Next steps for ecosystem service models: integrating complex interactions and beneficiaries

A common approach for modelling multiple ES in a single study has been to consider them as parallel noninteracting processes (Schulp et al. 2012; Schirpke et al. 2013). In this approach, each ES is modelled and mapped individually, allowing identification of areas of overlap where multiple services are provided. This is common at large spatial scales, such as national assessments of ES with examples applied to Germany (Grunewald et al. 2016), the United Kingdom (Bateman et al. 2013), and Israel (Lotan et al. 2018). While these studies offer in-depth assessments of the ecological health of systems and identify which services are available where, not considering interactions between services make predicting responses to changes difficult.

Some studies have started looking at spatial interactions between ES using ES bundles. ES bundles are defined as repeatable positive (synergies) and negative (trade-offs) in ES associations in space and time (Raudsepp-Hearne et al. 2010b). While this approach does not focus on mechanistic dynamics that lead to tradeoffs, it allows detection of spatial associations and disassociations between ES. Usually, ES bundles rely on land use and land cover data (LULC) by associating bundles of services to landscape classes (Queiroz et al. 2015; Baró et al. 2017). While this approach works well at large spatial scales where resolution is low, it presents strong limitations when working at the local scale, where diversity of management and processes within LULC types becomes one of the main drivers of ES bundle diversity. Vannier et al. (2019) proposed a social–ecological conceptual framework, integrating stakeholder and expert knowledge, to analyze ES bundles at small spatial scales. They used high-resolution data in combination with stakeholder-defined ES interests to breakdown LULC categories into relevant subclasses to map bundles across their landscape. While not directly integrating complex interactions between ES in the modelling process, such studies have highlighted the interconnected nature of ES, identifying trade-offs and synergies among services (Turner et al. 2014; Yang et al. 2015). Spake et al. (2017) presented a detailed evaluation of current methods used in ES bundle modelling. The use of ES bundles is a major first step in the integration of complexity in ES models.

Modelling multiple ES has been facilitated by the emergence of simulation platforms offering pre-existing intuitive models to simulate a wide suite of services. These spatially explicit tools allow users to model ES supply at a wide variety of scales. Bagstad et al. (2013b) reviewed 17 decision-support tools for ES quantification and valuation, each evaluated for several key criteria such as time requirements, capacity for independent application, scalability, and generalizability. One of the most popular platforms is InVEST (Integrated Valuation of Ecosystem Services and Tradeoffs, Sharp et al. 2015) developed by Stanford’s Natural Capital Project team. InVEST offers a suite of models to map a wide variety of ES such as crop pollination (Groff et al. 2016), carbon storage (Pechanec et al. 2018), and water yield (Redhead et al. 2016). InVEST relies on geographic information sources and produces maps as outputs to help inform decision-making. While users can model as many as 18+ ES in InVEST, these are considered as independent noninteracting entities and do not account for complex interactions. Another promising tool for modelling multiple ES is ARIES (Artificial Intelligence for Ecosystem Services, Villa et al. 2014). ARIES relies on common semantics that allow users to upload, share, or utilize others’ models and data, effectively reducing the time necessary for new users to be able to run models. ARIES uses artificial intelligence to identify suitable models and data sets available within its system to identify appropriate ES models in any given location or context (Martínez-López et al. 2019). However, ARIES is still limited in the number of ecosystem models already available compared with InVEST. A detailed early comparison between InVEST and ARIES using a case study can be found in Bagstad et al. (2013c), which highlights that while model outcomes between InVEST and ARIES can differ, the direction of change is typically in agreement between the two approaches. The time to develop models in both platforms was roughly equivalent, the time allocated to parameterize, run, and test the models was faster in ARIES.

Several studies have started pushing for integrating more complex interactions in ES modelling to produce more robust models for decision-support. Recently, Field and Parrott (2017) proposed a spatial network theory approach to highlight flow between ES at the local level. By building a network with links between spatial nodes of ES provisioning, they can represent the flow of ES within a landscape and the inter-connections between ES. This approach is a first step towards modelling the interactions among multiple ES on a landscape and permits the identification of spatial locations that may be important because of their role in facilitating connectivity of the network.

A similar approach, more social–ecologically oriented, is the Petri net approach (Rova et al. 2019). Petri nets are graphical and mathematical modelling tools that consist of two kinds of nodes: places and transitions. Places represent conditions, resources, or items (ES, resource systems, governance systems), while transitions represent actions or events (activities, processes, ecosystem functions). Although this model presents promising capabilities for exploring the potential impacts of general scenarios such as climate change on multiple ES within a social–ecological system, the lack of spatial inputs makes it difficult to inform precise spatial decision-making usually considered at smaller scales by stakeholders.

More generally, a wide variety of approaches exist for modelling the actual processes involved in ES and their interactions. Bayesian belief networks are probabilistic graphical models representing random variables as nodes with links connecting them (McVittie et al. 2015; Gonzalez-Redin et al. 2016). Although this probabilistic approach can account for uncertainty that can be overwhelming when modelling complex systems, interactions between ES are only represented as unilateral links and thus cannot integrate feedback effects. Causal loop diagrams and stock-flow models, which are part of wider group of models called system dynamics models, can also be used to model interactions (Robinson et al. 2013; Coletta et al. 2021). These approaches build representations of a system through variables that interact with each other in positive or negative ways. Modellers can integrate feedbacks by creating loops between variables. Still, it can be difficult to integrate nonlinear patterns such as threshold effects or nonlinear relationships more broadly.

While most studies focus on mapping and modelling the supply of ES, where and how services are used, consumed, or enjoyed by human populations is often not addressed. How people benefit from an ES is represented most often either as realized use or consumption of an ES, or by economic demand for the service. In general, economic demand is not readily observable to the researcher with simple administrative or observational data. Aggregate economic demand is defined as the quantity of a particular good (service) desired at a particular price. The higher the price, typically the lower the quantity demanded; the lower the price the higher the quantity demanded. This results in a typical downward-sloping demand curve for normal goods. For most ES, we often only observe some aggregate quantity that is consumed at a prevailing “price”, but rarely how consumption or use changes with a (substantial) change in price. Thus, demand in a strict sense is often difficult to observe. A concept closely linked to demand and use is ES flow, which describes how an ES that comes from a landscape (the supply) is captured or flows to recipient beneficiaries. In practice, empirical studies rarely cast their measurement as flow, but functionally this would most likely be similar if not indistinguishable from use or consumption. Therefore, we focus on distinctions between economic demand and use/consumption.

In some cases, specific economic valuation studies are carried out to estimate demand either through travel cost or willingness-to-pay studies. These typically combine individual survey data into an aggregated level of demand. These estimates can be used for valuing changes in ES supply, usually to estimate the economic impacts of land use and land cover changes, for example, on increases in water quality (e.g., Polasky et al. 2011). Castro et al. (2014) focused on a suite of ES (crop production, climate regulation, water flow maintenance, control of erosion and maintenance of habitats) in eastern Andalusia in Spain, aggregating demand at the local level. While aggregation allows easier valuation of changes, such an approach negates the possibility to integrate variance, and thus fine-scale spatial relationships between supply and demand.

From the travel cost literature, aggregate demand has also been estimated from publicly available social media databases to estimate visitation rates of natural areas. For example, Keeler et al. (2015) used data from Flickr, an online photo-sharing platform, to estimate visitations of Minnesota and Iowa lakes, and the travel costs for individual visitors were used to estimate demand for lake-based recreation in this region.

Aggregate demand has limitations, however, especially when the ES of concern are place-based and do not move. Further, disaggregating impacts of ES on populations has become of interest specifically for understanding how different segments of society depend and rely upon various ES, especially for vulnerable populations (Daw et al. 2011; Howe et al. 2012; Mace et al. 2012; Robinson et al. 2019). Several recent studies have spatially disaggregated demand primarily by using land cover maps. For example, Schulp et al. (2014) applied such an approach to estimate pollination services across Europe where demand was modelled using crop cover maps with each crop having its own demand characteristics. Tao et al. (2018) quantified ES demand for different land cover classes through an expert-based matrix model using comparable semi-quantitative units, as first proposed by Burkhard et al. (2012).

Others represent demand spatially with population distribution maps and assuming demand is homogeneous across the population. For example, Baró et al. (2016) examined the relationship between air quality and outdoor recreation demand in the Barcelona metropolitan region. Per-capita demand is assumed constant, and population density maps are then compared to ES supply maps. Others disaggregate demand across populations. For example, Ma et al. (2019) considered demand at the human settlement level, estimating the spatial–temporal variance of supply and demand of water security across the Miyun reservoir near Beijing, China. Including a temporal scale in this study allowed identification of areas that are more (or less) vulnerable to ES changes. By disaggregating demand at fine spatial scales, ES supply and demand can be overlapped to identify which areas have a positive or negative balance regarding specific ES (Wang et al. 2019). Finally, demand has also been modelled at the household level, which then allows for separating ES use and dependency into different groups of beneficiaries. Robinson et al. (2019) conducted a household-level assessment with detailed household-level survey data, analyzing how various beneficiary groups depend on ES for their well-being. Such a fine-scale approach allows one to not only identify vulnerable households but also anticipate how policy decisions would potentially impact each group of beneficiaries.

ES bundles have also been used to simultaneously analyze supply and demand across multiple services (García-Nieto et al. 2013; Baró et al. 2017; Zoderer et al. 2019). These studies map ES supply and demand to analyze the linkages among multiple ES to classify spatial units into ES bundles. Schirpke et al. (2019) used an ES bundle approach, integrating supply, demand, and flows at a municipality level to identify five different ES bundles. They then used a random forest analysis to explain ES bundle distribution and found that 12 socio-ecological variables explained 81% of the municipalities to the ES bundles. Identifying bundles that integrate flow and demand, and thus highlight vulnerable areas, could facilitate policymaking to target specific services and areas.

Human demand and interactions with the ecological system have also been modelled using agent-based models (Bagstad et al. 2013a). Humans are often represented as either groups or individuals that can influence the system either by directly interacting with resources or making decisions that will influence the modelled ecological processes. These approaches have been applied to model a wide variety of ES such as tourism (Balbi et al. 2013), forest timber (Habib et al. 2016), and livestock (Yan et al. 2019). These models allow exploration of the potential consequences of management scenarios on both supply and demand of ES.

Some recent modelling frameworks have emerged with the goal of linking ES supply to beneficiaries. Chaplin-Kramer et al. (2019) proposed a conceptual framework for calculating nature’s contribution to people. By modelling demand using population distributions in relation to ecosystem supply models, their framework identifies areas where benefits are realized (supply meets demand), realized with a gap (demand exceeds supply), and unrealized (absence of supply). This framework was applied at the global scale and highlights the importance of considering demand to identify management scenarios. Johnson et al. (2019) developed a modelling toolkit, “Mapping Ecosystem Services to Human well-being” (MESH) that integrates models from InVEST to estimate the tradeoffs and synergies between five ES and 10 associated sustainable development goals (United Nations 2015). These goals, developed by the United Nations, aim at achieving a better and more sustainable future by addressing global challenges such as poverty, inequality, climate change, and environmental degradation. In the context of ES, this can only be done by modelling flow of ES to beneficiaries (Bagstad et al. 2014).

As illustrated in our review, most ES models that consider interactions do so as either simple spatial relationships (ES bundles) or mechanistic approaches that fail to capture complex interactions such as threshold effects or nonlinear relationships. While simplifying interactions into simple positive or negative relationships facilitates straightforward assessments, overlooking these complex dynamics can lead to problematic decision-making.

In our case study, sea otter populations are known to have clear direct effects to some ES (Fig. 2a) For example, sea otters feed on shellfish and have an overall negative impact on shellfish populations, consequently impacting commercial shellfish fisheries dependent on those resources (Carswell et al. 2015). However, through consumption of sea urchins, otters can help kelp forest recover, thus increasing carbon sequestration (Ling et al. 2015). When left unchecked, sea urchin populations can take over kelp forests and create sea urchin “barrens”—places where nothing else grows (Filbee-Dexter and Scheibling 2014). On the SES framework, these interactions would be represented by direct and indirect interactions between resource units. This simple linear approach to modelling ES would lead to simple decision-support outcomes such as increasing carbon sequestration then protecting sea otter populations or increasing shellfish production then letting sea otter populations decline.

However, the relationships among sea otters, carbon sequestration, and shellfish are more complex than the simplified depiction of their interaction implies (Fig. 2b). For example, abalone, a type of sea snail, has historically been important for trade, food, and ceremonial regalia for many British Columbia First Nations (Sloan 2003, 2004). Abalone had been sustainably fished under traditional stewardship by First Nations for at least 2000 years before European colonization. With the introduction of commercial fisheries and under the governance of the federal Department of Fisheries and Oceans Canada (DFO), abalone populations collapsed leading to the closure of abalone fisheries in 1990 (Lee et al. 2019a). Abalone cannot thrive in urchin barrens. They rely on kelp forest as a primary habitat and thus are usually present in places alongside sea otters. Abalone are also a food source for sea otter populations, and high densities of sea otters can drive down abalone populations to near extirpation. Abalone’s relationship in this system is nonlinear and exhibits threshold effects related to sea otter population densities. Too few or too many otters can lead to the extirpation of abalone from the system (Raimondi et al. 2015). Thus, while the simplistic approach might yield a decision to remove sea otters in favor of shellfish as an ES, integrating these complex relationships highlight that this could come at the expense of a more diversified shellfish community that contains abalone (see also Lee et al. 2019b).

As highlighted by our example, integrating these complex interactions in ES modelling would allow for the identification of more nuanced decision-support outcomes. While these might result in the same or similar decisions, policies can be tailored to more precisely identify “losers” among ES outcomes and mobilize ways to counterbalance these losses. These complex interactions between resource units are at the foundation of ES provisioning. Thus, while resource units are often considered individually in the SES framework, when complex interactions are important, we encourage moving beyond this and defining networks of resource interactions in SES (Fig. 2c). This would allow us to consider such complex interactions from the beginning of model conceptualization, moving away from the current dominant approach that tends to envision services as independent systems operating in isolation, which may overlook important complexity.

As highlighted in our review, the integration of flows to beneficiaries in ES models is still at its infancy. ES models have clearly lagged in representing humans in SES. In the SES framework, humans are integrated into both the governance systems and actors boxes. Governance systems can refer to a complex set of institutions, norms, or rules, and their study has received much attention in the natural resources management literature (Paavola and Hubacek 2013; Baird et al. 2014; Loft et al. 2015). The actors, however, also can include a wide diversity of entities. In the early iterations of the SES framework (e.g., Ostrom, 2009), this component was initially labeled as “users”, referring to individuals who use natural resources in diverse ways for sustenance, recreation, or commercial purposes. This label later referred to actors that not only include users but also expand the category to any potential individual or group participating in the system. This change is more inclusive and allows a wider range of potential applications of the SES framework, adjusting the focus from direct users of a resource to a broader representation of people that interact with the system.

Excluding recognition of beneficiary groups can be problematic when building ES models. Related to the sea otter – sea urchin case, Gregr et al. (2020) estimated the economic benefits of sea otter populations in the region focusing on four key ES and beneficiaries: tourism, invertebrate fisheries, finfish fisheries, and carbon sequestration (Fig. 3a). Overall, while sea otter populations induce economic losses for invertebrate fisheries (CA$−7.3 million per year), Gregr et al. (2020) estimated the economic benefits of sea otters through the three other services to be at around CA$53.6 million per year. These conclusions suggest there are strong benefits to protecting sea otters along the coastal areas of British Columbia since sea otter recovery is reported to be highly valuable in these areas.

However, this study does not identify specific winners and losers but estimates the overall social–economic impact of the sea otter population. While the general benefits of sea otter populations may seem uncontestable at the scale of the whole system, identifying the beneficiaries of these services is necessary to better comprehend the potential inequities in the benefits experienced by different groups (Fig. 3b). For example, local Indigenous Peoples in this system presently rely heavily on shellfish fisheries not only as a source of revenue, but also mainly as a food source and may be negatively impacted by decreased shellfish populations. This economic impact may be potentially catastrophic for certain marginalized populations. However, salmon and herring, both cultural keystone species and traditionally significant food sources for Indigenous Peoples, require kelp forests as nurseries and may benefit from sea otter reintroduction, leading to longer-term recovery of inshore fisheries and cultural practices (Gauvreau et al. 2017). Indigenous Peoples also identified erosion protection as a valuable benefit from sea otter presence, which was not incorporated into the Gregr et al. (2020) study. If focusing on other groups of beneficiaries such as commercial shellfish fisheries or commercial finfish fisheries, the overall economic consequences of sea otter populations can vary drastically both in ES of interest and directionality. Identifying these beneficiary-specific impacts can lead to much different management decisions. Further, policies could more effectively target (or not) specific groups or areas when negative impacts are projected. This would be greatly welcomed in this system, where governance is centralized around government agencies such as the DFO and the British Columbia Fish and Wildlife Branch (Pinkerton et al. 2019). Local Indigenous populations, which historically had complex governance protocols to maintain sea otter populations (Salomon et al. 2015), were completely left out of decisions around sea otter reintroductions. Many studies have highlighted the importance of decentralizing governance to actively include these actors in governance (Plagányi et al. 2013; Salomon et al. 2018; Pinkerton et al. 2019; Burt et al. 2020) and the abalone example highlights some of the catastrophic consequences of failing to do so (Chadès et al. 2012; Lee et al. 2019a).

Our example highlights the potential disparities in decision outcomes that can emerge between aggregating or disaggregating beneficiaries within ES models. Representing beneficiaries as explicit groups within the SES framework instead of being aggregated into the actors box (Fig. 3c), can help identify such disparities. Listing all possible beneficiary groups and explicitly recognizing their relationships to ES within the SES framework would be an important first step toward their integration within ES models, leading to more informative outcomes for decision-support. A simple but powerful step further would be to quantify the flows of services to these disparate groups.