Enabling conservation Theories of Change

Global Theories of Change (ToCs), such as the post-2020 Global Biodiversity Framework (GBF), provide broad, overarching guidance for achieving conservation goals. However, broad guidance cannot inform how conservation actions will lead to desired outcomes. We provide a framework for translating a global-scale ToC into focussed, ecosystem-specic ToCs that consider feasibility of actions, as determined by national socioeconomic and political context (i.e., enabling conditions). We demonstrate the framework using coastal wetland ecosystems as a case study. We identi�ed six distinct multinational pro�les of enabling conditions (‘enabling pro�les’) for coastal wetland conservation. For countries belonging to enabling pro�les with high internal capacity to enable conservation, we described plausible ToCs that involved strengthening policy and regulation. Alternatively, for enabling pro�les with low internal enabling capacity, plausible ToCs typically required formalising community-led conservation. Our ‘enabling pro�le’ framework could be applied to other ecosystems to help operationalise the post-2020 GBF.


Introduction
Theories of Change (ToCs) describe how conservation interventions can achieve desired outcomes 1 .The Convention on Biological Diversity's post-2020 Global Biodiversity Framework (GBF) has an overarching ToC for achieving a 2050 vision of 'humans living in harmony with nature' 2 .Operationalising this global ToC will require conservation actions to be implemented by a diverse set of actors working internationally, regionally and locally, including NGOs, governments, and communities 1,3 .Ultimately, these actors will need well-de ned ToCs that state how action can address drivers of ecosystem loss and degradation, dependent on socioeconomic and political factors that in uence conservation feasibility (hereafter referred to as 'enabling conditions').Enabling conditions are fundamental to the development of a meaningful ToC, as ToCs will not be valid unless social, economic, and political mechanisms are in place to enable conservation action 4 .A rst step towards operationalising a global ToC can therefore be to generate multiple, nested ToCs that identify appropriate actions based on enabling conditions and drivers of ecosystem loss and degradation 1 .
Vegetated coastal wetland ecosystems -mangroves, seagrass, and saltmarsh -provide important services that support global environmental goals, such as action to regulate climate (Lovelock et al.,   2017; Zeng et al., 2021), and preserve biodiversity 7 .However, pressure on coastal ecosystems is increasing in all regions of the world 8 , degrading these services and creating an urgent need for their conservation.Coastal wetlands were under-represented in global ecosystem assessments that informed the Convention on Biological Diversity's previous global ToC, the 'Strategic Plan for Biodiversity' 9 .Furthermore, while global goals can inspire action to conserve and restore services provided by coastal wetlands, they will need to be translated into tangible actions.Developing nested, ecosystem-speci c ToC's could provide a rst step towards establishing the strategic direction needed to unlock funding and support for the conservation of these important ecosystems.
Here, we propose a framework for translating a global ToC into nested, ecosystem-speci c ToCs that are informed by enabling conditions.We focused on coastal wetland ecosystems as a case-study for demonstrating the proposed framework.Our approach for developing nested ToCs for coastal wetland ecosystems involved three steps (Fig. 1): 1) Identify and understand enabling pro les by compiling a database of national socioeconomic and political indicators relevant to coastal wetland conservation, and then classify countries with similar indicator values into groups that represent distinct enabling condition contexts for conservation, 2) Identify drivers of ecosystem loss and degradation within each enabling pro le, and 3) Describe plausible, nested ToCs for enabling pro les, i.e., conservation implementation pathways.In our case study, we only describe ToCs for seagrass and mangroves because global data on drivers of saltmarsh loss were lacking.Our framework for operationalising a global ToC has the potential to offer multiple bene ts including: 1) facilitating coordinated actions across multiple actors involved in implementing a global ToC, 2) encouraging knowledge sharing, and 3) providing a basis for 'experimental adaptation' whereby conservation actions are tested under different enabling pro les.Our case study results are most relevant to actors working internationally, as we consider how national enabling conditions can inform the development of multinational enabling pro les and associated ToCs.However, our framework could be applied sub-nationally to develop ToCs relevant to actors working at a local scale.

Identify and understand enabling pro les
From a database of 19 national socioeconomic and political enabling condition indicators (Supplementary Table S1), we used cluster analysis to identify 6 multinational enabling pro les for coastal wetland ecosystems (Fig. 2A).We then used classi cation trees to determine the relative importance of national indicators in differentiating enabling pro les (Fig. 2B), and how individual indicators de ne each pro le (Fig. 2C).To aid interpretation, we categorised the 19 national indicators into the following groups: 1) Policy -policy commitments and governance frameworks to facilitate conservation work (including international treaties), 2) Regulation -active management of pressures and impacts to the environment, 3) Engagement -active engagement with conservation, either through nancial investment (domestic or foreign) or social interest (Fig. 2B&C).Key indicators differentiating enabling pro les were the regulation of wastewater pollutants, regulation via environmental tax, the number of biodiversity-related projects funded by international aid, domestic conservation spending, Ramsar management, and commitment to international climate policy (Nationally Determined Contributions -NDCs) (Fig. 2B).Post-hoc hierarchical cluster analysis revealed that enabling conditions in Pro les 1 & 2 were more similar relative to Pro les 3 & 4 and Pro les 5 & 6 (Fig. 2A).The majority, i.e., 91%, of countries in Pro les 1 & 2 were high-income countries, 77% of countries in Pro les 3 & 4 were middle-income countries, and 52% of countries in Pro les 5 & 6 were low or lower middle-income countries (see Supplementary Fig. S1 for country income-status and enabling pro le designation).
Pro les 1 & 2 had high capacity to enable conservation through policy, regulation, and domestic conservation investment relative to other enabling pro les, however mangroves, seagrass and saltmarsh were not included in their NDC climate mitigation and adaptation policy strategies (Fig. 2C).Pro le 2 also had relatively low protection of vegetated coastal wetlands via the Ramsar convention, although implementation of management plans in Ramsar protected areas was high (Fig. 2C).Pro les 3, 4, 5, & 6 generally had higher capacity for enabling conservation through engagement mechanisms linked to foreign aid and social interest, although Pro le 1 had relatively high NGO-support for environmental projects and social interest in biodiversity.Conversely, policy and regulatory capacity in Pro les 3, 4, 5 & 6 was typically lower, with the exception of NDC climate mitigation and adaptation strategies and Ramsar protection.
There were clear differences in the policy and regulatory capacity of Pro les 3, 4, 5, & 6 (Fig. 2C).
Speci cally, Pro le 3 had moderate to high capacity for most policy and regulation indicators, whereas Pro le 4 had moderate to low capacity on most of these indicators (Fig. 2C).Pro le 5 had relatively low policy capacity and moderate regulatory capacity (Fig. 2C).Pro le 6 included countries affected by internal con ict (e.g., Somalia) and international sanctions (e.g., North Korea) (Fig. 2A), and had moderate to low capacity for most policy and regulation indicators, with the exception of including vegetated coastal wetlands in NDC climate change mitigation and adaptation strategies (Fig. 2C).

Identify Drivers Of Ecosystem Loss
We identi ed drivers of ecosystem loss within enabling pro les for mangroves and seagrass only, as global data on drivers of saltmarsh loss were not available.For mangroves, the main drivers of loss within enabling pro les were non-productive conversion or erosion, although for Pro le 3, agri/aquaculture accounted for a substantial proportion of loss (Fig. 3B).Pro le 2 countries did not intersect with the global distribution of mangroves and so are absent from Fig. 3.
For seagrass, catchment processes (e.g., coastal development, erosion, ooding) were a driver of loss common to all enabling pro les, while boating-related losses were unique to Pro le 1 (Fig. 4).
Climate/storms were also a driver of loss for Pro les 1 & 3, aquaculture and shing drove seagrass loss in Pro les 1, 2, 3, & 4, and disease drove seagrass loss only in Pro les 1 & 4 (Fig. 4).Pro le 5 countries did not intersect with the global distribution of seagrass and so are absent from Fig. 4.

Describe Plausible, Nested ToCs
We described a plausible ToC for conserving mangroves or seagrass in each enabling pro le.Our ToC descriptions were formalised as causal statements of how action can address drivers of loss, and lead to desired conservation outcomes (sensu Qiu et al., 2018 4 ; Fig. 5 and see Supplementary Table S2 for a detailed description of all ToCs and case-studies providing qualitative validation).In enabling pro les 1 & 5, non-productive conversion (e.g., vegetation dieback from nearby human development such as mines and roads, harvesting of mangrove trees for timber) was a main driver of mangrove loss, but ToCs differed (Fig. 5; Supplementary Table S2).In Pro le 1, improved monitoring of indirect negative effects on mangroves could inform improved policy and regulations to reduce mangrove dieback 12,13 (Fig. 5; Supplementary Table S2).Alternatively, in Pro le 5, mangrove clearing for fuel or timber could be reduced if NGOs are engaged to support the development of community-based sustainable management of mangroves, and ensure this is recognised in government policy 14 (Fig. 5).For seagrass, ToCs to address loss driven by aquaculture or shing differed between Pro les 2 & 4 (Fig. 5).In Pro le 2, policy could be established to ensure aquaculture is not placed near seagrass 15 .In Pro le 4, external support and funding to establish payments for seagrass ecosystem services could provide an alternative source of income that incentivises the reduction of destructive shing practices that negatively impact seagrass 16 (Fig. 5).

Discussion
Our framework for operationalising a global Theory of Change (ToC) ensures that enabling conditions underpin pathways for implementing conservation, thereby increasing the likelihood of achieving desired outcomes 4 .Enabling pro les offer a platform for knowledge transfer between pro les and countries that share drivers of ecosystem loss and degradation.In an era of rapid and complex global change, sharing knowledge on how to effectively implement conservation is important.Our framework could also encourage testing of conservation actions under similar or different enabling condition contexts, thereby encouraging experimental adaptation of ToCs 17 .

Theories Of Change For Coastal Wetland Ecosystems
We identi ed six distinct enabling pro les to inform nested ToCs for globally coordinated conservation of coastal wetland ecosystems.Pro les 1 & 2 generally had high capacity to enable conservation via policy, regulation, and domestic funding relative to other pro les.Many countries in Pro les 1 & 2 belong to the European Union (EU) where multilateral environmental agreements (e.g., the Water and Marine Strategy Framework Directives) have improved water quality and led to recovery of lost seagrass 18 .Alternatively, in Pro les 3, 4, 5 & 6, capacity to leverage support for conservation via engagement with external actors was relatively high (see Supplementary Table S2 for a detailed description of enabling conditions in each pro le).In the past, external actors such as NGOs have played an important role in prompting governments of countries in these pro les, e.g., Mexico and South Korea, to effectively implement Ramsar protection or wetlands 19 .
We used real-world examples of conservation interventions to validate proposed implementation pathways for each nested ToC (see Supplementary Table S2).We recommend that, where possible, pathways for implementation be tested quantitatively by relating enabling conditions to conservation outcomes, thereby ensuring ToC are robust (sensu Williamson et al., 2018 20 ).However, in data-sparse contexts, real-world examples of conservation interventions provide a qualitative alternative for justifying proposed pathways.Depending on enabling conditions and drivers of loss, it may be possible to de ne severable plausible implementation pathways that could become testable hypotheses, forming a basis for experimental adaptation of ToCs 17 .Where experimentation cannot be used to choose from competing implementation pathways, the heuristic 'Mitigation and Conservation Hierarchy' could help differentiate priority actions (i.e., refrain, reduce, restore, renew) 3 .The development of robust, nested ToCs may be limited by information available on enabling conditions and drivers of loss for individual ecosystems, and these should be made transparent to stakeholders (see Supplementary Table S3 for a complete description of the limitations of our case study on coastal wetland ecosystems).
There is no 'one-scale-ts-all' ToC.For example, the nested ToCs that we have described may not have su cient detail or local context for actors working to implement conservation on the ground.To overcome this, the Intergovernmental Science-Policy Platform on Biodiversity and Ecosystem Services (IPBES) has recognised the need for multi-scale conservation planning 21 .Our framework could be extended to support multi-scale conservation planning by establishing multi-level, hierarchical enabling pro les that represent enabling condition contexts operating at different spatial scales (e.g., sub-national enabling pro les nested hierarchically within multinational enabling pro les).In our case-study, ToCs were informed by enabling conditions operating at the national-scale and are therefore most relevant to actors developing and coordinating conservation actions internationally.To be relevant to local-scale conservation practitioners, ToCs could be further developed using a participatory framework (sensu Reed et al., 2022 22 ) that engages actors working across sectors and scales, from local practitioners to international policymakers, and ensures ToCs are just and equitable.It is also important to recognise that human behaviour can play an important role in whether ToCs will achieve desired outcomes.Tacit working models of how human behaviour and conservation relate to one another could be used to integrate this understanding into ToC development 23 .
Our framework provides a rst-step towards translating a global ToC into actionable implementation pathways.As countries implement conservation actions and monitor outcomes, the dissemination of knowledge and learnings could help other countries de ne ToCs or inform adaptation of those that are already established.ToCs by their very nature will be dynamic, requiring adaptation as enabling conditions and drivers of loss change through time.Therefore, ToCs should not be considered static entities, but instead should be adapted as enabling conditions and drivers of loss change.For example, rapidly developing middle-and low-income countries may acquire greater internal capacity for facilitating conservation and rely less on international aid 24 .Their ToCs could be adapted based on what has or has not worked in other enabling pro les with similarly high internal capacity for conservation.

Conclusion
Global conservation planning and mapping has been criticised for lacking a clear ToC 25 .It is true that a global ToC is at risk of failing to achieve overarching goals if it is not effectively translated into tangible and discrete pathways for implementing action.Our framework for operationalising a global ToC makes national and international capacity for conservation transparent and ensures that this is at the forefront of developing robust implementation pathways.

Compile a database of national enabling condition indicators
We compiled a database of national values for 19 policy, regulation, and engagement indicators representative of enabling conditions for vegetated coastal wetland conservation in 138 countries, i.e., 70% of all countries with oceanic coastline.To identify countries with these wetlands, we intersected the EEZ boundaries of countries that have oceanic coastline 26 with the global distributions of mangroves 27 , seagrass 28 , and saltmarsh 29 .
We used national indicators to classify countries into global enabling pro les that represent similar policy, regulation, and engagement settings for conservation.We rst used a Bayesian latent variable Where indicators were found to violate the 'missing at random' assumption we t an additional LVM without these indicators to check our predictions were robust to missing data.
Eleven indicators had missing values that were interpolated (see Supplementary Table S1 for the percentage of missing values for each indicator, ranging from 0-46%; and see Supplementary Fig. S2 for assessment of model t).The 'Conservation spending' indicator violated the 'missing at random' interpolation assumption of the LVM because data were only available for countries that were signatories to the Convention on Biological Diversity or the Sustainable Development Goals 31 .However, only one country (the United States) was missing a value for this reason, and all other missing values were due to insu cient data 31 .The 'Ramsar Management' indicator also violated this assumption because countries without coastal wetland Ramsar sites were designated as 'NA' (Supplementary Table S1).However, predictions from LVMs t with and without each indicator were positively correlated (Supplementary Fig. S3 & S4), demonstrating that parameter estimates were robust.Supplemental methods for tting models are also provided in Supplementary Appendix 1.

Classify countries into global enabling pro les
We performed a cluster analysis on the gap-lled, standard-normal indicator values obtained from the LVM to group countries into enabling pro les.Speci cally, we used k-medoid clustering with the 'partitioning-around-medoids' algorithm on a Euclidean distance matrix of indicator values.Standardnormal indicator values were rescaled to the minimum and maximum values of the indicator with the narrowest range before clustering to reduce leverage of indicators with exceptionally large ranges (i.e., binomial response variables: NDC commitment, NDC adaptation, NDC mitigation, and Ramsar protection).We investigated a range of clustering con gurations (n = 5 to 10) to identify the number of clusters that best represented country-level variability in indicator values, while also identifying general patterns useful for informing coastal wetland conservation.We used average silhouette width 32 to measure the quality of each clustering con guration (i.e., cluster cohesion and separation).All con gurations were of similar quality, so we chose 6 clusters as the nal con guration because it best balanced national indicator variability with generalisable patterns across countries.We assessed the robustness of clusters by re-evaluating the cluster analysis across the full distribution of indicator values predicted by the LVM (see Supplementary Fig. S5 and S6 for an assessment of the robustness of the nal clustering con guration).Finally, we used post-hoc hierarchical cluster analysis of cluster medoids to group and order enabling pro les by their similarity, and we used principal components analysis to visualise country-level variability within enabling pro les.

Determine how indicators de ne global enabling pro les
We classi cation trees to determine 1) the relative importance of national indicators in the classi cation of enabling pro les, and 2) how individual indicators de ne each pro le.Classi cation trees are non-parametric, supervised machine-learning models that use recursive partitioning to generate decision rules that relate predictor variables (i.e., indicator values) to response variables (i.e., enabling pro les) 33 .Observations are repeatedly split into sub-groups by predictor variables, aiming to minimize heterogeneity of observations in each sub-group of the nal tree 33 .
To measure the relative importance of indicators, we t a classi cation tree using indicator values as predictors of enabling pro les.Indicator importance was measured as the sum of the Gini goodness of split measure where the indicator was a primary splitting variable in the classi cation tree.Gini goodness of split is measured as the inverse of Gini impurity, an estimate of the probability of misclassi cation 34 .We also used decision rules generated by individual classi cation trees, where each indicator was the sole predictor of enabling pro les, to identify indicator thresholds that de ne each pro le.To minimize the in uence of outliers on threshold de nition, we t individual classi cation trees using only indicator values within the interquartile range of each enabling pro le.Threshold values were re-scaled from 0 to 1 to provide a relative measure of indicator scores de ning each pro le, where 0 = low and 1 = high.

Identify drivers of ecosystem loss in each enabling pro le
We identi ed drivers of coastal wetland loss and degradation in each enabling pro le using 1) data on drivers of mangrove areal loss derived from satellite data 10 , and 2) data on the drivers of seagrass areal loss from a meta-analysis of in-situ and remote sensing data 11 .Global data on drivers of saltmarsh loss was not available 9 .We use the term 'drivers' to refer to environmental stressors (both human and natural) that can cause ecosystem loss and degradation.This is unlike the well-known DPSIR (Driver-pressuresstate-impact-response) framework, rst elaborated in the European Environment Agency (EEA) programme and later on adopted for other environmental issues in Europe 35 , which refers to human environmental stressors as 'pressures'.However, our terminology is consistent with the literature for mangroves 10 and seagrass 11 .
Global drivers of mangrove loss from 2010 to 2016 were: erosion, extreme weather events, commodities (i.e., agriculture or aquaculture), non-productive conversion (including clearing and dieback from indirect effects of human development), and human settlement 10 .We calculated the proportion of mangrove loss attributed to each driver in each country, and then averaged these proportions across enabling pro les.
This statistic standardizes for differences in overall mangrove area across different countries.Seagrass study locations from Dunic et al., 2021 11 were intersected with country EEZ and enabling pro le boundaries, and drivers of trends were identi ed for each enabling pro le and continent to determine opportunities for conservation.
Seagrass data were not globally comprehensive and so the identi cation of ecosystem loss drivers was limited to countries where peer-reviewed studies identi ed drivers of trends in seagrass meadow area.Primary drivers were identi ed from original sources in one of two ways: 1) attribution by visual (aerial imagery or graphical) or inferential (statistical) methods or 2) the driver that was described and discussed most frequently 11 .We opted to exclude the 'invasive species' driver from our analysis because invasive fauna, such as tunicates, crabs, and lugworm disturbance, were not reported in the peer-reviewed literature, which may mis-represent the distribution and in uence of this driver.Note that absence of these invasive fauna in the peer-reviewed literature may be due to lack of classi cation/nomenclature.For example, lugworm disturbance has been identi ed as a driver of seagrass loss, but the lugworm was not classi ed as an invasive species 11 .A complete description of mangrove and seagrass drivers is provided in Supplementary Table S4.
Describe nested ToCs for each enabling pro le We described nested ToCs for each enabling pro le as causal statements that de ne how actions can lead to desired conservation outcomes for mangroves and seagrass.We used case-study examples to qualitatively validate nested ToCs.
model (LVM) to gap-ll missing indicator values prior to classi cation 30 .The LVM estimates correlations among all indicators across countries and leverages these correlations to interpolate missing values.The model assumes that values are 'missing at random', which we evaluated for each indicator by determining whether the probability of missing values was likely to be dependent upon both observed and unobserved information.The model was formulated: where µ ij is the mean response at country i for indicator j, is the indicator-speci c intercept, z i are vectors of latent variables, and are their corresponding indicator-speci c coe cients 30 .We set the number of latent variables in our model to 9 (approximately half the number of indicators), which provided accurate estimates of indicator responses (see Appendix 1 in the supplementary materials for a detailed description of model settings).Prior to tting the LVM, continuous indicator response variables were log-transformed and z-score standardised (mean = 0, standard deviation = 1).We then used Eq. 1 to predict indicator values to all countries, including interpolating to those countries with missing values.