The AusTraits plant dictionary.

Traits with intuitive names, a clear scope and explicit description are essential for all trait databases. The lack of unified, comprehensive, and machine-readable plant trait definitions limits the utility of trait databases, including reanalysis of data from a single database, or analyses that integrate data across multiple databases. Both can only occur if researchers are confident the trait concepts are consistent within and across sources. Here we describe the AusTraits Plant Dictionary (APD), a new data source of terms that extends the trait definitions included in a recent trait database, AusTraits. The development process of the APD included three steps: review and formalisation of the scope of each trait and the accompanying trait description; addition of trait metadata; and publication in both human and machine-readable forms. Trait definitions include keywords, references, and links to related trait concepts in other databases, enabling integration of AusTraits with other sources. The APD will both improve the usability of AusTraits and foster the integration of trait data across global and regional plant trait databases.

Biodiversity monitoring for a just planetary future.

Data that influence policy and major investment decisions risk entrenching social and political inequities.

Pretty Darn Good Control: When are Approximate Solutions Better than Approximate Models.

Existing methods for optimal control struggle to deal with the complexity commonly encountered in real-world systems, including dimensionality, process error, model bias and data heterogeneity. Instead of tackling these system complexities directly, researchers have typically sought to simplify models to fit optimal control methods. But when is the optimal solution to an approximate, stylized model better than an approximate solution to a more accurate model? While this question has largely gone unanswered owing to the difficulty of finding even approximate solutions for complex models, recent algorithmic and computational advances in deep reinforcement learning (DRL) might finally allow us to address these questions. DRL methods have to date been applied primarily in the context of games or robotic mechanics, which operate under precisely known rules. Here, we demonstrate the ability for DRL algorithms using deep neural networks to successfully approximate solutions (the "policy function" or control rule) in a non-linear three-variable model for a fishery without knowing or ever attempting to infer a model for the process itself. We find that the reinforcement learning agent discovers a policy that outperforms both constant escapement and constant mortality policies-the standard family of policies considered in fishery management. This DRL policy has the shape of a constant escapement policy whose escapement values depend on the stock sizes of other species in the model.

Bridging adaptive management and reinforcement learning for more robust decisions.

From out-competing grandmasters in chess to informing high-stakes healthcare decisions, emerging methods from artificial intelligence are increasingly capable of making complex and strategic decisions in diverse, high-dimensional and uncertain situations. But can these methods help us devise robust strategies for managing environmental systems under great uncertainty? Here we explore how reinforcement learning (RL), a subfield of artificial intelligence, approaches decision problems through a lens similar to adaptive environmental management: learning through experience to gradually improve decisions with updated knowledge. We review where RL holds promise for improving evidence-informed adaptive management decisions even when classical optimization methods are intractable and discuss technical and social issues that arise when applying RL to adaptive management problems in the environmental domain. Our synthesis suggests that environmental management and computer science can learn from one another about the practices, promises and perils of experience-based decision-making. This article is part of the theme issue 'Detecting and attributing the causes of biodiversity change: needs, gaps and solutions'.

The forecast trap.

Encouraged by decision makers' appetite for future information on topics ranging from elections to pandemics, and enabled by the explosion of data and computational methods, model-based forecasts have garnered increasing influence on a breadth of decisions in modern society. Using several classic examples from fisheries management, I demonstrate that selecting the model or models that produce the most accurate and precise forecast (measured by statistical scores) can sometimes lead to worse outcomes (measured by real-world objectives). This can create a forecast trap, in which the outcomes such as fish biomass or economic yield decline while the manager becomes increasingly convinced that these actions are consistent with the best models and data available. The forecast trap is not unique to this example, but a fundamental consequence of non-uniqueness of models. Existing practices promoting a broader set of models are the best way to avoid the trap.

Grazer behaviour can regulate large-scale patterning of community states.

Ecosystem patterning can arise from environmental heterogeneity, biological feedbacks that produce multiple persistent ecological states, or their interaction. One source of feedbacks is density-dependent changes in behaviour that regulate species interactions. By fitting state-space models to large-scale (~500 km) surveys on temperate rocky reefs, we find that behavioural feedbacks best explain why kelp and urchin barrens form either reef-wide patches or local mosaics. Best-supported models in California include feedbacks where starvation intensifies grazing across entire reefs create reef-scale, alternatively stable kelp- and urchin-dominated states (32% of reefs). Best-fitting models in New Zealand include the feedback of urchins avoiding dense kelp stands that can increase abrasion and predation risk, which drives a transition from shallower urchin-dominated to deeper kelp-dominated zones, with patchiness at 3-8 m depths with intermediate wave stress. Connecting locally studied processes with region-wide data, we highlight how behaviour can explain community patterning and why some systems exhibit community-wide alternative stable states.

Bifurcation or state tipping: assessing transition type in a model trophic cascade.

Ecosystems can experience sudden regime shifts due to a variety of mechanisms. Two of the ways a system can cross a tipping point include when a perturbation to the system state is large enough to push the system beyond the basin of attraction of one stable state and into that of another (state tipping), and alternately, when slow changes to some underlying parameter lead to a fold bifurcation that annihilates one of the stable states. The first mechanism does not generate the phenomenon of critical slowing down (CSD), whereas the latter does generate CSD, which has been postulated as a way to detect early warning signs ahead of a sudden shift. Yet distinguishing between the two mechanisms (s-tipping and b-tipping) is not always as straightforward as it might seem. The distinction between "state" and "parameter" that may seem self-evident in mathematical equations depends fundamentally on ecological details in model formulation. This distinction is particularly relevant when considering high-dimensional models involving trophic webs of interacting species, which can only be reduced to a one-dimensional model of a tipping point under appropriate consideration of both the mathematics and biology involved. Here we illustrate that process of dimension reduction and distinguishing between s- and b-tipping for a highly influential trophic cascade model used to demonstrate tipping points and test CSD predictions in silico, and later, in a natural lake ecosystem. Our analysis resolves a previously unclear issue as to the nature of the tipping point involved.

Revealing biases in the sampling of ecological interaction networks.

The structure of ecological interactions is commonly understood through analyses of interaction networks. However, these analyses may be sensitive to sampling biases with respect to both the interactors (the nodes of the network) and interactions (the links between nodes), because the detectability of species and their interactions is highly heterogeneous. These ecological and statistical issues directly affect ecologists' abilities to accurately construct ecological networks. However, statistical biases introduced by sampling are difficult to quantify in the absence of full knowledge of the underlying ecological network's structure. To explore properties of large-scale ecological networks, we developed the software EcoNetGen, which constructs and samples networks with predetermined topologies. These networks may represent a wide variety of communities that vary in size and types of ecological interactions. We sampled these networks with different mathematical sampling designs that correspond to methods used in field observations. The observed networks generated by each sampling process were then analyzed with respect to the number of components, size of components and other network metrics. We show that the sampling effort needed to estimate underlying network properties depends strongly both on the sampling design and on the underlying network topology. In particular, networks with random or scale-free modules require more complete sampling to reveal their structure, compared to networks whose modules are nested or bipartite. Overall, modules with nested structure were the easiest to detect, regardless of the sampling design used. Sampling a network starting with any species that had a high degree (e.g., abundant generalist species) was consistently found to be the most accurate strategy to estimate network structure. Because high-degree species tend to be generalists, abundant in natural communities relative to specialists, and connected to each other, sampling by degree may therefore be common but unintentional in empirical sampling of networks. Conversely, sampling according to module (representing different interaction types or taxa) results in a rather complete view of certain modules, but fails to provide a complete picture of the underlying network. To reduce biases introduced by sampling methods, we recommend that these findings be incorporated into field design considerations for projects aiming to characterize large species interaction networks.

Rebuilding global fisheries under uncertainty.

Current and future prospects for successfully rebuilding global fisheries remain debated due to uncertain stock status, variable management success, and disruptive environmental change. While scientists routinely account for some of this uncertainty in population models, the mechanisms by which this translates into decision-making and policy are problematic and can lead to unintentional overexploitation. Here, we explicitly track the role of measurement uncertainty and environmental variation in the decision-making process for setting catch quotas. Analyzing 109 well-sampled stocks from all oceans, we show that current practices may attain 55% recovery on average, while richer decision methods borrowed from robotics yield 85% recovery of global stocks by midcentury, higher economic returns, and greater robustness to environmental surprises. These results challenge the consensus that global fisheries can be rebuilt by existing approaches alone, while also underscoring that rebuilding stocks may still be achieved by improved decision-making tools that optimally manage this uncertainty.

Resolving the Measurement Uncertainty Paradox in Ecological Management.

Ecological management and decision-making typically focus on uncertainty about the future, but surprisingly little is known about how to account for uncertainty of the present: that is, the realities of having only partial or imperfect measurements. Our primary paradigms for handling decisions under uncertainty-the precautionary principle and optimal control-have so far given contradictory results. This paradox is best illustrated in the example of fisheries management, where many ideas that guide thinking about ecological decision-making were first developed. We find that simplistic optimal control approaches have repeatedly concluded that a manager should increase catch quotas when faced with greater uncertainty about the fish biomass. Current best practices take a more precautionary approach, decreasing catch quotas by a fixed amount to account for uncertainty. Using comparisons to both simulated and historical catch data, we find that neither approach is sufficient to avoid stock collapses under moderate observational uncertainty. Using partially observed Markov decision process (POMDP) methods, we demonstrate how this paradox arises from flaws in the standard theory, which contributes to overexploitation of fisheries and increased probability of economic and ecological collapse. In contrast, we find that POMDP-based management avoids such overexploitation while also generating higher economic value. These results have significant implications for how we handle uncertainty in both fisheries and ecological management more generally.

From noise to knowledge: how randomness generates novel phenomena and reveals information.

Noise, as the term itself suggests, is most often seen a nuisance to ecological insight, a inconvenient reality that must be acknowledged, a haystack that must be stripped away to reveal the processes of interest underneath. Yet despite this well-earned reputation, noise is often interesting in its own right: noise can induce novel phenomena that could not be understood from some underlying deterministic model alone. Nor is all noise the same, and close examination of differences in frequency, colour or magnitude can reveal insights that would otherwise be inaccessible. Yet with each aspect of stochasticity leading to some new or unexpected behaviour, the time is right to move beyond the familiar refrain of "everything is important" (Bjørnstad & Grenfell ). Stochastic phenomena can suggest new ways of inferring process from pattern, and thus spark more dialog between theory and empirical perspectives that best advances the field as a whole. I highlight a few compelling examples, while observing that the study of stochastic phenomena are only beginning to make this translation into empirical inference. There are rich opportunities at this interface in the years ahead.

Making ecological models adequate.

Critical evaluation of the adequacy of ecological models is urgently needed to enhance their utility in developing theory and enabling environmental managers and policymakers to make informed decisions. Poorly supported management can have detrimental, costly or irreversible impacts on the environment and society. Here, we examine common issues in ecological modelling and suggest criteria for improving modelling frameworks. An appropriate level of process description is crucial to constructing the best possible model, given the available data and understanding of ecological structures. Model details unsupported by data typically lead to over parameterisation and poor model performance. Conversely, a lack of mechanistic details may limit a model's ability to predict ecological systems' responses to management. Ecological studies that employ models should follow a set of model adequacy assessment protocols that include: asking a series of critical questions regarding state and control variable selection, the determinacy of data, and the sensitivity and validity of analyses. We also need to improve model elaboration, refinement and coarse graining procedures to better understand the relevancy and adequacy of our models and the role they play in advancing theory, improving hind and forecasting, and enabling problem solving and management.

The principles of tomorrow's university.

In the 21st Century, research is increasingly data- and computation-driven. Researchers, funders, and the larger community today emphasize the traits of openness and reproducibility. In March 2017, 13 mostly early-career research leaders who are building their careers around these traits came together with ten university leaders (presidents, vice presidents, and vice provosts), representatives from four funding agencies, and eleven organizers and other stakeholders in an NIH- and NSF-funded one-day, invitation-only workshop titled "Imagining Tomorrow's University." Workshop attendees were charged with launching a new dialog around open research - the current status, opportunities for advancement, and challenges that limit sharing. The workshop examined how the internet-enabled research world has changed, and how universities need to change to adapt commensurately, aiming to understand how universities can and should make themselves competitive and attract the best students, staff, and faculty in this new world. During the workshop, the participants re-imagined scholarship, education, and institutions for an open, networked era, to uncover new opportunities for universities to create value and serve society. They expressed the results of these deliberations as a set of 22 principles of tomorrow's university across six areas: credit and attribution, communities, outreach and engagement, education, preservation and reproducibility, and technologies. Activities that follow on from workshop results take one of three forms. First, since the workshop, a number of workshop authors have further developed and published their white papers to make their reflections and recommendations more concrete. These authors are also conducting efforts to implement these ideas, and to make changes in the university system.  Second, we plan to organise a follow-up workshop that focuses on how these principles could be implemented. Third, we believe that the outcomes of this workshop support and are connected with recent theoretical work on the position and future of open knowledge institutions.

Skills and Knowledge for Data-Intensive Environmental Research.

The scale and magnitude of complex and pressing environmental issues lend urgency to the need for integrative and reproducible analysis and synthesis, facilitated by data-intensive research approaches. However, the recent pace of technological change has been such that appropriate skills to accomplish data-intensive research are lacking among environmental scientists, who more than ever need greater access to training and mentorship in computational skills. Here, we provide a roadmap for raising data competencies of current and next-generation environmental researchers by describing the concepts and skills needed for effectively engaging with the heterogeneous, distributed, and rapidly growing volumes of available data. We articulate five key skills: (1) data management and processing, (2) analysis, (3) software skills for science, (4) visualization, and (5) communication methods for collaboration and dissemination. We provide an overview of the current suite of training initiatives available to environmental scientists and models for closing the skill-transfer gap.

After the games are over: life-history trade-offs drive dispersal attenuation following range expansion.

Increased dispersal propensity often evolves on expanding range edges due to the Olympic Village effect, which involves the fastest and fittest finding themselves together in the same place at the same time, mating, and giving rise to like individuals. But what happens after the range's leading edge has passed and the games are over? Although empirical studies indicate that dispersal propensity attenuates following range expansion, hypotheses about the mechanisms driving this attenuation have not been clearly articulated or tested. Here, we used a simple model of the spatiotemporal dynamics of two phenotypes, one fast and the other slow, to propose that dispersal attenuation beyond preexpansion levels is only possible in the presence of trade-offs between dispersal and life-history traits. The Olympic Village effect ensures that fast dispersers preempt locations far from the range's previous limits. When trade-offs are absent, this preemptive spatial advantage has a lasting impact, with highly dispersive individuals attaining equilibrium frequencies that are strictly higher than their introduction frequencies. When trade-offs are present, dispersal propensity decays rapidly at all locations. Our model's results about the postcolonization trajectory of dispersal evolution are clear and, in principle, should be observable in field studies. We conclude that empirical observations of postcolonization dispersal attenuation offer a novel way to detect the existence of otherwise elusive trade-offs between dispersal and life-history traits.

Optimal management of a stochastically varying population when policy adjustment is costly.

Ecological systems are dynamic and policies to manage them need to respond to that variation. However, policy adjustments will sometimes be costly, which means that fine-tuning a policy to track variability in the environment very tightly will only sometimes be worthwhile. We use a classic fisheries management problem, how to manage a stochastically varying population using annually varying quotas in order to maximize profit, to examine how costs of policy adjustment change optimal management recommendations. Costs of policy adjustment (changes in fishing quotas through time) could take different forms. For example, these costs may respond to the size of the change being implemented, or there could be a fixed cost any time a quota change is made. We show how different forms of policy costs have contrasting implications for optimal policies. Though it is frequently assumed that costs to adjusting policies will dampen variation in the policy, we show that certain cost structures can actually increase variation through time. We further show that failing to account for adjustment costs has a consistently worse economic impact than would assuming these costs are present when they are not.

Avoiding tipping points in fisheries management through Gaussian process dynamic programming.

Model uncertainty and limited data are fundamental challenges to robust management of human intervention in a natural system. These challenges are acutely highlighted by concerns that many ecological systems may contain tipping points, such as Allee population sizes. Before a collapse, we do not know where the tipping points lie, if they exist at all. Hence, we know neither a complete model of the system dynamics nor do we have access to data in some large region of state space where such a tipping point might exist. We illustrate how a Bayesian non-parametric approach using a Gaussian process (GP) prior provides a flexible representation of this inherent uncertainty. We embed GPs in a stochastic dynamic programming framework in order to make robust management predictions with both model uncertainty and limited data. We use simulations to evaluate this approach as compared with the standard approach of using model selection to choose from a set of candidate models. We find that model selection erroneously favours models without tipping points, leading to harvest policies that guarantee extinction. The Gaussian process dynamic programming (GPDP) performs nearly as well as the true model and significantly outperforms standard approaches. We illustrate this using examples of simulated single-species dynamics, where the standard model selection approach should be most effective and find that it still fails to account for uncertainty appropriately and leads to population crashes, while management based on the GPDP does not, as it does not underestimate the uncertainty outside of the observed data.