General rules for environmental management to prioritise social ecological systems research based on a value of information approach

1. Globally, billions of dollars are invested each year to help understand the dynam ics of social ecological systems (SES) in bettering both social and environmental outcomes. However, there is no scientific consensus on which aspect of an SES is most important and urgent to understand; particularly given the realities of limited time and money. 2. Here we use a simulation‐based “value of information” approach to examine where research will deliver the most important information for environmental management in four SESs representing a range of real‐life environmental issues. 3. We find that neither social nor ecological information is consistently the most important:


| INTRODUC TI ON
, annual global spending on environmental management was close to US$20 billion (Waldron et al., 2013).
Traditionally, the focus of this spending has been on developing an understanding of ecosystems or single species (McRae, Dickson, Keitt, & Shah, 2008;Simberloff, 2003). This narrow research approach has delivered mixed benefits because it omits relevant wider system dynamics (Liu et al., 2007). For example, the Convention on International Trade in Endangered Species of Wild Fauna and Flora (CITES) has often neglected the social motivations and pressures that drive the endangered species trade, and this has contributed to its failure to protect many species (Lenzen et al., 2012;Wolch & Emel, 1998). Equally, many marine protected area networks are designed on purely biological grounds despite their long-term success being heavily affected by social factors such as social acceptability and financial capacity (Christie, 2004;Gill et al., 2017;Mascia et al., 2003).
To improve management outcomes, the characteristics and dynamics of the wider social ecological system (SES) are receiving increasing research attention (Bodin, 2017). However, the dynamics of these coupled SES systems-how social actors interact over time, and the connections between and within social and ecological processes-create additional dimensions of uncertainty for environmental managers (Ostrom, Burger, Field, Norgaard, & Policansky, 1999).
Research can improve managers' understanding of these more complex SES models, but this requires funding, and can delay management that is often urgently needed. For example, next to nothing was known about the "extinct" Australian night parrot Pezoporus occidentalis, which was rediscovered in 2013 (Pyke & Ehrlich, 2014).
Five years later, and despite numerous programs of research, managers are still far from understanding how to protect it (Leseberg, nd;Pyke & Ehrlich, 2014). Furthermore, under economic constraints, increasing investment in ecological research may come at the cost of reduced investment in social research and vice versa. Therefore, knowledge acquisition must be prioritised based on an evaluation of its expected benefits and costs (Canessa et al., 2015;Grantham et al., 2008;Grantham, Wilson, Moilanen, Rebelo, & Possingham, 2009;Li et al., 2017;Runge, Converse, & Lyons, 2011).
To improve environmental management outcomes, a crucial research question is-what is the value of knowing more about each component (hereafter "uncertain system component") of coupled SES? We apply a simulation-based version of formal "value of information" theory (Raiffa & Schlaifer, 1961) to calculate the expected value of perfect information regarding each uncertain SES component (EVPXI, where X is each uncertain SES component) (sensu Yokota & Thompson, 2004). We then calculate which informationsocial or ecological-will deliver the greatest improvement in management outcomes.
Within the broad field of environmental management, different disciplines have argued for research effort to be concentrated on different system elements. For example, environmental research has traditionally focused on understanding uncertain ecological system components (Chadès et al., 2011). Meanwhile, both environmental and social fields of research have debated whether research effort should focus on understanding the dynamics of either nodes (e.g. the management functions of social actors; Marín & Berkes, 2010) or interactions (e.g. ecological processes such as connectivity; Pulliam, 1988;Urban & Keitt, 2001), with few studies contrasting the two (Sanchirico & Wilen, 1999). Other approaches, such as structured decision making (e.g. Martin, Runge, Nichols, Lubow, & Kendall, 2009), highlight the importance of understanding and clearly defining management objectives; as opposed to system dynamics. Our contribution in this research compares each of these options, and particularly focuses on the relative value of gathering ecological versus social information. We also assess whether it is more important to research the system components that managers are aiming to change (i.e., their objectives), or the system components that they plan to act upon.
We construct a general model of an SES that is simple enough to be analytically tractable, while also containing all the fundamental elements of an SES (Bodin & Tengö, 2012) (see Figure 1).
Our system construction builds on recent research proposing network 'motifs'-simplified, but non-trivial patterns of interconnections (Bodin & Tengö, 2012), which can be described as the basic building blocks of most networks (Milo et al., 2002). To discover which information is most valuable, we consider four different management problems-two from fisheries and two from sustainable agriculture (Figure 2). We designed our four problems to investigate the influence of all permutations of action and management objectives on research priorities ( Figure 2). To give two F I G U R E 1 Stylised representation of our social ecological system (SES) network structure. This structure has two social groups (nodes S 1 & S 2 ) who interact with two ecological populations (nodes E 1 & E 2 ). Arrows indicate the direction of potential interactions between social groups and ecological populations. These dynamics are determined by the behaviour of both the nodes and interactions, including the social ecological interactions that couple the social and ecological systems examples of these permutations: 'what is the most important element of an SES to understand when managers are intervening in a social system to achieve an ecological outcome?' And: 'what is the most important element of an SES to understand when managers are intervening in an ecological system to achieve an ecological outcome?' F I G U R E 2 Four different environmental management problems with different management actions and objectives (see Figure 1 for description of each uncertain social ecological system component). Image credits: Saxby, T.; Mulloway, D.T.; Nastase, E.; Hawkey, J.; Integration and Application Network, University of Maryland Center for Environmental Science (ian.umces.edu/imagelibrary) 2 | MATERIAL S AND ME THODS

| Overview of analyses
Our aim is to determine the relative expected value of different types of information in socio-ecological systems (SESs), to guide research to inform management actions. We explore this question using four SESs with different management objectives and actions (see Figure 2).
We frame these questions using network theory (note, however, that we do not perform any network analysis). Below, we provide a brief overview of value of information theory before describing our SESs in more detail and then providing details of the modelling approach.

| Value of information
Before commencing a management project, managers must decide whether to reduce uncertainty in a given aspect of the system (i.e., to learn about I xy , q x , H x , C xy or r x ). We measure the expected value of information as the improvement in an outcome when particular information is known with certainty, compared to when that information is unknown (Canessa et al., 2015;Runge et al., 2011). We define the value of information following Raiffa and Schlaifer (1961) and Yet, Constantinou, Fenton, and Neil (2018). Our SES model consists of a set of possible management actions A and a set of uncertain SES parameters with joint probability distribution P( ). For each management action a ∈ A the model aims to predict the utility of a denoted by U(a, ). The expected utility of each management action a is If the value of the SES parameters is unknown, we can calculate the expected utility of each management action and identify which action a yields the highest expected utility (Yokota & Thompson, 2004), that is: Alternatively, if it were possible to gather perfect information about all uncertain SES parameters, then managers could select the action that would maximise the value of the management outcome. Expected utility under this 'perfect information' setting is given by: The difference between maximum expected utility with perfect information (Equation 3) and maximum expected utility (Equation 2) is the expected value of perfect information (EVPI): In our research setting, we are interested in understanding the value of information for individual SES parameters to improve management outcomes. We therefore calculate the expected value of perfect information on 'X' (EVPXI). EVPXI is the difference in expected utility of an optimal action taken when the exact value of an uncertain model input ( x ) is known (Equation 5) compared to one taken knowing only prior information (Equation 2) (Yokota & Thompson, 2004). We illustrate this case with the following example. Consider a division of our uncertain SES parameters into parameter of interest x , and the rest of the parameters −x . The expected benefit if we have perfect information about x is: That is, we are only calculating the expectation over the unknown parameter values of −x , given the value of x . This formulation allows us to define EVPXI as follows: Essentially, EVPXI measures the relative benefits of resolving uncertainty in any one particular uncertain parameter, e.g. ecological or social nodes.

| Systems overview
Our four SESs exhaustively describe potential management actions and objectives for a very simple SES network motif. We replicate the finite set of motifs described in Bodin and Tengö (2012) by replicating the basic network structure (two social and two ecological nodes), and assuming uncertain priors-which allows us to assess the expected value of information (described above) across all parameters. We intend for our results and conclusions to be as generalizable as possible; hence, the system models we employ are as generic as possible (see Appendix S1 for a detailed description of objectives). Managers engage with nodes as this is the most commonly observed management intervention, e.g. payments for environmental services (Ferraro & Kiss, 2002), or restocking wild populations (Aprahamian, Martin Smith, McGinnity, McKelvey, & Taylor, 2003). This means that we are not considering systems where managers act on interactions. Thus, for example, a manager might revegetate an ecological habitat patch, but they would not revegetate linear habitat to improve dispersal (Jellinek, Parris, McCarthy, Wintle, & Driscoll, 2014). The fundamental components of each system model are (a) a discrete-time, continuous-state model of the ecological dynamics; (b) a management objective function; and (c) a discrete-time, discrete-state model of the social dynamics, based on the influence of the management action. Below, we outline these components for each of the four systems we consider. The range of uncertainty considered in each parameter is described in Table 1.
The fishery populations grow logistically with growth rates r 1 and r 2 , and there is dispersal between the populations that is likely ). An environmental manager is acting in the system, and their goal is to maximise the equilibrium size of the harvested fishery, even if this reduces yields. They might associate a higher stock abundance with superior ecosystem functioning or believe that it makes the stock better-equipped to deal with environmental change. To achieve their objective, managers intervene in one of the two social groups to encourage or incentivise a permanent harvest reduction of D = 0.25. Variations on this management setting have been observed around the globe.
For example, in Chile, a co-management program was instituted to stimulate the recovery of shellfish populations. As in our example, fishers in Chile voluntarily participate in observing catch limits (Castilla & Defeo, 2001). Community based fisheries management also features in Japan, where fishing rights apply to the entire sea area adjacent to a given fishing village (Yamamoto, 1995).

| Ecological dynamics
The abundance of the fishery population at patch j at time t is denoted E j,t and changes through time as: Since we are interested in long-term outcomes, we calculate and use the system equilibrium: the vector of E * j values where all E j,t+1 = E j,t = E * j . The growth rates r j drive increases in abundance. The stocks are demographically linked by the dispersal parameters (C jg ), which denote the proportional exchange of individuals. For example, C 12 describes the proportion of adult fish who move from ecological population 1 to ecological population 2. The harvest terms (H i ) indicate the proportion of the adult population removed from each population by social group i each time step. We constrained the value of H i between 0 and 0.4 to limit catches below the maximum sustainable yield (Punt, Smith, Smith, Tuck, & Klaer, 2014) (see Table 1), as the system would otherwise not contain a persistent population. These harvest parameters connect the social groups with the ecological populations, and therefore encapsulate the social ecological interactions. The binary variable u p,k = {0,1} describes the influence of management, as described below.

| Management objective
In system 1, managers aim to engage the social system to achieve an ecological objective-maximise the equilibrium size of the ecological metapopulation: To do so, managers must choose to intervene with social group one (a = 1), or group two (a = 0). The time and resources required for the managers to intervene mean that only one group can be targeted. As we outline below, the dynamics of the SES are probabilistic, depending on the response of the social groups to the management intervention. Managers are therefore actually attempting to maximise the expected value.

| Social dynamics in response to management
The state of the social system is described by the vector u p , where p = {1, 2, 3, 4}, whose binary elements u k = {0,1} describe whether group k engages in the environmental action. The values of u p are determined by the management action A, which (as described above) denotes whether the managers engage with group one (a = 1) or group two (a = 0).
We model engagement as a random process, where each social group has a probability q i of engaging with a management intervention targeted at that group. For example, in this SES, if the managers choose to engage with social group one by choosing a = 1, then that group will undertake the desired action (reducing their harvest rate) with probability q 1 . Once the group makes its decision, it may influence group two to similarly engage, with probability I 12 . This influence network is characterised by the matrix I: The willingness of each group to engage (q i ), therefore defines the dynamics of the social nodes, and the inter-group influence I if defines the dynamics of the network.
Given a particular management action a ∈ A, the response of the communities is defined by a discrete probability distribution over u p : Thus, to calculate the expected performance of the SES given a particular management action a, we calculate B as a function of all u p , weighted probabilistically using Equations 10-13: We assume that the management action carries either no net cost for the group (either has no cost, or is subsidised by the manager), or generates positive benefits, but that the groups will not instigate this action without management intervention. This could be because the groups are unaware of the benefits (assuming they exist), are reluctant to instigate the action, or lack the capacity to begin implementation (Pannell et al., 2006).

| System 2
The management setting for system 2 is a recreational fishery, where managers are planning on restocking a salmon population. In practice, this can occur because managers want to (a) facilitate colonization of new habitats; (b) restore spawning biomass in severely depleted populations, (c) compensate for major environmental disturbances such as hydroelectric development, or (d) augment an existing fishery to enable larger catches (Ritter, 1997;Ward, 2006). For example, in Florida in the USA, saltwater recreational fishing is a multi-billion dollar (US) industry, and fish stocking is used to restore depleted stocks (Tringali et al., 2008). In our system example there are two salmon populations that fisher group may encourage the other fisher group to also increase their salmon harvest with probability I if . The structure of the SES in system 2 is the same as in system 1, however, instead of a social intervention, the managers intervene in the system through the ecological populations. The differences in the dynamics of system 2 relative to system 1 are outlined below.
The ecological dynamics in system 2 account for the restocking of the ecological population by managers. If managers restock ecological population E j , then group S i will respond by increasing their harvest with probability q i , rather than decreasing it. The discrete probability distribution for action vector u p , p = {1, 2, …, 6} whose binary elements u k = {0, 1} describe whether group k engages in the environmental action and depends on the management action are as follows: (10) Pr u 1 = 1,1 = Aq 1 I 12 q 2 + 1 − A q 2 I 21 q 1 (11) Pr u 2 = 0,1 = 1 − A q 2 1 − I 21 q 1 (12) Pr u 3 = 1,0 = Aq 1 1 − I 12 q 2 (13) Cumulative proportion of combined social group Cumulative share of ecological metapopulation

| System 3
System 3 is an agricultural production system where a biological pest (E 1 & E 2 ) negatively affects the utilities of two farmers (Silverstein, 1981). Previous studies have estimated that pests and diseases lower crop production by 30%-40% (Thomas, 1999). To control agricultural pests, managers often try to encourage farmers to adopt integrated pest management practices, for example, biopesticides or resistant cultivars (Parsa et al., 2014).
In The structure of the SES in system 3 is largely the same as in systems 1 and 2. In system 3, however, there is a social objective, and managers intervene in the ecological system by removing a proportion M of the ecological population in node E 1 or E 2 . If managers reduce the biological pest population E j , then group S i will respond by decreasing their own removal with probability q i , rather than in- Farmers' utility is measured as a composite of the inverse size of the pest metapopulation (assessed at equilibrium, E * j ), and a Gini coefficient (Dorfman, 1979)  The discrete probability distribution for management vector u p is as described in system 2.

| System 4
System 4 is based on a non-timber forestry products (NTFP) extraction system. NTFP harvest has been shown to affect ecological processes (Ticktin, 2004), including forest structure and composition (Ndangalasi, Bitariho, & Dovie, 2007). However, commercial NTFP harvest has been promoted as a conservation strategy because it offers local rural people with economic alternatives to destructive land uses such as logging and cattle ranching (Ticktin, 2004). This is the management setting in which we base system 4. In this system, two social groups (S 1 & S 2 ) can extract NTFP from their local forest (E 1 & E 2 ) or convert the land to agriculture by clearing the forest (H 1 & H 2 ) (Chopra, 1993). The management action is social-managers offer incentives to social groups to decrease land clearing for agriculture (by D), and the objective is social-to equitably maximise the communities' utility by increasing non-timber forest products (through increasing the size of the forest patches). Groups will engage with managers-decreasing their land clearing-with probability q i , and influence the other social group to similarly engage with probability I if . Note that in this system, social interactions can amplify the benefits of intervention, while in the pest management system social interactions could reduce the benefits of an intervention.
The ecological dynamics are the same as specified in system 1. The social objective was assessed as specified in system 3 (Equation 23), except that these managers seek to equitably increase the equilibrium forest metapopulation, while the managers in system 3 sought to equitably reduce the pest population.
The discrete probability distribution for this action vector u p is the same as in system 1.

| Model discretisation
We conducted two EVPXI analyses (described in Section 2.2) in each of our four SES. In the first analysis we individually assessed the EVPXI of each of the individual five uncertain model inputs I xy , q x , r x , C xy and H x . In the second analysis we assessed the EVPXI of pairs of uncertain model parameters, grouped according to their character: the social inputs (I xy and q x ), the ecological inputs (C xy and r x ), and the socio-ecological inputs (H x ). The parameterisation of these inputs is described in Table 1.
e.g. I 12 and I 21 ; or q 1 and q 2 ). Each input has a discrete, uniform distribution, which samples the full range of possible input values at n equally spaced intervals. This creates n 2 discrete combinations (e.g. combinations of n values of I 12 and I 21 ). We then specified b replications of the other uncertain model inputs, −x , −x ∈ {1, 2, …, 10}. Using random number setting "twister" in Matlab, we gener-

| RE SULTS
We evaluated the EVPXI of each type of social ecological information in four SES with different management actions and objectives ( Figure 2). Systems 1 and 2 are fisheries examples; in both cases managers have an ecological objective, but in system 1 managers influence the network through the social group while in system 2 managers influence the network through the ecological population.

Systems 3 and 4 are sustainable agriculture examples. Managers
pursue social (primarily economic) objectives-in system 3 they undertake a social action, while in system 4 they undertake an ecological action.
Our EVPXI analyses showed that neither ecological nor social information is inherently more valuable for management: social information is most valuable in systems 1 and 4 (Figure 4), and ecological information is most valuable in systems 2 and 3. Regardless of the SES model, action or management objective, the highest F I G U R E 4 Expected value of partial information, or perfect information regarding X (EVPXI, where X is an uncertain social ecological system model component), in four social ecological systems (see Figure 2). Social uncertain components: q is willingness of a social group to participate in management, I is the influence of one social group on the other; Ecological uncertain components: r is the ecological population growth rate; C is ecological connectivity; Social ecological uncertain components: H is the interaction of social groups with ecological populations. Error bars describe upper and lower quartiles. When interpreting the relative value of information, note that the ranking past the first EVPXI is uninformative. Subsequent rankings would require the re-evaluation of subsequent EVPXIs once the first EVPXI has been addressed

| D ISCUSS I ON
This is the first evaluation of the value of information in a dynamic SES network. Our results can be summarised as: 'learn about the system-lever that you plan to pull'. Although intuitive, our results are at odds with current, widespread research practices: not to consider management actions (or even, necessarily, objectives) when deciding where to prioritise research effort. For example, research effort into the aforementioned night parrot first concentrated on improving understanding of the bird's biology (Pyke & Ehrlich, 2014), without reference to specific management actions that might make use of such information.
Neither the social nor ecological components in our analysis consistently displayed a higher value of information. Their relative importance depended on which was being targeted by management actions. This result can be explained by the conditional nature of information. For example, in our first fishery problem (System 1), the management action involves engaging with a fishing group to reduce harvests. Only if engagement is successful (with probability q x ), can the effects spread to the non-engaged fishing group (with probability I xy ). This latter process of social influence (I xy ) will only ever be relevant if the initial group engages with the managers (q x ), and I xy is therefore only conditionally important. By a similar argument, the socio-ecological connection (H x ) is also irrelevant if the intervention fails. The highest EVPXI is therefore associated with q x , not I xy or H x . If the initial action fails, then the process by which interventions propagate through the social network (its secondary impact) is not important. The more distant the parameter from the point of intervention, the lower its EVPXI. In general, this means that when management actions are social, then the highest EVPXI will be social; when management actions are ecological, the highest EVPXI will be ecological.
Unlike previous analyses that focused on the value of information in social or ecological systems in isolation from each other (Barnes, Lynham, Kalberg, & Leung, 2016;Chadès et al., 2011), our approach considers research priorities in a coupled SES. This explicit comparison allows us to conclude that neither social nor ecological information is more important than the other per se. Thus, contrary to a historical focus on environmental information (Clarke, 1995;Fahrig & Merriam, 1994;Noss, 1990), and a more recent push to consider social information (Dickman, 2010;Mascia et al., 2003); environmental managers need to understand the actions available to them before they can identify which information to prioritise. Our results can also be framed in terms of the primary versus secondary impacts of policy-we find that the primary effects of policy should be given first research priority. This finding is at odds with recent research in the field of economics, which has focused on the secondary impacts of policies, for example, the 'rebound effect' in energy markets and 'equilibrium sorting' models in housing markets (Herring & Roy, 2007;Kuminoff, Smith, & Timmins, 2013). Our result has further relevance for policy decisions, as technological and logistical constraints will limit managers' ability to change their actions in the short term (Cundiff, Fike, Parrish, & Alwang, 2009) whereas policy priorities (objectives) are often subject to greater short-term variability (Rodríguez et al., 2006).
On average, we find that the interactions in SES networks are relatively unimportant because they represent secondary processes.
Our findings are at odds with the perceived importance of 'influence' in social network analysis, as a specific component of 'social capital' (Lin, 1999). In our analysis, influence (social interaction, I) is always secondary in importance to q, which determines how likely a social group is to engage in environmental management. Although our model captures the essential elements of a connected SES, it is possible that the simple four-node geometry undervalues the importance of connections. In particular, large interaction networks can exhibit highly nonlinear dynamics that our model may not capture, such as the percolation threshold observed in complex networks (Newman & Watts, 1999). Such dynamics could dramatically increase the value of information associated with interactions, or with specific parts of the interaction network. However, our results suggest that, on average, information on interactions would still be less important than information on nodes. The value of information reflects its ability to alter management decisions. Because management will generally alter the characteristics of the nodes, rather than interactions, events at the first node will still have to occur for the large number of interactions to matter. Moreover, unless we assume interactions are all the same strength (a strong assumption), each must be learned about separately. Therefore, while collectively interactions may be important, individually they may remain of secondary importanceeven in complex networks. limited nodes and interactions. We limited our analysis to a tractably small SES because multi-dimensional value of information analysis is computationally intensive; our current analysis is already five-dimensional. However, the general interpretation of our results can be extrapolated to inform EVPXI in larger systems without the need for a formal value of information analysis-if managers can determine their available management actions and key system nodes. In the case of more complex SES, inferences about the EVPXI of system components may still be possible using motifs (Milo et al., 2002). If managers can identify dominant motifs in their SES, these motifs could form the basis of an EVPXI assessment.
In our analysis we estimated EVPXI. As with previous analyses (Costello et al., 2010;Johnson, Jensen, Madsen, & Williams, 2014;Runge et al., 2011), we focus on the benefits of additional information, rather than the costs of acquiring that information (but see Essington, Sanchirico, & Baskett, 2018). However, for environmental managers to make an informed decision regarding which uncertain component to investigate further, the relative costs of acquiring different types of information must be considered. These costs may include financial or time costs associated with gathering sample data. Other relevant factors that may increase overall cost include sampling feasibility, the level of expertise required to sample, and the reliability or reproducibility of sampling. These costs can be easily incorporated into the existing analysis by proportionally diminishing the EVPXI of information that is more expensive to collect.
Alternatively, it is possible to calculate the expected value of partial sample information, which calculates the benefits of acquiring imperfect information on each model component.
In our EVPXI analysis, we assume a uniform prior distribution for all uncertain system components. This assumption implies managers know nothing about the components or structure of the system and allows us to imitate complete uncertainty. However, this approach may overstate the amount of uncertainty typically present in an SES, as decision makers may have more information about some SES elements and less about others. It is also possible that uncertainty may be endogenous with respect to SES structure (i.e. the motif). For example, if managers are intervening in the social system, they may have more information about social system elements: reducing the EVPXI for social parameters. This indicates that the more that is known about a given SES, the more nuance is required when assessing EVPXI. Future efforts should test the impact on management outcomes of uncertain SES components in situations where informative priors are available. Priors could be elicited in a workshop context to capture expert's knowledge. Coupled with a sensitivity analysis, this approach would allow analysts to study the influence of prior information on the expected value of information.
Alternatively, informative priors could be incorporated through the use of motifs (Bodin & Tengö, 2012;Milo et al., 2002). As described previously, motifs can be considered the basic building blocks of a network. However, it is worth noting that the relevance of any single motif may be limited, since its contribution to the overall dynamics will be moderated by its interactions with the broader SES network.
We show that to improve environmental management, research should systematically focus on improving understanding of the uncertain SES component that is most directly affected by management actions. Contrary to the assumptions of different subfields of environmental management, the value of information is not intrinsic to the character (social or ecological) of system nodes or interactions.
Similarly, in contrast with the orthodoxy of structured decision-making, value of information is not related to management objectives.
Thus, our results show that when managers are undertaking social actions (e.g., engaging with fishers to increase stock levels in a fishery) their highest research priority should be understanding the dynamics of social groups. Alternatively, when manipulating ecological systems (e.g., controlling invasive species), it will be most important to understand the dynamics of ecological populations. Our insights provide fundamental and practical decision support for addressing ever-present uncertainty that impedes effective environmental decision making worldwide.