International standards for ecosystem accounting have only recently become a focus of interest (Edens and Hein 2013; Obst et al. 2013; Schröter et al. 2014; United Nations et al. 2014a; Vardon et al. 2016). Their purpose is to support the consideration of ecosystems in decisions by providing a coherent framework for codifying information on ecosystem extent, condition, services and benefits and for linking ecosystem services to human benefits. Ecosystem services assessments have been conducted at local (Maynard et al. 2010, Schröter et al. 2014), national (Schröter et al. 2016, DEFRA 2011) and global (MA 2005, Peh et al. 2013, TEEB 2010, Landers and Nahlik 2013, Kinzig et al. 2007, Costanza et al. 1997, de Groot et al. 2012, Díaz et al. 2015) scales. The national and global perspectives provided by ecosystem accounting encourage a broader consideration of the scales of drivers, ecological phenomena, institutions and stakeholders (Hein et al. 2006). It allows coherent monitoring, reporting, priority identification and trade-off analysis at scales and scopes that reflect national and international policy objectives and mandates. A global view further facilitates international comparisons and benchmarking, such as addressing the 2030 Agenda for Sustainable Development (SDGs) (United Nations 2015). Ecosystem accounting, thus, should ideally be able to aggregate data from across local areas and countries.

Fostering national and international agreement on measurement systems requires convergence amongst value systems (see Saner and Bordt 2016), but it also requires attention to the statistical principles to produce rigorous classifications of both ecosystems and ecosystem services.

The question of “Which ecosystems provide which services?” should be understood as a search for priorities. One can argue that ecosystems carry out many processes that are linked, directly or indirectly, to many ecosystem services—one may even claim that “all ecosystems provide all services.” This answer, however, does little to focus ecosystem services studies on priority ecosystems or priority ecosystem services in a study area.

Considering that existing ecosystem services studies implicitly or explicitly answer the question by identifying ecosystems and ecosystem services of interest, one may think that unified classification systems should already exist. Such systems would ideally provide a comprehensive and objective understanding of (a) which ecosystems potentially provide which services and (b) which services are potentially provided by which ecosystems. Detailed, rigorous and internationally-accepted classifications of both ecosystems and ecosystem services would provide a foundation for comparability across studies. While progress is being made on these (Bordt 2015, Bordt 2016, Chan et al. 2016, Saarikoski et al. 2015, Uhde et al. 2015), many studies are based either on meta-analyses of existing studies, expert judgement*1 or primary research on specific ecosystems or specific ecosystem services. While these are essential inputs, none alone can generate unbiased, generalisable, comprehensive and coherent classifications of ecosystems and ecosystem services.

When planning an ecosystem services study, one could begin with identifying ecosystems in the study area and then determining which services they provide. Alternatively, one could begin with identifying priority services and then determining which ecosystems are most likely to provide them. Either approach requires an understanding of which ecosystems provide which services. Such an understanding could be developed through exhaustive field research or by meta-analysis of existing knowledge. For example, a wetland in one location may have already been studied and determined to provide priority services of water purification, habitat and flood control. When studying a nearby wetland, one could gather basic information to verify the importance of these services, then focus new field research on measuring additional services such as food production and erosion protection. However, such local knowledge is often incomplete and primary field research is expensive and time-consuming. Furthermore, a highly local and contextual approach could directly contravene the global goal of data commensurability and aggregation. It is preferable, thus, to attempt a compromise that satisfies the need for aggregation based on all available knowledge that has been derived locally, nationally and globally.

Existing global ecosystem accounting frameworks provide a starting point. The System of Environmental Economic Accounting – Experimental Ecosystem Accounting (SEEA EEA) (United Nations et al. 2014b, United Nations Statistics Division 2017), the Economics of Ecosystems and Biodiversity (TEEB 2013) and the Intergovernmental Panel on Biodiversity and Ecosystem Services (IPBES) (Díaz et al. 2015) are three distinct, but overlapping, approaches to address the issue of larger-scale ecosystem services studies. However, they provide little guidance to national agencies seeking to focus their information collection and policies on priority ecosystems or priority ecosystem services. A table of standard ecosystem types by standard ecosystem services and some indication of the importance of the linkages would serve as a useful starting point.

Integrating local and global knowledge into coherent ecosystem accounts requires an overarching concept of the “global whole” (all ecosystem types and all ecosystem services) within which results of local, detailed studies can be combined. To develop this concept of the “global whole”, we compare and integrate nine selected input studies that range in scope from local to global. We combine their insights into a classification proposal, evaluate the level of consensus on the relationships between specific ecosystems and specific ecosystem services and conclude with recommendations for practitioners to use, test and extend these concepts.

We are aware of recent advances in applying such systematic approaches to ecosystem assessments, such as those described by Schröter et al. (2016) and Remme et al. 2015. Despite these advances, many national studies still apply classifications and standards suggested by Costanza et al. (1997)and MA (2005), for example, China's Gross Ecosystem Product (referred to in Rockström et al. 2016). This paper takes a case study approach to deepen our understanding of selected well-known examples and integrate what we have learned from them. The intent is to suggest a method for developing a practical global approach, rather than conducting a definitive analysis.

We selected the following nine input studies to provide the source material for the analysis. Selection was based on (a) their importance in the literature and (b) their inclusion ofan assessment of the relative importance of multiple ecosystem services being provided by multiple ecosystem types. If such an assessment were not explicitly presented in a table, as was the case for the the MA (2005), a table was constructed from analysis of the statements in the document (See Annex Table 1 in Suppl. material 1).

Meta-analytical studies (input studies 1-2)

Global assessments (input studies 3-7)

Local and national assessments (input studies 8-9)

By integrating the nine input studies described above, we develop and discuss in this section an ecosystem classification that should provide a compromise between the need for detail at the local scale and the need for universality at national and global scales. While we do not claim to have achieved a “final” classification, it provides insights into further improving ecosystem classifications for ecosystem accounting. Our ecosystem superset contains 48 categories.

Ecosystem classifications in the input studies were based on different principles including reporting category (MA 2005, Maynard et al. 2010), ecosystem type (Kinzig et al. 2007, Peh et al. 2013, TEEB 2010), biome type (Costanza et al. 1997, de Groot et al. 2012), habitat type (DEFRA 2011) or environmental sub-class (Landers and Nahlik 2013). Where input studies provided precise definitions, these were taken into account in the comparison.

According to Hancock (2013), “…a statistical classification is a set of discrete, exhaustive and mutually exclusive categories…” That is, detailed classes aggregate to higher levels, do not overlap and cover the entire spectrum of possibilities. One way of ensuring mutual exclusivity and exhaustiveness is to base the classification on well-defined criteria and rules. The SEEA Central Framework (SEEA CF) (Annex I, Section C in United Nations et al. 2014a) suggests a high-level classification of 14 mutually-exclusive land cover types. This is not entirely satisfactory for an ecosystem classification since, at this level of aggregation, one land cover type could represent several different ecosystem types. For example, “Tree covered areas” could exist in tundra, boreal, temperate or tropical forest or even in deserts and urban areas (as parks or woodlots). Several input studies (Costanza et al. 1997, de Groot et al. 2012, Maynard et al. 2010) further differentiate between types of forest (tropical, temperate, sclerophyll etc.).

Land cover-based classifications of ecosystems are inadequate to represent elevations of terrestrial ecosystems and depths of aquatic ecosystems. This also raises questions about how to classify non-surface ecosystems such as those existing under water, in caves and in soil. Several input studies differentiate mountain ecosystems (Kinzig et al. 2007, MA 2005, Maynard et al. 2010, TEEB 2010, DEFRA 2011), depths of water (Maynard et al. 2010) or underwater features such as seagrass beds and coral reefs (Costanza et al. 1997, Maynard et al. 2010, TEEB 2010).

The Québec Centre for Biodiversity Science (QCBS) Working Group 14, in collaboration with the European Space Agency (ESA), created a detailed classification of land cover specifically to support ecosystem accounting*4 (Uhde et al. 2015). Although this classification is based on rigorous classification principles, it focuses on earth observation-detectable (satellite data and aerial photography) land cover types expected in Québec. Extending the SEEA CF classification, it adds distinctions between:

1. Dense (impermeable) and open (permeable) artificial surfaces;
2. Annual and perennial crops;
3. Treed wetlands and forest, highlighting the concern that wetlands are often not detectable from remote sensing;
4. Coniferous, deciduous and mixed forest and three density categories for each;
5. Several types of wetlands; and
6. Deep and shallow freshwater bodies.

The proposed superset of ecosystem types (Table 1; see Annex Table 5 -Suppl. material 1 for definitions) further extends the QCBS/ESA classification with details obtained from the input studies reviewed in this paper. It is based on land cover as a primary criterion and adds elevation/depth as a secondary criterion when appropriate. Table 1 also notes whether the ecosystem type is easily detectable from earth observation. Those that are not easily detectable may require ground-truthing to identify. That is, distinguishing “Fens” from “Bogs” would require field observations to supplement satellite imagery. The table also notes where further detail on the vertical dimension would be beneficial in distinguishing ecosystems at different elevations (mountain versus lowland) and depths (pelagic versus benthic). For example, “Very dense coniferous forest” on lowlands are often considered distinct from those on mountains (MA 2005).

Substantial modifications to the Uhde et al. (2015) classification were necessary to incorporate the details of the input studies into a superset:

• 01.03 Dams*5 was added, since Maynard et al. (2010) attribute services to urban dams (artificial water bodies created for the storage of water). This highlights the question of how urban features should be considered in an ecosystem classification. Several input studies distinguish greenspace within urban systems as providers of ecosystem services.
• 07 Mangroves was added here to maintain high-level compatibility with the SEEA CF. However, this type would be better classified as a subset of 14.02 Intertidal water bodies, since mangroves exist uniquely in saline coastal habitats (Valiela et al. 2001).
• 13 Inland water bodies was expanded to distinguish “Rivers and streams” from “Lakes and ponds”, since several input studies (Landers and Nahlik 2013, Maynard et al. 2010, TEEB 2010) identify different ecosystem services from these types.
• 14 Coastal water bodies and inter-tidal areas was expanded to include the different types of coastal water bodies and inter-tidal areas used in the input studies. For example, Maynard et al. (2010) distinguish “Pelagic” (surface) from “Benthic” (sea bottom) coastal water bodies. For “Inter-tidal areas”, several authors distinguish types, such as “Rocky shores”, “Beaches”, “Coral reefs”, “Seagrass beds”, “Estuaries” and “Coastal dunes”.
• 15 Open ocean was added, since most input studies included an open ocean or marine type. This was further differentiated into “Pelagic” and “Benthic” by the author.
• 16 Atmosphere is used only by Landers and Nahlik (2013) as an environmental sub-class. It is not, in fact, an ecosystem (Cooter et al. 2013). "Atmosphere" is, however, of interest since it is not only an integral part of almost all ecosystems, it also engages in processes, such as airflow, that are distinct from the ecosystems with which it interacts.
• 17 Groundwater is also used only by Landers and Nahlik (2013). "Groundwater" is neither a surface feature, nor an ecosystem type, but is also of interest due to its distinct processes that interact with ecosystems.
• 18 Soil (United States Department of Agriculture 2016) was added to emphasise that distinct ecosystems exist in soil (Brady and Weil 2010), but are not commonly considered in ecosystem services assessments.

The development of the proposed superset of ecosystem types (Table 1) revealed some issues that should be disclosed with the aim of furthering the discussion to develop a universal classification.

The input studies did not always include detailed definitions of the ecosystem classification. For example, TEEB (2010) mentions “Coastal areas”, but does not distinguish further detail. In cases such as this, the more general type was corresponded to all appropriate detailed types in the superset. For example, TEEB’s “Coastal areas” was corresponded with 14.01 Coastal water bodies, 14.02.02 Rocky shores, 14.02.03 Beaches and 14.02.06 Estuaries.

Furthermore, several input studies included ecosystem types that were not specifically land cover types, but were distinguished by location (tropical vs. temperate forest), conditions (e.g. tundra, desert, urban) or elevation (mountain). Specifically:

1. Several input studies (Costanza et al. 1997, de Groot et al. 2012, Kinzig et al. 2007, TEEB 2010) include deserts or tundra, which were corresponded to both 10 Sparsely natural vegetated areas and 11 Terrestrial barren land.
2. Some input studies (Costanza et al. 1997, de Groot et al. 2012, Kinzig et al. 2007, TEEB 2010) distinguish tropical from temperate forest, both of which were corresponded to 06.02 Forest.
3. "Urban areas" were included in most of the input studies, but definitions ranged from hard surfaces only to greenspace only. When definitions were available, urban areas were corresponded to the appropriate ecosystem type. For example, the Landers and Nahlik (2013) category of “Created greenspace” (parks and lawns) was corresponded to 05 Grasslands and all three types of open forest.
4. Mountain ecosystems (Kinzig et al. 2007, MA 2005, Maynard et al. 2010, TEEB 2010) were corresponded to 10 Sparsely natural vegetated areas and 11 Terrestrial barren land, since the intent was to distinguish areas of limited ecosystem function. This is not entirely satisfactory, since most ecosystem types (forests, rivers & streams, lakes & ponds, grasslands etc.) also exist on mountains.

Broad ecosystem types, described in the input studies, required corresponding to several types in the superset. For example:

1. The Millennium Ecosystem Assessment (MA 2005) reporting category of “Natural grasslands/savannah/shrublands” was corresponded to 05 Grassland, 08 Shrub covered areas and 10 Sparsely natural vegetated areas.
2. Similarly, the FEGS-CS (Landers and Nahlik 2013) environmental class “Scrublands/Shrublands” was corresponded to 08 Shrub covered areas and 10 Sparsely natural vegetated areas.

The Maynard et al. (2010) ecosystem reporting categories were more detailed than the superset. They distinguish between rainforests, sclerophyll forests, native plantations, exotic plantations and native regrowth. All detailed forest types were corresponded to 06.02 Forests.

To manage the complexity of the comparisons, the resulting superset excludes some vertical (e.g. mountain vs. lowland) and latitudinal (e.g. tropical vs boreal) distinctions, “Islands” as a distinct type and local ecosystem types.

As with the superset of ecosystem types, the objective of this section is not to develop an ideal comprehensive classification of ecosystem services. Instead, we aim at comparing the ecosystem services classifications used in the nine input studies. Following CICES V4.3, we use 48 categories of ecosystem services which is, coincidentaly, the same number that emerged from the analysis of ecosystems.

Defining and classifying ecosystem services has progressed since many of the input studies were published. Boyd and Banzhaf (2007) systematised the definition of ecosystem services, to focus on “final” ecosystem services as “components of nature, directly enjoyed, consumed, or used to yield human well-being.” Two classification systems have since emerged that are largely consistent with this definition: FEGS-CS*6 (Landers and Nahlik 2013) and CICES (2013)*7, the Common International Classification of Ecosystem Services. However, they apply different interpretations of “directness” of use and linkage to ecosystem processes.

FEGS-CS links “final” ecosystem goods and services (FEGS) to specific environmental classes and beneficiaries. Since it applies a conservative approach to identifying and classifying “final” ecosystem services, it excludes several that are often considered ecosystem services in the input studies, such as "Cultivated crops" and "Animals from in-situ aquaculture", since these are not “self-sustaining in the environment” Landers and Nahlik (2013). Further, FEGS-CS excludes most “Regulation and maintenance” ecosystem services, such as "Carbon sequestration" and "Flood control", which are less directly “enjoyed, consumed or used”. FEGS-CS also applies a broad scope of environmental classes, which includes not only groundwater and atmosphere, but also FEGS such as land, water and air as media and conditions for human activities.

CICES V4.3 (CICES 2013, Haines-Young and Potschin 2013) describes ecosystem outputs as they directly contribute to human well-being by providing a framework in which “information about supporting and intermediate services can be nested and referenced.” (CICES 2013) It incorporates ecosystem services that have been applied since the Millennium Ecosystem Assessment (MA 2005) and therefore includes “Regulation and maintenance services” that are less directly “enjoyed, consumed and used”. As well, it includes “Provisioning services” from components of nature that are not self-sustaining.

Since the 48 “Classes” (detailed ecosystem service types) of CICES V4.3 aligned well with the nine input studies, we used this existing classification as the ecosystem services superset to compare the nine input studies in this paper (Table 2).

Different levels of detail between the input studies and CICES required corresponding ecosystem service to CICES classes. For example:

• CICES Classes 01.01.02.01 Surface water for drinking, 01.01.02.02 Groundwater for drinking, 01.02.02.01 Surface water for non-drinking purposes and 01.02.02.02 Groundwater for non-drinking purposes were frequently not distinguished in the input studies. Most input studies used one ecosystem service type for water supply, whether it was for drinking or for non-drinking purposes.

• CICES Class 01.02.01.01 Fibres and other materials from plants, algae and animals for direct use or processing was represented by de Groot et al. (2012) as three services (“Raw materials”, “Medicinal resources”, “Ornamental resources”), by Landers and Nahlik (2013) as four categories (“Fiber”, “Natural materials”, “Timber”, “Fungi”) and by Kinzig et al. (2007) as three ecosystem services (“Fiber”, “Biochemicals and pharmaceuticals”, “Ornamental resources”).

• CICES Classes 02.01.01.01 Bio-remediation by micro-organisms, algae, plants and animals, 02.01.01.02 Filtration/sequestration/storage/accumulation by micro-organisms, algae, plants and animals, and 02.01.02.01 Filtration/sequestration/storage/accumulation by ecosystems are also not distinguished in the input studies.

• CICES Class 02.03.01.02 Maintaining nursery populations and habitats was distinguished in further detail byde Groot et al. (2012) (into “Nursery service” and “Genetic diversity”) and TEEB (2010) (into “Habitat for species” and “Maintenance of genetic diversity”). Classes used by other authors for “Wild species diversity” (DEFRA 2011) and “Iconic species” (Maynard et al. 2010) were corresponded to this class.

• CICES Classes 02.03.03.01 Weathering processes and 02.03.03.02 Decomposition and fixing processes were not distinguished in the input studies. “Soil formation” and “Nutrient cycling” (Costanza et al. 1997), “Erosion prevention and maintenance of soil fertility” (TEEB (2010)), “Soil quality” (DEFRA 2011), “Arable land” and “Productive soils” (Maynard et al. 2010) were corresponded to this class.

• CICES Classes 03.01.01.01 Experiential use of plants, animals and land-/seascapes in different environmental settings and 03.01.01.02 Physical use of land-/seascapes in different environmental settings were also not distinguished in the input studies. Several input studies did provide more detailed ecosystem services types: FEGS-CS (Landers and Nahlik (2013) suggest “Presence of the environment”, “Open space”, “Viewscapes”, “Sounds and scents”; (DEFRA 2011) suggests “Environmental settings: landscapes/seascapes”, and “Environmental settings: Local places”; suggest “Iconic species”, “Inspiration”, “Sense of place”, “Iconic landscapes” and “Therapeutic landscapes”.

• CICES Classes 03.02.02.01 Existence and 03.02.02.02 Bequest were not specified in any input study. However, the very general class of “Cultural” in Costanza et al. (1997) and “Presence of the environment” in FEGS-CS (Landers and Nahlik 2013) were corresponded to these classes.

The consensus matrix combines the ecosystems superset (48 categories) as rows and the ecosystem services superset (48 categories) as columns. The content of this matrix is provided by the nine input studies. Each input study includes statements (such as measures of monetary and physical values or expert judgements) of the importance of specific ecosystem types to specific ecosystem services. The number of input studies that consider a specific ecosystem type as important to providing a specific ecosystem service class indicates the degree of consensus on that linkage. Given that there are nine input studies, the degree of consensus, what we call “Consensus level”, can range from Consensus Level 0 (no study considers the linkage as important) to Consensus Level 9 (all nine studies consider the linkage as important). In our study, Consensus Level 8 was the maximum observed.

Statements about ecosystem/ecosystem service linkages in each input study were first corresponded to the ecosystem and ecosystem service supersets. For example, the Costanza et al. (1997) statement about “Tropical forest” providing “Climate regulation” ($223 per hectare) was corresponded to ecosystem type 06.03 Forest and ecosystem service class 02.03.05.01 Global climate regulation by reduction of greenhouse gas concentrations in the superset.

Secondly, since metrics, used by the input studies to state the importance of ecosystem/ecosystem service linkages, differed between input studies, this required a means of selecting the “important” linkages from each input study. Table 3 shows the criteria used for this selection. For example, for Costanza et al. (1997), values above the median ($68 per hectare) were selected as indicating a linkage was important. Therefore, the statement about “Tropical forest” providing “Climate regulation” ($223 per hectare) counted towards the consensus that forests provide global climate regulation. However, their value of $47 per hectare for “Food production” arising from “Swamps/Floodplains” did not count towards consensus on 06.01 Treed wetlands or 09. Shrubs and/or herbaceous vegetation, aquatic or regularly flooded providing 01.01 Nutrition since it was below the median.

To facilitate comparison of input studies with different levels of granularity (i.e. coarseness of classifications), if there were no statements about lower level (more detailed) classes, then statements about higher-level (less detailed) classes were attributed to lower levels. For example, if an input study included statements about 06.02 Forests and not about lower levels (such as 06.02.01 Coniferous forest), the same statement about forests was attributed to all lower levels.

The summary consensus matrix (Fig. 1) shows the Consensus Level—the number of input studies agreeing on the importance of a given ecosystem/ecosystem service linkage. The detailed consensus matrix (Annex Table 6 - Suppl. material 1) shows the input studies represented for each ecosystem/ecosystem service combination.

Figure 1.  

Summary consensus matrix shoring the Consensus Level—the number of input studies agreeing on the importance of a given ecosystem/ecosystem service linkage.

Note: Ecosystem services (columns) are CICES V4.3 classes, ecosystem types (rows) are SEEA classes with additional detail. See Tables 1, 2 for interpretations of the detailed codes.

At first glance, Fig. 1 seems to indicate that, indeed, almost all ecosystems provide all services, since 88% of all mathematically possible linkages were considered important by at least one input study. There are, however, many fewer cells for which there is consensus amongst two or more input studies. There was no full consensus (Consensus Level 9); that is, not a single specific linkage was considered “important” by all nine input studies. In some cases, this is due to the criteria for inclusion as "important". For example, although eight input studies agree that 06.01 Treed wetlands were important providers of 01.01.01.04 Wild animals and their outputs, the value for this combination was below the threshold in Costanza et al. (1997)*8 and therefore Consensus Level 9 was never achieved.

Consensus Level 8 was achieved on the importance of (a) wetlands providing wild animals and aesthetic services and (b) dense forests providing fibres and other materials (Table 4, below).

Consensus Level 4 (see Annex Table 7 -Suppl. material 1) was selected for further analysis since there is consensus on more than one-third (890 of 2,304) of the mathematically possible linkages*10. Annex Table 7 is simplified and summarised in Fig. 2, which serves as a summary checklist. Low consensus is evident for ecosystem services provided by ecosystem types 01 Artificial surfaces, 12 Permanent snow and glaciers, 16 Atmosphere, 17 Groundwater and 18 Soil. There is also low consensus on which ecosystem types provide the services 01.03.02 Mechanical energy and 03.02.02 Other cultural outputs.

Figure 2.  

Mid-level consensus summary of “Which ecosystems provide which services?”: Consensus level 4 (4 of 9 studies agree that the ecosystem type is an important provider of the ecosystem service).

Ecosystem types providing the greatest number of services include 06 Tree covered areas and 09 Shrub and/or herbaceous vegetation, aquatic or regularly flooded. At this level of aggregation, both include wetlands.

Ecosystem services that are provided by a majority of the ecosystem types include:

• 01.01.01 Biomass (Nutrition),
• 01.02.01 Biomass (Materials),
• 03.01.01 Physical and experiential interactions,
• 03.01.02 Intellectual and representative interactions and
• 03.02.01 Spiritual and/or emblematic.

Input studies

Superset of ecosystem types

Superset of ecosystem service types

Limitations of the study

Recommendations on ecosystem classification

Recommendations on ecosystem services classification

Conclusions

The first author would like to thank the World Bank, United Nations Statistics Division, The Government of Canada and the Québec Centre for Biodiversity Sciences (QCBS) Working Group 14 for providing opportunities to be engaged in the growing community of practice of ecosystem accounting. He is also indebted to my PhD Examination Committee (Konrad Gajewski, Anthony Heyes, Jackie Dawson, Kai ML Chan) for their many constructive comments on earlier drafts of this paper.

Department of Geography, Environment and Geomatics; University of Ottawa, 75 Laurier Ave E, Ottawa, ON K1N 6N5, Canada