A Unified Life Cycle Impact Assessment Framework for Global Terrestrial Biodiversity Decline

Abstract

Life cycle impact assessment (LCIA) translates emissions and resource use into a set of environmental impact indicators, forming a key component of life cycle assessments. Current LCIA methods quantify biodiversity loss per pressure-impact pathway separately and at various spatial scales. We developed a new, spatially explicit and scale-consistent LCIA framework and applied it to derive characterization factors (CFs) for the impacts of land use, nitrogen emissions, and greenhouse gas (GHG) emissions on global terrestrial plant diversity. The framework combines the local potentially disappeared fraction of species (LPDF) due to the selected pressures with a multipressure species–area relationship (mp-SAR) model and with information on species endemism to estimate the global PDF (GPDF). For GHG emissions, we provide CFs for 250 GHGs and 20- and 100 year time horizons. Ecoregion-specific CFs of 6 land use types varied by 7 orders of magnitude, while emitting-region-specific CFs of ammonia and nitrogen oxides varied by 2 orders of magnitude. For land use, CF variability was driven almost entirely (>99%) by ecoregion. For nitrogen emissions, emitting region and compound each explained about one-third of the variability. Overall, our LCIA framework allows for consistent scaling of local species loss to global extinction while considering multiple impacts categories.

1. Introduction

Human activities such as agriculture, urbanization, and energy production pose increasing threats to biodiversity due to, among others, changes in land use, exploitation of resources, pollution, and global warming. These pressures can negatively impact biodiversity at different spatial scales. For example, they can alter local habitats, resulting in lower habitat quality and eventually a local loss of species. However, pressures can also affect species diversity at a regional scale, for instance by reducing habitat connectivity or inducing climate-driven shifts in species distributions. , Ultimately, local and regional species losses can result in global extinctions of species. To mitigate these impacts, it is important to accurately characterize the effects of human activities and resulting pressures on global biodiversity.

One way to assess the impacts of environmental pressures due to anthropogenic activities on biodiversity is via life cycle assessment (LCA). In LCA, the environmental impact of a product or services over its entire life cycle is assessed. In the life cycle impact assessment (LCIA) step of an LCA, characterization factors (CFs) are used that quantify the environmental impacts of different components of products or services on the different areas of protection, including human health, ecosystem quality, and resource scarcity. More specifically for ecosystem quality, LCIA methods aim to translate environmental pressures, such as land use, nitrogen emissions and greenhouse gas emissions, into potential impacts on biodiversity.

Many methods to quantify biodiversity loss in LCIA have been developed over the past decades. Eco-Indicator 99 was the first method to explicitly include biodiversity impacts at the end point level, accounting for the effects of land use, ecotoxicity, acidification, and eutrophication. The EI99 method used the potentially disappeared fraction of species (PDF) as an indicator. This method was refined in ReCiPe2008, which introduced improved damage pathways and incorporated the impacts of climate change on global biodiversity. More recent methods, such as Impact World+, LC-IMPACT, and ReCiPe2016, , have further advanced the representation of biodiversity impacts by increasing spatial resolution and including the impacts of water use and ozone formation on biodiversity. Impact World+ and LC-IMPACT continued using the PDF as a metric, whereas ReCiPe transitioned to time-integrated, absolute species loss (species·yr^–1) to improve integration of effects across terrestrial, freshwater, and marine ecosystems. Additionally, LC-IMPACT introduced the use of the species-area relationship (SAR) and species vulnerability scores to calculate global extinction risks from local and regional species losses, offering a first attempt to harmonize the spatial scales across impact categories.

The Global Guidance for Life Cycle Impact Assessment (GLAM) initiative of the UN environment aims to provide guidance and facilitate scientific consensus on LCIA indicators. , Within the GLAM initiative, several advancements on LCIA indicators have been made, including, but not limited to, the development of the first spatially explicit CFs for climate change for terrestrial and marine ecosystems; inclusion of land use intensities and fragmentation into land use CFs; and new regional CFs for terrestrial acidification.

Despite these recent developments, current LCIA methods lack a consistent and integrated approach for quantifying biodiversity effects across different impacts categories for two main reasons. First, because each impact category is modeled separately, species loss can be counted multiple times over the different impact categories. This can result in an overestimation of the predicted effects on local and global biodiversity. Second, the spatial scale at which biodiversity effects are assessed can differ between impact categories. For example, in ReCiPe2016, , biodiversity impacts of land and water use are quantified based on local biodiversity loss, whereas climate change impacts represent global species extinctions. This is problematic because local species loss is not directly comparable to global species loss. Although LC-IMPACT provides global CFs for different impact categories, it applies different methods depending on the impact, e.g. using local effects combined with SAR and vulnerability scores for land use, regional effects and vulnerability scores for acidification, and direct global extinction probabilities for climate change.

Here, we develop a new, spatially explicit and scale-consistent LCIA framework to estimate CFs based on global species loss, and applied it to derive a new set of CFs for the impacts of land use, nitrogen deposition, and climate change on global terrestrial plant diversity in a consistent and integrated way. We focus on plants because they fulfill a crucial role in terrestrial ecosystems, among others through their central function in primary production and regulation of the carbon cycle. The unified LCIA framework first assesses local species losses and then aggregates these to regional loss using a multipressure SAR model (mp-SAR). Regional loss is quantifed at the level of terrestrial ecoregions, as they represent relatively homogeneous ecological systems for which regional SARs can be applied. Subsequently, it uses the relative endemism richness of ecoregions to estimate global extinction threat as a proxy of global species loss, and it derives new CFs for global terrestrial biodiversity impacts.

2. Methods

2.1. Description of the Framework

2.1.1. Characterization Factors

Our unified framework quantifies characterization factors (CFs) based on the change in global potentially disappeared fraction of species (GPDF) per unit of emission (e.g., kg CO₂) or intervention (e.g., m²·yr cropland occupation) x. Each CF is composed of a fate factor (FF) and an effect factor (EF) and is calculated as follows

$$\mathrm{C}{\mathrm{F}}_{x,e}=\sum _r(\mathrm{F}{\mathrm{F}}_{x,e\to r}·\mathrm{E}{\mathrm{F}}_{x,r})$$ 1

where e is the emitting region, i.e. the region in which the emission or intervention x takes place, and r are the receiving regions, which are spatially differentiated at the ecoregion level. See Supporting Information Table S1 for an overview of the spatial delineation used in the framework.

The FF is defined as a change in pressure x (p _x ) at receiving region r due to a change in emission or intervention i _x at emitting region e

$$\mathrm{F}{\mathrm{F}}_{x,e\to r}=\frac{\mathrm{\Delta }{p}_{x,r}}{\mathrm{\Delta }{i}_{x,e}}$$ 2

The EF is defined as a change in GPDF due to a change in pressure x at receiving region r

$$\mathrm{E}{\mathrm{F}}_{x,r}=\frac{\mathrm{\Delta }\mathrm{G}\mathrm{P}\mathrm{D}{F}_{x,r}}{\mathrm{\Delta }{p}_{x,r}}$$ 3

For land use two different types of CFs are calculated, one for land occupation (CF_{lu_occ}; GPDF/m²) and one for land relaxation (CF_{lu_rel}; GPDF × yr/m²). Land relaxation refers to the recovery phase after land use is terminated, during which biodiversity is assumed to recover to its natural state. , The CF_{lu_occ} is calculated using eq , whereas CF_{lu_rel} is calculated as the product of CF_{lu_occ} and half the recovery time in the receiving region r (t _reg,r; years), following Koellner and Scholz

$${\mathrm{C}\mathrm{F}}_{\mathrm{l}\mathrm{u},\mathrm{r}\mathrm{e}\mathrm{l},\mathrm{r}}={0.5\times t}_{\mathrm{r}\mathrm{e}\mathrm{g},\mathrm{r}}{\times \mathrm{C}\mathrm{F}}_{\mathrm{l}\mathrm{u},\mathrm{o}\mathrm{c}\mathrm{c},\mathrm{r}}$$ 4

2.1.2. Quantifying the Global Potentially Disappeared Fraction of Species

The EFs are derived by calculating the average difference in GPDF _x,r between the current state and the natural state per unit of pressure x increase (eq ). The GPDF _x is calculated from the total GPDF of the combined pressures and the relative contribution of pressure x to the total local potentially disappeared fraction of species (LPDF) across the area (A; m²) of all grid cells i in ecoregion j

$$\mathrm{G}\mathrm{P}\mathrm{D}{F}_{x,j}=\mathrm{G}\mathrm{P}\mathrm{D}{F}_j·\frac{\sum _i({\mathrm{L}\mathrm{P}\mathrm{D}F}_{x,i,j}·A_{i,j})}{\sum _i(\sum _x(\mathrm{L}\mathrm{P}\mathrm{D}{F}_{x,i,j})·A_{i,j})}$$ 5

The GPDF _j is the total global PDF in ecoregion j, calculated using eq

$$\mathrm{G}\mathrm{P}\mathrm{D}{F}_j=R{E}_j·{\mathrm{R}\mathrm{P}\mathrm{D}F}_j$$ 6

where RE _j is the relative endemism richness of vascular plant species in ecoregion j and RPDF _j the regional PDF of ecoregion j.

The relative endemism richness of the ecoregions is taken from Moreira et al., where it is calculated as

$$R{E}_j=\frac{\sum _{\mathrm{s}}{n}_{\mathrm{s},\mathrm{j}}^{-1}}{S}$$ 7

where, n _s,j is the number of ecoregions n over which each species s in ecoregion j is present, and S is the total number of species globally.

The RPDF _j is calculated as

$${\mathrm{R}\mathrm{P}\mathrm{D}F}_j=1-{\left(\frac{\sum _i{h}_{i,j}·A_{i,j}}{{\sum }_i{A}_{i,j}}\right)}^{z_j}$$ 8

where, h _i,j is the environmental suitability for the species group of concern of grid cell i in ecoregion j, A _i,j the area of grid cell i in ecoregion j (m²), and z _j the slope of the mp-SAR in ecoregion j.

The environmental suitability h is determined by the combined effect of multiple pressures on species richness using a multipressure SAR model (mp-SAR). The mp-SAR assumes that pressures act independently, and that response addition can be used to calculate the combined environmental suitability. , In this way, each pressure affects only the fraction of species surviving the other pressures, ensuring that cumulative impacts never exceed affecting 100% of the species pool. The environmental suitability h is calculated as

$$h_{i,j}={(\prod _{x}1-{\mathrm{L}\mathrm{P}\mathrm{D}F}_{x,i,j})}^{1/z_j}$$ 9

where, LPDF _x,i,j is the local potentially disappeared fraction of species in grid cell i within ecoregion j caused by pressure x, and z _j is the slope of the mp-SAR in ecoregion j.

2.2. Application of the Framework

The framework was applied to calculate characterization factors (CF) for the effects of GHG emissions, land use (occupation and relaxation), and nitrogen emissions on global vascular plant species diversity. Below, we describe how the components of in the framework were parametrized for this application.

2.2.1. Upscaling

2.2.1.1. Z-Values

Biome-specific z-values of Gerstner et al. were linked to the ecoregions by assigning the same biome-specific z-value to all ecoregions within a given biome (Supporting Information Table S2). Gerstner et al. do not report a z-value for the mangrove biome, so the ecoregions that fall within the mangrove biome were assigned a global average z-value (z = 0.179).

2.2.1.2. Relative Endemism

The relative endemism richness (ER _j ) of vascular plant species of 818 ecoregions were obtained from Moreira et al. Additionally, pressure-specific data were used, which are described in more detail below.

2.2.2. Land Use

2.2.2.1. Fate Factor

For land use, the pressure in receiving region r is the area of land use (m²), which equals the elementary flow collected in the life cycle inventory (LCI) phase, and therefore no fate factor is applicable, i.e. $F{F}_{x,e\to r}$ equals 1 for each land use type.

2.2.2.2. Local Effect Data

LPDF responses (LPDF/m² land use type x) for six land use types x (urban land, forestry, pasture minimal intensity, pasture high intensity, cropland minimal intensity, and cropland high intensity; see Supporting Information Table S3 for detailed description of land use types) were calculated based on data underlying the meta-analysis of Gallego-Zamorano et al. These data were used to quantify, per land use type and per underlying data source, a remaining species richness value as RSR_l, s = max­(SR_l,s/SR_n,s, 1), where RSR is the remaining species richness in land use type l reported in data source s, SR_l,s is the plant species richness in land use type l in data source s, and SR_n,s is the plant species richness in a natural reference situation in data source s. The values were truncated at 1 to avoid increases in species richness. Next, a meta-analytical beta regression model was fitted in order to obtain an RSR value per land use type. Finally, LPDF was obtained from the fitted RSR values as 1-RSR. More details on the model fitting can be found in the Supporting Information methods. The new LPDF responses for land use can be found in Supporting Information Table S4.

A global land use map (10 arc-sec) for the year 2015 was obtained from the GLOBIO model. The land use categories of the GLOBIO map were harmonized with the land use categories of Gallego-Zamorano et al. following the same approach as Moreira et al. (Supporting Information Table S3). Additionally, the Δp _x,r was defined as the total area (m²) occupied by land use type x in receiving region r (i.e., in the specific ecoregion) as indicated on the GLOBIO map.

2.2.2.3. Land Use Recovery Times

For land relaxation, the ecoregion-specific t _reg values were determined based on biomes-realm specific t _reg values. Following De Baan et al., the t _reg values for recovery after intensive land use were used for land use types “agriculture minimal intensity”, “agricultural high intensity”, “urban”, and “forestry” in nonforest ecoregions, and “pasture minimal intensity” and “pasture high intensity” in forest ecoregions. For “forestry” in forest ecoregions and “pasture minimal intensity” and “pasture high intensity” in nonforest ecoregions, the t _reg values for recovery after extensive land use were used.

2.2.3. Greenhouse Gas Emissions

2.2.3.1. Fate Factor

The emission of a GHG has a global effect regardless of where the emission occurs. Therefore, the emitting region e can be any grid cell i and the receiving region r is the world. FF _x,t was used to describe the global mean temperature increase due to emission of greenhouse gas (GHG) x over time horizon t. In the present study, FFs (°C/kg GHG emission) for 250 GHGs and over two time horizons (20 and 100 years) were calculated as follows

$$F{F}_{x,t}=\mathrm{G}\mathrm{W}{P}_{x,t}·\mathrm{I}\mathrm{A}\mathrm{G}\mathrm{T}{P}_{\mathrm{C}{\mathrm{O}}_2,t}$$ 10

where, GWP _x,t is the global warming potential (kg CO₂-eq/kg GHG) for GHG x and time horizon t and $\mathrm{I}\mathrm{A}\mathrm{G}\mathrm{T}{P}_{\mathrm{C}{\mathrm{O}}_2,t}$ (°C·yr/kg CO₂) the time-integrated absolute global temperature potential for 1 kg of CO₂ over time horizon t. The $\mathrm{I}\mathrm{A}\mathrm{G}\mathrm{T}{P}_{\mathrm{C}{\mathrm{O}}_2,t}$ values for the 20- and 100 year time horizon were taken from Joos et al. and are 9.03 × 10^–15 and 4.76 × 10^–14 °C·yr/kg CO₂, respectively. For both time horizons, GWPs included climate-carbon feedbacks and were obtained directly from IPCC.

2.2.3.2. Local Effect Data

A mixed beta-regression model linking LPDF to global mean temperature increase since preindustrial time (T _i ; °C) was obtained from Schipper et al.

$${\mathrm{L}\mathrm{P}\mathrm{D}F}_{T,i}=1-\frac{1}{1+\exp (-(\beta 0+\beta 1·T_i))}$$ 11

where β0 (2.867) and β1 (−0.467) are the regression parameters, and the T _i of each terrestrial grid cell i is 1.1004 °C in the year 2014 (obtained from IPPC). T_i was used to define Δp _x,r in eq .

2.2.4. Nitrogen Emissions

2.2.4.1. Fate Factor

Fate factors (kg N deposition/kg NH₃ or NO _x emitted) were derived for 55 emitting source regions (Supporting Information Table S5) and two forms of nitrogen emission x, i.e. NO _x and NH₃. The $F{F}_{x,e\to r}$ describes the amount of nitrogen deposited at receiving location r caused by an emission of pressure x (NO _x or NH₃) in source region e. The FFs were determined using source-receptor matrices derived with the Tracer Model version 5 (TM5-FASST). The receiving (i.e., receptor) regions have a spatial resolution of 1° × 1°. The source-receptor matrices contain negative values reflecting small deviations (1.5%) in the mass balance of the source-receptor matrices and were set to zero to avoid negative nitrogen deposition flows.

2.2.4.2. Local Effect Data

A new relationship between LPDF and nitrogen addition was established using a beta-regression model and the data set of Midolo et al. Remaining species richness values were calculated based on the data by Midolo et al., and a beta regression model was fitted between RSR and nitrogen addition, following the approach of Gallego-Zamorano et al. Thereby considering the inclusion of three relevant moderators: 1) the duration of nitrogen addition, 2) the soil cation exchange capacity, and 3) the mean annual temperature. More details on the model fitting can be found in the Supporting Information methods. The most parsimonious beta-regression model estimates remaining local plant species richness due to nitrogen addition while accounting for the duration of nitrogen addition (in years) and local soil cation exchange capacity (centimole per kilogram). Following this model, the LPDF due to nitrogen deposition was calculated as follows

$${\mathrm{L}\mathrm{P}\mathrm{D}\mathrm{F}}_{\mathrm{N}\mathrm{d}\mathrm{e}\mathrm{p},i}=1-\{[1]/[1+\exp (-(2.699-1.087·\log 10({\mathrm{N}}_i)-1.61·\log 10(duration)+1.9·\log 10({\mathrm{C}\mathrm{E}\mathrm{C}}_i)\left]\right\}$$ 12

where N _i is the amount of nitrogen deposited (kg ha^–1 yr^–1) in grid cell i, duration is the duration of the nitrogen deposition (years), and CEC _i the cation exchange capacity (cmol/kg) of the soil in grid cell i.

As input for the model, we obtained global raster maps with spatial resolution of 0.25° with the annual nitrogen deposition between the years 1984 and 2015 and the CEC from Gallego-Zamorano et al. For the construction of these raster maps, Gallego-Zamorano et al. originally obtained spatial annual nitrogen deposition data (2 × 2.5°) from Ackerman et al. and the CEC data from Soilgrids (250 m resolution). For the duration of nitrogen deposition, we assumed 32 years, which equals the whole period nitrogen deposition data was available. Furthermore, we assumed that the LPDF could only be affected by nitrogen deposition in pristine vegetation. That is, we assumed that the additional nitrogen deposition in urban areas, forestry, croplands, and pastures is subordinate to the land use itself, and therefore does not affect the local species richness response. The average nitrogen deposition (kg ha^–1 yr^–1) in ecoregion j, determined using the raster maps of Gallego-Zamorano et al., was used to define Δp _x,r in eq .

2.2.5. Country-specific Characterization Factors

In addition to the ecoregion- and source region-specific CFs, we also calculated country-specific CFs for land use and nitrogen deposition. The country-specific CFs were calculated as the area-weighted averages of the ecoregion- or source region-specific CFs that intersect the countries. For countries overlapping ecoregions without a CF, the country-specific CF is derived solely from the ecoregions for which a CF is available.

2.2.6. Software

We performed the entire modeling framework and derivation of the CFs in R (version 4.4.1). We projected all input raster maps to Mollweide equal-area projection, and resampled rasters to a resolution of 10 arc-sec using bilinear interpolation for continuous variables and nearest neighbor interpolation for categorical variables.

2.3. Characterization Factor Comparison

We compared the newly derived CFs with the CFs used in the ReCiPe and LC-IMPACT methods. − Note that we could not compare CFs from ReCiPe directly, because a different unit is used, i.e., species loss per unit of intervention rather than PDF per unit of intervention. Therefore, we compared CFs based on the relative impact per unit of intervention compared to the impact of 1 kg of CO₂ (100 year time horizon) of the corresponding method. We made the comparisons based on country-specific CF values, except for the CFs for land use from ReCiPe as only global values were available. For land use, we distinguished different categories by ReCiPe, LC-IMPACT, and the present study. An overview of land use type comparisons with each other can be found in Supporting Information Table S6.

3. Results and Discussion

3.1. Global Characterization Factors

3.1.1. Land Use

Figure shows that CFs for land occupation differ by multiple orders of magnitude. The lowest CFs were found for dessert, steppe, and tundra ecoregions, whereas the largest CFs were found for forest ecoregions. Note that not for each ecoregion and land use type a CF could be determined due to the absence of certain land types in certain ecoregions in the underlying land use map. This occurred mainly for the land use types “cropland high intensity” and “pasture high intensity”. The CFs for land occupation and land relaxation were strongly correlated across ecoregions (r > 0.82, Figure , Supporting Information Figure S1, Supporting Information Table S7), indicating that ecoregions with relatively high CFs for land use occupation also tend to have high CFs for land use relaxation. Via variance decomposition analysis (Supporting Information methods), we found that the geographical location, i.e. the ecoregion, was the dominant factor (>99%) contributing to the variability between land use CFs, compared to the land use type (<0.05%). Similarly, Moreira et al. found that the global extinction threat of vascular plant species depends more on the relative endemism richness, than differences in local relative richness responses. Both findings indicate that, with respect to global biodiversity impacts, the geographical location is substantially more important than the type of land use. This is because impacts in regions with high endemic richness pose substantially greater global extinction risks than impacts in regions with few endemic species. Additionally, within ecoregions, the highest CFs (land use occupation and relaxation) were found for high intensity croplands and the lowest for urban areas.

1.

Ecoregion-specific land use occupation characterization factors (CF; GPDF/m²) for six land use types. Gray colors indicate the absence of a CF for the specific ecoregion. This occurs when a particular land type is not present in that ecoregion in the GLOBIO land-use map, making it impossible to calculate a corresponding EF (eq ).

3.1.2. Greenhouse Gas Emissions

Figure shows the CFs for the four most important GHGs for a 20- and 100 year time horizon. CFs for the 246 other GHGs are given in Supporting Information 2. The largest CF was found for sulfur hexafluoride (SF6), which is around 4 orders of magnitude larger than the CF of CO₂ (Supporting Information 2). In general, the CFs for 100 year time horizon are up to a factor of 7 larger than those for the 20 year time horizon.

2.

Global characterization factors (CF; GPDF × year/kg) for four greenhouse gases considering a 20- (A) and 100 year (B) time horizon.

3.1.3. Nitrogen Emissions

Figure shows the CFs per emitting region for NH₃ and NO _x . The CFs for NH₃ were on average three times larger than the CFs for NOx. For both compounds, CFs were highest in in Central and South America, the Mediterranean, and South Africa. In contrast, CFs were lowest in northern Europe, Central Africa, and Chili. A variance decomposition analysis (Supporting Information Methods) showed that the emitting region and nitrogen compound explained 16% and 50% of the variability in the CFs, respectively.

3.

Global characterization factors (CF; GPDF × year/kg) for NH₃ (a) and NO _x (b) emissions per emitting region.

3.2. Method Comparison

In this study, we presented a unified framework to derive CFs for impacts on global biodiversity. The novel methodology allows for a consistent scaling of local species loss to global extinction across impacts categories (here shown for greenhouse gas emissions, land use, and nitrogen emissions), which is lacking in current LCIA methodologies. − The approach results in a single global end point, i.e. GPDF, which is comparable among impact categories, resolving the currently inconsistent comparison and aggregation of biodiversity impacts across different impact pathways in current LCIA methodologies. In addition, we used a novel method, the mp-SAR, to combine multiple local species losses into a combined local species loss before estimating global extinctions. This method ensures that biodiversity effects are not overestimated, avoiding double counting of global species losses between impact categories. This is in contrast with other recently developed single-pressure methods, such as Scherer et al. which include more land categories and taxonomic groups, and account for land fragmentation, but their approach addresses only one impact category and does not combine multiple pressures. Finally, we implement a novel approach to estimate global species loss based on regional species loss and regional endemism, which advances on the current state of the art of using species’ vulnerability scores , to estimate potential global extinctions.

Figure compares the impact of the CFs for land use and nitrogen emissions (acidification) per unit of intervention relative to the CF of 1 kg CO₂ (100 year time horizon) for different LCIA methodologies. For land use, the relative impacts compared to the impact of 1 kg CO₂, are similar for the new CFs, LC-IMPACT and ReCiPe. For the emission of 1 kg of NH₃ and NO _x , the relative impacts of the three methods fall within the same range. However, for the new CFs there is a shift toward stronger relative impacts of NH₃ and NO _x emission, with impacts that are 10 to 1000 times larger than those of CO₂. Furthermore, the regional variability between our country-specific CF (NH₃: σ² = 31,990 and NO _x : σ² = 2265, N = 166) is smaller than the variability between the country-specific CFs of LC-IMPACT (NH₃: σ² = 80,019, and NO _x : σ² = 29,318, N = 166). Compared to the country-specific CFs of ReCiPe (NH₃: σ² = 344, and NO _x : σ² = 7728, N = 166) the variability in CFs is larger for NH₃ and smaller for NO _x .

4.

Relative impact of characterization factors (CF) for land use occupation and nitrogen emissions per unit of intervention. Relative CFs are calculated relative to the global CF of CO₂ over a 100 year time horizon. Dots represent only the countries for which CFs are available for all three methods (N_lu = 170 and N_nd = 166), with the exception of ReCiPe global average land-use CFs. An overview of land use types used per method is provided in Supporting Information Table S6.

The differences in the relative impacts of NH₃ and NO _x between the CFs derived in the present study and those from LC-IMPACT and ReCiPe are attributable to methodological differences. First, different source-receptor matrices were used to calculate FFs. More importantly, we based our EFs on a direct causal link between nitrogen addition and local plant species loss by establishing a new response model based on a global database containing quantitative data from nitrogen addition experiments. Previous methods used a soil sensitivity indicator in combination with an EF derived from a species sensitivity distribution (SSD) , to link NH₃ and NO _x emissions to PDF. The SSDs developed by Gade et al. and used by Lebrun et al. were derived by overlaying records from the Global Biodiversity Information Facility (GBIF) and soil grid data from the Soil Reference Information Centre (ISRIC), which comes with the substantial drawback that species occurrences and the soil pH are not measured at the same time and location. Here, we used experimental data that allows to attribute species loss to nitrogen addition at the exact same location and time, thereby reducing uncertainty in the CFs for NH₃ and NO _x emissions.

We found more pronounced differences between the relative impact of NH₃ and NO _x in LC-IMPACT compared with both the present method and ReCiPe. LC-IMPACT showed substantially greater variation between the CFs for NH₃ and NO _x . These differences are primarily driven by the approach used to scale regional species loss to global species loss. Both LC-IMPACT and the present study scale regional to global species loss, but follow different approaches. LC-IMPACT applies global extinction probabilities to scale regional to global species loss, which can vary by 5 orders of magnitude because species are weighted by their IUCN threat status. In our study, regional species loss is scaled to global species loss using relative endemism scores, which vary by 3 orders of magnitude. ReCiPe does not scale regional to global species loss.

3.3. Model Challenges

We encountered several modeling challenges in the parametrization and application of the framework. First, we relied on spatially generic pressure-effect relationships to estimate LPDF, even though species responses to pressures may be spatially heterogeneous. For example, species responses to temperature increase are expected to differ between arctic, temperate, and tropical regions. , However, because Schipper et al. found a more parsimonious fit for the general pressure-effect relationship compared to a model distinguishing between tropical and nontropical regions, we used the general pressure-effect relationship to derive the EFs. Nevertheless, local temperature increases due to GHG emissions also vary spatially. Currently, EFs are based on regressions linking global temperature increase to LPDF. In future research, the current framework allows, if data availability is sufficient, to update the pressure-effect relationships to link local temperature increase to LPDF. In that case, also new fate factors linking greenhouse gas emissions to local temperature increase should be developed.

Second, unlike Scherer et al., we did not further subdivide the six land use types from Gallego-Zamorano et al. into more specific management types. Further subdividing the land use types reduces the sample size per land use type and decreases the robustness of the pressure-effect relationships. Furthermore, mapping the global distribution of detailed land management types is challenging due to lack of empirical data. − Yet, studies have shown that biodiversity impacts can be significantly different between land management varieties. − Future research is needed to distinguish the plant diversity responses to different forest management types and to identify the spatial distribution of the different forest management types.

Third, the mp-SAR applied in the framework assumes that different pressures act independently, not accounting for potential interactions among stressors. Several studies reported responses consistent with additive effects of stressors, − whereas other studies have identified synergistic and antagonistic effects of multiple stressors on biodiversity. − Similar to other LCIA methodologies, the current framework does not capture synergistic and antagonistic effects of pressures in the CFs, which may result in under- or overestimation of biodiversity impacts. Further research is needed to better understand local biodiversity responses to different combinations of stressors before implementation in LCIA is feasible.

Finally, all input parameters of the proposed framework, such as the FFs for GHG and nitrogen emissions, LPDF parametrizations, z-values, and relative endemism data are subject to uncertainty. It is important to address this uncertainty. For GHG emissions, the uncertainty in the IAGTP of CO₂ is ±40% (referring to 90% confidence interval) for a 20 year time horizon and ±60% for a 100 year time horizon. Additionally, in the 6 assessment report of the IPCC, the total uncertainty in GWPs of a number of GHGs and the uncertainty in the underlying components of these GWPs are reported. Based on this information, we calculated the uncertainty in the GWPs, excluding the uncertainty in the CO₂ component of the GWP, as this uncertainty is already included in the IAGTP uncertainty ranges (Supporting Information Table S8). The uncertainty in the GWPs, in terms of a 90% confidence interval, is ranging from ±19% for PFC-14 to ±40% for N₂O. For nitrogen emissions, the fate factors are based on source-receptor matrices, without reported confidence intervals. For the LPDFs, uncertainty was the highest for land-use, with the widest 90% confidence interval for intensive pasture (median = 0.31; 90% CI: 0.08–0.69), and the narrowest for intensive cropland (0.53, 0.36–0.69) and minimal pasture (0.08; 0.01–0.34). Uncertainty in the LDPF was much lower for climate change at 1.1 °C warming (0.09; 90% CI: 0.07–0.11). For nitrogen deposition, the uncertainty in the LDPF was relatively low at small deposition rates (0.02; 90% CI: 0.01–0.05 at 0.1 kg/ha/yr), and increases at high deposition rates (0.23; 0.14–0.34 at the highest reported deposition rate of 27.7 kg/ha/yr). See Supporting Information Table S9 and Figure S2. Furthermore, Gerstner et al. do not report confidence intervals for the z-values, although the uncertainty bands shown for the predicted species-area curves are relatively small. Additionally, Scherer et al. found that the z-value of the SAR relationship is less decisive for their CFs. The relative endemism richness is based on GBIF species occurrences, and therefore no confidence intervals or uncertainty measures are available.

3.4. Outlook

In future research, the framework can be extended to include other impact categories (water use, photochemical ozone formation, and terrestrial ecotoxicity), other terrestrial species groups (e.g., birds and mammals) and other ecosystems (freshwater and marine). However, such an extension requires 1) the availability of local response data, 2) a global distribution map of the pressure, 3) the scaling factor of the SAR, and 4) information on species composition at the ecoregion level. Local response data for land use and GHG emissions is, for instance, available for terrestrial vertebrates. Distribution maps of pressures are currently already available for other pressures, as for example photochemical ozone formation. Additionally, biome-specific scaling factors (z-values) for vertebrate species (amphibians, birds, and mammals) are available. , For freshwater ecosystem, species discharge relationships can be used to large river segments instead of species-area relationships. Finally, species composition data at the ecoregion level is available for terrestrial, freshwater and marine ecosystems for vertebrates (IUCN) and plants (GBIF). , Additionally, if updated global pressure maps come available, the characterization factors provided in this study can relative easily be updated by rerunning the biodiversity framework.

The inputs, outputs, and R scripts of the model and subsequent analysis are publicly available at Zenodo (10.5281/zenodo.17,474,004).

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.est.5c15422.

• Additional methods for land use and nitrogen deposition model parametrization and variance decomposition analysis; additional tables describing spatial delineations, z-values, land use types, local potentially disappeared fraction of species (LPDF) responses to land use, emitting regions of the TM5-FASST model, land use types used for characterization factor (CF) comparison, correlations between CFs, uncertainty estimates for global warming potentials (GWPs) and LPDFs, and beta regression models for land use and nitrogen deposition; and additional figures showing ecoregion-specific CFs for land use relaxation, pressure–response relationships, and residual diagnostics for land use and nitrogen deposition models (PDF)

• Calculated characterization factors for land use, greenhouse gas emissions, and nitrogen emissions: Supplemental_data_2_characterization_factors (XLSX)

Conceptualization: K.J.J.K, A.M.S., and M.A.J.H. Formal analysis: S.D.J., H.M. Methodology: S.D.J., K.J.J.K., H.M., A.M.S., R.v.Z., and M.A.J.H. Writingoriginal draft: S.D.J. Writingreviewing and editing: S.D.J., K.J.J.K., H.M., A.M.S., R.v.Z., and M.A.J.H.

The authors declare no competing financial interest.