Is It Worth It? Potential for Reducing the Environmental Impact of Bitumen Roofing Membrane Production

Abstract

Between 51% and 72% of a bituminous roofing membrane used for structural waterproofing consists of organic material, predominantly bitumen—a derivative of crude oil refining—highlighting the strong dependence of this product on fossil resources. Considering that several tonnes of these membranes must be replaced every 30 to 50 years, substantial potential exists for emission reduction through the establishment of circular material systems. This study investigates this potential by analysing 26 Environmental Product Declarations (EPDs) and life cycle datasets from across Europe covering the period from 2007 to 2023. To ensure comparability, all data were normalised to a declared unit of 1 kg of roofing membrane. The reinforcement layers were categorised into glass and polyester & glass composites, and their differences were examined using Welch’s t-tests. Correlative analyses and linear as well as multiple regression models were then applied to explore relationships between environmental indicators and the shares of organic and mineral mass fractions. The findings reveal that renewable energy sources, although currently representing only a small share of total production energy, provide a major lever for reducing nearly all environmental impact categories. The type of reinforcement layer was also found to influence the demand for fossil resources, both materially and energetically. For most environmental indicators, only multiple regression models can explain at least 30% of the variance based on the proportions of organic and mineral inputs. Overall, the study underscores the crucial importance of high-quality, transparently documented product data for accurately assessing the sustainability of building products. It further demonstrates that substituting fossil energy carriers with renewable sources and optimising material efficiency can substantially reduce environmental burdens, provided that methodological consistency and clarity of indicator definitions are maintained.

1. Introduction

Bitumen is among the oldest known construction materials, with evidence of its extraction in Mesopotamia dating back several millennia. Historical records indicate that additives were used even then to modify its melting point. In addition to its role in construction, bitumen also found applications in early medicine [1]. Today, bitumen is predominantly employed in road surfacing and waterproofing applications.

The relevance of sustainability in construction materials is growing, driven by climate change, limited raw materials, and the need to reduce greenhouse gas emissions [2,3,4]. Bitumen, the key component of bituminous membranes, is derived from petroleum and therefore strongly linked to fossil energy consumption [5]. In recent studies, the C${\mathrm{O}}_2$ equivalents of bitumen-based materials have been reported to vary significantly depending on regional production conditions and refining technologies, with some studies suggesting that previous estimates may have underestimated total emissions [6] (p. 60). Moreover, the worldwide increase in building renovation and replacement of waterproofing membranes contributes to large quantities of bituminous waste. Within the European Union, millions of tons of bitumen-based materials are produced annually, and their disposal or recycling poses a growing environmental challenge. At the same time, initiatives such as the EU project “From Roof to Road” demonstrate practical approaches to recycle roofing felt waste into road construction materials, reflecting a broader move toward circular economy models [7].

In Germany alone, the demand for bituminous sealing materials reached 63 million square metres in 2023 [8]. Given that approximately 60% to 70% of flat roofs in the country are covered with bituminous membranes, this figure is expected to rise [9]. While there is a growing demand for housing and usable space, the service life of waterproofing systems remains limited due to environmental exposure, UV radiation, and natural ageing. Consequently, replacement is typically required every 30 to 50 years [10].

Accurate and up-to-date figures regarding the annual demand for bituminous construction materials are difficult to obtain, and available data primarily rely on secondary literature. The Monitoring Report on the Construction Industry published by the German Environment Agency provides only a separate assessment of mineral construction materials, estimating the total construction waste at approximately 200 million tonnes per year [11]. A market analysis conducted by the Technical University of Darmstadt in 1998 indicated an annual volume of approximately 200,000 t of bituminous membranes classified as hazardous waste [12]. Corporate data from Nordic Waterproofing Holding report a production volume of 52 million square metres of roofing membranes in 2014, of which 58% were used for renovation purposes. When scaling this market share to the estimated annual turnover of the largest roofing membrane producers in Europe, a total annual production of approximately 3.7 billion ${\mathrm{m}}^2$ can be inferred [13] (pp. 31–46). This may lead to a high demand for recycling technologies.

Bitumen offers significant material reuse potential. This potential is already being explored through the recycling of road construction bitumen [14,15,16]. Life Cycle Assessments (LCAs) serve as a key methodology in evaluating such potential, analysing environmental impacts across the entire product lifecycle. These assessments are instrumental in comparing different materials and identifying opportunities for environmental optimisation, as demonstrated in studies on polymer-based roofing membranes [17]. In addition to methodological considerations, ref. [18] emphasised the capacity of LCAs to investigate trends—illustrated here through the example of roofing in the construction sector—and to identify opportunities for further development. Reusing bituminous mass from roofing membranes presents a logical and environmentally beneficial approach, as the majority of emissions are attributed to the initial bitumen production, followed by transport and energy consumption [19,20].

Fundamentally, bituminous roofing membranes are composed of four principal components: the bitumen binder (typically comprising oxidised bitumen with polymer additives); mineral fillers such as limestone, quartz or basalt powder; protective surface layers (e.g., sand or slate) that guard against mechanical and thermal stress; and a reinforcement layer made from materials such as synthetic fibres, glass fibre, aluminium, or felt, which imparts specific technical properties to the membrane [21].

From the perspective of sustainability and resource conservation, reuse should be preferred over recycling of end-of-life products [22,23]. Since shredding roofing membranes and use them as a substitute in road surfacing or even incinerating them is referred to as downcycling or disposal in terms of product life cycles, the question arises as to how high the savings potential could be if the material were “simply” recycled.

Goal and Scope

In order to provide a contextual framework preceding the detailed analyses, estimations were conducted based on the available sample to elucidate observed trends and to support a robust understanding of the environmental determinants linked to bituminous sealing materials. This study deals with mass-specific indicators. All of them and also material parameters are therefore referenced to one kilogram of roofing membrane mass, which allows them to be considered in a more generic way. Although this does not correspond to the functional unit as roofing membranes are otherwise traded in the construction industry, it is also important to take a closer look at the real data and examine the actual data situation in more detail.

The almost lowest mass-related fraction is brought by the different kinds of carrier material used like glass fibre, polyester or aluminium. Carrier layers play a key role in defining the mechanical properties of the membranes, tailored to meet different application requirements. These materials, however, also entail varying production methods, which may influence the overall environmental impact due to differences in energy demand and emissions. Although comprehensive production-specific data for these materials is limited, it appears promising to investigate whether distinct combinations of carrier materials correspond to significant variations in environmental indicators. These significance tests may help determine whether the selection of a particular carrier material contributes measurably to the ecological performance of the membrane system.

To explore a more comprehensive understanding of the environmental impacts caused by the production of bituminous membranes, and to investigate whether meaningful relationships exist between the input-related indicators of abiotic depletion potential of elementary and fossil materials (ADPE and ADPF), the primary renewable and non-renewable energies (PERT and PENRT) and output-related impact categories, a statistical examination of potential correlations was undertaken. Some environmental indicators, which will be introduced in more detail in the following chapter on Section 2, partly reflect the same chemical constituents, which is why correlative relationships are already known in advance [24]. Anyway this line of inquiry leads to a central question: Is there a measurable interaction between these indicators during the production process between the life cycle stages A1 to A3 (Raw material supply, transportation, and manufacturing according to ISO 14044 [25]), and if so, how strong is it?

In addition to the collection of data across 15 environmental indicators, more detailed information was tried to obtained regarding the composition of the membranes. When considering the compounds of bituminous roofing membranes, the largest proportion by mass is accounted for by the coating material. Depending on the product specification, this is followed by an inert mineral filler, the granules, and finally the mechanical carrier layer. Oils, plasticizers, tire powder, and other additives make up the smallest fractions [21].

However the data available on material composition is very ambiguous which is actually a common problem when creating LCAs. Although the EPDs contain a binding description of the product composition, unfortunately this information is not uniform and varies greatly in terms of accuracy. For example, the bitumen content is often—but not always—based on the total mass of polymer-modified coating material, including synthetics and oils. Similarly, the term “minerals” commonly encompasses fillers such as limestone, quartz, or basalt powder, along with the sprinkling material, which makes it difficult to distinguish the individual components and their respective functions based solely on the manufacturer’s data sources. This enables us to include two additional factors in our investigations: the total coating mass of the polymer-modified bitumen material and the minerals used as filler and surface coating. This allows us to take the macroscopic influences of the organic and mineral fractions into account when attempting to predict the environmental impacts by regression models.

To this end, the study examines whether robust correlations between the quantities of these main components and the resulting environmental indicators can be established by applying suitable regression models. These correlations are essential for determining possible changes in environmental impacts due to variations in the material variations.

2. Methodology

2.1. Impact Categories

The current study takes into account fifteen impact categories to quantify the relationships between production process and the emission of chemical substances into the environment. These categories are based on the requirements of EN 15804 [26] (Annex C), and are further explained by [27] (p. 200 ff.) and [28] (p. 109 ff.).

Abiotic depletion potentials are differentiated into two categories: elements (ADPE) and fossil resources (ADPF). These summarise the consumption of finite resources measured in kg Sb-eq. for ADPE and MJ (megajoules) for ADPF. ADPE is significantly influenced by rare elements such as gold, tellurium, platinum, silver, and minerals while ADPF mainly comprises fossil fuels such as coal, natural gas, and crude oil.

Nitrogen-based emissions are captured primarily under the acidification potential (AP) and the eutrophication potentials for terrestrial (EP-terrestrial) and marine (EP-marine) ecosystems, including nitrogen oxides, ammonia, and nitrate compounds. The acidification potential (AP) also includes sulphur dioxide and related compounds, while the freshwater eutrophication potential (EP-freshwater) predominantly reflects phosphorus-based emissions. The commonly applied units are mol ${\mathrm{H}}^{\mathrm{+}}$-eq. for AP, mol N-eq. for EP-terrestrial, kg N-eq. for EP-marine, and kg P-eq. for EP-freshwater. As noted by [27] (p. 224), “Point sources in the form of wastewater treatment plants for households (e.g., from polyphosphates in detergents) and industry as well as fish farming are important sources of phosphorus and nitrates.”

The most prominent impact category in the construction sector is the global warming potential (GWP). Common greenhouse gases contributing to GWP include carbon dioxide, methane, nitrogen oxides, and halogenated hydrocarbons, all standardised to the reference unit kg C${\mathrm{O}}_2$-eq. In order to better trace the sources of emissions, the GWP is subdivided into three categories: biogenic, fossil, and land use and land-use change (LULUC). These reflect emissions from biomass use and decay, fossil carbon oxidation, and carbon stock changes due to land use, respectively. GWP-total represents the sum of these subcategories and is calculated over a standardised time horizon of 100 years. As [27] (p. 207) state, “GHG emissions are attributable to almost any human activity. The most important contributing activities are: burning of fossil fuels and deforestation.”

The ozone depletion potential (ODP) accounts for substances that contribute to stratospheric ozone layer degradation, such as halogenated hydrocarbons, methane, nitrogen dioxide, bromine, and chlorine compounds. These are commonly associated with fire suppressants, insulating foams, and agricultural chemicals, and are measured in kg CFC-11-eq.

Photochemical ozone creation potential (POCP), often referred to as “summer smog”, describes the formation of ground-level ozone due to photochemical reactions involving volatile organic compounds (VOCs) and carbon monoxide in the presence of nitrogen oxides and sunlight. Relevant substances include alkenes, aldehydes, ketones, alkanes, and halocarbons, with impacts expressed in kg NMVOC-eq. (non-methane volatile organic compounds).

The cumulative energy demand of a product or process is represented by the categories PERT (primary energy from renewable sources, total) and PENRT (primary energy from non-renewable sources, total). Both are measured in MJ, with PERT comprising inputs from water, wind, solar, and biomass energy, while PENRT includes fossil fuels, biomass and uranium. It is important to note that while ADPF and PENRT are largely overlapping in scope, the latter additionally includes energy contributions from biomass and uranium, while biomass like wood is also accounted for in PERT.

Finally, the water depletion potential (WDP) quantifies the relative potential of freshwater withdrawals to cause water stress, considering both availability and local demand. WDP is measured in cubic metres (m³) of water, whereby a high value indicates that water consumption is significant or takes place in water-scarce areas.

For analytical purposes, the indicators used here can be grouped into two general categories. On the one hand, ADPE, ADPF, PENRT, and PERT are classified as input indicators, which quantify raw material consumption and energy demand, representing indirect environmental pressure. On the other hand, the remaining indicators are outputbased impact categories that reflect direct environmental consequences of emissions or resource use.

2.2. Database

All data used to investigate the manufacturing processes of different bituminous roofing membranes were sourced primarily from publicly available Environmental Product Declarations (EPDs), with the exception of five membranes from the Ecoinvent database [29].

Initial information was drawn from the Ecoinvent and Ökobaudat databases [30]. These were subsequently supplemented with increasingly specific datasets from industry benchmarks provided by European Waterproofing Association (EWA), Hasse GmbH & Co. KG, and the roofing membrane manufacturers Phønix Tag Materialer (PTM) and Danosa. In total, 26 datasets were compiled (see

Table A1. EPD findings of bitumen roofing sheets of different sources. The coating mass contains bitumen, plastics, and oils for modification and can be summed up as organic mass. Minerals include all stone powders used for filling and surface sprinkling.

Source	Product	Carrier	Density [$\frac{\mathbf{kg}}{{\mathbf{m}}^2}$]	Thickness [mm]	Organic Ratio	Mineral Ratio	Year Ref.	Geography	Database
Hasse	Yearly production	Glasfiber, Polyester fleece, Aluminium	5.1	2.6	0.7	0.26 [7]	2021	DE	Ecoinvent 3.9
Ökobaudat	G200 S4	Glas fiber	5.0	4.0	-	-	2018	DE	GaBi
	PYE PV200 S4	Polyester fleece	5.21	4.0	-	-	2018	DE	GaBi
	PYE PV200 S4 ns	Polyester fleece	6.20	4.0	-	-	2023	DE	GaBi
	V60	Glas fleece	5.0	4.0 [4]	-	-	2023	DE	GaBi
Ecoinvent	Alu80	Aluminium	1.9 [2]	1.5 [2]	0.66	0.26	2007	RER [1]	Ecoinvent 3.10
	EP4	Polyester fleece	4.7 [2]	4.00	0.64	0.26	2007	RER [1]	Ecoinvent 3.10
	PYE PV200 S5	Glas fiber	6.02 [2]	5.00	0.70	0.19	2007	RER [1]	Ecoinvent 3.10
	V60 S4	Glas fleece	5.76 [2]	4.00	0.71	0.28	2007	RER [1]	Ecoinvent 3.10
	VA4	Glas fleece / Aluminium	5.33 [2]	4.00	0.72	0.26	2007	RER [1]	Ecoinvent 3.10
Phønix Tag Materialer	AeroTæt PF2000	Polyester fleece	2.24	2.10	0.54	0.42	2021	DK	GaBi 10.6.1.35
	Aero Tæt PF3200	Polyester fleece	3.39	2.90	0.54	0.42	2021	DK	GaBi 9.2.1.68, Ecoinvent 3.6, Eurobitume LCI 2019
	BituFlex PF5000 SBS	Polyester fleece	5.00	4.30	0.59	0.365	2021	DK
	BituFlex Kombi PF/GF5000 SBS	Polyester fleece / Glas fleece	5.30	4.40	0.59	0.365	2021	DK
	DuraFlex PF3500 SBS	Polyester fleece	3.30	2.90	0.59	0.365	2021	DK
	DuraFlex Kombi PF/GF3500 SBS	Polyester fleece / Glas fleece	3.30	2.90	0.59	0.365	2021	DK
	Topmembran PF4600 SBS	Polyester fleece	5.00	4.70	0.715	0.235	2021	DK
	Flammespærre GF3000	Glas felt	2.54	2.00	0.54	0.42	2021	DK
	Bundmembran PF4500 SBS	Polyester fleece	5.10	4.70	0.715	0.235	2021	DK
European Waterproofing Association	Benchmark [6]	Polyester and glas fleece	5.30	4.30	0.565	0.355	2019	BE BY DK FI FR DE IT LT NL NO PT RU ES SE	SimaPro 9, Ecoinvent 3.6, Plastics Europe 2014
Danosa [3, 5]	System NTV2/EXT1	Polyester fleece, Glas fiber	8.64	5.0	0.515	0.46	2021	ES	Ecoinvent 3.8, SimaPro 9.3
	System TPP1/NTG1	Polyester fleece, Glas fiber	7.96	5.8	0.515	0.46	2021	ES	Ecoinvent 3.8, SimaPro 9.3
	System TVA1/TVH1/TPC1/TPC2	Polyester felt/Glas fiber	9.04	6.15	0.515	0.46	2021	ES	Ecoinvent 3.8, SimaPro 9.3
	ESTERDAN PLUS 50/GP ELAST	Polyester fleece	6.0	3.5	0.515	0.46	2021	ES	Ecoinvent 3.8, SimaPro 9.3
	POLYDAN PLUS FM 50/GP ELAST	Polyester felt/Glas fiber	5.6	3.5	0.515	0.46	2021	ES	Ecoinvent 3.8, SimaPro 9.3
	System NTV6	Polyester fleece	7.84	5.0	0.515	0.46	2021	ES	Ecoinvent 3.8, SimaPro 9.3

Declarations for datatable Table A1: [1] Geographical allocation according to the ecoinvent model. [2] Average of comparable products based on independent research. [3] Bituminous membranes with emissions <10% are aggregated by the manufacturer. [4] Estimate based on density. [5] Substitution of approximately 18.72% bitumen (recycled oil compounds & tank residues), 100% ash (biomass), and 50% polyester (recycled LDPE). [6] Values converted from a 105-year service life to 35 years (without renewal). [7] Estimate based on similar bitumen-to-mineral ratios found in sealing membranes with comparably high bitumen content.

and

Table A2. Research findings for the production process A1 to A3 of bitumen roofing sheets of different sources. The functional unit is a square meter of sheet membrane. The freely available original data from the research work is presented here. These values are converted into a mass-specific unit for further investigation.

Quelle	Product Name	ADPE	ADPF	AP	EP-Freshwater	EP-Marine	EP-Terrestrial	GWP-Biogenic	GWP-Fossil	GWP-Luluc	GWP-Total	ODP	POCP	PENRT	PERT	WDP
		[kg Sb eq.]	[MJ]	[Mole H+ eq.]	[kg P eq.]	[kg N eq.]	[Mole N eq.]	[kg CO2 eq.]	[kg CO2 eq.]	[kg CO2 eq.]	[kg CO2 eq.]	[kg CFC-11 eq.]	[kg NMVOC eq.]	[MJ]	[MJ]	[m3]
Hasse	Jahresproduktion	3.68 × 10−5	1.78 × 102	2.25 × 10−2	6.36 × 10−4	4.60 × 10−3	4.07 × 10−2	1.41 × 10−2	5.19	3.33 × 10−3	5.21	8.53 × 10−7	3.22 × 10−2	1.78 × 102	3.25	1.23
Oekobaudat	G200 S4	5.286 × 10−7	193.2	0.009556	0.000004574	0.002042	0.02229	0.03031	3.301	0.002774	3.334	1.11 × 10−11	0.008925	167.6	5.694	0.09493
Oekobaudat	PYE PV200 S4	6.523 × 10−7	253.6	0.009933	0.00001524	0.002631	0.02835	0.0297	5.764	0.002741	5.796	2.197 × 10−11	0.01356	227.2	11.07	0.1462
Oekobaudat	PYE PV200 S4 ns	6.539 × 10−7	252.9	0.01005	0.00001526	0.002696	0.02908	0.02915	5.765	0.003105	5.798	2.203 × 10−11	0.01365	252.9	11.12	0.1485
Oekobaudat	V60	5.021 × 10−7	188	0.008386	0.000004493	0.001874	0.02058	0.01533	2.889	0.002736	2.907	1.306 × 10−11	0.008311	188	6.235	0.06025
Ecoinvent	Alu80	2.27 × 10−5	87.6	0.0222	0.00115	0.00012	0.037	0.082	3.82	0.0078	3.83	9.24 × 10−7	0.021	98.4	4.7	1.00
Ecoinvent	EP4	5.88 × 10−5	179.1	0.0218	0.00137	0.00020	0.037	0.201	4.92	0.0041	4.93	1.36 × 10−6	0.036	201.9	6.6	1.78
Ecoinvent	No Name (PYE PV 200 S5)	6.28 × 10−5	269.2	0.0325	0.00109	0.00026	0.057	0.383	8.56	0.0034	8.57	1.46 × 10−6	0.046	293.2	7.6	4.67
Ecoinvent	V60 (S4)	7.08 × 10−5	214.5	0.0223	0.00127	0.00021	0.036	0.183	4.90	0.0039	4.91	1.61 × 10−6	0.043	240.1	6.5	1.25
Ecoinvent	VA4 (wie V60 S4 + Al)	7.11 × 10−5	221.0	0.0287	0.00233	0.00031	0.048	0.395	6.11	0.0078	6.13	2.17 × 10−6	0.045	260.7	13.1	2.34
PTM	AeroTaet PF2000 (Dampfspaerre)	3.47 × 10−7	6.32 × 101	1.45 × 10−3	5.42 × 10−5	8.10 × 10−4	8.74 × 10−3	7.98 × 10−3	7.40 × 10−1	1.04 × 10−3	7.49 × 10−1	8.92 × 10−9	1.26 × 10−3	6.64 × 101	7.35	9.86 × 10−2
PTM	AeroTaet PF3200 (Dampfspaerre)	3.98 × 10−7	1.06 × 102	1.71 × 10−3	6.90 × 10−5	1.19 × 10−3	1.29 × 10−2	1.33 × 10−2	1.00	1.20 × 10−3	1.01	1.08 × 10−8	1.50 × 10−3	1.12 × 102	7.66	1.30 × 10−1
PTM	BituFlex PF5000 SBS	4.98 × 10−7	1.56 × 102	3.56 × 10−3	7.54 × 10−5	1.90 × 10−3	2.07 × 10−2	2.82 × 10−3	2.07	1.73 × 10−3	2.07	1.07 × 10−8	4.20 × 10−3	1.64 × 102	9.33	3.11 × 10−1
PTM	BituFlex Kombi PF/GF5000 SBS	5.27 × 10−7	1.63 × 102	3.82 × 10−3	7.98 × 10−5	1.98 × 10−3	2.15 × 10−2	1.09 × 10−2	2.13	1.80 × 10−3	2.14	1.15 × 10−8	4.24 × 10−3	1.70 × 102	9.55	3.14 × 10−1
PTM	DuraFlex PF3500 SBS	4.67 × 10−7	1.17 × 102	2.77 × 10−3	7.15 × 10−5	1.46 × 10−3	1.58 × 10−2	7.70 × 10−3	1.51	1.25 × 10−3	1.52	1.10 × 10−8	3.05 × 10−3	1.22 × 102	9.04	2.32 × 10−1
PTM	DuraFlex Kombi PF/GF3500 SBS	1.03 × 10−6	1.12 × 102	3.35 × 10−3	1.11 × 10−4	1.48 × 10−3	1.67 × 10−2	1.18 × 10−2	1.39	1.23 × 10−3	1.41	3.07 × 10−8	2.82 × 10−3	1.17 × 102	8.67	3.54 × 10−1
PTM	Topmembran PF4600 SBS	6.62 × 10−7	2.18 × 102	5.31 × 10−3	1.05 × 10−4	2.65 × 10−3	2.89 × 10−2	3.28 × 10−2	3.16	1.74 × 10−3	3.19	1.61 × 10−8	6.72 × 10−3	2.28 × 102	9.47	4.96 × 10−1
PTM	Flammespaerre GF3000	1.88 × 10−7	6.18 × 101	1.68 × 10−3	1.49 × 10−5	7.58 × 10−4	8.20 × 10−3	−1.51 × 10−2	4.79 × 10−1	9.04 × 10−4	4.65 × 10−1	1.45 × 10−12	8.00 × 10−4	6.53 × 101	6.92	4.00 × 10−2
PTM	Bundmembran PF4500 SBS	5.59 × 10−7	2.17 × 102	4.97 × 10−3	8.49 × 10−5	2.57 × 10−3	2.81 × 10−2	2.07 × 10−2	2.99	1.56 × 10−3	3.02	1.16 × 10−8	6.42 × 10−3	2.27 × 102	8.98	4.67 × 10−1
EWA	Bitumen-Durchschnitt [7]	3.68 × 10−6	6.13 × 101	9.17 × 10−3	4.27 × 10−5	1.42 × 10−3	1.58 × 10−2	−7.98 × 10−2	1.28	1.51 × 10−3	1.20	9.76 × 10−7	6.44 × 10−3	6.30 × 101	2.43	5.43
Danosa	System NTV2/EXT1	1.22 × 10−5	2.36 × 102	3.35 × 10−2	5.03 × 10−5	5.88 × 10−3	6.46 × 10−2	3.45 × 10−3	5.67	1.47 × 10−2	5.69	6.53 × 10−7	1.86 × 10−2	2.53 × 102	5.86	4.36
Danosa	System TPP1/NTG1	1.22 × 10−5	2.36 × 102	3.35 × 10−2	5.03 × 10−5	5.88 × 10−3	6.46 × 10−2	3.45 × 10−3	5.67	1.47 × 10−2	5.69	6.53 × 10−7	1.86 × 10−2	2.53 × 102	5.86	4.36
Danosa	System TVA1/TVH1/TPC1/TPC2	1.57 × 10−5	2.66 × 102	3.84 × 10−2	5.89 × 10−5	6.72 × 10−3	7.39 × 10−2	4.22 × 10−3	6.32	1.50 × 10−2	6.34	7.16 × 10−7	2.13 × 10−2	2.84 × 102	6.13	4.94
Danosa	ESTERDAN PLUS 50/GP ELAST	1.04 × 10−5	1.39 × 102	2.08 × 10−2	3.55 × 10−5	3.68 × 10−3	4.05 × 10−2	2.99 × 10−3	3.51	7.98 × 10−3	3.52	3.98 × 10−7	1.18 × 10−2	1.49 × 102	3.32	2.63
Danosa	POLYDAN PLUS FM 50/GP ELAST	1.04 × 10−5	1.39 × 102	2.08 × 10−2	3.55 × 10−5	3.68 × 10−3	4.05 × 10−2	2.99 × 10−3	3.51	7.98 × 10−3	3.52	3.98 × 10−7	1.18 × 10−2	1.49 × 102	3.32	2.63
Danosa	System NTV6	1.57 × 10−5	2.66 × 102	3.84 × 10−2	5.89 × 10−5	6.72 × 10−3	7.39 × 10−2	4.22 × 10−3	6.32	1.50 × 10−2	6.34	7.16 × 10−7	2.13 × 10−2	2.84 × 102	6.13	4.94

Declarations for datatable Table A2: [7] Estimate based on similar bitumen-to-mineral ratios found in sealing membranes with comparably high bitumen content.

). The data collected are broadly representative of the European market and were published between 2007 and 2021. A EPD assess products or product groups using environmental indicators in accordance with ISO 14044 [25].

Accordingly, when conducting a life cycle assessment (LCA) or preparing an environmental product declaration (EPD), it is essential to clearly define all framework conditions and model boundaries in the target and investigation definition. This includes defining the life cycle stages (e.g., raw material extraction, manufacturing, use, disposal or modules A–D according to EN 15804 [26], determining the system boundaries (which processes, inputs and outputs are included or excluded) and defining the geographical, technological and temporal scope of the data. In addition, methodological assumptions, data bases, uncertainties and cut-off criteria must be documented to ensure transparency, reproducibility and comparability. Only through this precise preliminary definition can it be ensured that the product system under consideration is realistically represented and that the environmental impacts can be reliably interpreted.

So these declarations represent the highest benchmark in the industry, which we therefore want to examine about their informational content. These are commonly used to ensure the highest possible degree of comparability, despite the considerable methodological flexibility permitted in the compilation of product declarations [18].

The environmental impact categories (as outlined in Section 2.1) used for assessing the manufacturing phase (modules A1–A3) are calculated as synthetic balances of material and energy inputs and corresponding emissions. The chemical outputs are classified into categories, weighted according to their relative environmental impact, and summarised across 13 emission-based indicators and two cumulative energy indicators. Each is reported in a standardised unit per kilogram of material. To ensure comparability between products of varying thickness, all indicators were normalised by dividing the impact values by the area-related mass (expressed in kg per ${\mathrm{m}}^2$). This approach yields, for example, a unit of kg C${\mathrm{O}}_2$-equivalent per kg of material in the case of the GWP, effectively isolating the influence of sheet thickness on the results.

All applied methods as follows and the corresponding datasets used for the analysis are summarised in

Table 1. Overview of analysed Datasets and EPDs with the applied statistical methods to different source sets. Extract from Appendix A Table A1.

Source	Product	Carrier	Overview	Correlation Matrix	Test of Significance	Multiple Regression
Hasse [31]	Yearly production	Gfib, Pf, Alu	x	x		x
Ökobaudat [30]	G200 S4	Gfib	x	x
	PYE PV200 S4	Pf	x	x	x
	PYE PV200 S4 ns	Pf	x	x	x
	V60	Gfl	x	x
Ecoinvent [29]	Alu80	Alu	x
	EP4	Pf	x
	PYE PV200 S5	Gfib	x
	V60 S4	Gfl	x
	VA4	Gfl/Alu	x
Phønix Tag Materialer [32]	AeroTæt PF2000	Pf	x	x	x	x
	Aero Tæt PF3200	Pf	x	x	x	x
	BituFlex PF5000 SBS	Pf	x	x	x	x
	BituFlex Kombi PF/GF5000 SBS	Pf/Gfl	x	x	x	x
	DuraFlex PF3500 SBS	Pf	x	x	x	x
	DuraFlex Kombi PF/GF3500 SBS	Pf/Gfl	x	x	x	x
	Topmembran PF4600 SBS	Pf	x	x	x	x
	Flammespærre GF3000	Glas felt	x	x		x
	Bundmembran PF4500 SBS	Pf	x	x	x	x
European Waterproofing Association [33]	Benchmark	Polyester/Gfl	x	x	x	x
Danosa [34]	System NTV2/EXT1	Pf, Gfib	x	x	x	x
	System TPP1/NTG1	Pf, Gfib	x	x	x	x
	System TVA1/TVH1/TPC1/TPC2	Pfelt/Gfib	x	x	x	x
	ESTERDAN PLUS 50/GP ELAST	Pf	x	x	x	x
	POLYDAN PLUS FM 50/GP ELAST	Pfelt/Gfib	x	x	x	x
	System NTV6	Pf	x	x	x	x

Pf = Polyester fleece; Gfib = Glas fiber; Gfl = Glas fleece; Alu = Aluminium; Pfelt = Polyester felt.

. This table provides an overview of the methodologies employed in the study, alongside the specific data sources utilised in processing the bitumen roofing membrane data. Since Hasse’s results used many different carrier materials, these were excluded from the significance tests, as were two of the Ökobaudat roofing membranes. The second of these were excluded because glass was significantly underrepresented in the sample. The results from the Ecoinvent database represent scientifically reliable values, but they have not been verified in accordance with standards and were therefore only used to provide an overview. Since the Ökobaudat platform does not publish any information on the composition of roofing membranes, this information was also excluded in the regression analyses.

2.3. Overview

To provide an initial overview of the dataset and to estimate characteristics of the underlying population, descriptive statistical measures including mean values, standard deviations, and 95% confidence intervals were calculated based on the full sample of 26 products (N = 26) [35] (p. 353 ff.). These statistical methods are well established in empirical research and represent the minimum standard required to adequately describe and interpret survey data.

2.4. Pearson Correlation Coefficients

Correlation analysis provides an efficient method for identifying both the strength and direction of relationships between two variables. When dealing with multiple variables, it is common practice to present all pairwise correlations within a matrix format to allow for a clear and descriptive overview.

In this study, Pearson correlation coefficients (r) were calculated for all combinations of the selected environmental indicators. These coefficients are presented in a triangular matrix format for improved clarity and interpretability. In addition, significance testing was conducted to assess the likelihood that the observed correlations occurred by chance. The thresholds proposed by Cohen [36] (p. 77 ff.) (Effect: r = 0.1: small|r = 0.3: medium|r = 0.5: strong) were applied to interpret the strength of correlations, and statistical significance was evaluated using two-tailed t-tests.

Due to the generic nature and relatively low specificity of the Ecoinvent dataset, its results are considered highly variable [20,37]. Consequently, only data from Hasse, Ökobaudat, Phønix Tag Materialer (PTM), EWA, and Danosa (see Table 1) were used for the correlation analysis. This adjustment reduced the effective sample size to 21 roofing membrane products.

2.5. Statistical Significance Testing

The remaining carrier materials were grouped into four categories: aluminium (n = 1), glass fibre (n = 3), polyester (n = 10), and a glass-polyester composite (n = 7). As aluminium and glass fibre category comprises only one and three entries, they are not considered representative and lacks sufficient statistical power to yield meaningful results (Table A1). With the dataset reduced to two material categories—polyester and glass-polyester composites (N = 17)—a two-tailed significance test was conducted to compare group means and examine the null hypothesis (${\mathrm{H}}_0$) that no true difference exists between them [38]. Due to the unequal group sizes and potential heterogeneity of variances, Welch’s t-test was applied instead of the standard Student’s t-test [35] (pp. 452–453).

2.6. Regression Analysis

The variables representing the mass ratios of bitumen and limestone were included as independent variables in the following analyses. Precise formulations regarding plasticisers, tyre-derived additives, or oils were not available and could therefore not be considered. Similar to correlation analysis, linear regression was used to model the relationship between two variables and to quantify the proportion of variance explained by the fitted model. This is represented by the coefficient of determination (${\mathrm{R}}^2$), which indicates the percentage of the total variance in the dependent variable that is explained by the model [35] (p. 245 ff., p. 711 ff.). Before performing regression modelling, the degree of multicollinearity between the two independent variables—bitumen mass ratio and limestone (mineral) mass ratio—was examined to ensure the validity of subsequent interpretations [35] (p. 742).

In addition to assessing general trends, the regression models aim to determine which of the two materials has a greater predictive influence on specific environmental indicators or whether one exhibits an inverse relationship. To this end, three separate regression models were developed for each environmental indicator:

• Bitumen mass ratio vs. limestone mass ratio;
• Bitumen mass ratio vs. indicator;
• Limestone mass ratio vs. indicator.

Finally, a multiple linear regression model was constructed for each indicator using both bitumen and limestone mass ratios as predictors. This allows for the evaluation of whether the combined contribution of both materials provides a better explanation of variance than each single predictor alone. The statistical significance of each regression model was assessed using the F-statistic, which compares the variance explained by the model with the unexplained variance (i.e., the residual variance) relative to the overall mean of the data [35] (p. 500 ff.). Due to the limited material-specific data available in the Ökobaudat dataset and the high variability observed in the Ecoinvent values, the sample size available for variance comparisons is restricted. Consequently, the three variable groups—bitumen mass ratio, mineral mass ratio, and environmental indicator (k = 3)—were analysed based on the 17 roofing membrane products drawn from the datasets provided by Hasse, Phønix Tag Materialer (PTM), EWA, and Danosa (N = 17).

2.7. Calculation Software

Python (v3.11.8) was used for data handling, exploratory analysis, and correlation tasks. It offers an easily accessible and well established programming framework that can be used for a wide variety of projects involving data processing, analysis methods and machine learning. The significance testing and regression analyses were conducted using RStudio (v2024.12.1) which is another free and user-friendly programming environment with a graphical interface that is suitable for efficient calculations.

3. Results

3.1. Overview

Table 2. Description of the environmental indicators of the whole sample with mean value, standard deviation and 95% confidence interval in their specific unit according to 1 kg of membrane mass.

	dof	Mean	Standard Deviation	CI 95%
ADPE	25	0.0000031	0.0000046	[0.0    0.0]
ADPF		34.3	8.2	[30.9    37.6]
AP		0.0029	0.0024	[0.0019   0.0039]
EP-freshwater		0.000080	0.00015	[2.0 × 10−5    14.2 × 10−5]
EP-marine		0.00043	0.00025	[0.00033   0.00053]
EP-terrestrial		0.0065	0.0032	[0.0052    0.0078]
GWP-biogenic		0.011	0.021	[0.0023    0.020]
GWP-fossil		0.73	0.39	[0.56    0.89]
GWP-luluc		0.00092	0.00083	[0.00058   0.0012]
GWP-total		0.73	0.40	[0.56   0.89]
ODP		0.00000010	0.00000013	[0.0   0.0]
POCP		0.0030	0.0029	[0.0018   0.0043]
PENRT		35.9	8.7	[32.3   39.5]
PERT		1.6	0.81	[1.2   1.9]
WDP		0.29	0.28	[0.17   0.40]

Note. dof = degree of freedom; CI = confidence interval.

presents the results of the first-order location and dispersion measures across the entire dataset. Both ADPF and PENRT reflect the use of fossil energy carriers, regardless of whether these are employed for energetic or material purposes, which explains the close similarity of their values. The observed discrepancy of 1.6 MJ between the two indicators may result from the additional consideration of peat, uranium, and biomass in the PENRT category. In contrast, the contribution of renewable energy—summarised under PERT—remains comparatively low, with an average value of 1.6 MJ and a relatively high variation (ranging from 1.2 MJ to 1.9 MJ). These results indicate that renewable energy sources such as wind and hydropower currently play only a minor role in the production processes examined. Although PERT cannot directly capture the substitution of fossil fuels, it nevertheless provides valuable insight into the share of renewable energy within the overall primary energy demand. When comparing total energy inputs, this value should not be contrasted with the full PENRT, but only with a fraction of the fossil indicator, since energy contained in materials such as bitumen mass or polymers remains embedded in the product after manufacturing rather than being consumed during the process itself.

The input parameters ADPE, ADPF, PENRT, and PERT exhibit the highest mean values overall, yet also display considerable variability across their respective standard deviations. Particularly for abiotic depletion of elements, the variance exceeds the mean itself, although the absolute magnitude of this indicator remains very low for the product under study.

The extent of this effect can be described using the coefficient of variation, defined as the ratio of the standard deviation to the mean, which appears at comparable magnitudes in several output-side indicators. Naturally, this propagation of variation from the input parameters into the output indicators must be acknowledged. In addition to the relatively high dispersions observed for ODP and GWP-biogenic, the variance of EP-freshwater exceeds its mean by a factor of more than eight. This substantial spread in EP-freshwater can mainly be attributed to industrial wastewater discharges and processes within the chemical industry, where the quantification of phosphorus emissions often relies on assumptions and estimations. Similarly, the water demand represented by the WDP for the production of one kilogram of roofing membrane varies considerably between 170 L and 400 L, which can likewise be traced back to similar methodological and data-related uncertainties.

The comparison of mass-specific parameters with the functional unit of one square metre provides an analytical framework to capture recipe- and process-dependent dimensions. These, in turn, may be governed by fundamental factors such as material composition, transport, additives, and energy sources, or by limitations in data availability and measurability within the underlying life-cycle assessment.

3.2. Welch Test

As summarised in Table 1, the following analysis considers only 17 data entries for the comparison of the two most frequently represented carrier materials—glass (n = 10) and polyester + glass (n = 7). To determine whether different reinforcement materials result in measurable differences across the environmental indicators,

Table 3. Results of the Welch test about the hypothesis if the impact categories according to their carrier materials polyester and glas + polyester are distinguishable with focus on their mean values about 1 kg membrane mass.

	t-Value	dof	p-Value	CI 95%	M_ g	SD_ g	M_ pg	SD_ pg
ADPE	0.40	11.2	0.694	[−0.0    0.0]	0.0000016	0.0000035	0.0000011	0.00000067
ADPF	2.5	13.1	0.025 **	[ 1.37    16.87]	36.1	7.2	27.0	6.8
AP	−1.1	13.1	0.278	[−0.0    0.0]	0.0019	0.0016	0.0028	0.0015
EP-freshwater	1.1	10.5	0.311	[−0.0    0.0]	0.000039	0.000080	0.000012	0.0000094
EP-marine	−1.1	12.9	0.286	[−0.0    0.0]	0.00046	0.00019	0.00056	0.00018
EP-terrestrial	−0.58	10.9	0.576	[−0.0    0.0]	0.0056	0.0017	0.0062	0.0020
GWP-biogenic	1.8	15.6	0.088 *	[−0.0    0.02]	0.0068	0.012	−0.0011	0.0058
GWP-fossil	1.0	15.9	0.326	[−0.12   0.35]	0.65	0.27	0.54	0.17
GWP-luluc	−1.4	9.8	0.202	[−0.0    0.0]	0.00067	0.00049	0.0011	0.00068
GWP-total	1.0	15.9	0.317	[−0.12    0.35]	0.66	0.27	0.54	0.18
ODP	−0.86	15.8	0.400	[−0.0   0.0]	0.000000042	0.000000084	0.000000072	0.000000056
POCP	0.52	13.5	0.613	[−0.0    0.0]	0.0021	0.0019	0.0017	0.00067
PENRT	2.4	11.5	0.032 **	[0.91    16.7]	37.4	6.5	28.6	7.3
PERT	2.0	12.5	0.068 *	[−0.07   1.61]	1.9	0.75	1.1	0.75
WDP	−2.2	9.2	0.059 *	[−0.61    0.01]	0.17	0.20	0.47	0.30

Two-sided t test; ** p < 0.05, * p < 0.10; Note. N = 17|g = “glas”|pg = “polyester & glas”; dof = degree of freedom; CI = confidence interval.

presents the outcomes of the Welch t-tests for all categories, including the corresponding test statistics and their significance levels. To ensure full transparency, the table reports not only the t-values, p-values, and the mean and standard deviation estimates for both groups, but also the degrees of freedom and confidence intervals, thereby providing a more comprehensive insight into the statistical results.

At a significance level of 5.0%, two tests yield robust results: ADPF (${\mathrm{t}}_{\left(13.1\right)}$ = 2.5, p = 0.025) and PENRT (${\mathrm{t}}_{\left(11.5\right)}$ = 2.4, p = 0.032). The true means for both indicators fall within nearly identical confidence intervals and, with 95.0% probability, are located between 1.37 MJ and 16.9 MJ for ADPF and between 0.91 MJ and 16.7 MJ for PENRT. The influence of fossil resources in these categories primarily relates to the provision of process energy for fibre production and to transport requirements, which can differ substantially between glass and polyester-based fibres. For example, if glass fibre can be effectively substituted by polyester fibre, thereby reducing the overall energy demand, this could plausibly explain the observed difference between the group means. The direct material contribution of polyester, as an organic compound, appears comparatively minor in this context.

The WDP indicator (${\mathrm{t}}_{\left(9.2\right)}$ = −2.2, p = 0.059) narrowly misses statistical significance, yet exhibits a remarkably high degree of variation. For the glass group, this variation even exceeds the mean itself. Moreover, since the confidence interval ([−0.61–0.01]) includes the value 0.0, no clear mean difference can be asserted. Nonetheless, the mean difference of approximately 300 litres of water remains substantial and could result from process variations or estimated data with higher uncertainty.

A similar pattern is observed for the PERT indicator (${\mathrm{t}}_{\left(12.5\right)}$ = 2.0, p = 0.068), where the inclusion of zero within the confidence interval again implies non-significance, although the mean difference of 0.8 MJ per kilogram—higher for pure glass fibre—may still be of interpretive relevance.

As introduced in Section 3.1, the coefficient of variation defined as the ratio of standard deviation to mean, serves as a measure of relative dispersion. This coefficient is notably high for GWP-biogenic in polyester + glass (${\displaystyle \frac{0.0058}{0.0011}}=5.3$) and for ADPE in the glass reinforcement (${\displaystyle \frac{0.0000035}{0.0000016}}=2.2$). In the development of Environmental Product Declarations (EPDs), it is essential to establish well-founded assumptions and simplifications to ensure an accurate representation of the product system. The pronounced variation in these indicators may therefore be indicative of differing underlying assumptions, particularly regarding the origin or production processes of the respective reinforcement materials.

3.3. Pearson Correlation Coefficients

The triangular correlation matrix (dof = 19), excluding the ecoinvent datasets, is presented in the

Table A3. Correlation coefficients r over all environmental input and output categories for all 21 datasets according to the production of 1 kg of membrane mass.

	ADPE	ADPF	AP	EP			GWP				ODP	POCP	PENRT	PERT	WDP
				freshw.	mar.	terr.	bio.	fos.	luluc	tot.
ADPE	1.00
ADPF	−0.13	1.00
AP	0.70 ***	−0.17	1.00
EP-freshwater	0.79 ***	0.08	0.16	1.00
EP-marine	0.76 ***	0.11	0.91 ***	0.37 *	1.00
EP-terrestrial	0.63 ***	0.09	0.91 ***	0.2	0.98 ***	1.00
GWP-biogenic	−0.06	0.84 ***	−0.10	0.13	0.18	0.17	1.00
GWP-fossil	0.48 **	0.60 ***	0.59 ***	0.23	0.67 ***	0.63 ***	0.47 **	1.00
GWP-luluc	0.38 *	−0.22	0.89 ***	−0.18	0.78 ***	0.86 ***	−0.07	0.38 *	1.00
GWP-total	0.47 **	0.61 ***	0.58 ***	0.23	0.67 ***	0.63 ***	0.49 **	1.00 ***	0.38 *	1.00
ODP	0.73 ***	−0.56 ***	0.69 ***	0.4 *	0.54 **	0.48 **	−0.61 ***	0.19	0.45 **	0.17	1.00
POCP	0.90 ***	0.19	0.74 ***	0.64 ***	0.78 ***	0.67 ***	0.13	0.79 ***	0.39 *	0.79 ***	0.62 ***	1.00
PENRT	−0.12	0.97 ***	−0.15	0.1	0.17	0.16	0.84 ***	0.53 **	−0.16	0.54 **	−0.56 ***	0.14	1.00
PERT	−0.54 **	0.31	−0.80 ***	0.017	−0.60 ***	−0.61 ***	0.31	−0.39 *	−0.65 ***	−0.39 *	−0.74 ***	−0.59 ***	0.33	1.00
WDP	0.35	−0.56 ***	0.66 ***	−0.093	0.43 *	0.47 **	−0.69 ***	0.02	0.62 ***	0.01	0.85 ***	0.26	−0.63 ***	−0.74 ***	1.00

*** p < 0.01, ** p < 0.05, * p < 0.10; Note. N = 21.

due to its extensive size (15 × 15 dimensions).

Focusing initially on the first two columns which already contain two of the most important input parameters ADPE and ADPF, a number of key relationships can be observed. The almost perfect correlation between ADPF and PENRT (r = 0.97; p < 0.01) is not particularly surprising, given the interdependencies discussed in the previous chapter. Likewise, the strong positive correlations of both indicators with GWP-fossil (r = 0.60; p < 0.01) and GWP-biogenic (r = 0.84; p < 0.01) can be attributed to the role of raw materials in energy provision. The higher correlation coefficient of 0.84 is likely associated with the constant share of biomass in the energy mix, which is primarily used for energy generation. In contrast, portions of the fossil resources are used materially rather than energetically, and thus do not contribute to C${\mathrm{O}}_2$-equivalent emissions which is explaining the comparatively lower correlation of 0.60.

Furthermore, a negative correlation is observed with both ODP (r = −0.56; p < 0.01) and WDP (r = −0.56; p < 0.01), which may be indicative of extended process chains or the adoption of newer production technologies. Examining the first column of the matrix provides additional insight: the elemental resource use, primarily comprising rare metals and minerals, correlates positively and significantly with nearly all environmental impact categories (r = 0.47 to r = 0.90). This result is entirely plausible, since the extraction, refinement, and transportation of such materials are inherently linked to energy- and emission-intensive industrial processes—precisely the principle underpinning this indicator, which aims to represent the environmental burden of scarce material utilisation in LCA. Exceptions to this pattern include biogenic and land-use related greenhouse gas emissions as well as water depletion.

Interestingly, ADPE also exhibits a negative correlation with PERT (r = −0.54; p < 0.05), suggesting in purely statistical terms that a higher share of renewable energy in the production of one kilogram of roofing membrane is associated with a reduction in the demand for rare metals and minerals. Consequently, this could lead to a decrease across most environmental impact indicators classified as output emissions in this analysis.

Examining the correlations for PERT in the penultimate row of the matrix reveals that both GWP-luluc (r = −0.65; p < 0.01) and WDP (r = −0.74; p < 0.01) are, unlike ADPE, directly and negatively associated with the use of renewable energy. A direct comparison of ADPF and PENRT correlations, on the other hand, yields no notable discrepancies, reflecting their conceptual similarity.

It should be emphasised that correlation coefficients do not imply causality. While the analysis demonstrates that various parameters within the dataset covary, the underlying causes are likely multifactorial and complex. Therefore, any interpretations drawn here remain hypothetical and could be refined further through path analysis or structural equation modelling to explore the directional relationships among variables in greater depth.

3.4. Regression Analysis

The principal mass fractions of the roofing membranes range from 54.0% to 71.5% coating mass and 19.0% to 46.0% mineral filler (Table A1).

The coating mass is primarily composed of organic substances such as bitumen, polymers, and oils, which can be directly associated with the environmental indicators ADPF and PENRT. In contrast, the mineral filler is not explicitly represented within any single indicator. While ADPE includes mineral compounds within its assessment of abiotic element depletion potential, it excludes commonly occurring materials such as calcium, quartz, and basalt. Given that both fractions together constitute 92% to 99% of the total mass of the membranes, the subsequent analysis aims to explore regression-based relationships between these two main categories of organic and mineral materials in order to identify potential dependencies. Because the EPD datasets often provide only approximate or aggregated composition data, and as manufacturers (understandably) rarely disclose their precise formulations, this approach offers an opportunity to infer potential environmental effects indirectly through chemical composition categories.

To assess potential collinearity between the two predictors, a multiple regression was first performed (

Table 4. Results of the regression model of organic mass ratio predicting the limestone proportion of 1 kg of membrane mass (dof = 15).

	Est.	Std. Err.	t-Value	p-Value	r2	F-Statistic
Intercept	1.02	0.026	39.1	<0.000 ***	0.98	596.7
Coefficient	−1.1	0.045	−24.4	<0.000 ***

*** p < 0.01.

). The results revealed an extremely high, nearly perfect correlation between the two variables (${\mathrm{r}}^2$ = 0.98), indicating a negative linear relationship (Coefficient = −1.1, Intercept = 1.02). The term “filler” for the mineral component therefore proves to be quite apt: as the bitumen content increases across the 17 datasets, the proportion of mineral matter decreases by approximately the same magnitude, and vice versa. The type of mineral filler appears to have little influence on this relationship, which can be attributed to its inert functional role—a characteristic also applicable to the surface granules included within the mineral fraction parameter. The intercept value of 1.02 in the model theoretically indicates that, in the absence of any bitumen, the formulation would consist almost entirely of mineral content.

Such high collinearity substantially complicates the interpretation of results. Because both independent variables are so closely interdependent, it becomes difficult to attribute causal effects with confidence. Even more advanced analytical approaches, such as partial correlation analysis or the examination of standardised regression coefficients, fail to yield meaningful distinctions. Consequently, the following analysis focuses on simple regression models employing only a single predictor variable at a time.

3.4.1. Prediction by Organic Mass

Table 5. Results for the linear regression models of predicting environmental indicators by ratio of organic mass (dof = 15) according to 1 kg of membrane mass.

	Intercept	Coefficient	p-Value	${\mathrm{r}}^2$	F-Statistic
ADPE	−0.0000011	0.0000039	0.540	0.03	0.4
ADPF	−8.34	67.49	0.005 ***	0.42	10.8
AP	0.0068	−0.0080	0.189	0.11	1.9
EP-freshwater	−0.00011	0.00022	0.018 **	0.32	7.1
EP-terrestrial	0.0083	−0.0044	0.554	0.02	0.4
EP-marine	0.00061	−0.00012	0.869	0.00	0.0
GWP-biogenic	−0.013	0.023	0.177	0.12	2.0
GWP-fossil	0.18	0.62	0.435	0.04	0.6
GWP-luluc	0.0039	−0.0054	0.012 **	0.35	8.2
GWP-total	0.17	0.65	0.426	0.04	0.7
ODP	0.00000011	−0.00000010	0.660	0.01	0.2
POCP	−0.00072	0.0042	0.417	0.04	0.7
PENRT	−5.2	64.8	0.011 **	0.36	8.5
PERT	0.081	2.5	0.466	0.04	0.6
WDP	1.3	−1.7	0.095 *	0.17	3.2

*** p < 0.01, ** p < 0.05, * p < 0.10.

presents the 15 regression models based on the organic mass-specific components of the roofing membranes, derived from 17 datasets provided by Hasse, PTM, EWA, and Danosa. The table includes the intercepts, the slopes of the linear regressors, as well as the p-values, coefficients of determination (${\mathrm{r}}^2$), and F-statistics for each model. The coefficient of determination serves as a measure of how much of the variance in the dependent variable can be explained by the predictor. To evaluate the explanatory strength of these relationships, the classification proposed by Cohen (1988) is applied, according to which an ${\mathrm{r}}^2$ value greater than 0.13 indicates a moderate level of explained variance, and values exceeding 0.26 represent a high level of explanatory power [36].

When examining the input-side indicators, clear relationships can be observed between the organic material content and both ADPF (${\mathrm{r}}^2$ = 0.42, ${\mathrm{F}}_{(1,15)}$ = 10.8, p = 0.005) and PENRT (${\mathrm{r}}^2$ = 0.36, ${\mathrm{F}}_{(1,15)}$ = 8.5, p = 0.011), both of which demonstrate statistically robust results at the 5.0% significance level. Variations in the ADPF indicator can therefore be attributed to differences in the amount of fossil-based material used in the products, explaining approximately 42.0% of the observed variance, while the remaining 58.0% likely stem from other process-related factors such as energy supply, transportation, or infrastructure. The regression for PENRT shows a slightly weaker relationship due to its inclusion of uranium, peat, and biomass, which account for roughly 6.0% of the variance difference between the two models (Difference between both ${\mathrm{r}}^2$ values, used to explain the discrepancy related to their constituent materials). In contrast, elemental resource depletion and renewable energy utilisation show no notable correlation with the organic mass content of the roofing membranes within this dataset.

Regarding the output-side environmental indicators, the models for EP-freshwater (${\mathrm{r}}^2$ = 0.32, ${\mathrm{F}}_{(1,15)}$ = 7.1, p = 0.018) and GWP-luluc (${\mathrm{r}}^2$ = 0.35, ${\mathrm{F}}_{(1,15)}$ = 8.2, p < 0.012) exhibit significant results. The regression for EP-freshwater, illustrated in Figure 1 (left), plots the individual data points against the organic mass fraction within the membrane formulations. The regression line fits particularly well within the range below 0.60 kg organic content, a region dominated by six datasets from Danosa with a mean organic share of 0.515 (see Table A1). More scattered values, showing higher deviations, are observed for two datasets from PTM and one benchmark product from Hasse, all of which exceed an organic fraction of 0.70. The eutrophication of freshwater is characterised by phosphorus emissions, which mainly arise during the production phase from mining effluents, refinery operations, chemical processing, and energy generation. According to the correlation matrix (Table A3), EP-freshwater shows no substantial relationship with ADPF (r = 0.08) or PENRT (r = 0.10), yet it can still be effectively explained through the mass-specific proportion of organic material.

The C${\mathrm{O}}_2$ emissions represented by the GWP-luluc indicator refer to carbon dioxide released through land-use and land-use change, such as the conversion of forested areas into industrial or urban zones. However, organic substances such as sawn timber or bio-based oils may counterbalance this effect in the A1–A3 life-cycle stages, as the carbon initially sequestered in biomass temporarily offsets emissions. These carbon compounds are only re-released at the end-of-life stage during disposal or energy recovery. The computed regression model indicates that a higher share of organic material in the membrane mass is associated with a reduction in C${\mathrm{O}}_2$ emissions from land-use and land-use change, explaining 35% of the variance observed in the sample. The triangular correlation matrix (Table A3) also reveals a similar, albeit non-significant, negative tendency for ADPF (r = −0.22). This suggests that GWP-luluc emissions are likely linked more strongly to material-related factors rather than to energy provision itself. In this regard, the use of bio-based oils, packaging materials, and organic additives emerges as a potentially influential driver in reducing land-use-related greenhouse gas emissions.

3.4.2. Prediction by Mineral Mass

Analogous to Section 3.4.1,

Table 6. Results of the linear regression models of predicting environmental indicators by ratio of limestone mass according to 1 kg of membrane mass (dof = 15).

	Intercept	Coefficient	p-Value	r2	F-Statistic
ADPE	0.0000019	−0.0000020	0.728	0.01	0.1
ADPF	50.50	−51.95	0.021 **	0.31	6.6
AP	−0.0012	0.0089	0.099 *	0.17	3.1
EP-freshwater	0.00009	−0.00019	0.029 **	0.28	5.8
EP-terrestrial	0.0031	0.0069	0.297	0.07	1.2
EP-marine	0.00038	0.00040	0.541	0.03	0.4
GWP-biogenic	0.006	−0.014	0.378	0.05	0.8
GWP-fossil	0.63	−0.24	0.740	0.01	0.1
GWP-luluc	−0.0013	0.0055	0.003 ***	0.46	12.6
GWP-total	0.63	−0.25	0.733	0.01	0.1
ODP	0.00000002	0.00000007	0.725	0.01	0.1
POCP	0.00260	−0.0024	0.612	0.02	0.3
PENRT	50.9	−49.0	0.039 **	0.25	5.1
PERT	2.540	−2.6	0.403	0.05	0.7
WDP	−0.2	1.4	0.137	0.14	2.5

*** p < 0.01, ** p < 0.05, * p < 0.10.

presents the 15 regression models based on the mineral mass fractions of the roofing membranes, derived from 17 datasets supplied by Hasse, PTM, EWA, and Danosa. The table lists the intercepts, slopes of the linear regressors, as well as the p-values, coefficients of determination (${\mathrm{r}}^2$), and F-statistics for each regression model to provide a comprehensive overview of the results.

According to the modelled regressions, an increasing proportion of mineral filler leads to a reduction in the fossil-based input indicators—ADPF (${\mathrm{r}}^2$ = 0.31, ${\mathrm{F}}_{(1,15)}$ = 6.6, p = 0.021) and PENRT (${\mathrm{r}}^2$ = 0.25, ${\mathrm{F}}_{(1,15)}$ = 5.1, p = 0.039)—by approximately 50 MJ per kilogram. This relationship can be clearly attributed to the high collinearity between organic and mineral base materials shown in Table 4 (${\mathrm{R}}^2$ = 0.98; Coefficient = −1.1), which explains why the strength of association is similar in both directions, albeit with an inverse trend compared to the regressions in Table 5. This further underlines the mass-equivalent substitution of coating mass with inert mineral fillers, which act primarily as neutral, volumetric components.

This reciprocal behaviour also helps to validate the output-side indicators. Both EP-freshwater (${\mathrm{r}}^2$ = 0.28, ${\mathrm{F}}_{(1,15)}$ = 5.8, p = 0.029) and GWP-luluc (${\mathrm{r}}^2$ = 0.46, ${\mathrm{F}}_{(1,15)}$ = 12.6, p = 0.003) display values comparable to those of the models in Table 5, but with inverse predictive tendencies, likely reflecting the direct substitution of organic materials. The second of these models is visualised in Figure 1 (right), which illustrates the individual data points along with the regression line. The GWP-luluc values are distributed consistently across the different proportions of mineral content, forming two small clusters at 0.365 and 0.46, corresponding to datasets from PTM and Danosa, respectively.

A noteworthy observation when comparing both model sets is that the degree of explained variance, as expressed by ${\mathrm{r}}^2$, depends on the type of base material. In predicting significant environmental indicator values, the regression models in Table 5 and Table 6 reveal that the organic materials provide a better predictive fit for EP-freshwater (${\mathrm{r}}^2$ = 0.32), whereas the mineral fillers achieve a higher explanatory power for GWP-luluc (${\mathrm{r}}^2$ = 0.46). However, according to the correlation matrix (Table A3; r = −0.18), no significant correlative relationship exists between these two indicators themselves. Whether these statistical associations also share common causal mechanisms cannot be determined based on the present regression models alone.

It is also important to highlight the non-significant findings. Although Table A3 shows that the input indicators ADPF and PENRT correlate strongly with GWP-fossil and GWP-biogenic, and negatively with ODP and WDP, no reliable predictions for these indicators can be derived from the models—despite the high mass-specific share of organic substances in the membrane formulations (54.0% to 71.5%). This can be explained by the fact that the ADPF indicator not only reflects the material input directly incorporated into the roofing membrane, but also encompasses process-related energy consumption, transportation, packaging, and cleaning agents, which are not explicitly represented within the coating mass data.

Furthermore, the strong collinearity between the two predictors—organic and mineral components—does not imply that their relationships with environmental indicators are transferable. Changes in elemental and mineral resources within the ADPE indicator cannot be explained by variations in the two main constituents, and therefore no mutual dependency is evident. Similarly, renewable energy sources, represented by PERT, exhibit no predictive potential in this context.

3.4.3. Multiple Regression Model

To conclude the discussion on regression analysis, a multiple regression model is now introduced in order to predict the output-side environmental indicators based on the organic and mineral main constituents of the roofing membrane mass. As noted at the beginning of this chapter, a very high degree of collinearity exists between these two predictors. This can make interpretation exceedingly difficult and, as [35] (p. 742) points out, it is generally advisable in such cases to introduce latent variables or reduce model complexity, a methodological issue also discussed in other research domains [39]. However, in the present context, the model is not intended to uncover causal relationships, but merely to provide adequate predictive capability for the given data.

In the Appendix A,

Table A4. Models of multiple regression about the independent variables of organic and mineral mass for each impact category (do${\mathrm{f}}_{\mathit{between}}$ = 2, do${\mathrm{f}}_{\mathit{within}}$ = 14). The level of significance of the intercepts and coefficients is shown by star markers and the adjusted determination coefficient is used in ${\mathrm{R}}^2$.

		Estimate	p-Value	R2 (adj.)	F(2,14)
ADPE	Intercept	−0.000062 *	0.197	0.09	1.8
	Prop. organics	0.000070 *
	Prop. minerals	0.000060 *
ADPF	Intercept	−339.7 ***	<0.000 ***	0.67	17.4
	Prop. organics	425.3 ***
	Prop. minerals	325.8 ***
AP	Intercept	−0.067 **	0.030 **	0.31	4.6
	Prop. organics	0.072 **
	Prop. minerals	0.073 **
EP-freshwater	Intercept	−0.00058	0.042 **	0.27	4.0
	Prop. organics	0.00073
	Prop. minerals	0.00046
EP-terrestrial	Intercept	−0.12 ***	0.002 ***	0.51	9.5
	Prop. organics	0.13 ***
	Prop. minerals	0.12 ***
EP-marine	Intercept	−0.012 ***	0.003 ***	0.49	8.8
	Prop. organics	0.013 ***
	Prop. minerals	0.012 ***
GWP-biogenic	Intercept	−0.29 ***	0.001 ***	0.56	11.3
	Prop. organics	0.33 ***
	Prop. minerals	0.28 ***
GWP-fossil	Intercept	−12.9 ***	0.003 ***	0.50	9.1
	Prop. organics	14.7 ***
	Prop. minerals	12.8 ***
GWP-luluc	Intercept	−0.026 ***	<0.000 ***	0.63	14.4
	Prop. organics	0.027 ***
	Prop. minerals	0.030 ***
GWP-total	Intercept	−13.2 ***	0.002 ***	0.52	9.7
	Prop. organics	15.1 ***
	Prop. minerals	13.1 ***
ODP	Intercept	0.00000078	0.795	0.11	0.23
	Prop. organics	−0.00000082
	Prop. minerals	−0.00000066
POCP	Intercept	−0.057 *	0.103	0.17	2.7
	Prop. organics	0.065 **
	Prop. minerals	0.055 *
PENRT	Intercept	−360.5 ***	<0.000 ***	0.63	14.8
	Prop. organics	448.5 ***
	Prop. minerals	349.3 ***
PERT	Intercept	14.7	0.597	0.06	0.5
	Prop. organics	−13.2
	Prop. minerals	−14.4
WDP	Intercept	6.9	0.160	0.12	2.1
	Prop. organics	−7.8
	Prop. minerals	−5.5

*** p < 0.01, ** p < 0.05, * p < 0.10.

lists, for all 15 indicators, the model parameters, including the intercepts and the respective coefficients for the organic and mineral mass components. As in the previous regressions, model performance is assessed using the p-value, the coefficient of determination (${\mathrm{R}}^2$)—or, in this case, the adjusted ${\mathrm{R}}^2$ to prevent overestimation of model fit (In multiple regression analysis, it is standard practice to report the adjusted ${\mathrm{R}}^2$ in order to account for the number of predictors)—and the F-statistic.

When first examining the input-side indicators, a strong level of significance again emerges for ADPF (${\mathrm{R}}^2$ = 0.67, ${\mathrm{F}}_{\left(\mathrm{2,14}\right)}$ = 17.4, p =< 0.000) and PENRT (${\mathrm{R}}^2$ = 0.63, ${\mathrm{F}}_{\left(\mathrm{2,14}\right)}$ = 14.8, p < 0.000). The explained variance resulting from variations in the composition of organic materials and mineral fillers is thus considerably higher than in the single-parameter regression models, even though both predictors account for identical shares of variance when considered individually (${\mathrm{r}}^2$ = 0.98). In the case of PENRT, a suppressor effect can be identified, as the combined model explains more variance than the sum of both single regressions from Section 3.4.1 and Section 3.4.2 (0.36 + 0.25 < 0.63). This indicates an information gain resulting from the interaction of both predictors within the multiple model.

In contrast to the earlier regression analyses, several output-side impact categories show significant results when both parameters are included. Only ODP, POCP, and WDP remain non-significant, though, according to Cohen [36] (p. 79 ff.), their ${\mathrm{R}}^2$ values above 0.13 still represent moderate levels of explained variance. The variance of EP-freshwater (${\mathrm{R}}^2$ = 0.27, ${\mathrm{F}}_{(2,14)}$ = 4.0, p = 0.042) can thus be explained by approximately 27.0%, although this indicator is actually better predicted by organic content alone (${\mathrm{r}}^2$ = 0.32). All other significant indicators in the multiple regression show adjusted coefficients of determination ranging from 0.31 to 0.63, which are consistently higher than those observed in their corresponding single-parameter models, indicating improved model performance when both predictors are considered simultaneously.

4. Discussion

The substantial variations observed for EP-freshwater and WDP illustrate a fundamental challenge in the interpretation of EPD data: assumptions regarding geographical allocation, cut-off criteria, data gaps, or differences in background databases can lead to considerable uncertainties and deviating results. These inconsistencies result in distortions of indicator values, which in turn limit the comparability of products and datasets. A valid interpretation of such categories therefore requires both a transparent documentation of the underlying data basis and a methodological harmonisation of the assumptions applied [40].

A central aspect in the interpretation of results which is also emphasised by [41], is the precise definition and delineation of the individual environmental indicators, as this is essential for an accurate assessment of the outcomes. The indicators used in this study reveal partly similar yet conceptually distinct focuses. On average, ADPF (34.3 MJ) and PENRT (35.9 MJ) per kilogram of roofing membrane show very similar tendencies, differing primarily through the inclusion of biomass, peat, and uranium within the PENRT category.

The absolute share of renewable primary energy represented by PERT is comparatively low, averaging 1.6 MJ, which highlights the predominantly fossil-based nature of current production processes. However, this indicator must be interpreted with greater specificity: PENRT includes not only the energetic use of fossil resources but also a significant proportion of organically derived material inputs—approximately 42.0%, according to the linear regression model (Table 5). When considering only the energetic components, the ratio between the pure energy category PERT and the estimated energetic share of PENRT nearly doubles to 7.7% (${\displaystyle \frac{1.6\mathrm{MJ}}{35.9\mathrm{MJ}·(1-0.42)}}=0.077$) indicating that the relative contribution of renewable energy sources is somewhat higher when adjusted for non-energetic fossil inputs.

Finally, the ADPE indicator also deserves mention. This category relates to the depletion of metallic and mineral resources, yet it does not include common minerals such as limestone, quartz, or basalt, which are typically used as fillers or surface granules in roofing membranes. This exclusion is of particular importance for the interpretation of results, as it implies that a considerable portion of the mineral mass employed in the products remains unrepresented within this indicator, thereby influencing the overall assessment of resource-related environmental impacts.

4.1. Predicting Carrier Material

The fossil resource indicator (ADPF) demonstrates a clear potential to differentiate between the carrier reinforcements made of glass and those composed of glass combined with polyester. Various methods of raw material sourcing and fibre production are conceivable, each of which could be reflected in the overall energy demand, leading to a significant mean difference of between 1.4 MJ and 16.9 MJ in the indicator values. It is also plausible that outsourced process chains play an important role, since in regions with less stringent environmental regulations, the associated energy supply may rely more heavily on fossil resources. Conversely, as suggested by [42], the higher energy efficiency of polyester fibre production could make the substitution of glass fibres an environmentally favourable alternative. However, as discussed by [43], considerable uncertainties in process modelling, along with the use of highly generic assumptions, may also influence these results and compromise the accuracy of the estimated energy mix.

4.2. Correlation Matrix

The results of the correlation analysis indicate that an increase in the share of renewable energy within the PERT category is associated with a reduction in the consumption of rare materials represented by ADPE. Moreover, ADPE exhibits significant relationships with nearly all examined output-side environmental indicators, suggesting an overall decrease in environmental burden with a higher share of renewable energy use. This leads to the well-founded conclusion that the integration of renewable energy sources into the production process can potentially reduce the total emissions of the environmental impacts, a finding that aligns with observations regarding production load and energy mix reported by [44,45]. This relationship becomes particularly evident in the context of metallic and mineral resource use, underlining the existing dependence on scarce raw materials and the resulting ecological challenges, as also emphasised by [46].

4.3. Regression Models

A prediction based on a single variable with regard to the environmental indicators yields significant results only for EP-freshwater and GWP-luluc. These can be positively associated with the use of organic and mineral-based materials, respectively. Specifically, 32% of the variance in EP-freshwater can be explained by the presence of organic substances, while 46% of GWP-luluc is attributable to mineral inputs. Neither indicator shows a direct correlation with each other nor with the primary input parameters ADPF and PENRT, which underlines their direct dependence on the type of material resources used.

For GWP-luluc, the influence of mineral masses may point towards land-use change activities such as deforestation for mining operations or the extraction of new mineral resources. In contrast, approximately one-third of freshwater eutrophication is linked to organic matter, which may plausibly stem from industrial wastewater generated during the extraction and refining of crude oil. This also helps explain the previously mentioned high variability associated with this indicator, as upstream processes often rely on assumptions and estimations, which can lead to significant distortions in the resulting values.

Although the fossil-based indicators ADPF and PENRT are strongly correlated with GWP-fossil and GWP-biogenic, and negatively correlated with ODP and WDP, there is no significant relationship between the total mass of roof membranes and these output indicators. This holds true even though the proportion of organic materials in the investigated membranes varies between 54% and 71.5%. This suggests that more complex interactions—stemming from specific processes related to energy production, transportation, packaging, or the detailed material composition—play a dominant role and cannot be adequately captured through this macro-level analysis alone.

Organic matter and mineral substances in the roofing membranes exhibit a strong inverse covariance, with an ${\mathrm{R}}^2$ value of 0.98 and a slope of −1.1, indicating an almost 1:1 substitution rate by mass. Fillers composed of minerals such as limestone powder, fine quartz sand, or basalt, as well as surface granules serving as a technical top-layer protection, can be identified in this sample as inert fillers. These serve as substitutes for the organic matrix, a relationship also highlighted in the studies by [47,48].

The results of the multiple regression analysis show that the indicators ODP, POCP, and WDP cannot be significantly explained by either the individual or combined variation of the independent variables proportion of organic and mineral materials. In contrast, almost all other input and output indicators, with the exception of EP-freshwater, exhibit a markedly increased explanatory power—ranging from 31% to 63%—when both material components are simultaneously included in the model. EP-freshwater, however, achieves its highest predictive accuracy (${\mathrm{R}}^2$ = 0.32) solely through the influence of organic materials.

The findings for the indicator PENRT are particularly noteworthy. A slight increase in explained variance is observed when both predictors are considered together, even though each variable alone shows only limited explanatory power. The high degree of collinearity between organic and mineral content complicates the interpretation of this effect. However, it is plausible that the inclusion of mineral inputs better reflects the increased energy demands associated with the extraction, transport, and processing of raw materials. Similar patterns have been observed [49,50].

4.4. Summary

In summary, it can be concluded that it is indeed possible to relate different carrier materials to specific environmental indicators. At least with regard to the fossil resource input represented by ADPF, a statistically significant distinction can be made depending on whether glass or a polyester-glass composite has been used in the roofing membranes. On the output side, however, no clear pattern emerges, although this may become more evident with further data collection and larger sample sizes.

The interactions within the impact categories of roofing membranes are highly diverse and allow for several insightful conclusions. Firstly, it becomes apparent that renewable energy sources are currently only marginally represented in the production process. Nonetheless, their increased usage is clearly associated with reductions in environmental emissions.

Secondly, by jointly considering the organic and mineral mass components in the membrane formulations, at least one third of the variation in environmental indicators can be predicted. This enables evidence-based estimations to be made during the early development stages of roofing membranes, allowing the environmental impacts of material compositions to be assessed more precisely. The near-perfect covariance between the two mass components further emphasises their direct relationship, although this simultaneously complicates the identification of causal effects.

Indicators that could not be predicted within the current model are likely influenced by more complex factors that cannot be sufficiently captured by a simple analysis of input materials. In this context, uncertainties in data acquisition, data gaps, assumptions, variations in background data, and simplifications within Environmental Product Declarations (EPDs) all play a substantial role and contribute significantly to the variability observed in the results. Additionally, it is understandable that manufacturers may be reluctant to disclose the detailed composition of their products, which in turn limits the extent to which more granular correlations and insights can be explored.

5. Conclusions

Climate change necessitates a more sustainable approach to the use of fossil resources, which remain a fundamental component of conventional bituminous waterproofing membranes in the construction sector.

Returning to the initial question, we can now conclude that “Yes, it’s worth it!”. The particular use of renewable energy sources and the efficient utilisation of raw materials can significantly mitigate potential environmental impacts. The high technical performance and durability of bituminous roofing membranes are difficult to reconcile with a reduction in organic content unless more complex material interactions are taken into account. In this regard, the substitution of fossil energy carriers with renewable alternatives represents an effective strategy for improving the sustainability of production processes. However, the feasibility of such a transition depends strongly on infrastructural prerequisites, the regional availability of renewable energy, and the electrification level of industrial operations—factors that will be critical for the future development of this product category.

EPDs are robust frameworks with precisely formulated scopes and cut-off criteria for the creation of product declarations in accordance with normative standards. Nonetheless, their validity is highly dependent on the quality of the underlying data and the assumptions were made. It remains unclear to what extent differences between products or their production processes can be consistently captured through EPD results alone. The findings of this study, however, suggest that meaningful insights can indeed be gained. Also, interrelationships between indicators can further reveal underlying patterns that support a more comprehensive understanding of environmental performance which are strongly contingent on the level of detail and effort invested in data collection and analysis. In practice, it is particularly important that planners and builders consciously compare EPDs and generic data sets and assess them appropriately.

Future research could build on these findings by expanding sample sizes, uncovering more detailed product-specific information, and developing more nuanced models capable of capturing the complexity of manufacturing processes. Pathway analyses may also serve as promising methods to assess material impacts over the entire life cycle and help identify viable alternatives or potential recycling strategies, thereby supporting the sustainable transformation of material systems in the construction sector. Further studies should also address the potential for the genuine recycling of bituminous membranes.