Understanding Energy Demand of the Tertiary Sector by Energy Carriers and End-Uses: An Integrated Bottom-Up and Top-Down Model Taking Germany as the Example

Abstract

Despite its high share in national energy demand, the tertiary sector is an undifferentiated residual in the energy balances of most countries. To develop effective policy measures for enhancing energy efficiency, policymakers need a sound understanding of how the energy demand is structured. To date, only a few studies assess energy demand in the tertiary sector quantitatively. This paper aims to close this gap by introducing an integrated model that transparently breaks down the energy demand according to statistical subsectors, energy carriers and end-uses. The model combines a technology-based bottom-up with a top-down approach and applies it to a case study on Germany based on survey data from companies. Model validity is analyzed using a set of indicators. The results for Germany show that the model is consistent with the national energy balance showing an aggregate deviation of only 1%. The results for 2019 indicate that electricity demand in Germany’s tertiary sector is dominated by lighting (35%), information and communication technologies (33%) and the provision of mechanical energy (9%), while fuel demand is mainly due to space heating (78%), followed by mechanical energy (15%).

1. Introduction

Current progress in energy and climate policies is insufficient to achieve the 1.5 °C target for global warming that has been aimed for by the Paris climate agreement. Even limiting global warming to under 2 °C is becoming increasingly more difficult [1]. Therefore, there needs to be a change in the economy, governance and technology toward a sustainable, resource-efficient, climate-neutral and circular world [2,3].

Policymakers set the framework for this course. To define and implement effective policy measures to limit global warming, they need to understand the structure of energy demand. Since detailed data about real-world energy systems is usually limited or missing, policymakers regularly need to rely on knowledge gained from energy system models. Energy system models are specific representations of real-world systems based on detailed assumptions and system boundaries [4,5]. These models have several purposes: In addition to supporting the provision of official energy balances, energy system models may further subdivide energy demand by energy end-use [6]. The models allow for the development of scenarios that describe the outcome of an energy system model run [4]. Also, energy efficiency activities can be understood, and the best technology opportunities and new policies can be identified [6,7]. Another purpose is the investigation of the costs and benefits of policy measures. By adjusting the underlying input and assumptions, the effects and impacts of policy measures and long-term decarbonization scenarios can be analyzed, best strategies can be highlighted, and greenhouse gas emission reduction targets can be set [4,6,8].

Energy system models are often structured by official energy statistics which distinguish the supply side and the demand side. Ideally, the results of such models match the official energy balances, i.e., they have a top-down fit with statistics, and they also explain the elements that constitute energy demand from a disaggregate level, i.e., a bottom-up perspective. In doing so, the bottom-up view adds additional information on the structure of energy generation, energy end-use, improvement measures and finally carbon dioxide reduction options [9,10,11]. A key challenge in doing so is to provide a consistent and transparent fit of the top-down and the bottom-up values.

Demand side models are often distinguished by the sector they address along with the energy balances, i.e., households, industry, transport, a residual and the tertiary or service sector. In official energy balances, the tertiary sector is a rather undifferentiated residual, and data on energy end-use by energy carriers and subsectors is generally not published by statistical offices [12]. From an energy system perspective, the tertiary sector still holds a substantial share in energy demand. In Germany, for example, it had a non-negligible share of 14.7% of the country’s entire energy consumption in 2019 [12,13]. The sector includes the operation of public and private services, trade, commerce, manufacturing, schools, universities, hospitals, hotels and public swimming pools [14]. The definition of the tertiary sector differs between countries, with Germany including agriculture, construction and small industries, for example, in contrast to the EU [15].

Like any other sector, the tertiary sector’s final energy consumption needs to be reduced effectively and sustainably, and thus, transparency on its energy demand, its potential contribution to reducing greenhouse gas emissions and potential levers for policymaking are required.

Despite this relevant and dynamic role, the number of studies analyzing the energy demand of the tertiary sector is relatively small. Altogether, there is a gap related to transparently modeling the sector’s energy demand consistent with statistical data and containing up-to-date information.

Against this background, the aim of this paper is to develop and apply an integrated top-down and bottom-up model for the tertiary sector. The associated research question is as follows: How can an integrated model transparently break down tertiary sector energy demand by end-use, energy carrier and subsectors according to the statistical classification of the economic activities in the European Community (NACE) as a bottom-up perspective to provide policymakers with a better understanding whilst still remaining consistent with the energy balance from a top-down view?

To answer this question, this paper is structured as follows: Section 2 provides an overview of studies addressing energy demand in the tertiary sector. Section 3 introduces a methodology to quantify energy demand that allows for breaking down energy demand into energy carriers and end-use using a combined bottom-up and top-down approach. To illustrate and validate this model, it is applied to a case study of Germany’s tertiary sector using national statistics, results from a dedicated survey among companies and other sources. This is followed by a presentation and discussion of the results and methodological aspects (Section 4). Finally, conclusions are provided (Section 5).

2. Energy Demand in the Tertiary Sector

2.1. Investigations of Energy Demand in the Tertiary Sector

The literature includes various works specifically relating to the tertiary sector’s dynamic structure. In developed countries, the tertiary sector is the most important economic sector [16]. Studies for the US, China and other OECD countries demonstrated the connection between economic growth and a growing tertiary sector [16,17,18,19,20,21,22].

However, activities, structure, productivity, substitution and intensity affect the total energy consumption [18]. The dynamics of consumption are manifold. This is visible in various works. Activity is expected to shift towards cooling end-uses at a high rate of growth worldwide [16,23]. Space heating and cooking were influenced by structural changes and energy efficiency improvements [18]. The share of electricity end-use may also rise due to the expansion of the use of heat pumps [24]. As far as the potential of renewable resources is concerned, a study conducted in the Netherlands shows that the tertiary sector plays a major role in the generation of solar and wind energy [25]. The internal structure in China, for example, is changing from traditional subsectors such as hotels and restaurants to finance and real estate [22]. In addition to the high share of the tertiary sector in the total final energy consumption, the sector also has a major impact on the energy consumption within buildings (34%) [24]. Policies need to attract buildings, but there is a big difference between Europe and China, with China’s policies needing to address the increasing proportion of new buildings [16]. In summary, structural changes in recent decades have led to a shift of processes to other sectors (outsourcing effects), and new technologies have influenced and continue to influence the energy demand structure of the sector. Therefore, energy demand and emissions do not reflect the full amount of emissions due to direct and indirect emissions from supply chain effects and goods produced in other sectors, as well as strong pull effects on other sectors [26,27,28].

There are several other studies on the tertiary sector. To obtain a more detailed understanding of the tertiary sector, an econometric analysis was implemented in [24] to examine the dependency of energy consumption and various variables like employees, worked hours, total floor area and gross domestic product (GDP) or gross value added (GVA). Other studies addressed issues such as the inelasticity of changes both in price and demand for electricity [29]. The main findings were that demand is inelastic in the short term, but the reverse is true in the long term, and that pricing policies appear to be effective. Energy efficiency measures also had a long-term impact due to income elasticity [29]. Other investigations found high theoretical and technological potential and a willingness of companies in the food retail and restaurant subsectors to participate in demand response [30]. Further analysis showed that supermarkets, hotels and office buildings could play an important role in compensating for grid instabilities [31].

These dynamics demonstrate the need for up-to-date energy demand modeling for the sector.

2.2. Modeling Energy Demand in the Tertiary Sector

2.2.1. Energy Demand Modeling

Energy system models are used to understand and describe the dynamics of energy demand quantitatively. They can be distinguished in different ways. The paper of van Beeck [32] describes an approach to classify the modeling approaches into different categories (

Table 1. Classification of energy system models based on [32,33] with modifications. Underlined entries indicate the suggested model design.

Purposes:	general: description, prediction, forecasting and backcasting specific: ernergy demand, supply side or impact
Analytical approach:	top-down, bottom-up approach or hybrid
Methodology:	econometric, macroeconomic, economic equilibrium, optimization, simulation or multi-criteria methodologies
Mathematical approach:	accounting framework, linear programming, mixed integer programming or dynamic programming
Geographical coverage:	global, regional, national, local or project
Sectoral coverage:	single-sector or multi-sectoral models
Time horizon:	status quo, short, medium or long term
Data requirements:	quantitative or qualitative technical or monetary

). Starting from the general and specific purpose to the characteristics such as model structure, analytical approach, underlying methodology, mathematical approach, geographical coverage, sectoral coverage, time horizon and data requirements [32]. The sectoral coverage can be divided into a partial equilibrium when considering only one sector or equilibrium models, if the whole economy is considered [5]. The target can either be forecasting, exploring or backcasting [32]. Taking a closer look at the analytical approach, a typical distinction is made between the aforementioned top-down and bottom-up models [32].

Some of the aforementioned aspects are particularly relevant to this investigation and are therefore addressed in more detail. Regarding the analytical approach used in this paper, there are top-down, bottom-up and hybrid approaches. With a bottom-up approach, energy end-use can be determined from each sub-process [10,11]. This is important to focus on current and future technological details to find suitable policies and levies [9]. The strength of bottom-up models, which have a more detailed technical characterization of technologies and end-uses, lies in the representation of a significant number of different technology options, the coverage of energy sources at the primary and final energy levels, the substitution of processes and the demonstration and comparison of efficiency [9,34,35,36]. Top-down models, on the other hand, take a more macroeconomic, broader view of economic sectors and agents [9,35]. Compared to bottom-up models, the technological details are absent, which limits their representation of the energy system [9,34]. Top-down models are useful for examining the macroeconomic impact of energy policies [34,36]. In addition to using the two modeling approaches independently, combining them as ‘hybrid’ or ‘integrated solutions’ allows us to combine their benefits. [9]. This can be the key to connecting the detailed technology-based bottom-up approach with the systematic macroeconomic perspective of a top-down approach [9,34]. On the other hand, such integrated models also face the challenge of matching both bottom-up and top-down perspectives in a consistent manner.

In addition to the analytical approach, the available data are also important in the development of energy system models. These require a thorough knowledge of today’s energy consumption, which serves as the baseline year and provides the possibility to establish a benchmark for future energy consumption [37]. Setting the baseline year and gathering information for energy system models is a very important part of energy system modeling. After setting the baseline boundaries and the year, the energy use data can be determined [37]. In the literature, there are energy system models using the energy balance or activity, intensity and efficiency data for the baseline year, which project future energy demand, for example. An example, therefore, is PRIMES which uses data from Eurostat as a baseline reference [38,39], or POLES which uses the databases of Enerdata [40]. Other energy consumption predicting models like FORECAST are based on energy system models for the tertiary sector [41].

2.2.2. Specific Energy System Models

Few energy system models deal specifically with the tertiary sector (

Table 2. Energy system models for the tertiary sector subdivided into energy carriers and energy end-use.

Purpose of Energy Model	Analytical Approach And Underlying Methodology	Geographical Coverage	Sectoral Coverage	Time Horizon	Disaggregation	Year of Publication	Ref.
current electricity demand	static BU method consisting of basic drivers and specific energy consumption indicators	Europe (29 countries)	trade, hotels/restaurants, traffic, finance, health, education, public administration, waste, sport, social service, real estate	2007	electricity (lighting, ventilation/cooling, circulation pumps/heating auxiliaries, ICT, data center, hot water, space heating, laundry, cooking, refrigeration/freezing, miscellaneous building technologies, street lighting, elevators)	2010	[43]
future electricity demand (FORECAST)	BU method consisting of basic drivers and specific energy consumption indicators	Europe (29 countries)	trade, hotels/restaurants, traffic, finance, health, education, public administration, waste, sport, social service, real estate	2012–2035 (outlook to 2025)	electricity (lighting, electric heating, ventilation/cooling, refrigeration, cooking, data centers)	2012	[41]
detailed data on energy consumption (EL-TERTIARY)	electricity metering, survey and analysis	EU (12 countries, 123 selected buildings)	offices, schools, universities, kindergartens, hotels, supermarkets, hospitals	2006–2008	electricity (central IT, heating, hot water, motor drives, office equipment, refrigeration, ventilation, air condition/cooling, lighting)	2007/2008	[44,45]
energy carrier and end-use	extrapolation from survey in 2004 and 2006	Germany	construction, offices, small manufacturing, retail trade, hospitals/schools/pools, hotels/restaurants, food, laundries, agriculture, horticulture, textile	2006–2011	electricity and fuels (lighting, mechanical energy, hot water, process heat, process cooling, air conditioning, ICT, space heating)	2009	[46]
energy carrier and end-use	extrapolation from survey in 2008 and 2010	Germany	construction, offices, small manufacturing, retail trade, hospitals/schools/pools, hotels/restaurants, food, laundries, agriculture, horticulture, textile	2006–2011	electricity and fuels (lighting, mechanical energy, hot water, process heat, process cooling, air conditioning, ICT, space heating)	2013	[47]
energy carrier and end-use	extrapolation from survey in 2012	Germany	construction, offices, small manufacturing, retail trade, hospitals/schools/pools, hotels/restaurants, food, laundries, agriculture, horticulture, textile	2011–2013	electricity (lighting, mechanical power, hot water, process heat, process cooling, air conditioning, ICT, space heating), fuels (mechanical power, hot water, process heat, space heating)	2015	[48]

). Following the reviews by Chatterjee et al. [42] and Prina et al. [36], the following selection criteria for the literature review are set:

• Articles must describe their methodology for a baseline year, not only using energy balances such as Eurostat, i.e., how energy consumption is subdivided into energy carriers and energy end-use [39];
• Articles should address the tertiary sector.

Previous related publications include the work of Fleiter et al. [43] and Jakob et al. [41]. Their works introduce a static bottom-up (BU) method, consisting of basic energy service drivers and specific energy consumption indicators, calculating the share of the energy end-use in the tertiary sector [41,43]. This model uses typical values in the literature for the installed power and the annual utilization rate, achieving results within a range of ±10% compared to Eurostat statistics [43]. Another method for monitoring electricity consumption in the tertiary sector was adopted by Gruber et al. [44] and Plesser et al. [45] by installing electricity meters in buildings in the tertiary sector. For this purpose, 123 representative buildings in 12 EU countries were equipped with electric meters for a certain period of time, scaling the electricity consumption measured to an annual consumption of the electric appliances such as lighting, office equipment, ventilation and air conditioning [45]. The findings from these European studies have also been used in Hungary and Brazil, for example [49,50]. The most recent empirically based analysis relies on survey data from 2012 which also served as a basis for more recent works on the sector [46,47,48]. From the survey, the average electricity and fuel consumption per subsector was determined and related to the average number of employees in the surveyed companies [51]. The final electricity or fuel consumption is calculated by multiplying these survey values by the actual number of employed persons or other statistically available drivers in the respective year [51]. However, in this analysis, a detailed procedure is not available to the public and a transparent structure of the model is lacking.

To sum up, in the present energy system models for the tertiary sector, there is some room for improvement in terms of transparency of the models; most models use data that are more than a decade old which do not account for more recent dynamics in the sector, and subdividing the energy consumption into subsectors based on the NACE code, energy end-use and energy carriers is only carried out in few cases. There is also a lack of consideration for renewable energy sources.

3. Methodology: Design and Implementation of an Integrated Model

3.1. Design of the Tertiary Sector Model

Against this background and in contrast to existing models, the tertiary model suggested here seeks the following:

3.

To match the energy demand of national sectoral statistics from a top-down-perspective with a technology-oriented energy demand using a bottom-up approach to yield a consistent dataset;

4.

To subdivide energy demand by subsectors, energy end-use and energy carriers to point out major areas of energy demand;

5.

To use recent data and statistics to provide an up-to-date analysis;

6.

To yield a transparent calculation method to enhance the credibility of results but to also underline the dependency of input data and assumptions.

With reference to the model classification in Table 1, the suggested model can be described as follows: its purpose is to cover energy demand using a hybrid approach, i.e., a combined top-down and bottom-up approach.

Figure 1 shows the methodology of the model which follows a multi-level approach consisting of eight steps. Steps (1) to (4) cover setting up the mathematical model for determining energy demand, whereas steps (5) to (7) refer to how data are included in the model. The ulterior step (8) serves to perform quality control for the analysis.

3.1.1. Step (1): End-Use Level

The first step aims to determine the final energy demand by end-uses and energy carriers. Following AG Energiebilanzen, the end-uses include lighting, information and communication technology (ICT), mechanical energy, hot water, space heating, process heating, process cooling and air conditioning [52]. Furthermore, in keeping consistent with the official energy balances, the following energy carriers are covered: coal, motor gasoline, jet fuel kerosene, diesel oil, fuel oil light, liquified petroleum gas, natural gas, liquid and solid biomass, biogas, environmental heat, solar (thermal) heat, geothermal energy, district heating and electricity [13]. Waste heat is not included as it is not a tertiary sector consumption category in AG Energiebilanzen [13]. Environmental heat is based on combined heat and power plants.

The determination of final energy demand is based on three generic bottom-up end-use calculation types that originate from considerations about reasonably obtainable data on the company level as well as energy demand features of certain applications across companies. The following calculation types are distinguished:

Type I is used where typical average values for energy demand per end-use device can be used to determine energy demand. It is mainly used for standardized devices, e.g., ICT equipment, and computed by the following:

$${}^{I}e_{m,j,k,BU}=n_{m,j}·p_{m,j,k}·c_{m,j}·h_{m,j}$$

(1)

with

${}^{I}e_{m,j,k,BU}$ as the final bottom-up (BU) energy demand of company m for the end-use j per energy carrier k [kWh/a];

$n_{m,j}$ as the number of end-use devices operated by company m for end-use j [n];

$p_{m, j,k }$ as the average nominal rated energy consumption per end-use device of company m for end-use j and energy carrier k [kW/n];

$c_{m,j}$ as the capacity factor per end-use device of company m for end-use j [%];

$h_{m,j}$ as the average annual operating hours of company m per end-use j [h/a].

Type II is used where installed overall capacities are more easily determined than individual number and powers of devices, i.e., it is computed by the following:

$${}^{II}e_{m,j,k,BU}=i_{m,j}·c_{m,j}·h_{m,j}$$

(2)

with

$i_{m,j,k}$ as the installed capacity of company m for end-use j and energy carrier k [kW].

Type III is characterized by a generic driver combined with driver-specific consumption. It is mainly used for space heating and hot water and depends on specific heating demand according to the type and age of the heating system used in the company:

$${}^{III}e_{m,j,k,BU}=d_{m,j}·s_{m,j}·{\eta }_{m,j}$$

(3)

with

$d_{m,j}$ as the driver value of company m for end-use j [e.g., L, m²];

$s_{m,j,k}$ as the driver-specific consumption value of company m for end-use j and energy carrier k [e.g., kWh/L, kWh/m²].

The sum over all types is a company’s bottom-up energy demand per end-use and energy carrier:

$$e_{m,j,k,BU}={}^{I}e_{m,j,k,BU}+{}^{II}e_{m,j,k,BU}+{}^{III}e_{m,j,k,BU}$$

(4)

3.1.2. Step (2): Company Level

While the bottom-up demand reflects the technological data of a company, its sum will usually deviate from the bottom-up demand for various reasons. The reasons for this are diverse and include errors and aggregation issues in bottom-up data, misaligned assumptions and/or end-uses. Differences may also arise if the boundaries between business and private use are unclear, for example, if the company owner’s home and place of work are located in the same building [53]. Also, buildings may supply one or more workplaces with, for example, heating and hot water or mechanical equipment as specified, but the associated power data are missing and estimates have to be used [53].

To ensure that the bottom-up results $e_{m,j,k,BU}$ mirror the company’s overall demand and neither over- nor underestimate it, an alignment with the company’s top-down demand is needed when a value is given. For this, the bottom-up values per energy carrier are calculated from the end-use models $e_{m,j,k, BU}$ for each company by the following:

$$E_{m,j,k}=e_{m,j,k,BU}·{\displaystyle \frac{{\sum }_{j=1}^J{e}_{m,j,k,BU}}{e_{m,k,TD}}}$$

(5)

with

$E_{m,j,k}$ as the final energy consumption of company m for end-use j per energy carrier k;

$e_{m,j,k, BU}$ as the final bottom-up model (BU) energy consumption of company m for end-use j per energy carrier k;

$e_{m,k, TD}$ as the final top-down (TD) energy demand of company m across all end-uses per energy carrier k.

3.1.3. Step (3): Subsectoral Level

The energy demand for the companies is then extrapolated to an aggregated level across all companies per subsector by using subsectoral drivers (Equation (6)). These drivers are introduced to avoid structural distortions from treating small and large companies alike. For this, the model uses subsectoral drivers from statistics quantified on the sectoral level by $b_i$ (Table 3) and on the company level by $b_m$. These drivers include the number of employees, students and annual guests in accommodation or hospital beds. The calculation of the subsectoral demand is made by the following:

$$E_{i,j,k}=b_i·{\displaystyle \frac{1}{n_i}}·\sum _{m=1}^M\left({\displaystyle \frac{E_{m,j,k}}{b_m}}\right)$$

(6)

with

$E_{i,j,k}$ as the final energy consumption of the subsector i with the end-use j per energy carrier k;

n_i as the number of companies in the sample belonging to subsector i;

$b_i$ as the value of the subsectoral driver in subsector i;

$b_m$ as the value of the driver of company m;

3.1.4. Step (4): Aggregated Level

In the fourth scaling step, the energy consumption determined is calibrated to the energy balance of the tertiary sector. Again, this is necessary to ensure consistency with the overall values:

$$E_{i,j,k, total}={\displaystyle \frac{E_{i,j,k}}{E_{k,balance }}}·\sum _{j=1}^J{E}_{i,j,k}$$

(7)

with

$E_{i,j,k,total}$ as the total energy consumption of the subsector i with the end-use j per energy carrier k;

$E_{i,j,k}$ as the final energy consumption of the subsector i with the end-use j per energy carrier k;

$E_{k,balance }$ as the final energy consumption per energy carrier k from the energy balance.

The model implementation is carried out in Excel.

3.2. Implementation for a Case Study of the German Tertiary Sector

To illustrate and validate the model, it is applied to a case study of the German tertiary sector. The structure of the sector as well as the drivers of the companies obtained from the survey and the value of the subsectoral drivers mainly from Statistisches Bundesamt [54] can be seen in Table 3. Drivers taken from the survey are calculated using the sum of owners, full-time employees and half of the values given for part-time employees.

Table 3. Structure of the tertiary sector according to the NACE code in Germany [55].

NACE	Subsector of Tertiary Sector	Drivers	${\mathit{b}}_{\mathit{i}}$	${\displaystyle \sum _{\mathit{m}=1}^{\mathit{n}}}{\mathit{b}}_{\mathit{m}}$
A	Agriculture, forestry and fishing	Employees	510,000	3505
C*1	Manufacturing	Employees	435,295	22,304
F	Construction	Employees	1,990,542	6182
G	Wholesale and retail trade; repair of motor vehicles and motorcycles	Employees	5,010,932	12,113
H	Transporting and storage	Employees	2,030,773	1803
		Passengers	247,800,000	*2
I	Accommodation and food service activities	Employees	1,182,067	1830
		Overnight stays by guests	495,600,000	5,076,154
J	Information and communication	Employees	1,140,294	4962
K	Financial and insurance activities	Employees	852,350	2738
L	Real estate activities	Employees	363,275	*3
M	Professional, scientific and technical activities	Employees	2,310,911	7096
N	Administrative and support service activities	Employees	3,075,934	482
O	Public administration and defense; compulsory social security	Employees	1,872,717	9076
P	Education	Students	11,217,933	29,168
Q	Human health and social work activities	Hospital beds	494,326	19,740
		Employees	2,912,807	946
R	Arts, entertainment and recreation	Employees	420,219	340
S	Other services activities	Employees	1,154,029	1015

*¹ only companies with less than 20 employees are covered here. *² airports are not covered by the survey. *³ real estate activities are not covered by the survey.

Table 3. Structure of the tertiary sector according to the NACE code in Germany [55].

NACE	Subsector of Tertiary Sector	Drivers	${\mathit{b}}_{\mathit{i}}$	${\displaystyle \sum _{\mathit{m}=1}^{\mathit{n}}}{\mathit{b}}_{\mathit{m}}$
A	Agriculture, forestry and fishing	Employees	510,000	3505
C*1	Manufacturing	Employees	435,295	22,304
F	Construction	Employees	1,990,542	6182
G	Wholesale and retail trade; repair of motor vehicles and motorcycles	Employees	5,010,932	12,113
H	Transporting and storage	Employees	2,030,773	1803
		Passengers	247,800,000	*2
I	Accommodation and food service activities	Employees	1,182,067	1830
		Overnight stays by guests	495,600,000	5,076,154
J	Information and communication	Employees	1,140,294	4962
K	Financial and insurance activities	Employees	852,350	2738
L	Real estate activities	Employees	363,275	*3
M	Professional, scientific and technical activities	Employees	2,310,911	7096
N	Administrative and support service activities	Employees	3,075,934	482
O	Public administration and defense; compulsory social security	Employees	1,872,717	9076
P	Education	Students	11,217,933	29,168
Q	Human health and social work activities	Hospital beds	494,326	19,740
		Employees	2,912,807	946
R	Arts, entertainment and recreation	Employees	420,219	340
S	Other services activities	Employees	1,154,029	1015

*¹ only companies with less than 20 employees are covered here. *² airports are not covered by the survey. *³ real estate activities are not covered by the survey.

3.2.1. Step (5): Company Data Acquisition via Survey

To obtain data on the company level, a survey among companies in Germany’s tertiary sector was carried out. It was based on a questionnaire used for previous analyses [46,47,53,56] but adjusted to major recent dynamics and new applications. In particular, questions on ICT equipment, data networks and LED lighting were added. The original questionnaire followed a structure from Geiger et al. [53] which grouped comparable companies and institutions for the survey and identified cross-cutting technologies. For the present analysis, the structure of the tertiary sector was based on NACE code classifications. Furthermore, the energy demand of several large, formerly state-owned institutions such as Deutsche Bahn (railway provider), Deutsche Telekom (telecommunications providers), Deutsche Post (postal service) and airports were allocated to the tertiary sector [57]. Military services are not considered.

The questionnaire itself is structured into a general part for all companies and specific questions for individual subsectors. It covers items relating to the energy consumption of the targeted companies, including exact information about appliances, equipment and machines used in the companies, e.g., in terms of numbers and power demand. The general part targeted information about heating systems, hot water provision, air conditioning and ventilation, vehicles, lighting systems and ICT such as servers, computers, monitors, copy machines, networks, projectors and data centers. Also, dishwashers, freezers and vending machines were covered as well as lunches that were sold. The specific part added detailed sectoral questions on energy demand, e.g., on drivers and compressor units, thermal engineering processes and other special industry-specific systems. As an example for hospitals, the specific part asked questions about the number of hospital beds, the equipment of the room and the hospital, such as swimming pools, diagnostic systems, laundries, disinfection and sterilization.

The survey was implemented via phone interviews from autumn 2020 to spring 2021 and covered the year 2019, i.e., the last ‘typical’ year preceding before the COVID pandemic. The phone interviews typically took between approximately 18 and 60 min to complete (10 to 90% quantile of respondents). The obtained answers were checked for plausibility, and obviously incorrect values and answers were removed prior to the calculation. In total, 1451 completed questionnaires were allocated to the subsectoral split indicated in Table 3.

Due to limits to the length of the survey and to considerations that some values such as the power demand of applications might be obtained more easily and reliably from the literature, additional values were taken over from the literature to complement the data. Depending on the case under consideration, this included the various data inputs for the previously introduced bottom-up model types I, II and III.

There are several areas and energy carriers that require special treatment for various reasons: Airports were handled differently due to their limited number. Therefore, the top-down values of the energy carriers are derived from values for the main German airports which were scaled by assigning a representative scaling key consisting of 70% office-type operations and 20% logistic-type and 10% retail-type businesses. For NACE ‘Real estate activities [L]’ not covered in the survey, office-type operations were used as well as additionally for, ‘Professional, scientific and technical activities [M]’, ‘Administrative and support service activities [N]’, ‘Arts, entertainment and recreation [R]’ and ‘Other service activities [S]’. Furthermore, since the current survey does not allow for conclusions about fuels for propulsion consumption in construction, the fuel ratio in agriculture to the construction industry from the previous survey was adopted for the construction subsector. Overall, the fuel reported in the AGEB statistics has been allocated proportionately to motor gasoline and diesel fuels and jet fuel kerosene entirely to airports. Geothermal energy was not inquired about in the 2019 survey, so the demand according to AGEB statistics was allocated to sectors and applications based on the distribution of environmental and solar heating. A similar approach was taken for biofuels; they were distributed proportionally similarly to the other fuels. It should also be noted that air conditioning does not add to the environmental heat.

3.2.2. Step (6): Subsectoral Statistical Values

Drivers for transferring company statistics to the subsectoral level are mainly the number of employees. These are available from the German Federal Statistical Office [54,58]. In the case of the aforementioned state-owned institutions, insurance and finance as well as airports and data centers, employment statistics from the companies’ or associations’ annual reports or specific sectoral statistics are used [59,60,61,62,63,64,65]. In some areas, employee numbers are less meaningful and other drivers appear more promising: the number of overnight stays is used for ‘Accommodation and food service [I]’ [66], the number of students in schools and universities for ‘Education [P]’ [67,68] and the number of hospital beds for hospitals under ‘Human Health [Q]’ [69] (see Table 3).

3.2.3. Step (7): Energy Balance of Tertiary Sector

The sectoral top-down values in this investigation are based on the annual energy balance published by AG Energiebilanzen [13]. It details energy demand by energy carrier for the sector in its entirety. Energy consumption is subdivided for each year into energy carriers including environmental heat, solar heat and renewable energies [70]. In areas that are not covered by the official statistics, the energy balance is based on sales data (e.g., oil), and the consumption of storable energy carriers is derived from market research results based on changes in stock levels [13].

3.2.4. Step (8): Validation and Quantification of Tertiary Model

For validation and quality control purposes, two indicators are introduced. The indicator I₁ (Equation (8)) is used to quantify and compare the uncalibrated energy demand over all subsectors per energy carrier $E_{k,uc}$ (refer to step (3) in Figure 1) to the energy consumption per energy carrier according to official sector statistics $E_{k,balance}$. I₁ mirrors how well the calculated bottom-up value on the sectoral level meets the actual top-down value in the interval [0, 1]. The closer the value to 1, the better the match.

$$I_1={\displaystyle \frac{E_{k,uc}}{E_{k,balance}}}$$

(8)

Here,

$I_1$ is the indicator for validation of the tertiary model;

$E_{k,uc}$ is the uncalibrated energy demand per energy carrier k (Step (3) in Figure 1);

$E_{k,balance}$ is the reference value for the energy demand per energy carrier k from the national energy balance.

In addition to the base model for the subsectoral aggregation described in step (3) of Figure 1, an alternate approach (2) can also be used to calculate the subsectoral demand by the following:

$$E_{i,j,k}={\displaystyle \frac{\sum _{m=1}^n{E}_{m,j,k}}{\sum _{m=1}^n{b}_m}}·b_i$$

(9)

with

$E_{i,j,k,\left(2\right)}$ as the final energy consumption of the subsector i with the end-use j per energy carrier k with the second modeling approach (2).

The difference to the first approach (Equation (6)) is that the energy consumption of the companies per subsector is added together and divided by the total driver levels from the survey in that subsector. This means that an average value is first calculated for each subsector. This value is then multiplied by the sum of the respective drivers for this subsector.

4. Results and Discussion

4.1. Baseline Model Output

Figure 2 (for the numerical values, see Appendix B 

Table A1. Calibrated energy consumption in the tertiary sector in TJ by energy end-use and subsector for the year 2019.

2019		Fossil Fuels							Renewable Energies					Electricity	Total
		Coal	Motor Gasoline	Jet Fuel Kerosene	Diesel Oil	Light Fuel Oil	Liquified Petroleum Gas	Natural Gas	Biomass	Environmental Heat	Solar–Thermal	Geothermal	District Heat	Electricity	All
A	Agriculture, Forestry and Fishing	69	6037	-	89,807	13,943	158	26,150	86,795	25	379	171	2	21,924	245,461
C	Manufacturing (only companies with less than 20 employees)	-	-	-	-	8863	25	13,894	5491	51	5	24	158	17,514	46,024
F	Construction	-	602	-	8958	4552	5981	11,034	8005	2827	60	1225	89	18,149	61,481
G	Wholesale and Retail Trade; Repair of Motor Vehicles and Motorcycles	-	-	-	-	18,803	14,906	57,757	297	22	670	294	3632	74,002	170,381
H	Transporting and Storage	-	142	3746	2114	2905	1	8259	355	96	10	45	2041	16,626	36,339
I	Accommodation and Food Service Activities	-	-	-	-	10,853	-	62,775	9826	581	100	289	14,373	64,472	163,271
J	Information and Communication	-	-	-	-	479	-	7159	87	6	-	3	689	83,959	92,382
K	Financial and Insurance Activities	-	-	-	-	3149	-	4396	51	-	-	-	185	5239	13,020
L	Real Estate Activities	-	-	-	-	1166	37	3834	159	0	-	-	224	2731	8151
M	Professional, Scientific and Technical Activities	-	-	-	-	7335	513	24,004	977	4	-	2	1415	85,155	119,404
N	Administrative and Support Service Activities	-	132	-	1957	10,259	282	29,923	1451	13	-	4	1708	12,382	58,111
O	Public Administration and Defense; Compulsory Social Security	-	-	-	-	5676	-	24,750	1177	-	-	-	1505	15,327	48,435
P	Education	-	-	-	-	11,085	-	48,383	28	153	48	85	4239	30,529	94,549
Q	Human Health and Social Work Activities	-	-	-	-	12,917	-	39,414	402	385	0	164	6867	35,224	95,374
R	Arts, Entertainment and Recreation	-	-	-	-	1283	40	5281	253	14	36	21	398	5675	13,002
S	Other Services Activities	-	-	-	-	4476	111	13,601	479	6	-	2	725	30,118	49,519
	Tertiary Sector	69	6913	3746	102,836	117,744	22,054	380,613	115,834	4184	1309	2329	38,248	519,026	1,314,905

) shows the aggregated model output by subsector for the case study of Germany’s tertiary sector. The dominant energy carriers vary by subsector. In the subsector ‘Agriculture [A]’, biomass is most commonly used, for example, for transportation and space heating. In the ‘Trade [G]’ and ‘Accommodation and food service [I]’ sectors, as well as in ‘Information [J]’ and in most of the other less energy-intensive sectors, electricity is the most predominant energy carrier.

The percentage breakdown of applications for electricity and other energy sources is shown in Figure 3. Lighting (35%) and ICT (33%) are the main end-uses for electricity, as are space heating (78%) and mechanical energy (15%) for the other energy carriers.

Figure 4 shows the percentage of electricity and fuel consumption in the different application areas according to the NACE code shown in Table 3 (Figure 4 and Appendix B 

Table A2. Calibrated energy consumption in the tertiary sector by energy end-use and subsector for the year 2019.

2019		Energy Consumption		Share of Energy End-Use
		Electricity	Fuels	Electricity								Fuels
		Total	Total	Lighting	ICT	Mech. Energy	Hot Water	Process Heat	Space Heating	Process Cooling	Air Cond.	Lighting	ICT	Mech. Energy	Hot Water	Process Heat	Space Heating	Process Cooling	Air Cond.
NACE Code		TJ	TJ	%								%
A	Agriculture, Forestry and Fishing	21,924	223,536	35	7	37	1	4	7	0	10	-	-	46	1	1	52	-	-
C	Manufacturing (only companies with less than 20 employees)	17,514	28,511	40	11	19	1	19	6	1	3	-	-	0	3	15	82	-	-
F	Construction	18,149	43,332	65	20	8	0	0	2	1	2	-	-	22	3	0	75	-	-
G	Wholesale and Retail Trade; Repair of Motor Vehicles and Motorcycles	74,002	96,380	38	33	8	1	0	10	8	3	-	-	-	3	0	97	-	-
H	Transporting and Storage	16,626	19,714	33	41	7	0	9	6	2	2	-	-	31	7	0	61	-	-
I	Accommodation and Food Service Activities	64,472	98,799	29	6	17	3	27	8	2	7	-	-	-	9	8	83	-	-
J	Information and Communication	83,959	8423	9	69	1	0	0	1	0	20	-	-	-	7	0	92	-	-
K	Financial and Insurance Activities	5239	7781	25	56	5	0	1	10	1	1	-	-	-	8	0	92	-	-
L	Real Estate Activities	2731	5420	39	46	5	0	1	3	1	5	-	-	-	5	0	95	-	-
M	Professional, Scientific and Technical Activities	85,155	34,249	37	36	8	1	2	8	6	3	-	-	-	5	0	95	-	-
N	Administrative and Support Service Activities	12,382	45,729	33	37	7	0	11	6	1	2	-	-	5	5	0	90	-	-
O	Public Administration and Defense; Compulsory Social Security	15,327	33,108	44	43	5	0	1	-	1	6	-	-	-	4	0	96	-	-
P	Education	30,529	64,020	55	26	9	2	1	2	1	3	-	-	-	10	1	89	-	-
Q	Human Health and Social Work Activities	35,224	60,149	51	18	4	3	5	14	2	3	-	-	0	8	2	90	-	-
R	Arts, Entertainment and Recreation	5675	7327	32	33	6	0	10	4	2	11	-	-	-	7	12	81	-	-
S	Other Services Activities	30,118	19,401	37	33	9	1	1	9	7	3	-	-	4	5	6	85	-	-
	Tertiary Sector	519,026	795,879	35	33	9	1	6	7	3	7	-	-	15	5	2	78	-	-

). The results generally underline the change compared to the previous investigation of the tertiary section in 2012 [47]. Lighting is no longer the dominant application in electricity consumption [47]. Rather, lighting and ICT are both at about 30% of the electricity demand. This is mainly due to the ongoing replacement of incandescent lamps with energy-saving LED lighting and the increasing demand for ICT. Additionally, space heating has the largest share of energy demand for most of the subsectors. The exceptions are the sectors ‘Agriculture [A]’ with a large share of mechanical energy and ‘Information [J]’ with a major proportion of ICT and air conditioning. Apart from this, all thermally conditioned areas in agriculture are attributed to space heating. Process heat is mainly used in the ‘Accommodation And Food [I]’ sector. In general, the mechanical energy turns out to be relatively small as the largest part is allocated to the transport sector.

4.2. Comparison of Bottom-Up Results with the German Top-Down Energy Balance

4.2.1. Validation Results for the Baseline Assumptions

The modeled energy consumption by subsector and energy carrier in 2019 prior to the calibration to the top-down values according to the AGEB energy balance indicates in a transparent way how well the bottom-up model matches the statistics (A similar investigation for the bottom-up/down-top comparison on company level using the survey data from this investigation is available in [71]). The results for indicator I₁ (

Table 4. Values for indicator I₁ by energy carrier comparing the uncalibrated bottom-up results to the national top-down energy balance for the year 2019 and the impact of a sensitivity analysis modifying the baseline input data. The last row shows the results if an alternative subsectoral model is used *.

Case	Energy Carrier							Total
Indicator I1	Coal	Liquified Petroleum Gas	Light Fuel Oil	Natural Gas	District Heat	Renewable Energies	Electricity	All
Baseline values	0.07	0.09	1.53	0.82	2.36	0.81	0.92	1.01
Sensitivity: +10%	0.07	0.09	1.53	0.82	2.36	0.75	0.92	1.02
Sensitivity: −10%	0.07	0.09	1.53	0.81	2.36	0.75	0.92	1.01
Alternate model	0.02	0.01	0.34	0.45	1.99	0.51	0.63	0.61

* Since the 2019 survey does not allow for conclusions to be drawn about the type of transport fuel and geothermal energy, this is not considered in Table 4.

‘Baseline Values’) show that the model tends to underestimate energy consumption for almost all energy carriers. The coal and the liquid gas share is substantially lower than in the energy balance [70], and the district heating values are substantially above the statistics.

Overall, the bottom-up approach yields results that are consistent in terms of total energy consumption with the national energy balance, only showing a deviation of 1% across all energy carriers.

4.2.2. Sensitivity Analysis

To understand the impact of modifying the literature values and estimates as an input to the model, these values are subjected to a sensitivity analysis (Table 4 ‘Sensitivity +10% and −10%’). In each calculation on the end-use levels ${}^{I}e_{m,j,k,BU}$, ${}^{II}e_{m,j,k,BU}$ or ${}^{III}e_{m,j,k,BU}$ coupled with the further literature or estimated values, the sensitivity is considered only once; otherwise, the analysis would be duplicated. The values are changed in the range between +10% and −10%. The latter occurs only if directly coupled with the further literature values or estimated values; otherwise, there would be a double consideration both for energy consumption and power. The values of efficiency rates, specific heat capacity, specific energy content of the fuels and capacity factors are not subjected to a sensitivity analysis.

Some results from the sensitivity analysis can be explained by the design of the model. It can be determined that the results do not vary by more than 0.7% when the specified values are modified within a range of ±10%. This underlines a stronger impact of survey data compared to the literature values or estimates and to the top-down calibration in the model.

4.2.3. Alternative Subsectoral Approach

Table 4 also contains the results if the alternative subsectoral approach is used (Table 4 ‘Alternate model’). The results show that the alternate model only achieves an overall 61% match with the overall energy balance. This means that the baseline approach yields a better overall result with the data used. It also improves the mapping of the company level, as the intermediate values are weighted by individual companies rather than by the reported consumption totals per subsector. The weighting of the individual companies is stronger than in the first approach in which the entire sector is considered. Due to the better match, the baseline approach was chosen as the more appropriate way for extrapolating the final energy consumption to Germany’s tertiary sector as a whole.

4.3. Limitations of the Tertiary Model

While the overall baseline model yields a good fit of bottom-up data and top-down statistics, it has various limitations. They include the dependence on external data which usually needs to be collected via an effortful data collection process. Though a survey setting via phone calls, as in this case, helps to minimize data errors or misunderstandings as a human may directly respond to interpretation issues or point out implausible answers, inconsistencies in data collection cannot be ruled out. Also, the use of a survey might also limit the detection and identification of new trends. The data of the survey were checked for each company regarding inconsistencies. However, the possibility to provide a validation with actual documentation or metered data is not given. In addition, the bottom-up approach cannot represent the overall energy consumption in every regard due to necessary measures of simplifying assumptions and extrapolations. The two-fold top-down calibration on a company level and on a sectoral level can help to limit the impact of faulty bottom-up approaches; yet, errors will still be reflected in the overall results. In addition, the scaling factor in Equation (7) is the same for all subsectors per energy carrier, and no error differentiation is made in the extrapolation per subsector.

4.4. Implications for Policymaking

The model and its result may help policymakers in the preparation of measures to address energy demand in the tertiary sector, thereby helping to curb the sector’s greenhouse gas towards the 1.5 °C target. As underlined, the results for 2019 show that the distribution of applications using electricity in the sector is about 35% lighting, 33% ICT, 9% mechanical energy, 1% hot water, 6% process heat, 7% space heating, 3% process cooling and 7% air conditioning. This underlines that lighting and ICT are particularly energy-relevant applications and therefore may offer a high potential for savings. Such findings allow policymakers to consider targeted measures and to subsequently understand their potential impact. In the field of lighting, for example, with appropriate promotion and information, existing incandescent lamps could be replaced with more energy- and cost-efficient LEDs. In the other major savings area of ICT, especially with the growing number of data centers, policies could focus on efforts to maximize the energy efficiency of products and services. Applications using fuels are 15% mechanical energy, 5% hot water, 2% process heat and 78% space heating. Consequently, the main lever for energy savings, especially for natural gas, is in the heat sector.

Under the consideration of the discussed limitations, policymakers can use such analyses as a basis for in-depth analysis of specific applications and energy efficiency measures, and for the development of policies in areas such as heat pumps, biomass heating, district heating, solar thermal and hydrogen technologies. The more detailed distinction between subsectors allows for even more detailed insights and possibilities for measures for policymakers.

5. Conclusions

In line with the aim of this paper, a hybrid top-down and bottom-up model was developed which allows us to differentiate energy demand in the tertiary sector by end-uses, energy carriers and subsectors. To illustrate the model, a case study for the German tertiary sector combines the literature values, estimates and survey results from 1451 companies. The results are in line with the official German energy statistics of the sector, and aggregated subsectoral bottom-up data show a very good fit with the overall top-down statistics, with a deviation across all energy carriers of 1%.

The approach and results yielded here can serve as a starting point for policymakers when seeking to decarbonize the tertiary sector which is often treated as a residual in official energy balances. Future efforts complementing this work may focus on further investigations of sectors and end-uses with a high relevance for energy demand. The model can serve as a basis for further development, offering the opportunity to uncover saving potentials, identify relevant efficiency measures and design policy measures for the tertiary sector, which can then be implemented and analyzed. The savings potential can also be used to consider the non-energy benefits associated with energy efficiency improvements that go beyond the energy aspects [6,72,73].