh values of the IDC and GEAK buildings can be used for calibration purposes, they are further normalized on a common climatic reference (arbitrarily Geneva), by way of the standard degree days of the associated weather station:

𝑞_{𝑠ℎ,𝑟𝑒𝑓} =^𝐷𝐷_𝐷𝐷^𝑟𝑒𝑓𝑞_𝑠ℎ (7)

With this algorithm, the SITG and GEAK datasets allow to compute 15’588 and 11’499 qsh,ref calibration values.

Kemmler (*) Heeren (**) Model (***)

MF residential n.a 270 321

SF residential n.a 230 259

Total 509 500 580

Mixed use n.a n.a 63

Non residential n.a n.a 59

Total 245 122

Total 754 702

*) (Kemmler et al., 2014) **) (Heeren et al., 2009) ***) Current study

Table 4: Comparison of total heated surface AE with other studies

2.4.2 Statistical analysis per building category and age

For the sake of statistical analysis, preceding data is grouped into the four main building categories of Table 1 (multi-family residential, single-family residential, non-residential, mixed-use). For each of these categories, the data is further divided into 12 construction periods, and analyzed in terms of statistical distribution (Figure 3 for the IDC dataset,

(12)

Annex 1 for the GEAK dataset).

Figure 3: Boxplot distribution of space heating demand qsh,ref by construction period (IDC database)

At this stage, it is worthwhile noticing that the variability within each construction period is much higher than between the periods, as previously put forward by Khoury (2014).

Finally, for each of the building categories and construction periods, we define the average space heating demand which will be used in the extrapolation process of the next section:

𝑞_{𝑠ℎ,𝑎𝑣𝑔}=^{∑ 𝑞}^{𝑠ℎ,𝑟𝑒𝑓}_{∑ 𝐴} ^𝐴^𝐸

𝐸 (8)

This analysis is done separately for the SITG and GEAK datasets, the respective values being used for reconstruction of the entire GWR building stock located in Geneva, respectively in the rest of the Swiss territory.

2.4.3 Final energy for heat production

For each building of the GWR database, the floor specific final energy for heat production is estimated by: i) the average SH demand of its corresponding category and construction

(13)

period, taking into account the climatic correction of the building location; ii) the corresponding DHW demand; iii) the efficiency of the heat production system of the building:

𝑒 = (_𝐷𝐷^𝐷𝐷

𝑟𝑒𝑓𝑞_{𝑠ℎ,𝑎𝑣𝑔}+ 𝑞_𝑑ℎ𝑤) 𝜂⁄ (9)

In preceding equation the DD must be estimated for each building of the GWR database.

Therefore we use a linear regression that fits the DD of all Swiss reference meteorological climate stations, as function of their altitude and latitude. Figure 4 compares the modelled and the real DD values, as given by the SIA 2028 standard (SIA, 2010). The coefficient of determination R² reaches 0.95, while the maximum and average relative differences are 14% and 4%. So as to use this model for the buildings of the GWR database, their altitude is calculated by overlapping with the Swiss numeric elevation model (Swiss topo, 2015b).

Figure 4: Model DD versus SIA 2028 DD for all Swiss meteorological stations

2.4.4 Geo-dependent database and confidence intervals

For each building of the GWR database, the estimated building specific final energy for heat production Ebld is derived from AE and e:

𝐸_𝑏𝑙𝑑 = 𝐴_𝐸 𝑒 (10)

(14)

By summing up of the Ebld values of all buildings contained in each pixel, we generate a GIS database for all pixel of the Swiss territory:

𝐸_{𝑝𝑖𝑥𝑒𝑙} = ∑ 𝐸_𝑏𝑙𝑑 = ∑ 𝐴_𝐸 𝑒 (11)

The procedure is repeated for various pixels sizes, ranging from 100 x 100 m to 1.6 x 1.6 km.

For each of the above pixels, we compute a confidence interval around the estimated heat demand value. It is computed by way of a bootstrap resampling algorithm (Efron & Tibshirani, 1993), which consists in generating a heat demand distribution that replicates the dispersion of the calibration datasets. Therefore, for each building of the pixel, the heated surface AE is replaced by (1 + ∆𝐴_𝐸)𝐴_𝐸, where AE is a randomly picked relative error of the corresponding building category. Similarly, a new e value is generated by replacing the qsh,avg of equation (9) by a randomly picked value of the corresponding calibration dataset of same building category and construction period.

The replicated AE and e values are fed into equations (10) and (11) generating a replicated heat demand of the entire pixel. Repeating of this procedure a 1000 times per pixel generates a heat demand probability distribution of each pixel. This allows to compute a 10% level confidence interval, defined by the 5% and 95% percentiles of this distribution. A similar procedure computes the 5% and 1% confidence intervals. Lastly we apply a bias correction to this simulated distribution to correct the bias induced by the difference between the average of the randomly picked qsh values and the surface weighted average qsh,avg.

2.5 Web service and GIS maps

The results of the model, which cover the entire Swiss territory, are made available by way of a dedicated web service (Schneider, Assouline, Mohajeri, & Scartezzini, 2015).

The delivered data contains for each pixel the estimated final energy demand and the total estimated heated surface together with confidence intervals for several confidence levels.

On the basis of these results, we finally build a series of heat demand GIS maps, with more or less aggregated information, depending on the zoom level. Figure 5 shows an example of such a heat density map for an urban district. A color ramp indicates the final energy consumption per m² of territory, for each pixel of 200 x 200 m.

(15)

Figure 5: Sample map of final energy for heat production, district of Zurich (MJ per m² of territory, 200 x 200 m pixel).

2.6 Statistical distribution at cantonal or national scale

As has just been seen for the above described GIS database, when territorializing data at pixel level, the model uses average values (depending on building category and age), and the integration of the statistical distribution is handled by way of an associated confidence interval.

When working at an integrated and non-localized level, as for example at national or cantonal scale, the model can also be used for generating a statistical distribution of building heat demand (in terms of final energy e, or of SH demand qsh). In this case, for each building of the GWR database in the territory under consideration, qsh is randomly picked in the corresponding calibration dataset and corrected in terms of degree days of the building location, see equation (7), while e is generated by replacing the qsh,avg of equation (9) by the randomly picked qsh value. Similarly, a replicated AE is calculated by way of (1 + ∆𝐴_𝐸)𝐴_𝐸 and a randomly picked AE value.

3 Results and discussion

3.1 Heat demand at national level

Around 90% of the 1.93 million buildings listed in the 2013 version of the GWR have an indication on ground surface and/or dwelling surface. For these buildings sufficient

(16)

data is available to apply the appropriate regression model of Table 2. Summing up all estimated AE values for the Swiss building stock leads to a total heated surface

estimation of 702 million m². In Table 4, these figures are compared with two other studies (Kemmler et al., 2014; Heeren, Gabathuler, & Wallbaum, 2009).

Construction periods Total / (%)

(All constr. periods) Before 1945 1945-1980 After 1980 Unknown

Total number of buildings in FRBD

MF 110'436 127'381 118'319 61 356'197 (18%)

SF 384'627 405'943 464'848 594 1'256'012 (65%)

MU 39'175 24'003 15'906 79 79'163 (4%)

NR 38'841 46'961 51'000 103'644 240'446 (12%)

Total 573'079 604'288 650'073 104'378 1'931'818 Nb. of buildings with heated surface estimation

MF 109'428 125'832 116'987 47 352'294 (20%)

SF 379'976 402'058 460'062 527 1'242'623 (71%)

MU 37'605 23'023 15'090 58 75'776 (4%)

NR 20'130 26'579 23'216 9'979 79'904 (5%)

Total 547'139 577'492 615'355 10'611 1'750'597 Total heated surface AE (millions of m²)

MF 69.64 121.92 129.08 0.04 320.7 (46%)

SF 88.71 75.17 94.91 0.07 258.9 (37%)

MU 22.38 22.25 18.20 0.04 62.9 (9%)

NR 11.58 20.77 23.31 3.62 59.3 (8%)

Total 192.3 240.1 265.5 3.8 701.7

Total final energy demand for space heating and DHW (TWh/year)

MF 10.36 18.41 12.77 0.00 41.5 (44%)

SF 13.42 12.40 9.48 0.01 35.3 (38%)

MU 4.17 3.77 2.62 0.01 10.6 (11%)

NR 1.33 2.68 1.89 0.37 6.3 (7%)

Total 29.3 37.2 26.8 0.4 93.7

MF: Multi family residential MU: Mixed use SF: Single family residential NR: Non residential

Table 5: Aggregated results of GIS heat demand bottom-up model

In the case of the residential buildings, for which the total dwelling surface amounts to 410 million m²according to the GWR, the total heated surface of our model (580 million m²) is significantly higher than the one of the other studies (509 and 500 million m² see Table 4). Note that latter estimations lead to an average fdwell of 1.2, i.e. much lower than the ratios used in this study (see Table 2), which seem rather plausible since they are based on a large number of observations. In the case of the remaining

buildings, the total heated surface of our model (122 million m²) is significantly lower than the one of Kemmler et al. (2014) (245 million m²). As pointed out before, this is mainly due to the missing buildings in the GWR database.

(17)

Table 5 contains the number of buildings in the GWR (total and those for which the heated surface could be estimated), the heated surface and the final energy demand for a reference climatic year, per building type and construction period. For the year 2013, the total aggregated final energy demand sums up to 93.7 TWh/year (76.8 TWh/year for residential buildings and 16.9 TWh/year for mixed use and non-

residential ones). Kemmler et al. (2014) estimate this demand to 91.1 TWh/year, (59.5 TWh/year for residential buildings and 31.6 TWh/year for mixed use and non-

residential ones) for the year 2013. Although the two models have a good concordance for global demand, significant differences exist when considering the residential and non-residential sectors separately, mainly due to the differences between the heated surfaces.

Building sector Energy carrier

Residential Entire stock

(Model*) (Kemmler**) (Model*) (SFOE***)

Coal 0.1 0.1 0.2 0.1

Electricity 4.0 5.2 4.4 7.6

Other renewable 2.3 1.9 2.5 4.8

District heating 2.7 2.2 4.1 5.0

Wood 8.3 5.5 9.3 11.0

Gas 14.4 13.8 18.1 16.5

Oil 44.0 27.9 52.2 45.8

Other 1.0 3.0 2.9 2.9

Total (TWh) 76.8 59.5 93.7 93.8

*) Current study **) (Kemmler et al., 2014) ***) (SFOE, 2013) assuming that 13% of electricity and 50% of Gas is used for SH and DWH.

Table 6: Comparison of total final energy use by carrier

Figure 6 gives an insight of the share of final energy consumption per building type and construction period. Of the final energy consumed by the Swiss building stock, 82% are due to residential buildings (with approximately even shares between SF and MF); for all categories, 71% are due to buildings constructed before 1980 (which also have the highest specific heat demand per square meter).

Figure 6: Final energy demand by building type and construction period

(18)

3.2 Comparing per energy carrier

The GWR indicates for each building the main energy carrier used for space heating.

Unfortunately it is not known if this information is up to date and if the same carrier is used also for DHW production. For this reason our model may underestimate the used electricity, since in some cases DHW is produced by an auxiliary electric device.

While bearing this uncertainty in mind, it is possible to use this information to estimate aggregated demand per carrier and to compare it with other existing

estimations based on national energy balances statistics.

Canton Demand [GWh] Model* [GWh] 

LU (Residential only) ** 2'996 3'219 7.4%

TI (Residential only) *** 3'778 3'214 -14.9%

VS (all categories) **** 4'240 4'811 13.5%

*) Current study **) (Kulawik and Bucher 2013) ***) (Pampuri et al. 2017)

****) (Ruiz and Pernet 2016)

Table 7: Comparison of final energy demand at regional level

Table 6 compares the total amount of final energy per carrier for the residential sector only and for the whole building stock. For the residential building sector the estimation of Prognos (Kemmler et al., 2014) is significantly lower than ours. This is certainly a consequence of the difference on heated surface estimations of Table 4. Surprisingly this mismatch concerns mainly the oil carrier.

Since Kemmler et al. (2014) does not provide a similar decomposition for the entire building stock, we compare our model to the national statistics of final energy consumption (SFOE, 2013), which is available per energy carrier and main activity sectors (residential, industry, services and mobility). To estimate the share of final energy used for SH and DHW, we use the following assumptions: (i) of the 33 TWh of final gas consumption, 50% are used for SH and DHW, which corresponds to the share of gas used by the residential sector according to Figure 5 of SFOE (2013); (ii) of the 58 TWh of final electricity consumption, 13% are used for SH and DHW, which

corresponds to the preliminary conclusions of the ElectroWhat model (Schneider, Le Strat, & Patel, 2017); (iii) all 45.8 TWh of heating oil is used for SH and DHW. These assumptions lead to a total demand of 93.8 TWh which is close to the one of our model.

Furthermore there is also a good match with the decomposition per energy carrier.

Again at overall level our estimation is in line with national statistics, but the decomposition per building category or per energy carrier remains subject to discussion, due to uncertainties in our model (energy carrier not up to date, incomplete

nonresidential building stock, non-differentiated main and secondary residences), but probably also to uncertainties when dealing with existing statistics, in matching between energy carriers and energy services.

(19)

Finally, since the initial purpose of the model is to generate a GIS heat demand atlas, the disaggregation of total final energy consumption per energy carrier is not pushed further in this study.

3.3 Statistical analysis of the model error

The purpose of this section is to test the robustness of the confidence interval around the estimated demand at pixel level and to analyze the relation between model error and aggregation level. For this sake we use the data available within the SITG information system (SITG, 2015), for which 206 pixels of 200 by 200 (m) contain the actual demand of all buildings.

For all of these pixels we check whether the actual demand is between the lower and upper bounds of the confidence interval as computed in section 2.4.4. For three computed confidence level α (10%, 5% and 1%), Table 8 shows the proportion of the pixels for which the actual value is outside the confidence interval, which turn out to be close to 𝛼, comforting that the bootstrap algorithm is adequate. The slightly higher values could be due to the limited number of observations.

Confidence level

NPixel  = 10%  = 5%  = 1%

206 15.5% 8.3% 2.4%

Table 8: Proportion of pixels for which the actual demand is outside the confidence interval of level 

As a next step, in Figure 7, we analyze the dependence between the model error |Δ𝐸_{𝑃𝑖𝑥𝑒𝑙}| and the number of buildings in a pixel, where |Δ𝐸_{𝑃𝑖𝑥𝑒𝑙}| is given by the difference between the estimated and the actual aggregated demand 𝐸_{𝑃𝑖𝑥𝑒𝑙} , in absolute value. As expected, the error decreases with the number of buildings in a pixel, since over and underestimations compensate with increasing aggregation level. This decrease is inversely proportional to the square root of the number of buildings, as predicted by the central limit theorem.

(20)

Figure 7: Relation between average model error and number of buildings within a pixel

3.4 Energy saving potential of the Swiss building stock

As seen in Figure 6, at national level, 71% of the final energy demand for SH and DHW production relates to buildings built before 1980. These buildings, which are about to enter into a retrofit cycle, are also the ones with the highest e values. This shows that retrofit has a major role to play for reducing this demand and achieving the goals of the 2050 energy strategy of the Swiss Federal council. The purpose of this section is to estimate the achievable heat demand saving potential of the Swiss building stock.

3.4.1 Realistic versus normed potential

As shown by Khoury et al. (2016), there is an important performance gap between the normed and the actual heat demand savings achieved by building retrofit.

The normed potential for space heating is defined as the difference between the specific heat demand before retrofit (qsh) and the limit value to be achieved by retrofit in normed use conditions (qsh, norm).

Δ𝑞_{𝑠ℎ,𝑛𝑜𝑟𝑚} = 𝑞_𝑠ℎ− 𝑞_{𝑠ℎ,𝑛𝑜𝑟𝑚} (12)

The latter value is calculated according to standard SIA 380/1 (SIA, 2009).

The realistic (net) saving potential is defined as the difference between the specific heat demand before retrofit (qsh) and the actual space heat demand after retrofit (qsh,net).

Δ𝑞_{𝑠ℎ,𝑛𝑒𝑡} = 𝑞_𝑠ℎ− 𝑞_{𝑠ℎ,𝑛𝑒𝑡} (13)

The analysis of ten representative case studies (Khoury et al., 2016) of recently retrofitted post war MF residential buildings allowed to characterize the performance gap by way of a statistical correlation between Δ𝑞_{𝑠ℎ,𝑛𝑜𝑟𝑚} and Δ𝑞_{𝑠ℎ,𝑛𝑒𝑡}.

(21)

Δ𝑞_{𝑠ℎ,𝑛𝑒𝑡} = 0.0009 (Δ𝑞_{𝑠ℎ,𝑛𝑜𝑟𝑚})²+ 0.17 Δ𝑞_{𝑠ℎ,𝑛𝑜𝑟𝑚} if Δ𝑞_{𝑠ℎ,𝑛𝑜𝑟𝑚} < 600 MJ/m² (14) This relation (R²=0.99) takes into account the entire retrofit process, from the design stage (choice of solutions and use of simulation software) all the way through to the use of the buildings by the occupants and the operation by the energy managers. It can be viewed as a characterization of the current retrofit and operation practices.

The identification of the most determinant factors and reasons behind the performance gap in building retrofit was carried out by Khoury, Alam Eddine and Hollmuller (2017) to examine the main factors behind the energy performance gap and to quantify their relative importance via sensitivity analysis.

3.4.2 Estimation of the normed saving potential

For each building of the GWR database, the estimation of the normed saving potential by the way of equation (12) requires the estimation of qsh and qsh,norm.

The quadratic term of equation (14) implies that qsh cannot be replaced by an average value and that the distribution of qsh values needs to be taken into account.

Therefore qsh is randomly picked in the corresponding calibration dataset as explained in section 2.6.

The limit value to be achieved by retrofit in normed use conditions qsh,norm is calculated according to standard SIA 380/1 (SIA, 2009) and depends on the building category, the geographic location and the building shape factor. The shape factor, which is defined as the ratio between the building envelope area Ath and its energy reference area AE, is taken from the standard SIA 2031 (SIA, 2016) for all the building categories, except for the MF residential buildings. For this category, the factors were estimated on the basis of the number of floors using the relation below that was derived from the GEAK database.

𝐴_𝑡ℎ 𝐴_𝐸

⁄ = 0.32 + 1.97

√𝑁𝑓𝑙𝑜𝑜𝑟𝑠

⁄ (15)

Finally the normed saving potential is computed using equation (12) and summed up over the entire building stock. The results are shown in

Figure 8 for the entire building stock and in Table 9 per building category. If all the buildings undergo deep energy retrofit according to SIA standard 380/1 (SIA, 2009), the normed saving potential for space heating is estimated to Qsh,norm= 38.1 TWh/year.

When focusing on the multi-family residential sector, this potential amounts to 16.2 TWh/year, representing almost half (42.5%) of the total normed potential. The rest is distributed as follows: 37% for one and two family houses; 12% for mixed use buildings and 8% for non-residential buildings.

(22)

3.4.3 Estimation of the realistic saving potential

For each building, the realistic (net) saving potential value is estimated by way of equation (14). For buildings with Δ𝑞_{𝑠ℎ,𝑛𝑜𝑟𝑚} above 600 MJ/m², outside of the building sample analyzed by Khoury et al. (2016), a more conservative value is used:

Δ𝑞_{𝑠ℎ,𝑛𝑒𝑡} = 0.7 Δ𝑞_{𝑠ℎ,𝑛𝑜𝑟𝑚} if Δ𝑞_{𝑠ℎ,𝑛𝑜𝑟𝑚}≥ 600 MJ/m2 (16) When summing up over the entire Swiss building stock, the realistic (or net) saving potential for space heating amounts to Qsh,net = 18.4 TWh/yr

Figure 8, horizontal hatched area). This represents almost half (48%) of the normed saving potential. Table 9 gives the normed and realistic saving potential for space heating per building category. The achievable fraction defined as the ratio between

Qsh,net and Qsh,norm does not vary much between categories and remains close to 50%.

(23)

Construction periods

Building categories Before 1946 1946 - 1980 After 1980 Total (incl. unknown)

MF residential

Qsh,norm 4.4 7.8 4.0 16.2

Qsh,net 2.1 3.6 1.7 7.4

AF 48% 46% 41% 46%

SF residential

Qsh,norm 5.1 5.3 3.7 14.1

Qsh,net 2.7 2.8 1.6 7.1

AF 52% 52% 45% 50%

Mixed use

Qsh,norm 1.8 1.7 1.0 4.6

Qsh,net 1.0 0.9 0.5 2.4

AF 52% 52% 48% 51%

Non residential

Qsh,norm 0.7 1.4 0.7 3.1

Qsh,net 0.4 0.7 0.3 1.5

AF 50% 51% 44% 49%

Total CH stock

Qsh,norm 12.1 16.3 9.4 38.1

Qsh,net 6.1 8.0 4.1 18.4

AF 51% 49% 44% 48%

Qsh,norm: Normed (gross) saving potential (TWh/yr)

Qsh,net: Realistic (net) saving potential (TWh/yr)

AF: Fraction of achievable potential (Qsh,net/Qsh,norm) (%)

Table 9: Theoretical (normed) and realistic (net) saving potential for space heating by building category

(24)

Figure 8: Retrofit potential of the Swiss building stock for space heating demand: gross potential (total hatched area) and net potential (horizontal hatched area)

4 Conclusions

This paper presents a statistical regression bottom up model for the spatial

characterisation of the final energy for space heating and domestic water production.

For each building of the Swiss national building and dwelling register, the model estimates the heated surface and the floor specific final energy demand for space heating and domestic hot water production, by way of average statistical indicators that were derived from two large calibration sets.

This approach, which is based on the distribution of measured consumptions, bears the advantage of including the variability between buildings of same category and age.

It is hence an alternative to commonly used calculation models based on building physics, which suffer from a lack of accuracy when used at large territorial scale, due to the limited available information on the characteristics of the building components.

For each pixel of territory, the estimated energy for heat production is summed over all contained buildings, and a bootstrap algorithm allows to estimate the confidence interval around the average value given by the model. The resulting database is shared using a web service, as first order input for territorial energy planning.

Note that the presented approach could be adapted to other countries or communities having a GIS database containing basic information of their building

(25)

stock, as well as statistical indicators concerning heated surface and floor specific final energy demand for heat production.

At national level, the total aggregated final energy demand sums up to 93.7 TWh/year (76.8 TWh/year for residential buildings and 16.9 TWh/year for mixed use and non-residential ones). The global demand is in good concordance with existing statistics at national level, but significant differences exist for the demand when considering residential and non-residential sectors separately.

The decomposition per building category or per energy carrier remains subject to discussion, due to uncertainties in our model (energy carrier not up to date, incomplete nonresidential building stock, non-differentiated main and secondary residences), but probably also to uncertainties when dealing with existing statistics, in matching between energy carriers and energy services. Since the initial purpose of the model is to generate a GIS heat demand atlas, the disaggregation of total final energy consumption per energy carrier is not pushed further in this study.

For a portion of territory for which the actual heat demand is known at building level (Canton of Geneva), we tested the robustness of the confidence interval around the estimated demand for three confidence levels (10%, 5% and 1%), for pixels of 200 by 200 m. In all three cases, the proportion of pixels for which the actual heat demand lies outside the computed confidence interval turns out to be close to the respective

confidence level, comforting that the bootstrap algorithm is adequate. Furthermore, we analyze the model error as a function of the number of buildings within a pixel. As expected, the error decreases inversely proportional to the square root of the number of buildings, as predicted by the central limit theorem.

Finally, since retrofit has a major role to play for reducing this demand and achieving the goals of the 2050 energy strategy of the Swiss Federal council, we estimate the achievable heat demand saving potential of the Swiss building stock. For this sake we reach back on previous results concerning the performance gap between the normed and the actual heat demand savings achieved by building retrofit, which reflects the current retrofit and operation practices. When applying the results of latter study to our model, we estimate that if the entire Swiss building stock would undergo deep energy retrofit, the normed saving potential for space heating would amount to 38.1 TWh/year. When taking into account the performance gap, the realistic (or net) saving potential for space heating reduces to 18.4 TWh/year, i.e. almost half of the normed saving potential.

5 Acknowledgements

This research is part of the activities of SCCER FEEB&D, which is financially

supported by the Swiss Commission for Technology and Innovation (CTI). The authors thanks the Swiss Federal Statistical Office for the free access to the GWR database. We also thank Professor Martin Kumar Patel of the University of Geneva for giving us access to CECB database (exchange of information within the scope of a collaboration with SCCER CREST).

(26)

6 References

Connolly, D., Lund, H., Mathiesen, B. V., Werner, S., Möller, B., Persson, U., … Nielsen, S. (2013). Heat Roadmap Europe: Combining district heating with heat savings to decarbonise the EU energy system. Energy Policy, 65(0), 475–489.

doi:10.1016/j.enpol.2013.10.035

De Wilde, P. (2014). The gap between predicted and measured energy performance of buildings: A framework for investigation. Automation in Construction, 41, 40–

49. doi:10.1016/j.autcon.2014.02.009

Efron, B., & Tibshirani, R. (1993). An introduction to the bootstrap. New York:

Chapman & Hall.

Eicher, H., Pauli, H., Erb, M., & Gutzwiller, S. (2011). Ausbau von WKK in der Schweiz WKK-Standortevaluation auf Basis einer GIS-Analyse. Dr.

Eicher+Pauli AG planer für Energie und Gebäude Technik.

Eicher, H., Pauli, H., Sres, A., & Nussbaumer, B. (2014). Weissbuch Fernwärme Schweiz – VFS Strategie, Langfristperspektiven für erneuerbare und

energieeffiziente Nah- und Fernwärme in der Schweiz. Dr. Eicher+Pauli AG planer für Energie und Gebäude Technik.

Heeren, N., Gabathuler, M., & Wallbaum, H. (2009). Gebäudeparkmodell SIA Effizienzpfad Energie Dienstleistungs- und Wohngebäude. Vorstudie zum Gebäudeparkmodell Schweiz Grundlagen zur Überarbeitung des SIA Effizienzpfades Energie. Zurich: TEP Energy GmbH, Zurich,Switzerland.

Kämpf, J. H., & Robinson, D. (2007). A simplified thermal model to support analysis of urban resource flows. Energy and Buildings, 39(4), 445–453.

doi:10.1016/j.enbuild.2006.09.002

Kemmler, A., Piégsa, A., Ley, A., Wüthrich, P., Keller, M., Jakob, M., & Catenazzi, G.

(2014). Analyse des schweizerischen Energieverbrauchs 2000 - 2013 nach Verwendungszwecken. Bundesamt für Energie Bern.

Khoury, J. (2014). Rénovation énergétique des bâtiments résidentiels collectifs: état des lieux, retours d’expérience et potentiels du parc genevois. Université de Genève, Thèse de doctorat N^o 4752, direction Bernard Lachal. Retrieved from

http://archive-ouverte.unige.ch/unige:48085

Khoury, J., Alam Eddine, Z., & Hollmuller, P. (2017). Understanding and bridging the energy performance gap in building retrofit (Submitted). In CISBAT 2017.

Presented at the CISBAT 2017, EPFL Lausanne, Switzerland.

Khoury, J., Hollmuller, P., & Lachal, B. M. (2016). Energy performance gap in building retrofit : characterization and effect on the energy saving potential. In 19. Status- Seminar. Presented at the 19. Status-Seminar «Forschen für den Bau im Kontext von Energie und Umwelt», Zurich: ETHZ. Retrieved from http://archive-

ouverte.unige.ch/unige:86086

Kirchner, A., Bredow, D., & Ess, F. (2012). Die Energie Perpectiven für die Schweiz by 2050. Prognos AG.

Majcen, D., Itard, L., & Visscher, H. (2013). Actual and theoretical gas consumption in Dutch dwellings: What causes the differences? Energy Policy, 61, 460–471.

doi:10.1016/j.enpol.2013.06.018

(27)

Mavromatidis, G., Orehounig, K., Richner, P., & Carmeliet, J. (2016). A strategy for reducing CO2 emissions from buildings with the Kaya identity – A Swiss energy system analysis and a case study. Energy Policy, 88, 343–354.

doi:10.1016/j.enpol.2015.10.037

OFS. (2014). Eidgenössisches Gebäude- und Wohnungsregister,Web

Services,Technisches Dossier zum Datenaustausch via WebServices.

Bundesamt für Statistik BFS.

Orehounig, K., Mavromatidis, G., Evins, R., Dorer, V., & Carmeliet, J. (2014). Towards an energy sustainable community: An energy system analysis for a village in Switzerland. Energy and Buildings, 84(0), 277–286.

doi:10.1016/j.enbuild.2014.08.012

Perez, D. (2014). A framework to model and simulate the disaggregated energy flows supplying buildings in urban area. EPFL, THÈSE No 6102 (2014). Retrieved from http://infoscience.epfl.ch/record/197073

Perez, D., Kämpf, J. H., Wilke, U., Papadopoulou, M., & Robinson, D. (2011).

CitySimsimulation: the case study of Alt-Wiedikon, a neighbourhood of Zürich City. In Proceedings of CISBAT 2011 - CleanTech for Sustainable Buildings, pp. 937–940.

Persson, U., Möller, B., & Werner, S. (2014). Heat Roadmap Europe: Identifying strategic heat synergy regions. Energy Policy, 74(0), 663–681.

doi:10.1016/j.enpol.2014.07.015

Quiquerez, L., Faessler, J., Lachal, B. M., Mermoud, F., & Hollmuller, P. (2015). GIS methodology and case study regarding assessment of the solar potential at territorial level: PV or thermal? doi:10.5278/ijsepm.2015.6.2

Saner, D., Vadenbo, C., Steubing, B., & Hellweg, S. (2014). Regionalized LCA-Based Optimization of Building Energy Supply: Method and Case Study for a Swiss Municipality. Environmental Science & Technology, 48(13), 7651–7659.

doi:10.1021/es500151q

Schneider, S., Assouline, D., Mohajeri, N., & Scartezzini, J.-L. (2015). Geo-dependent energy demand and supply Web-Service. Retrieved November 27, 2015, from http://wisesccer1.unige.ch/

SFOE. (2013). Schweizerische Gesamtenergiestatistik 2013 (No. 805.006.13). CH-3003 Bern: Swiss Federal Office of Energy. Retrieved from

http://www.bfe.admin.ch/themen/00526/00541/00542/00631/index.html?lang=fr

&dossier_id=00763

SIA. (2007). Norme SIA 416/1:2007. Indices de calcul pour les installations du bâtiment. Société suisse des ingénieurs et des architectes (SIA).

SIA. (2009). Norme SIA 380/1:2009. L’énergie thermique dans le bâtiment. Société suisse des ingénieurs et des architectes (SIA).

SIA. (2010). Norme SIA 2028: Données climatiques pour la physique du bâtiment, l’énergie et les installations du bâtiment. Société suisse des ingénieurs et des architectes.

SIA. (2016). Norme SIA 2031: Certificat énergétique des bâtiments. Société suisse des ingénieurs et des architectes.

(28)

SITG. (2014). Geneva territorial information system (SITG). Retrieved April 14, 2014, from http://ge.ch/sitg/sitg_catalog/sitg_donnees

SITG. (2015). SITG Indice de dépense de chaleur (IDC) des bâtiments (LEn >

5.8.2010) - moyenne sur 2 ou 3 ans. Retrieved June 19, 2015, from http://ge.ch/sitg/sitg_catalog/sitg_donnees/i?keyword=IDC

Swan, L. G., & Ugursal, V. I. (2009). Modeling of end-use energy consumption in the residential sector: A review of modeling techniques. Renewable and Sustainable Energy Reviews, 13(8), 1819–1835. doi:10.1016/j.rser.2008.09.033

Swiss topo. (2015a). Swiss topo Topographic Landscape Model TLM. Retrieved February 25, 2015, from https://www.swisstopo.admin.ch/de/wissen- fakten/topografisches-landschaftsmodell.html

Swiss topo. (2015b). Swiss topo digital height model DHM25. Retrieved August 25, 2015, from

https://shop.swisstopo.admin.ch/fr/products/height_models/dhm25200

(29)

7 Tables and figures

Table 1. Building categories and number of buildings

Table 2. Relation coefficients between heated floor surface and gross surface (fgross) and between heated floor surface and dwelling surface (fdwell), as observed on the IDC database

Table 3. Estimated efficiencies of the heat production system. Source (Khoury, 2014) Table 4. Comparison of total heated surface AE with other studies

Table 5. Aggregated results of GIS heat demand bottom-up model Table 6. Comparison of total final energy use by carrier

Table 7. Proportion of pixels for which the actual demand is outside the confidence interval of level 

Table 8: Comparison of final energy demand at regional level

Table 9. Normed (gross) and realistic (net) saving potential for space heating by building category

Figure 1. Model overview

Figure 2. Heated floor area AE as a function of gross surface Agross and Adwel for multifamily buildings

Figure 3. Boxplot distribution of space heating demand qsh,ref by construction period (IDC database)

Figure 4. Model DD versus SIA 2028 DD for all Swiss meteorological stations

Figure 5. Sample map of final energy for heat production, district of Zurich (MJ per m² of territory, 200 x 200 m pixel).

Figure 6. Final energy demand by building type and construction period

Figure 7. Relation between average model error and number of buildings within a pixel Figure 8. Retrofit potential of the Swiss building stock for space heating demand: gross potential (total hatched area) and net potential (horizontal hatched area)

(30)

Annex 1: Boxplot distribution of space heating demand qsh,ref by construction period (GEAK database)