Forest risks and benefits to human health are widely recognised. Yet, variation across forest types and their ecological characteristics driving health effects remain underexplored. Based on empirical data from an interdisciplinary European forest network, we developed a Bayesian Belief Network to quantify seven causal pathways relating different forest types to physical and mental health. Results show that forests always generate net health benefits regardless of their ecological characteristics. Forest canopy density and tree species diversity emerge as key drivers, but their effect size and directionality are strongly pathway-dependent. Changes in forest canopy density can generate trade-offs. For example, forests optimised for heat buffering and air pollution mitigation may compromise medicinal plant yield and enhance Lyme disease prevalence. Tree diversity effects were weaker but more consistently positive. Forest management should therefore account for such trade-offs to tailor forest biodiversity and functioning to local public health needs of priority.
Article
Forest biodiversity and structure modulate human health benefits and risks
https://doi.org/10.21203/rs.3.rs-4669329/v1
This work is licensed under a CC BY 4.0 License
Version 1
posted
You are reading this latest preprint version
Dr.FOREST
ecosystem service
nature-based solution
environmental health
environmental psychology
A growing body of scientific literature hails contact with nature for its physical, mental and social wellbeing benefits1–5. With 74% of deaths worldwide caused by non-communicable diseases (e.g., cardiovascular and respiratory conditions)6 and an increasing global burden of mental health disorders7, preventive nature-based solutions can counterbalance multiple public health issues. Increasing the surface area of publicly accessible greenspaces has wide-ranging benefits observable at the scale of cities including, for example, reduced sales of mood disorder and cardiovascular medication8, heat mortality mitigation9 and, more generally, annual preventable death reductions by 0.2–5.5%10.
Not all greenspaces are equal, and identifying greenspace characteristics that optimise health outcomes can help offsetting the limitation posed by space restrictions1. Firstly, the vegetation type will modulate health outcomes11,12. Reviews comparing greenspaces such as forests, grasslands and others suggest that forests are more strongly linked with a range of physical and mental benefits1,12. With their high level of biomass and tall canopies, trees and forests may create immersive environments that are more strongly decoupled from ambient conditions than low-stature vegetation. Forests thereby foster cooler13 and subjectively quieter3 environments with – in most circumstances – better air quality14 that benefit physical and mental health. A global evidence assessment affirms the supportive role of forests in the physical, mental, social and spiritual health of humans throughout all life stages3.
Compared to greenspace quantity, the quality of a given type of greenspace also affects human health but is much less studied11,12. Quantity is commonly assessed using well-defined proxies such as surface area of vegetated land-use classes or the Normalised-Difference Vegetation Index (NDVI)9,10, representing greenspaces as relatively homogenous entities1,11. In contrast, greenspace quality is often inconsistently quantified and can encompass aspects of accessibility, safety, amenities and biodiversity11,12,15. In health sciences, the latter is often proxied by indices that are merely indirectly related to biodiversity including, for instance, perceived bird, plant and butterfly species richness16 or ‘naturalness’5. These are poorly aligned with standards for rigorous ecological inventories established in the natural sciences11,15,17. Yet, thorough quantification of biodiversity holds great promise. Five decades of biodiversity-ecosystem functioning (BEF) research have demonstrated that ecologically-relevant characteristics such as species diversity, functional traits and structural complexity greatly influence the functions and services that ecosystems provide18–20 – ultimately shaping health outcomes17. Until now, the integration of BEF into biodiversity-health research has been limited, accentuating the need for interdisciplinary approaches.
The biodiversity-health relationship is increasingly recognised, including by the Convention on Biological Diversity (CBD) and the World Health Organisation (WHO)4,19,21. Still, the majority of nature-health studies to date overlook biodiversity effects and fail to unravel underlying mechanisms11,12. Reported greening effects on reductions in heat mortality9, antidepressant prescriptions22 and overall mortality10 are compelling and invaluable to illustrate the magnitude of nature’s disease-prevention potential at large scales. However, the many factors co-varying with biodiversity in broad-scale longitudinal datasets limit attribution of causal effects11,12. These scientific efforts should be complemented with experimental studies with proper control conditions to understand the full complexity of nature-health relationships and their inherent context-dependency1,17,18. Identifying mechanisms and context-dependencies may inform how and when management interventions can optimise health outcomes. For example, increasing tree canopy cover may reduce heat and air pollution levels in rural settings, but may aggravate these environmental stressors in cities depending on specific building and street design by locally trapping heat and particulate matter23,24. Interindividual differences can also alter health outcomes, with perceived rather than actual biodiversity generally better predicting mental wellbeing outcomes16,25. Subjectivity may also affect physical wellbeing, with forest environments reducing perceived heat26 and perceived noise27 more strongly than expected from objective measurements. Nature and forests may also cause harm by enhancing exposure to disease-bearing vectors28 and potentially allergenic pollen29. Such ecosystem disservices are rarely considered in some disciplines17,21, though physical risks related to wildlife exposure are thoroughly covered in the field of tropical medicine3. At last, a handful of studies insinuate that nature-health outcomes may positively or negatively interact26,30. How health benefits can be optimised and risks minimised, leveraging synergies and avoiding trade-offs between pathways, is largely unknown.
Here, we use a unique continental-scale interdisciplinary database gathered in situ in 164 European forest stands (Fig. 1a) by a team composed of ecologists, environmental psychologists, neuroscientists and public health experts to examine seven distinct forest-health pathways: i) mental wellbeing via visual stimuli; ii) mental wellbeing via auditory stimuli; iii) thermal comfort; iv) polyphenol content of medicinal plants; v) mushrooms and nutrition, vi) air quality, and vii) ticks and Lyme disease. The latter represents an important ecosystem disservice in Europe. Forest ecological characteristics were assessed using well-established forest inventory methods, enabling the attribution of health outcomes to recognized ecological variables (Fig. 1b). The diversity in assembled datasets was synthesised using a Bayesian Belief Network (BBN), combined with an adapted version of the Ecosystem Services Cascade Model19,31 providing the mechanistic framework. This flexible statistical approach allowed for the modelling of complex interaction chains, the detection of synergies and trade-offs and the inclusion of crucial but unsampled variables.
Effects of forest characteristics on functions, services and health outcomes
The effect magnitude and direction of seven focal forest characteristics depended strongly on specific forest-health pathways and their considered stage along the Cascade Model (i.e. function, service or health outcome; Fig. 2). Canopy density was the most influential forest property overall, followed by stand density and tree diversity. Canopy height, proportion of deciduous trees and leaf size had relatively weak effects, and stem density was least influential. However, effect magnitude was highly dependent on the forest-health pathway. Increased tree diversity was a dominant driver for mental wellbeing via visual stimuli and mushroom pathways. Canopy density had a positive effect on the thermal comfort pathway yet a negative effect on medicinal plants. Increased canopy density led to increased risk of Lyme disease, mainly by strongly increasing relative humidity thereby supporting high tick population densities. Mental wellbeing via visual stimuli was almost not influenced by actual forest characteristics (i.e., in opposition to what is subjectively perceived).
Changes in canopy density, represented by Leaf Area Index (LAI), were explained by mean leaf size (6.2% of variance explained), followed by stand density (3.1%), tree diversity (1.3%) and canopy height (0.7%). We highlight these results for canopy density as a central driver, but statistics on associations between other forest characteristics can be found in the Supplementary Information (p. 11).
The more intermediate steps between two variables in the model, the less sensitive they were to each other. This dilution effect starts with the forest characteristics, proceeding via function to service and then to health outcome variables. However, this top-down approach largely conceals the predictive power of intermediate variables. This can be demonstrated with the mechanisms underlying mental wellbeing via visual stimuli. While tree diversity only explained 1.2% of the variance in perceived biodiversity, the latter was strongly explained by more intermediate variables: perceived density (6.7%) and perceived naturalness (5.4%). Furthermore, perceived biodiversity itself effectively determined restorativeness (13.9%), attention restoration (4.4%) and stress reduction (4.3%). For a complete understanding, sensitivities should thus be scrutinised case-by-case using the full model (Supplementary Fig. S1).
Trade-offs and synergies in health outcomes
Based on mutual information held among variables in function, service and health outcome stages, several synergies and trade-offs emerged. A quartet of pathways were consistently associated in function, service and health outcome stages: thermal comfort, medicinal plants, air quality and ticks and Lyme disease (Fig. 3). This means that forest characteristics that effectively reduced heat stress also improved air quality through reducing air particulate matter. Similarly, medicinal plant production and Lyme risk reduction also synergized. However, these two pairs of pathways were antagonistic to each other. Forests that provide a high level of medicinal plant production and regulation of disease (reduced risk in Lyme) had a reduced capacity for microclimate and air quality regulation. Some weaker synergies and trade-offs were found with respect to the three remaining pathways, where especially both mental wellbeing pathways were relatively disconnected from other health outcomes.
Comparing overall forest effects to inter-forest variability
When health outcomes were defined relative to non-forest conditions, forest presence generated substantial absolute health benefits – regardless of forest characteristics (Fig. 4a). Interpretation was more challenging when explicit controls were absent (Fig. 4b), requiring a case-by-case consideration. For example, mental wellbeing outcomes were compared before and after being exposed to visual or acoustic stimuli for the mental wellbeing pathways.
Overall forest effects were compared to relative differences caused by changing canopy density and tree diversity. Some health outcomes appeared to be nearly exclusively determined by forests in general, with very small changes generated by their ecological characteristics. This is most notably the case for anxiety reduction (visual), negative affect reduction (auditory), positive affect increase (visual and auditory) and PM10 risk mitigation. For cases where forest characteristics did matter, tree diversity was most influential for mushroom health benefits. Yet, canopy density was again the strongest driver of variation: most evidently in microclimate (thermal comfort) and disease (Lyme) regulation, followed by medicinal plants- benefits and PM2.5 risk mitigation.
Analogous analyses of contrasted stand types (young plantations vs. mature natural forests) were yielding even smaller differences than shown in Fig. 4. While changes in stand type affected canopy height very strongly, the latter had generally little influence on other variables (Fig. 2).
Drivers of forest functioning, services and health outcomes
We here report on the first trans-continental study linking forest biodiversity, functioning and health outcomes, capitalising on interdisciplinary approaches. For example, recent heat mitigation research reported which forest characteristics determine surface or air temperature and relative humidity of forest microclimates35, but few have connected this to indices relevant to human perception and health outcomes33. In parallel, heat mitigation by ‘homogenous’ greenspaces is already thoroughly reported9,36, but ignores ecological intricacies. By using a BBN to merge multiple data sources, this gap was effectively bridged for multiple forest-health pathways.
The two most influential and independent forest characteristics were forest canopy density and tree diversity. Canopies are key drivers of ecosystem functioning18, creating a physical, chemical and biological filter to ambient conditions37. Tree diversity is also a recognised determinant of ecosystem function, ecosystem service provision and non-tree biodiversity18,38. Beyond confirming these known consequences, we here show that they also shape human health impacts. Even though effect sizes are relatively small, it may make a substantial difference at the scale of populations at which greenspaces can prevent well over hundreds of deaths9,10.
The process of disentangling mechanistic pathways via the Ecosystem Service Cascade Model also revealed dependencies at spatiotemporal, societal and individual levels. At the spatiotemporal level, the importance of forests to human health was defined by the surrounding environmental conditions, where locations with high particulate matter pollution and periods with high heat loads benefited proportionally more from forest tempering effects13. Likewise, seasonality determines production of mushrooms39 and medicinal plants40, and the amount of songbird-generated biophony with potential mental benefits41. Societal context also plays a crucial role: local legislation and overall foraging culture are determinants for potential health benefits from non-timber forest products42. For example, while mushroom and berry collection is common in e.g. Poland, Germany and France42, it is legally not allowed in Flanders, Belgium. Mental wellbeing is also determined by societal realities: forests and dense vegetation can provoke fear of assault by criminals or wild animals43. This implies that our findings are best applicable to the world’s forests that are perceived as ‘safe’. At last, individual experiences may further shape health outcomes. Fear of crime varies with individual factors such as age, gender and culture43. Mental wellbeing was relatively insensitive to actual changes in forest characteristics in our analysis. This is partly explained by the pivotal role of perceived (visual and auditory) biodiversity, which can be based on cues like species richness, greenness, acoustic complexity or functional traits like colourfulness25 – thereby recurrently deviating from actual biodiversity21. Half of variation in thermal comfort is determined by factors other than the physical microclimate, including physiological acclimatisation, biological sex, thermal preference and expectations44.
Trade-offs bar all-round optimisation
Health outcome interactions were most clear for heat mitigation, medicinal plants, air quality and Lyme disease risk, with canopy density as the common denominator. The latter enhances mitigation of heat stress33,35 and particulate matter34, but reduces understory light availability45 which impedes production of medicinal plants. Closed canopies also create ideal conditions for tick activity and survival by maintaining relative humidity levels that rarely fall below 80%46. Whereas canopy density is the most influential forest characteristic, corroborating a recent study on ecosystem services47, it can inadvertently lead to unwanted side-effects. This does not apply to tree diversity: though less influential, it always returned positive effects – except for a small increase in Lyme disease risk. Note that the effect of increasing tree diversity and forest complexity on Lyme risk in Europe is contested, with reports of both negative48 and positive49 outcomes.
Some other, weaker, interactions were observed or theoretically expected. A separate analysis found enhanced levels of thermal and mental wellbeing to mutually reinforce each other, where forest environments incite this synergy26. This non-linear, subtle interaction did not clearly emerge here (Fig. 3). Other indirectly explored interactions include the increase in questing tick nymphs with higher levels of medicinal plants, because ticks use understory vegetation to reach vertebrate hosts28,49. At the same time, more lush understory vegetation affected perceived density, which was a driver for perceived biodiversity and thus potentially determines mental wellbeing. Greater bird diversity increases biophony, but also influences tick nymphal infection prevalence as birds are suitable hosts for multiple Borrelia strains50. Some interesting interactions could not be explored. High rainfall frequency and abundance may increase mushroom biomass39, but also improve the air filtering capacity of the forest by cleaning the leaves and rendering them more sticky51. Given that air pollution may worsen mental health52, filtering of particulate matter by forests may have additional mental health effects.
Again, this emphasises the need to consider results in the light of local needs. In a theoretical region with low-to-absent Lyme prevalence and foraging regulations restricting medicinal plant harvesting, silvicultural interventions may aim for closed canopies to foster the synergy in heat and particulate matter improvements. Prioritisation of target health outcomes should follow the health impact, the number of people affected and forest effect magnitude given the local context. For instance, in 2022, air pollution and heat caused an estimated 300 00053 and 60 00054 deaths in Europe, respectively, which could be effectively mitigated by forests9. In contrast, despite widespread malnutrition in Europe53, harvesting mushroom and medicinal plants may only have marginal health impacts compared to industrial sources of food and medicine production42. Still, these may have great importance locally and have recreational co-benefits not addressed here42.
Suboptimal forests are better than no forests
Despite the large variability, our findings indicate that forests generally have net benefits to health, regardless of their biodiversity and structure (Fig. 4). Mental wellbeing effects were especially little influenced by actual measures because they essentially revolve around perceived forest characteristics, whereas predicted heat mortality strongly varied between forest types. Yet, even the least optimal forest stands led to strong heat reductions. Indeed, even young plantations have been found to reduce perceived temperatures by 10°C under hot conditions33. In fact, the whole variety of forest conditions was generally unequivocally favourable to human health, except for ticks which are especially abundant in forests compared to any other habitat28.
Altering forest characteristics is thus less consequential than the mere presence of forests, which has two positive implications. First, studies considering forests as homogenous entities certainly ignore plenty of variation, yet their findings may be roughly applicable to a wide range of forest conditions. Second, increasingly prevalent reforestation efforts, despite being mostly constituted of young plantation forests55, may already yield substantial health benefits to people.
Strengths and weaknesses define avenues for future research
While our modelling effort brings about insights on largely uncharted scientific terrain, limitations require consideration before extrapolating our conclusions. From the technical side, BBNs were elemental to linking forest characteristics with their health effects owing to the capacity to model complex causal pathways with heterogeneous data sources. Yet, they have inherent weaknesses. First and foremost, the mandatory discretisation of continuous variables is disadvantageous for environmental modelling, where such variables are omnipresent. This leads to significant information loss that can partly be compensated for by increasing the number of states – though this ramps up the number of required observations to populate Conditional Probability Tables56. We compensated for this by using expert opinion as priors, which improves information content for state combinations that are not covered by empirical data31. Despite these measures, a clear effect size dilution was visible especially for long causal chains. While partly due to discretisation, this incidentally also reflects real-life processes: as the myriad of unmeasured confounders and aforementioned dependencies modulate nature-health relationships, small final effects are expected. Forest-health benefits and their modulation by forest characteristics should therefore not be exaggerated.
From the conceptual side, a major strength and singularity is the direct integration of multiple forest-health pathways. Yet, our selection of forest-health effects is not comprehensive. Others include, for example, reduction of noise stress, modulation of pollen allergy prevalence and fostering of spiritual wellbeing3. These may create additional synergies and trade-offs that we have not covered. Next, our analysis focuses on local scales (plot- or stand level) but some pathways require consideration of landscape-scale dynamics. Most notably, highly mobile vertebrates such as ungulates and birds are likely influenced by characteristics of whole forest complexes. At last, outcomes should be cautiously interpreted in light of assumptions listed in the Supplementary Information. In sum, pathways assume that users interact intimately with the forest environment: our ‘virtual guinea pig’ is consistently exposed to forest microclimate, atmosphere, soundscape and visual aspects and regularly consumes non-timber forest products. These assumptions may be naive in many cases, especially for urbanites, and resonates with the challenge of quantifying nature exposure, accounting for the frequency, duration and type of nature contact57.
In short, future research may turn towards following avenues: i) directly comparing multiple forest-health pathways and their interactions, ii) considering mechanistic processes via frameworks like the Cascade model and coupling statistical approaches like Structural Equation Models or BBNs, iii) incorporating relevant undesirable health effects, and iv) investigating how biodiversity-health relationships apply to (semi-)urban environments.
Management recommendations and conclusions
Despite the case-specificity and complex pathway interactions, general patterns stand out. Firstly, any European forest is likely to have beneficial effects regardless of how it is managed, certainly compared to urban environments5. Any type of forest is also likely to be more beneficial than low-stature vegetation for heat and particulate matter reduction, mushroom production and mental wellbeing, though forests host more ticks28.
Yet, forest managers and policy makers have several degrees of freedom to enhance health effects and mitigate risks, with potentially considerable impact at the public health level. Our findings suggest that targeting canopy density (leaf area index) and stand density (basal area) are the most effective tools to modulate physical health outcomes in particular. Management strategies aiming at densification such as continuous cover forestry may enhance heat buffering and air filtering capacities, but may lead to an increased risk of Lyme and reduced medicinal plant yield. This clear trade-off requires careful consideration of local priorities regarding these four health outcomes. Canopy cover changes will also affect forest functioning itself, with potential cascading effects. For example, thinning may lead to increased water availability for certain tree species under drought58, but rapid canopy opening may also perturb understory biodiversity35. In contrast, increasing tree diversity represents a relatively safe, no-regret intervention with net advantages1 and important co-benefits for biodiversity conservation4,59. In that sense, this may make it a compelling all-round management strategy in many situations with a win-win for biodiversity and human health, especially if local public health priorities are difficult to identify. Diverse forests are especially appealing given their higher resilience to multiple global change drivers and insect pests60, which may secure provision of health benefits in the longer term59. In sum, despite rather low direct health impacts, increasing forest tree diversity emerges as a robust strategy for promoting human health whereas altering forest density can be used to maximise specific benefits according to local contexts.
Study sites and design
We deployed an international network of ten forest sites along a West-East gradient in Central Europe (Fig. 1a), covering oceanic to sub-continental climates. Of these ten sites, four comprised mature forests, while six comprised young plantations within tree diversity experiments. The mature forest stands reflected natural conditions that varied in tree diversity levels. These included the Białowieża (PL), Hainich (DE) and MIXLOR (FR) sites61, and stands of the TREEWEB network62. All were designed to obtain an even distribution of species per species richness level, and to minimise variation in environmental factors such as soil properties and topography that could covary with species composition. The young plantations63 were aged nine to 17 years at the time of sampling. These were all located within a small perimeter using a uniform planting design that standardised the density and spatial arrangement of different species, where only tree diversity varied with negligible covariation of other environmental factors. These included B-TREE (AT), BIOTREE (DE), three FORBIO sites (BE) and ORPHEE (FR). Both site types were specifically designed to study BEF relationships, enabling the quantification of causal effects of forest ecological characteristics on human health.
Selection of forest stands revolved around tree species diversity as one of the major determinants supporting overall forest biodiversity and functioning18,38. At each site, an equal number of monospecific and mixed forest plots was selected. Mixed plots were always composed of a randomised selection of species present as monospecific stands. All mixed plots contained three target tree species, which was the diversity level that all sites had in common. Non-target species were often present in mature natural stands but were never dominant, having established after plot demarcation61. Our selection covers a total of 164 permanent forest plots and 18 target tree species. Participating researchers were encouraged to cover as many of these plots as possible to generate directly comparable forest-health data for identical plots (“all measurements on all plots” strategy). However, due to COVID-related fieldwork impediments, most researchers failed to cover the full selection (see Supplementary Information for number of plots per pathway).
The same set of forest ecological characteristics was sampled over all plots (Fig. 1b). Central herein was tree diversity, expressed as a Shannon index. Unlike species richness, Shannon Diversity accounts for species’ relative abundances, here based on species-specific basal area shares64. A second characteristic receiving special attention in our analyses is canopy density, proxied by the Leaf Area Index via indirect sampling (using hemispherical photographs). Details on remaining characteristics are found in the Supplementary Information.
Forest-health pathways
This meta-analysis synthesises the published and unpublished works of six separate doctoral studies that gathered empirical data and insights for seven forest-health pathways, covering both mental and physical outcomes that can be considered services or disservices. Pathways are listed and shortly described in Table 1 and described in more detail in the Supplementary Information.
Table 1 | Overview of submodels and their constituting studies. Submodels represent the seven forest-health pathways which have all been covered by multiple studies using interdisciplinary approaches. We briefly discuss the setup of studies of which data populated the final model, and some key findings. Included references refer to readily published studies done within the overarching project. The number of sites refers to the network shown in Fig. 1a and the symbols refer to forest-health pathways as shown in Fig. 1c.
|
Outcome |
Researchers |
Setup |
Key findings |
|
A1 |
KR, RRYO, MRM & ABON |
In situ experiment involving 223 participants subjected to non-forest baseline conditions and different peri-urban forests with contrasting levels of tree species richness32. Complemented with a lab-based EEG experiment. |
Biophysical structure has weak effects, with only tree diversity and understory productivity having limited influence on short-term mental health and wellbeing. Mainly subjectively perceived forest characteristics, and especially perceived biodiversity, fostered positive mental wellbeing effects25. We found little congruence to actual biodiversity. |
|
A2 |
TS, SM, KR, AGCR, RRYO, ABON & MSL |
Monitoring of bird song diversity and abundance using audiologgers at eight sites. Complemented with a lab-based experiment using similar psychological questionnaires to submodel A1. |
Forest characteristics, especially stand density, effectively explained variation in bird song diversity. Yet, further effects were strongly tempered since mental wellbeing is essentially driven by subjectively perceived acoustic diversity, analogous to submodel A1. |
|
B |
LG, PDF, DL, BM & KV |
A thermal comfort index (Physiologically Equivalent Temperature) was measured for 14 months in eight rural sites33 and in an urban setting65. How this was subjectively perceived by people was tested in a complementary experiment26. |
Cooling effects were very strongly explained by forest characteristics, with especially variables related to forest density and height being impactful, followed by multiple composition-related variables. Cooling capacity increased with increasing ambient heat stress. This was effectively experienced as such by participants. Mental wellbeing (A1) was found to positively interact with thermal comfort in forests26. |
|
C1 |
KS & BJ |
Inventory of medicinal plants occurring over eight sites. The biologically active substances, polyphenols, were measured in a selection of species, and extrapolated at forest plot scales. |
Polyphenol content is mainly determined by medicinal plant productivity rather than plant diversity, and canopy density crucially determined productivity by altering the light regime. Soil fertility had little effect. |
|
C2 |
KS & BJ |
Continuous and exhaustive sampling of mushrooms at 20 Białowieża plots over two years39. This was complemented by a single sampling campaign at five other sites. |
Because many macroscopic mushrooms are mycorrhizal and thus associate to specific tree species, tree diversity was a more influential variable than canopy cover. Health benefits are mostly driven by mushroom productivity and only little by mushroom diversity. |
|
D |
MS, DH, BM & DG |
Detailed case-study at B-TREE with quantification of particulate matter deposition for different size classes (<100, <10, < 2.5µm)34. Meta-analysis on tree characteristics driving deposition rates51. Complemented with direct LAI measures at eight sites and PM monitoring at four of them. |
Variation in PM pollution mitigation by forests was mainly influenced by the canopy closure and its roughness, whereas mitigation becomes stronger as ambient pollution levels increase – analogous to submodel B. |
|
E |
ABOU, TV & HJ |
Detailed case-study at ORPHEE where ticks were sampled and screened for various Borrelia strains49. A second sampling campaign covered several other sites. A meta-analysis28 on forest effects on ticks further defined submodel conceptualisation. Complemented with data from the Forester48 project, to quantify links between forest characteristics, mouse- and deer abundance and tick-related variables (DON, NIP and DIN). |
Forest characteristics were hypothesised risks posed by ticks and Lyme via three pathways: microclimate humidity (mediated by canopy density), understory vegetation and the vertebrate host community. Humidity levels and deer abundance strongly predicted tick densities (DON), whereas mouse abundance and understory productivity had only modest effects. |
The choice for Bayesian Belief Networks
A BBN consists of a Directed Acyclic Graph, composed of a set of nodes (i.e. variables) connected via causal links. If node B is the causal agent of node A, node A is the child node and node B is a parent node. Each node is partitioned into a limited set of states that the node can reach. The probability of a node being in a particular state is based on so-called Conditional Probability Tables (CPT) which store probability distributions for all state combinations of parent nodes. BBNs do not generate p-values. Instead, the certainty of a given state (outcome) is related to the narrowness of the probability distribution. See Jensen & Nielsen (2007)66 for a standard technical reference or Cain (2001)56 for an applied introduction.
We chose BBN as an analytical method as they provide a very flexible approach to integrate data from heterogeneous sources including expert knowledge, quantitative literature findings and empirical research31. This was essential given that empirical data from its constituting studies – despite interdisciplinary methodologies – sometimes failed to cover the whole mechanistic pathway from forest until health outcome. Other advantages include the capacity to handle small datasets and missing data, the highly visual and interactive model building process, the fast processing times, and the very intuitive way of dealing with uncertainty31,67. BBNs also have disadvantages: feedback loops and cyclic processes cannot be handled, complex models with long interaction chains lead to ‘dilution’ of effect sizes, random effects to account for e.g. spatial autocorrelation cannot be directly implemented, and generating child node CPTs becomes exponentially more cumbersome with each additional parent node31,67. Another limitation is the absent direct implementation of hierarchical data structures (e.g. random factors). Most importantly, continuous nodes need to be discretised into a limited number of states, leading to information loss67.
Model building steps
BBNs were built and adapted as individual research projects advanced. Model building was partitioned in three phases centred around three annual meetings (2021–2023) during which protocols were presented and discussed prior to application. Protocols were adapted from existing guidelines56,68. Phases overlapped in time due to continuously updating insights as fieldwork and data analysis progressed in parallel (Supplementary Information p.5). BBNs were developed in Netica (v7.11)69.
Phase I
The main goal of phase I was to create a Directed Acyclic Graph for each pathway/submodel. The framework we imposed was the Ecosystem Services Cascade Model by Haines-Young & Potschin19, previously adapted for BBNs31 and here applied to health outcomes. This forced partners to disentangle their forest-health pathway into a mechanistic conceptual model with four stages. The first stage groups nodes defining the biophysical structure of the ecosystem (e.g. tree diversity, canopy height). This defines ecosystem functioning (e.g. heat buffering, particulate matter retention), which influences (dis-)service provisioning (e.g. medicinal plant productivity, attention restoration), which will ultimately influence the variables defining the health outcome (e.g. Lyme disease risk, positive affect). Partners avoided feedback loops and more than three parent nodes for the same child node whenever possible68. They prioritized nodes being empirically sampled and respected relationship causality: “Draw arrows from node A to node B if changes in A will lead to changes in B, but not the opposite”. We acknowledge the more recent Nature’s Contributions to People framework70, but use ecosystem services to comply with the Cascade model.
Phase II
Phase II aimed at identifying interactions between submodels and to define node’s state thresholds. Forest characteristics that were shared in multiple submodels were jointly defined and constitute the overarching model structure shown in Fig. 1c. This represents current knowledge in forest ecological interactions. Next, we explored interactions between submodels to increase model cohesion and the potential to unveil trade-offs and synergies. At last, nodes were partitioning according to specific state thresholds. The following guidelines were provided for continuous variables: i) define the node’s unit and potential range, ii) define the minimum number of possible states that still makes logical sense (maximum five)68, iii) define threshold values based on published guidelines (e.g., WHO air quality guidelines), recognized classifications (e.g., heat stress categories) or logical reasoning (e.g., increasing vs. decreasing anxiety)71, and iv) detailed reasoning, define state names whenever possible, and list key references. When obvious threshold values were lacking, data percentiles were used. When a lot of empirical data was available and a child node had few parents, the state number limitation could be relaxed because the minimum data amount is alleviated.
Phase III
The last phase quantified and joined submodels. The four major classes that can be combined in BBNs are raw empirical data, quantitative literature results, logical equations or expert opinion31,56. Each has advantages and disadvantages. Empirical raw data are very easy to incorporate, especially when available for multiple nodes simultaneously, but the underlying counting algorithm requires sufficient data points for robust probability estimation. Quantitative literature findings can include formulae with regression coefficients, or estimates with error distributions. When available, priority was given to robust meta-analyses. They were most often used to quantify health outcomes, given that multiple of our studies only gathered empirical data until the service stage. Their advantage is the relatively easy implementation in Netica, which integrates numerous error distribution functions, but finding published results with the same predictor set becomes more challenging as the number of parent nodes increases. Logical equations usually link parent nodes via basic mathematical relationships. For example, the density of infected ticks (DIN) equals the density of questing ticks (DON) times the nymphal infection prevalence (NIP). They do not cause an effect dilution and are easy to incorporate, with few disadvantages. At last, expert opinion data are used when quantitative data are unavailable or not robust enough, which also represents their main advantage. We used a direct expert elicitation style based on recommendations by Kuhnert and colleagues72, assuming that our expert teams were made sufficiently familiar with probability theory through annual meetings and individual discussions. Here, experts had to estimate probability distributions for each combination of parent node states, thereby directly generating CPTs. The disadvantage is that this becomes a very cognitively-demanding endeavour as the number of parent nodes increases, which exponentially augments possible state combinations.
To support partners in identifying data classes, a custom-made decision tree was presented (Supplementary Information p. 8). The decision tree aims to circumvent aforementioned weaknesses or provides alternatives by combining classes. For example, using only raw empirical data is discouraged whenever there are less than ten observations per combination of parent states, which is a relaxed version of the untested73 recommendation by Cain (2001)56. Concretely, when one parent has three states and the second has four, the empirical dataset should have at least 120 observations for the child node. When this is not reached but empirical data are available, users are redirected to options where literature results or expert opinion data are used as priors72 before ‘belief updating’ is done with raw data as posteriors. The decision tree also implies a priority ranking in classes: whenever possible, own empirical data was prioritised over literature results, which was prioritised over expert opinion data.
The common forest structure was exclusively quantified using empirical data. However, since all submodels relied on it and the canopy density node had four parents, additional data were retrieved. Its eight constituting nodes (plus ‘understory productivity’ and ‘soil fertility’ nodes) were reinforced with data from international forest networks that were designed to study forest biodiversity effects on ecosystem functioning, analogous to our network. These include 225 FORMICA47 plots and the 41 TREEWEB62 plots that were not yet included in our network, for a total of 430 plots.
As a final submodel-specific procedure, a formal validation meeting was organised after submodels were operationalized in Netica. Results of sensitivity analyses and general submodel behaviour were scrutinised for inconsistencies by our expert partners67, allowing for final adaptations before merging them with other submodels. At last, all submodels were joined to the common structure and interactions were added. Some interactions were indicated as merely ‘implicit’ associations when two nodes were not directly comparable or already had the same parent node combinations.
General model insights and shared forest nodes
With 82 nodes (i.e. variables), 123 links and 3154 conditional probabilities, the final model is among the largest published BBNs in ecosystem service modelling31. Most nodes (n = 56; 68%) were exclusively quantified using our own empirical data. Ten were operationalized via logical equations and nine via quantitative literature findings. Expert opinion elicitation was used for eight nodes, but always as priors with posterior belief updating using empirical data when the minimal number of observations was lacking.
The model building exercise yielded an extensive Supplementary Information document that is appended for maximal transparency and comprehensibility of the numerous choices underlying submodels. It contains information on general methodology, submodel-specific summaries and limitations, and node-specific details including variable definitions, state thresholds and key references. As only part of the full complexity of our final BBN can be exhibited here, this reference document should accompany the model’s usage and interpretation.
Synthesis analyses
Once the complete BBN was assembled, it was subjected to three main analyses. First, the explanatory power of the common forest characteristics was quantified using built-in sensitivity analyses, represented by the variance reduction of the expected real value of the child node due to a finding in the parent69. Second, we extracted the mutual information between health outcome nodes to identify trade-offs and synergies. This is quantified as the reduction in entropy in one node, measured in computational bits, due to a finding in the other node69. Because health outcome nodes are separated by long interaction chains that cause effect dilutions56, this yielded low values. Threshold values were established to distinguish robust associations from the bulk of negligible ones. If remaining associations were visible between nodes of different submodels in each stage until health outcomes, they were considered the strongest and most consistent associations. Those only emerging in function or service stages were considered secondary associations. Third and last, the magnitude of health outcomes was expressed relative to non-forest conditions and among forests. To illustrate potential inter-forest differences, four scenarios were considered based on two of the most influential and independent forest characteristics: monospecific or mixed stands with open (LAI < 1) or closed (LAI > 3) canopies. Per scenario, 10 000 cases were simulated in Netica to obtain means and uncertainty distributions. In about half of the forest-health pathways, health outcomes were explicitly expressed relative to control conditions, which could be urban environments (mental wellbeing – visual) or ambient, non-forest conditions (thermal comfort and air quality). In the other cases, suitable control conditions were generally less evident. For mental mental wellbeing via auditory stimuli, the baseline were measures taken prior to the intervention, which was already accounted for by looking at mental state changes (Δ). For medicinal plants and mushrooms, baselines can be assumed to represent no intake. For Lyme disease, baselines would be environments with risk absence such as highly urbanised locations.
Acknowledgements
This research was funded by the ERA-Net BiodivERsA project Dr.FOREST, with the national funders German Research Foundation (DFG − 428795724, Germany), French National Research Agency (ANR, France), Research Foundation – Flanders (FWO, Belgium), Austrian Science Fund (FWF, Austria) and National Science Center (NCN, Poland, project no. 2019/31/Z/NZ8/04032), as part of the 2018–2019 BiodivERsA call for research proposals. PDF received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (ERC Starting Grant FORMICA 757833). DL was supported by a postdoctoral fellowship of the FWO. TV was supported by the UGent GOA project (BOF20/GOA/009). QP would like to thank the Walloon forest service (SPW – DNF) for its support to the maintenance of the FORBIO-Gedinne experiment in the frame of the 5-yr research programme ‘Plan quinquennal de recherche et de vulgarisation forestières’.
- Beute F et al (2023) How do different types and characteristics of green space impact mental health? A scoping review. People Nat 5:1839–1876
- Hartig T, Mitchell R, de Vries S, Frumkin H (2014) Nature and Health. Annu Rev Public Health 35:207–228
- Konijnendijk C, Devkota D, Mansourian S, Wildburger C (2023) Forests and Trees for Human Health: Pathways, Impacts, Challenges and Response Options. A Global Assessment Report. 232
- WHO. Nature, Biodiversity and health: an overview of interconnections. 42 (2021)
- Reyes-Riveros R et al (2021) Linking public urban green spaces and human well-being: A systematic review. Urban Forestry Urban Green 61:127105
- WHO (2022) Noncommunicable diseases progress monitor. World Health Organization
- Patel V et al (2018) The Lancet Commission on global mental health and sustainable development. Lancet 392:1553–1598
- Chi D et al (2022) Residential Exposure to Urban Trees and Medication Sales for Mood Disorders and Cardiovascular Disease in Brussels, Belgium: An Ecological Study. Environ Health Perspect 130:057003
- Iungman T et al (2023) Cooling cities through urban green infrastructure: a health impact assessment of European cities. Lancet 0
- Barboza EP et al (2021) Green space and mortality in European cities: a health impact assessment study. Lancet Planet Health 5:e718–e730
- van den Berg M et al (2015) Health benefits of green spaces in the living environment: A systematic review of epidemiological studies. Urban Forestry Urban Green 14:806–816
- Nguyen P-Y, Astell-Burt T, Rahimi-Ardabili H, Feng X (2021) Green Space Quality and Health: A Systematic Review. Int J Environ Res Public Health 18:11028
- De Frenne P et al (2019) Global buffering of temperatures under forest canopies. Nat Ecol Evol 3:744–749
- Diener A, Mudu P (2021) How can vegetation protect us from air pollution? A critical review on green spaces’ mitigation abilities for air-borne particles from a public health perspective - with implications for urban planning. Sci Total Environ 796:148605
- Ekkel ED, de Vries S (2017) Nearby green space and human health: Evaluating accessibility metrics. Landsc Urban Plann 157:214–220
- Dallimer M et al (2012) Biodiversity and the Feel-Good Factor: Understanding Associations between Self-Reported Human Well-being and Species Richness. Bioscience 62:47–55
- Aerts R, Honnay O, Van Nieuwenhuyse A (2018) Biodiversity and human health: mechanisms and evidence of the positive health effects of diversity in nature and green spaces. Br Med Bull 127:5–22
- Brockerhoff EG et al (2017) Forest biodiversity, ecosystem functioning and the provision of ecosystem services. Biodivers Conserv 26:3005–3035
- Haines-Young R, Potschin M (2010) The links between biodiversity, ecosystem services and human well-being. in Ecosystem Ecology (eds. Raffaelli, D. G. & Frid, C. L. J.) 110–139Cambridge University Press, 10.1017/CBO9780511750458.007
- Tilman D, Isbell F, Cowles JM (2014) Biodiversity and Ecosystem Functioning. Annu Rev Ecol Evol Syst 45:471–493
- Marselle MR et al (2021) Pathways linking biodiversity to human health: A conceptual framework. Environ Int 150:106420
- Marselle MR (2020) Urban street tree biodiversity and antidepressant prescriptions. Sci Rep 11
- Krayenhoff ES et al (2021) Cooling hot cities: a systematic and critical review of the numerical modelling literature. Environ Res Lett 16:053007
- Han D, Shen H, Duan W, Chen L (2020) A review on particulate matter removal capacity by urban forests at different scales. Urban Forestry Urban Green 48:126565
- Rozario K et al (2024) Perceived biodiversity: is what we measure also what we see and hear? 2024.04.03.587905 Preprint. at https://doi.org/10.1101/2024.04.03.587905
- Gillerot L et al (2024) Forests are chill: The interplay between thermal comfort and mental wellbeing. Landsc Urban Plann 242:104933
- Rey Gozalo G, Barrigón Morillas JM, Montes González D (2018) Atanasio Moraga, P. Relationships among satisfaction, noise perception, and use of urban green spaces. Sci Total Environ 624:438–450
- Bourdin A et al (2023) Forests harbor more ticks than other habitats: A meta-analysis. For Ecol Manag 541:121081
- Dudek T, Kasprzyk I, Dulska-Jeż A (2018) Forest as a place for recreation but also the source of allergenic plant pollen: to come or avoid? Eur J For Res 137:849–862
- Park B-J et al (2011) Relationship between psychological responses and physical environments in forest settings. Landsc Urban Plann 102:24–32
- Landuyt D et al (2013) A review of Bayesian belief networks in ecosystem service modelling. Environ Model Softw 46:1–11
- Rozario K et al (2024) The more the merrier? Perceived forest biodiversity promotes short-term mental health and well-being—A multicentre study. People Nat 6:180–201
- Gillerot L et al (2022) Forest structure and composition alleviate human thermal stress. Glob Change Biol 28:7340–7352
- Steinparzer M, Haluza D, Godbold DL (2022) Integrating Tree Species Identity and Diversity in Particulate Matter Adsorption. Forests 13:481
- Zellweger F et al (2020) Forest microclimate dynamics drive plant responses to warming. Science 368:772–775
- Wong NH, Tan CL, Kolokotsa DD, Takebayashi H (2021) Greenery as a mitigation and adaptation strategy to urban heat. Nat Rev Earth Environ 2:166–181
- Verheyen K et al (2024) Forest canopies as nature-based solutions to mitigate global change effects on people and nature. J Ecol
- Ampoorter E et al (2020) Tree diversity is key for promoting the diversity and abundance of forest-associated taxa in Europe. Oikos 129:133–146
- Stojek K, Gillerot L, Jaroszewicz B (2022) Predictors of mushroom production in the European temperate mixed deciduous forest. For Ecol Manag 522:120451
- Gori A, Nascimento LB, Ferrini F, Centritto M, Brunetti C (2020) Seasonal and Diurnal Variation in Leaf Phenolics of Three Medicinal Mediterranean Wild Species: What Is the Best Harvesting Moment to Obtain the Richest and the Most Antioxidant Extracts? Molecules 25:956
- Grinfeder E et al (2022) Soundscape dynamics of a cold protected forest: dominance of aircraft noise. Landsc Ecol 37:567–582
- Schulp CJE, Thuiller W, Verburg PH (2014) Wild food in Europe: A synthesis of knowledge and data of terrestrial wild food as an ecosystem service. Ecol Econ 105:292–305
- Jansson M, Fors H, Lindgren T, Wiström B (2013) Perceived personal safety in relation to urban woodland vegetation – A review. Urban Forestry Urban Green 12:127–133
- Lai D et al (2020) A comprehensive review of thermal comfort studies in urban open spaces. Sci Total Environ 742:140092
- Blondeel H et al (2020) Light and warming drive forest understorey community development in different environments. Glob Change Biol 26:1681–1696
- Gray JS (1998) Review The ecology of ticks transmitting Lyme borreliosis. Exp Appl Acarol 22:249–258
- Vanneste T et al (2024) Trade-offs in biodiversity and ecosystem services between edges and interiors in European forests. Nat Ecol Evol 1–8. 10.1038/s41559-024-02335-6
- Vanroy T et al (2024) The effect of forest structural complexity on tick-borne pathogens in questing ticks and small mammals. For Ecol Manag 562:121944
- Bourdin A et al (2022) Forest Diversity Reduces the Prevalence of Pathogens Transmitted by the Tick Ixodes ricinus. Front Ecol Evol. 10.3389/fevo.2022.891908
- Ostfeld RS, Keesing F (2012) Effects of host diversity on infectious disease. Annu Rev Ecol Evol Syst 43:157–182
- Steinparzer M, Schaubmayr J, Godbold DL, Rewald B (2023) Particulate matter accumulation by tree foliage is driven by leaf habit types, urbanization- and pollution levels. Environ Pollut 335:122289
- Cao Z, Zhou J, Li M, Huang J, Dou D (2023) Urbanites’ mental health undermined by air pollution. Nat Sustain 6:470–478
- OECD (2022) Health at a Glance: Europe 2022: State of Health in the EU Cycle. Organisation for Economic Co-operation and Development
- Ballester J et al (2023) Heat-related mortality in Europe during the summer of 2022. Nat Med 29:1857–1866
- Freer-Smith P et al (2019) Plantation forests in Europe: challenges and opportunities. 10.36333/fs09
- Planning improvements in natural resources management: Guidelines for using Bayesian networks to support the planning and management of development programmes in the water sector and beyond. (2001)
- Bratman GN et al (2019) Nature and mental health: An ecosystem service perspective. Sci Adv 5:eaax0903
- Sohn JA, Saha S, Bauhus J (2016) Potential of forest thinning to mitigate drought stress: A meta-analysis. For Ecol Manag 380:261–273
- Kilpatrick AM, Salkeld DJ, Titcomb G, Hahn MB (2017) Conservation of biodiversity as a strategy for improving human health and well-being. Phil Trans R Soc B 372:20160131
- Jactel H, Moreira X, Castagneyrol B (2021) Tree Diversity and Forest Resistance to Insect Pests: Patterns, Mechanisms and Prospects. Annual Review of Entomology 66, null
- Baeten L et al (2013) A novel comparative research platform designed to determine the functional significance of tree species diversity in European forests. Perspect Plant Ecol Evol Syst 15:281–291
- De Groote S et al (2017) Tree species identity outweighs the effects of tree species diversity and forest fragmentation on understorey diversity and composition. Plant Ecol Evol 150:229–239
- Paquette A et al (2018) A million and more trees for science. Nat Ecol Evol 2:763–766
- Nickmans H (2019) The nutrition of oak and beech trees along a tree diversity gradient
- Gillerot L, Landuyt D, De Frenne P, Muys B, Verheyen K (2024) Urban tree canopies drive human heat stress mitigation. Urban Forestry Urban Green 92:128192
- Jensen FV, Nielsen TD (2007) Bayesian networks and decision graphs. Springer
- Aguilera PA, Fernández A, Fernández R, Rumí R, Salmerón A (2011) Bayesian networks in environmental modelling. Environ Model Softw 26:1376–1388
- Marcot BG, Steventon JD, Sutherland GD, McCann RK (2006) Guidelines for developing and updating Bayesian belief networks applied to ecological modeling and conservation. Can J Res 36:3063–3074
- Norsys Software Corporation. Netica Application User’s Guide (1998)
- Pascual U et al (2017) Valuing nature’s contributions to people: the IPBES approach. Curr Opin Environ Sustain 26–27:7–16
- Uusitalo L (2007) Advantages and challenges of Bayesian networks in environmental modelling. Ecol Model 203:312–318
- Kuhnert PM, Martin TG, Griffiths SP (2010) A guide to eliciting and using expert knowledge in Bayesian ecological models. Ecol Lett 13:900–914
- Chen SH, Pollino CA (2012) Good practice in Bayesian network modelling. Environ Model Softw 37:134–145
There is NO Competing Interest.