Gabriel Dansereau

Optimal monitoring strategies should be designed to efficiently monitor all essential facets of biodiversity. Yet, species interactions are often overlooked in monitoring designs compared to spatial coverage and species richness, partly due to the inherent difficulty of sampling and monitoring interactions compared to species distributions. Here, we used simulations to test the efficiency of monitoring species interactions within Biodiversity Observations Networks (BONs). We tested several methods for designing and optimizing BONs to monitor species interactions and examined efficiency when increasing the number of monitored sites. We showed that the required sampling effort is considerably higher for interactions, especially when considering that interaction realization and detection realistically depend on species abundance. Thus, expectations to monitor a relevant proportion of interactions within a BON should be lower than expectations to monitor its species. However, from a single‐species perspective, optimizing monitoring designs based on a known species range proved an efficient strategy to retrieve its interactions, outperforming designs based on total species or interaction richness. Further efficiency gains required detailed knowledge of where realized interactions occurred, highlighting the need for better knowledge of where interactions are likely to take place and integration of factors influencing interaction realization, such as species abundances. Our results were consistent even when the total species range was over‐ or underestimated by 10%, which shows a tolerance in the precision of the information required for optimization. These findings highlight how attainable levels of information can lead to efficiency improvements when designing BONs. With the right target and optimization strategy, and readily available information, such as species ranges, we can therefore guide observation network designs to monitor species interactions as central biodiversity components.

Meeting the goals of the Global Biodiversity Framework (GBF) requires a substantial increase in biodiversity monitoring effort, motivating the development of a Global Biodiversity Observing System (GBiOS). However, GBiOS cannot be built without a practical framework for designing effective biodiversity monitoring networks that balances budgetary constraints with competing policy objectives, which we currently lack. Here we provide the foundations for a theory of optimal biodiversity monitoring design based on: (1) a hierarchical, multi-layer architecture for monitoring networks that matches the multi-scale nature of ecological processes and environmental policymaking, (2) a simulation-based framework for evaluating the efficacy of monitoring network design against the variety of plausible “ecological worlds”, and (3) an integrative toolbox for monitoring design, drawing on methods from spatial sampling theory, environmental monitoring design, and optimal decision theory. Together, these form a basis for building the most effective GBiOS possible to best inform policymaking and meet the goals and targets of the GBF.

Land use and climate change are increasing the risk of spillover of zoonotic disease into human populations. However, we lack actionable information about the prevalence of pathogens in wildlife populations for most of the globe, challenging our ability to implement strategies to prevent zoonoses. Even when this data exists, it has historically been sampled opportunistically and without guidance based on known geographic distributions of hosts of zoonotic pathogens. Biosurveillance is essential to mitigating zoonotic spillover risk, but given the expensive nature of monitoring pathogens in wildlife, we need to be strategic about deciding where and what to sample to obtain as much useful information as possible — particularly to identify regions where spillover is most probable and implement preventative measures to mitigate outbreak risk. The field of biodiversity monitoring has established many practices that can directly inform optimal biosurveillance efforts. One such concept is the Biodiversity Observation Network (or BON), which aims to select monitoring locations that most effectively and efficiently capture the status and trends of biodiversity. We present a protocol for integrating data on host biodiversity into sampling priority for wildlife disease surveillance based on host species distribution models, with optional potential to integrate pathogen prevalence data (if available). This protocol has the flexibility to target different forms of sampling (collecting host occurrence vs. pathogen prevalence data) to adapt to different levels of data availability, but still makes adaptive sampling recommendations based on a principled understanding of host distribution and pathogen biology. We illustrate this flexibility with two case studies, prioritizing sampling for Hanta- and Arenaviridae in rodents and shrews in India and South Korea, respectively representing data poor and data rich contexts. We view this framework as a basis for integrating long-term biosurveillance and biodiversity monitoring programs, and maximizing the useful information available for public health decision making.

Optimal monitoring strategies should be designed to efficiently monitor all essential facets of biodiversity. Yet, species interactions are often overlooked in monitoring designs compared to spatial coverage and species richness, partly due to the inherent difficulty of sampling and monitoring interactions compared to species distributions. Here, we used simulations to test the efficiency of monitoring species interactions within Biodiversity Observations Networks (BONs). We tested several methods for designing and optimizing BONs to monitor species interactions and examined efficiency when increasing the number of monitored sites. We showed that the required sampling effort is considerably higher for interactions, especially when considering that interaction realization and detection realistically depend on species abundance. Thus, expectations to monitor a relevant proportion of interactions within a BON should be lower than expectations to monitor its species. However, from a single-species perspective, optimizing monitoring designs based on a known species range proved an efficient strategy to retrieve its interactions, outperforming designs based on total species or interaction richness. Further efficiency gains required detailed knowledge of where realized interactions occurred, highlighting the need for better knowledge of where interactions are likely to take place and integration of factors influencing interaction realization, such as species abundances. Our results were consistent even when the total species range was over- or underestimated by 10%, which shows a tolerance in the precision of the information required for optimization. Therefore, our results highlight how attainable levels of information can lead to efficiency improvements when designing BONs. With the right target and optimization strategy, available information, such as species ranges, can guide observation network designs to monitor species interactions as central biodiversity components.

Decision-makers and managers urgently need strategies to halt biodiversity loss and maintain ecosystem resilience, while meeting socio-economic needs. Although forests provide essential socio-economic services and host important levels of biodiversity, biodiversity conservation practices in forestry remain largely focused on single species or habitat preservation. This likely overlooks the complexity of species interactions that sustain critical forest functions, some of which can feed back into socio-economic values -- e.g., pest-regulation, or lack thereof, can influence timber provisioning. This limits the ability to anticipate cascading effects of species loss, climate change, and altered disturbance regimes.Species interaction networks offer a powerful tool to address these gaps. Despite their potential and the fact that trophic network data are increasingly available, explicit integration of network-derived indicators into forest management is virtually nonexistent. This is in great part due to challenges of uncertainty, interpretability, and accessibility for practitioners. Yet, we argue that existing data sources—such as regional metawebs, species distribution models, and long-term monitoring programs—combined with established network indicators can provide a sufficient foundation to utilise network information to guide forest management now, at least at broad spatial scales.In this presentation, we will focus on how we could close this gap. We will outline criteria that indicators derived from interaction networks must meet to be relevant for decision-makers and managers, and will show how an existing network indicator, trophic network robustness, could be used presently to measure a forest system’s capacity to withstand cascading extinctions, using existing data. We will also discuss how existing data and methods can be tied with scenarios of climate and habitat change (the latter influenced by forestry practices) to forecast impacts on forest trophic networks and weigh these against changes in other forest properties, at large spatial scales. Although these solutions are not yet operationalizable at forest operation scales, they can help shift forest management and broad-scale decision-making from species-centric approaches and enhance our ability to meet forest biodiversity conservation goals.

Land use and climate change are increasing the risk of spillover of zoonotic disease into human populations. However, we lack actionable information about the prevalence of pathogens in wildlife populations for most of the globe, challenging our ability to implement strategies to prevent zoonoses. Even when this data exists, it has historically been sampled opportunistically and without guidance based on known geographic distributions of hosts of zoonotic pathogens. Biosurveillance is essential to mitigating zoonotic spillover risk, but given the expensive nature of monitoring pathogens in wildlife, we need to be strategic about deciding where and what to sample to obtain as much useful information as possible. The field of biodiversity monitoring has established many practices that can directly inform optimal biosurveillance efforts. One such concept is the Biodiversity Observation Network (or BON), which aims to select monitoring locations that most effectively and efficiently capture the status and trends of biodiversity. We present a protocol for integrating data on host biodiversity into sampling priority for wildlife disease surveillance based on host species distribution models, with optional potential to integrate pathogen prevalence data (if available). This protocol has the flexibility to target different forms of sampling (collecting host occurrence vs pathogen prevalence data) to adapt to different levels of data availability, but still makes adaptive sampling recommendations based on a principled understanding of host distribution and pathogen biology.We illustrate this flexibility with two case studies, prioritizing sampling for Hanta- and Arena-viridae in rodents in India and South Korea, respectively representing data poor and data rich contexts. We view this framework as a basis for integrating long-term biosurveillance and biodiversity monitoring programs, and maximizing the useful information available for public health decision making.

(1) Species distribution modeling requires to handle varied types of data, and benefits from an integrated approach to programming. (2) We introduce SpeciesDistributionToolkit , a Julia package aiming to facilitate the production of species distribution models. It covers various steps of the data collection and analysis process, extending to the development of interfaces for integration of additional functionalities. (3) By relying on semantic versioning and strong design choices on modularity, we expect that this package will lead to improved reproducibility and long-term maintainability. (4) We illustrate the functionalities of the package through several case studies, accompanied by reproducible code.

Representing species interactions probabilistically as opposed to deterministically conveys uncertainties in our knowledge of interactions. The sources of uncertainty captured by interaction probabilities depend on the method used to evaluate them: uncertainty of predictive models, subjective assessment of experts, or empirical measurement of interaction spatiotemporal variability. However, guidelines for the estimation and documentation of probabilistic interaction data are lacking. This is concerning because our understanding of interaction probabilities depend on their sometimes elusive definition and uncertainty sources. We review how probabilistic interactions are defined at different spatial scales. These definitions are based on the distinction between the realisation of an interaction at a specific time and space (local networks) and its biological or ecological feasibility (metaweb). Using host–parasite interactions in Europe, we illustrate how these two network representations differ in their statistical properties, specifically: how local networks and metawebs differ in their spatial and temporal scaling of interactions. We present two approaches to inferring binary interactions from probabilistic ones that account for these differences and show that systematic biases arise when directly inferring local networks from metawebs. Our results underscore the importance of more rigorous descriptions of probabilistic species interactions that specify their conditional variables and uncertainty sources.

(1) Species distribution modeling requires to handle varied types of data, and benefits from an integrated approach to programming. (2) We introduce SpeciesDistributionToolkit, a Julia package aiming to facilitate the production of species distribution models. It covers various steps of the data collection and analysis process, extending to the development of interfaces for integration of additional functionalities. (3) By relying on semantic versioning and strong design choices on modularity, we expect that this package will lead to improved reproducibility and long-term maintainability. (4) We illustrate the functionalities of the package through several case studies, accompanied by reproducible code.