Monitoring social–ecological networks for biodiversity and ecosystem services in human-dominated landscapes

Humans and nature are inextricably connected through ecosystem benefits and ecosystem services (ES(s) hereafter). The demands of a growing human population for provisioning ESs (e.g., food production, energy, timber) have triggered the worldwide erosion of biodiversity (Ceballos et al. 2015; IPBES 2019), the increase in species invasions, the loss of ecosystem functions (Cardinale et al. 2011), and trade-offs among the ESs we depend on (Cord et al. 2017a; Martín-López et al. 2014). Particularly in the past half century, 75% of basic ESs (i.e., provisioning, regulating, supporting, and cultural) are estimated to be in a degraded state (Díaz et al. 2019). Our capacity to make decisions that attenuate and reverse these impacts on biodiversity and ESs loss can be improved via the implementation of adaptive monitoring strategies (IPBES 2019; MA 2005).

The need to monitor change in biodiversity and ESs has been identified at regional and global scales (GEO BON 2017). Commitments to the international (i.e., Convention on Biological Diversity) and national biodiversity and sustainability goals (e.g., Canada’s Sustainable Development Goals) will likely see renewed investment in monitoring capacity in the coming decade. Despite the surge in investment in the monitoring of biodiversity and ESs across spatial scales, integrated monitoring schemes that evaluate the ecological and social dimensions of change at different spatial scales still remain scarce (Geijzendorffer and Roche 2013; Geijzendorffer et al. 2017b; Kühl et al. 2020). Likewise, while current ecological monitoring captures important dynamics in particular species and ES stocks, they generally fail to monitor the co-dependence, or connectivity, among ecosystems and human uses (e.g., ES flow, drivers of change, beneficiaries, demand; for review see Kluger et al. 2020), which is essential if we are to assess the long-term sustainability of biodiversity and ESs under changing or novel environmental and social conditions (Tallis et al. 2012).

Monitoring biodiversity and ESs in spatially extended social–ecological systems (hereafter SES) is a complex challenge (Dee et al. 2017; Bodin et al. 2019). It is clear no single strategy for ecosystem monitoring and management can solve complex and context-dependent environmental problems (e.g., Ostrom 2007; McGinnis & Ostrom 2014). Nevertheless, network-based approaches can help assess how changes within and between components of ecological and social systems affect ESs (Dee et al. 2017). Thus, a monitoring design based on networked SES can provide critical understanding of how changing interconnections in SES can generate risks for ES supply or human well-being (e.g., expressed as large temporal fluctuations; Duncan et al. 2015; Henderson et al. 2016; Tallis et al. 2012). Through this approach, monitoring can help identify early warning indicators of ecosystem degradation or collapse, tipping points in SES and rates at which irreversible loss is being approached (Bauch et al. 2016; Gonzalez et al. 2017; McCann et al. 2020; Scheffer et al. 2009; Tallis et al. 2012). Network-based monitoring can also be used to inform risk-mitigation strategies that keep SES networks within a sustainable operating space (Gonzalez et al. 2017). Yet, this present work stresses that network-based monitoring designs should be spatially explicit to characterize and quantify spatially heterogeneous responses to drivers of change (e.g., climate warming, increase in ES demand). Monitoring adaptively across spatial networks can account for nonstationary change and interdependencies and provide the information needed to attain short and long-term outcomes through adaptive management practices (Lindenmayer and Likens 2009).

The future of nature and human well-being relies on our capacity to embrace the complexities of SES and provide scalable evidence-based management strategies. One such initiative is NSERC ResNet (Natural Sciences and Engineering Research Council, nsercresnet.ca), a network for managing, modelling, and monitoring ESs across Canada. ResNet is structured across three themes (i.e., management, modelling, and monitoring) and six Canadian working landscapes: agriculture in the Bay of Fundy dykelands, the Prairies, and boreal Northwestern Territories; agriculture and peat mining in Québec; energy development in Northern British Columbia and Alberta; and fisheries in the Pacific coast. Landscapes share common ESs at the regional and federal levels and contribute with unique ESs at the provincial level (Table 1). The goal of maintaining long-term sustainability and resilience of biodiversity and ESs requires integrative monitoring that generates information to support management decisions across spatial and temporal scales. Hence, as a part of ResNet’s monitoring theme, we conceptualize a multilayer and multiscale network framework for monitoring biodiversity and ESs in SES. We illustrate our proposed framework with a conceptual case study on the provisioning ES of maple syrup production in the landscape of Québec. Yet, our framework can be applied to different socio-ecological contexts and sets of ESs across and beyond ResNet’s landscapes. We also outline the need to use monitoring, statistical, and modelling frameworks to integrate quantitative and qualitative expectations derived from experts and stakeholders to generate a causal understanding of change across sites of the SES. The future operationalization of our proposed framework will allow for the detection of unintended spatial and temporal feedback arising from interlinked social and ecological interactions. Establishing monitoring to integrate evidence into decision-making is essential if we are to achieve the long-term sustainability of biodiversity and ESs into an uncertain future, and here we provide a first step towards this goal.

Robust monitoring of biodiversity and ESs relies on rigorous design protocols and analytical methods to detect trends across scales, account for potential sources of error, and be adaptive as knowledge and objectives evolve (Larsen et al. 2001; Lindenmayer and Likens 2009; Yoccoz et al. 2001). Particularly for ESs, monitoring must also rely on a comprehensive description of the SES relative to the monitoring questions or objectives (Kluger et al. 2020; Kühl et al. 2020). Past studies have proposed different frameworks to conceptualize (Ostrom 2007; McGinnis & Ostrom 2014), analyze (Bodin et al. 2019), and operationalize (Dee et al. 2017) ES networks of SES under varying objectives (Table 2). For example, Bodin et al. (2019) discussed types of causal interactions in SES networks to understand how interdependencies drive changes, while Dee et al. (2017) presented steps to build a SES network for ES management. However, the challenge of monitoring the joint change in social and ecological networks has, to our knowledge, not been addressed in previous frameworks. Hence, we build upon former studies to conceptualize a spatial network framework for monitoring ESs in SES.

The spatial network monitoring framework we propose aims to establish a spatially explicit and cross-scale description of how drivers of change may impact biodiversity and ESs in SES. Our framework is envisioned to address monitoring questions pertaining to the sustainability of the SES by providing a comprehensive blueprint of the SES. The steps of our framework are: (step 1) identify the SES and its components relative to the ES(s) of concern (section 2.1); (step 2) define the SES network composition and spatially explicit network structure (section 2.2); (step 3) identify the scale mismatches in the monitoring design of an SES network (section 3.1); (step 4) identify the drivers of change in the SES (section 3.2); (step 5) determine the variables, scales, and frequencies of monitoring needed to capture the range of variability seen across the SES network (section 3.3); and (step 6) select appropriate analytical tools and integrate monitoring with management into an adaptive cycle (section 4). Here, we provide practical examples of how to implement steps 1–4 and discuss future avenues to address steps 5–6.

Detecting change in ESs variables across an SES network relies on monitoring the states of the ecological and social nodes and the interactions that connect nodes within and across network layers. For example, the relative species composition of a forest stand can be representative of different ecological states (e.g., high or low richness, diversity, or evenness), which will influence both ecological interactions and the ES supply to the social network. Not all nodes and links will be “active” at any one time, and the spatial pattern of nodes and links can be dynamic through time. Thus, network-based monitoring can be viewed as an adaptive layer of observations and measurements designed to infer the changing relationships mediating the dynamics of the SES network over space and time. Because network components occur and interact across different scales of space and time and change according to different drivers, monitoring requires identifying the scale mismatches in an SES network and the drivers of ES change.

Monitoring the ecological and social variables reflecting patterns of change in biodiversity and ESs is not sufficient to ensure effective interventions for sustainability (Metzger et al. 2020). First, monitoring schemes must be designed to support inference of the causal structure underlying the trends and fluctuations the SES network exhibits. Second, monitoring of the SES network needs to be adaptive, that is incorporate new information as it becomes available, whether information is developed independently of the monitoring design or by directly probing the system (Lindenmayer and Likens 2009). This approach is necessary to capture shifting trends in ES supply and demand over space and time and that may arise under novel environmental (e.g., climate regimes) and socio-economic conditions (e.g., prospective urban development). Thus, in the sixth step of our framework, we highlight how adaptive monitoring for causal inference can be achieved when monitoring is integrated with modelling and management.

The SES network provides a comprehensive and spatially explicit description of the system that serves as a blueprint of what to monitor and its spatial distribution and scales. But detecting and assessing change requires statistical and modelling methodologies to determine the causes of change at different scales, across the network’s nodes and links, within and across layers. We advocate for the use of causal inference analyses to discover and quantify the causal interdependencies in a system, that is to identify which components of a system cause others to change and by how much (Bodin et al. 2019; Ferraro et al. 2019). The network monitoring framework we emphasize here lends itself to new methods for causal network analysis that go beyond simple correlations among variables by employing time series of essential variables monitored across a SES network (Runge et al. 2019). In addition to estimating causal links, causal discovery methods can also account for spatial and temporal lags in complex systems (e.g., delayed effects of policy or climate change; Runge et al. 2019). Cause and effect relationships in SES networks are challenging to assess particularly when they rely on observational data, because they may involve many interacting variables (with direct and indirect effects) playing out over different dimensions of space and time. As presented in Ferraro et al. (2019), addressing human and ecological confounding variables will be important in identifying causal relationships and ruling out rival explanations. The advances in causal inference are promising, but given the limitations of different methods and one’s specific research objectives, simplifications about the SES can be made but their assumptions need to be clearly stated.

Through the integration of monitoring and modelling, statistically parsimonious models and competing models can be used to infer the changing strength of causality and support prediction of outcomes across the SES network (Runge et al. 2019). For instance, monitoring and modelling used to detect nonlinear feedbacks and lagged effects in a system can support the identification of early warnings (Bauch et al. 2016) or forecast the transition to alternative states (Nelson et al. 2009). Another example of a well-established method for causal inference is Bayesian belief networks (BBNs). BBNs are multivariate statistical models based on conceptualized causal webs of influence that encode conditional transition probabilities for each node state (Landuyt et al. 2013). A BBN model should be informed by the multilayer network, where the BBN nodes can depict subsets or aggregates of the SES network components (e.g., a group of beneficiaries of ES at the local scale). In turn, the BBN links convey hypothesized causal relationships between the BBN nodes and competing models can be contrasted (e.g., Alexander et al. 2020). In addition to its network properties, BBNs are capable of combining qualitative and quantitative knowledge from a diverse set of stakeholders and experts, supporting, for example, tests of alternative BBN models and its impact on predictions about the changing state of network nodes given a set of initial conditions and hypothesized causal constraints (Marcot et al. 2006). The growing accessibility to and flexibility of software for implementing Bayesian networks makes them suitable candidates for ESs monitoring and mapping across SES networks (Stritih et al. 2020). An important weakness of BBNs, however, is their limitation to handle feedback and time series. To model changes over time, others have suggested the use of dynamic BBNs where the model structure is replicated to depict feedback and time steps (Nyberg et al. 2006; Uusitalo et al. 2018). Nevertheless, dynamic BBNs often require large amounts of data for parameterization. Alternative approaches to dynamic BBNs include the addition of nodes to account for the time frame (e.g., Liedloff and Smith 2010) or representing a BBN as a difference model where each node is the difference between two time points and the links are the relationships between the differences (see section 5 for a conceptual example; Scutari et al. 2017).

Monitoring and modelling should furthermore be closely linked to management through an adaptive cycle. Adaptive monitoring schemes (Lindenmayer and Likens 2009) allow for the iterative adjustment of measurement effort, frequency, and coverage so that new information can be used to update competing causal models and decision-making (Dee et al. 2017; Lindenmayer and Likens 2009). Through the feedback between monitoring, model-based inference, and management we can iteratively learn about the outcomes we observe on the SES and that mediate the status of the ecosystems and the livelihoods of actors living within it. Altogether, it is only when we consider the interactions between ecological and social systems that monitoring, modelling, and management can truly assess the risks associated with the unintended consequences of human activities and identify management opportunities to safeguard the long-term sustainability of the SES.

To illustrate the networked SES monitoring framework described above and present a walkthrough of the framework steps, we conceptualize a case study based on an SES system of maple syrup provisioning in southern Québec, Canada (Fig. 1b). In the first framework step, the SES top-tier categories are identified relative to the focal ES, in this case maple syrup production. The ecological system is defined by mixed-deciduous forest stands (resource system) and sugar maple trees part of the system (resource units); the social system is defined as the Québec Maple Syrup Producers (governance system) and the maple syrup producers (actors). The mixed-deciduous forest stands interact with the sugar maple trees, which directly supply sap. The Québec Maple Syrup Producers establish the regulations by which the maple syrup producers can extract sap from the sugar maple trees to produce the ES of maple syrup. In turn, how much sap is extracted is expected to influence the sustainability of the ecological system, and cascade to impact the dynamics of supply and demand in the social system.

While the conceptual SES offers an aspatial description of the system (Fig. 3a), its monitoring requires the spatially explicit identification of the SES network configuration, achieved in the second framework step (Fig. 3b). In the maple syrup case study, land cover and field data can help identify the ecosystem nodes representing forest patches with productive maple trees, patches where sugar maple trees are absent altogether, and patches of sugar maple trees that are too young or old to be tapped for sap extraction. Nested within the ecosystem layer are the resource nodes, represented by patches of mature sugar maple trees which are themselves connected to the nested ES nodes of sap production. The state of both resource and ES nodes can be informed by field and census data. The ecological network layers are coupled to the social network layers through local interactions of sap extraction, linking ES nodes to maple syrup producer nodes. Maple syrup production arising from the ES–actor interactions can be monitored with census data and production data from the Québec Maple Syrup Producers. Alternatively, the Québec Maple Syrup Producers is a governance organization that does not necessarily have a spatially explicit definition (i.e., spatially diffuse), and can be represented as a layer.

In the third framework step the potential for scale mismatches in the SES network is explicitly assessed, highlighting the scales at which ecological and social nodes are not spatially and (or) temporally aligned (i.e., mismatched). Highlighting these scale mismatches helps understand the spatial scales and temporal frequencies over which monitoring needs to be done (i.e., yearly, with producer-level resolution). In this case study, an important source of scale mismatch is a driver of change, and its explicit identification is achieved in the fourth framework step. For instance, maple syrup production has increased over time, both at the local and regional scales (Fig. 3c, data available from Statistics Canada (2017) and Fédération des producteurs acéricoles du Québec (2016, 2019). However, inter-annual climate variability and warming trends have impacted sugar maple differentially across the region (Legault et al. 2019). In particular, potential geographic shifts in climate conditions and changes in the beginning of the tap season are expected to negatively impact the production of maple syrup in southern Québec, while expanding production and the tree line northwards (Rapp et al. 2019). Thus, the impact of climate change on ecological nodes is expected to cascade through the network affecting the social system differentially across space. For example, a local decline in maple syrup production would have negative implications for the local actor nodes who currently benefit from it (e.g., reduction in income), leading to a scale mismatch. Moreover, the decline in ES supply would lead to a growth in market value. Demand from long-distance markets (national or international) would result in a telecoupling system (Liu et al. 2013) that may not only drive an increase in the exploitation of currently extracted maple resource nodes, but also lead to the exploitation of new resource nodes not previously harvested. As a consequence, spatial variation in ES supply could result in geographically variable social outcomes and inequities in market access among actors (Pascual et al. 2014). Thus, the network-based monitoring of these interdependencies would support ecological management and social governance that are attune to cross-scale trends in maple syrup production and its feedback effects across the SES.

The SES network and its drivers of change provide a blueprint of what to monitor and its spatial distribution and scale, so that on the fifth framework step appropriate variables are selected and monitored. For instance, to monitor the effects of climate change on maple syrup production the ES supply of maple syrup production (pounds/ha) can be estimated by multiplying the maple syrup yield data at the interaction between ES and actor nodes (pounds/tap, available by administrative region from (Fédération des producteurs acéricoles du Québec 2016)) with the number of maple taps at the ES nodes (available by census consolidated subdivision from Statistics Canada (2017)). Data can be aggregated to a unit relevant to monitored SES network components. For example, we mapped maple syrup production (Fig. 4a) by aggregating data to hydroshed level 10 (Lehner and Grill 2013). The effect of climate change can be monitored at the ecosystem and resource nodes, using remote sensing data on precipitation, temperature, number and duration of freeze–thaw cycles and phenology data on day of year of green up. Dynamics at the ecosystem nodes can also be monitored remotely with data on forest cover, land use, and total above ground biomass. The resource nodes can be monitored with field data on sugar maple cover, stand age, and percent contribution to community composition. The social nodes can be monitored with census and national statistics. Actor nodes can be defined by the number and location of maple syrup producers and monitored with data on production values. Additionally, the percentage of the population reliant on maple syrup production for income can provide an estimate of the ES benefit. The governance nodes can be assessed by monitoring maple syrup demand and market value. The ES interest can be monitored with economic data on investment in maple syrup companies, either at the actor or governance nodes. The ES nodes can be assessed with census data on number of maple taps, providing an estimate of ES stock. The ES–actor interactions measuring ES supply can also be assessed regionally by the number of tapped trees and annual production of maple syrup.

On the sixth and final framework step, a modelling approach such as a BBN model can be used to understand the effect of climate change on maple syrup production and the overall SES network. In this case, the nodes and links of the SES network can be aggregated and then conceptualized as a BBN (Fig. 4b) describing the hypothesized causal relationships between climatic, ecological, social, and ES variables, where nodes depict the difference between two time points (Fig. 4a). Monitoring data can provide the values needed to determine the state of BBN nodes and the strength of causality links between BBN nodes (i.e., conditional transition probabilities), and qualitative knowledge from experts can also be incorporated to describe the conditional transition probabilities between BBN node states. Bayesian inference then propagates probabilities of network states through the Bayesian network to generate quantitative outputs for alternative management scenarios for ESs. Monitoring by purely mapping and assessing deviations from historical climate baselines do provide a relevant measure of change. Yet, an understanding of the causes of change relies on the pairing of cross-scale monitoring and causal inference modelling, resulting in more robust management and decision-making (e.g., Olander et al. 2018).

Here, we conceptualized a framework to monitor ESs across an SES, mainly focusing on the provisioning of a single ES. However as discussed, landscapes supply bundles of ESs and SES dynamics often result in trade-offs between ESs (Cord et al. 2017a; Galafassi et al. 2017). Hence, in the context of ResNet, our future work will operationalize our proposed ESs monitoring network, including a list of appropriate indicator variables and analytical tools designed to address (i) one or more ES(s) within a landscape, (ii) a common ES across landscapes, and finally (iii) multiple distinct ESs across landscapes. This approach will allow us to estimate the range of variability seen across ResNet’s landscapes in terms of their varying levels of human modification and complex dynamics of ES trade-offs occurring across spatial and temporal scales (Bennett et al. 2009). For example, the ES of maple syrup production interacts with tourism, carbon storage, water quality, agriculture, forestry and peat extraction, to mention a few. Moving towards a Canada-wide monitoring network also requires accounting for how SESs differ in their structures. For instance, the SES of energy development in British Columbia and Alberta is characterized by strong trade-offs between the components of the ecological and social system due to conflicts between conservation and development. Another case in the Bay of Fundy involves budgetary constraints for Dykelands restoration and other land use outcomes. Finally, we emphasize the need to integrate knowledge sources to produce relevant monitoring that supports local and regional decisions and incorporates the range of values, priorities, and concerns of multiple stakeholders in the present and into the future (Kühl et al. 2020).

Managing for the long-term sustainability of biodiversity and ESs requires integrated strategies for adaptive monitoring. We proposed a spatially explicit multilayer and multiscale framework to guide monitoring of networked SES while accounting for the different levels and types of variability in ES states seen in and across ecosystems. When operationalized, our proposed framework has the potential to identify change in SES network structure at different scales, while accounting for mismatches in the interactions (i.e., patterns of supply and demand) among network components. Monitoring can improve causal models over time, which in turn can support forecasts and adaptive adjustments to future monitoring activities. This iterative approach will be implemented within ResNet and its performance compared among ESs embedded in SES networks found in each of ResNet’s landscapes. Our framework provides the foundations to establish the well-designed long-term monitoring programmes required to achieve the targets for biodiversity and ESs sustainability and to inform policy and governance at several scales within and across landscapes in Canada and elsewhere.

We would like to thank the editors for the opportunity to contribute to this special issue, and the reviewers for the thoughtful feedback. We acknowledge the support of the Natural Sciences and Engineering Research Council of Canada (NSERC); (funding reference number NSERC NETGP 523374-18). Cette recherche a été financée par le Conseil de recherches en sciences naturelles et en génie du Canada (CRSNG); (numéro de référence NSERC NETGP 523374-18). M-JF is thankful for a NSERC of Canada Discovery Grant (#5134) and the NSERC Canada Research Chair (CRC). AG acknowledges the support of the Liber Ero Chair in Biodiversity Conservation, and the NSERC Discovery program.