REVIEW: Predictive ecology in a changing world

Harnessing Deep Learning in Ecology: An Example Predicting Bark Beetle Outbreaks

Addressing current global challenges such as biodiversity loss, global change, and increasing demands for ecosystem services requires improved ecological prediction. Recent increases in data availability, process understanding, and computing power are fostering quantitative approaches in ecology. However, flexible methodological frameworks are needed to utilize these developments towards improved ecological prediction. Deep learning is a rapidly evolving branch of machine learning, yet has received only little attention in ecology to date. It refers to the training of deep neural networks (DNNs), i.e. artificial neural networks consisting of many layers and a large number of neurons. We here provide a reproducible example (including code and data) of designing, training, and applying DNNs for ecological prediction. Using bark beetle outbreaks in conifer-dominated forests as an example, we show that DNNs are well able to predict both short-term infestation risk at the local scale and long-term outbreak dynamics at the landscape level. We furthermore highlight that DNNs have better overall performance than more conventional approaches to predicting bark beetle outbreak dynamics. We conclude that DNNs have high potential to form the backbone of a comprehensive disturbance forecasting system. More broadly, we argue for an increased utilization of the predictive power of DNNs for a wide range of ecological problems.

Seasonal difference in temporal transferability of an ecological model: near-term predictions of lemming outbreak abundances

Ecological models have been criticized for a lack of validation of their temporal transferability. Here we answer this call by investigating the temporal transferability of a dynamic state-space model developed to estimate season-dependent biotic and climatic predictors of spatial variability in outbreak abundance of the Norwegian lemming. Modelled summer and winter dynamics parametrized by spatial trapping data from one cyclic outbreak were validated with data from a subsequent outbreak. There was a distinct difference in model transferability between seasons. Summer dynamics had good temporal transferability, displaying ecological models' potential to be temporally transferable. However, the winter dynamics transferred poorly. This discrepancy is likely due to a temporal inconsistency in the ability of the climate predictor (i.e. elevation) to reflect the winter conditions affecting lemmings both directly and indirectly. We conclude that there is an urgent need for data and models that yield better predictions of winter processes, in particular in face of the expected rapid climate change in the Arctic.

Generality in multispecies responses to ocean acidification revealed through multiple hypothesis testing

Decades of research have demonstrated that many calcifying species are negatively affected by ocean acidification, a major anthropogenic threat in marine ecosystems. However, even closely related species may exhibit different responses to ocean acidification and less is known about the drivers that shape such variation in different species. Here, we examine the drivers of physiological performance under ocean acidification in a group of five species of turf-forming coralline algae. Specifically, quantitating the relative weight of evidence for each of ten hypotheses, we show that variation in coralline calcification and photosynthesis was best explained by allometric traits. Across ocean acidification conditions, larger individuals (measured as noncalcified mass) had higher net calcification and photosynthesis rates. Importantly, our approach was able to not only identify the aspect of size that drove the performance of coralline algae, but also determined that responses to ocean acidification were not dependent on species identity, evolutionary relatedness, habitat, shape, or structural composition. In fact, we found that failure to test multiple, alternative hypotheses would underestimate the generality of physiological performances, leading to the conclusion that each species had different baseline performance under ocean acidification. Testing among alternative hypotheses is an essential step toward determining the generalizability of experiments across taxa and identifying common drivers of species responses to global change.

Transferability of biotic interactions: Temporal consistency of arctic plant-rodent relationships is poor

Variability in biotic interaction strength is an integral part of food web functioning. However, the consequences of the spatial and temporal variability of biotic interactions are poorly known, in particular for predicting species abundance and distribution. The amplitude of rodent population cycles (i.e., peak-phase abundances) has been hypothesized to be determined by vegetation properties in tundra ecosystems. We assessed the spatial and temporal predictability of food and shelter plants effects on peak-phase small rodent abundance during two consecutive rodent population peaks. Rodent abundance was related to both food and shelter biomass during the first peak, and spatial transferability was mostly good. Yet, the temporal transferability of our models to the next population peak was poorer. Plant-rodent interactions are thus temporally variable and likely more complex than simple one-directional (bottom-up) relationships or variably overruled by other biotic interactions and abiotic factors. We propose that parametrizing a more complete set of functional links within food webs across abiotic and biotic contexts would improve transferability of biotic interaction models. Such attempts are currently constrained by the lack of data with replicated estimates of key players in food webs. Enhanced collaboration between researchers whose main research interests lay in different parts of the food web could ameliorate this.

Ecological research and environmental management: We need different interfaces based on different knowledge types

The role of ecological science in environmental management has been discussed by many authors who recognize that there is a persistent gap between ecological science and environmental management. Here we develop theory through different perspectives based on knowledge types, research categories and research-management interface types, which we combine into a common framework. To draw out insights for bridging this gap, we build our case by:We point out the complementarities as well as the specificities and limitations of the different types of ecological research, ecological knowledge and research-management interfaces, which is of major importance for environmental management and research policies.

Weak interspecific interactions in a sagebrush steppe? Conflicting evidence from observations and experiments

Stable coexistence requires intraspecific limitations to be stronger than interspecific limitations. The greater the difference between intra- and interspecific limitations, the more stable the coexistence, and the weaker the competitive release any species should experience following removal of competitors. We conducted a removal experiment to test whether a previously estimated model, showing surprisingly weak interspecific competition for four dominant species in a sagebrush steppe, accurately predicts competitive release. Our treatments were (1) removal of all perennial grasses and (2) removal of the dominant shrub, Artemisia tripartita. We regressed survival, growth, and recruitment on the locations, sizes, and species identities of neighboring plants, along with an indicator variable for removal treatment. If our "baseline" regression model, which accounts for local plant-plant interactions, accurately explains the observed responses to removals, then the removal coefficient should be non-significant. For survival, the removal coefficients were never significantly different from zero, and only A. tripartita showed a (negative) response to removals at the recruitment stage. For growth, the removal treatment effect was significant and positive for two species, Poa secunda and Pseudoroegneria spicata, indicating that the baseline model underestimated interspecific competition. For all three grass species, population models based on the vital rate regressions that included removal effects projected 1.4- to 3-fold increases in equilibrium population size relative to the baseline model (no removal effects). However, we found no evidence of higher response to removal in quadrats with higher pretreatment cover of A. tripartita, or by plants experiencing higher pre-treatment crowding by A. tripartita, raising questions about the mechanisms driving the positive response to removal. While our results show the value of combining observations with a simple removal experiment, more tightly controlled experiments focused on underlying mechanisms may be required to conclusively validate or reject predictions from phenomenological models.

Is structural sensitivity a problem of oversimplified biological models? Insights from nested Dynamic Energy Budget models

Many current issues in ecology require predictions made by mathematical models, which are built on somewhat arbitrary choices. Their consequences are quantified by sensitivity analysis to quantify how changes in model parameters propagate into an uncertainty in model predictions. An extension called structural sensitivity analysis deals with changes in the mathematical description of complex processes like predation. Such processes are described at the population scale by a specific mathematical function taken among similar ones, a choice that can strongly drive model predictions. However, it has only been studied in simple theoretical models. Here, we ask whether structural sensitivity is a problem of oversimplified models. We found in predator-prey models describing chemostat experiments that these models are less structurally sensitive to the choice of a specific functional response if they include mass balance resource dynamics and individual maintenance. Neglecting these processes in an ecological model (for instance by using the well-known logistic growth equation) is not only an inappropriate description of the ecological system, but also a source of more uncertain predictions.

Can species distribution models really predict the expansion of invasive species?

Predictive studies are of paramount importance for biological invasions, one of the biggest threats for biodiversity. To help and better prioritize management strategies, species distribution models (SDMs) are often used to predict the potential invasive range of introduced species. Yet, SDMs have been regularly criticized, due to several strong limitations, such as violating the equilibrium assumption during the invasion process. Unfortunately, validation studies–with independent data–are too scarce to assess the predictive accuracy of SDMs in invasion biology. Yet, biological invasions allow to test SDMs usefulness, by retrospectively assessing whether they would have accurately predicted the latest ranges of invasion. Here, we assess the predictive accuracy of SDMs in predicting the expansion of invasive species. We used temporal occurrence data for the Asian hornet Vespa velutina nigrithorax, a species native to China that is invading Europe with a very fast rate. Specifically, we compared occurrence data from the last stage of invasion (independent validation points) to the climate suitability distribution predicted from models calibrated with data from the early stage of invasion. Despite the invasive species not being at equilibrium yet, the predicted climate suitability of validation points was high. SDMs can thus adequately predict the spread of V. v. nigrithorax, which appears to be—at least partially–climatically driven. In the case of V. v. nigrithorax, SDMs predictive accuracy was slightly but significantly better when models were calibrated with invasive data only, excluding native data. Although more validation studies for other invasion cases are needed to generalize our results, our findings are an important step towards validating the use of SDMs in invasion biology.

How complex should models be? Comparing correlative and mechanistic range dynamics models

Criticism has been levelled at climate-change-induced forecasts of species range shifts that do not account explicitly for complex population dynamics. The relative importance of such dynamics under climate change is, however, undetermined because direct tests comparing the performance of demographic models vs. simpler ecological niche models are still lacking owing to difficulties in evaluating forecasts using real-world data. We provide the first comparison of the skill of coupled ecological-niche-population models and ecological niche models in predicting documented shifts in the ranges of 20 British breeding bird species across a 40-year period. Forecasts from models calibrated with data centred on 1970 were evaluated using data centred on 2010. We found that more complex coupled ecological-niche-population models (that account for dispersal and metapopulation dynamics) tend to have higher predictive accuracy in forecasting species range shifts than structurally simpler models that only account for variation in climate. However, these better forecasts are achieved only if ecological responses to climate change are simulated without static snapshots of historic land use, taken at a single point in time. In contrast, including both static land use and dynamic climate variables in simpler ecological niche models improve forecasts of observed range shifts. Despite being less skilful at predicting range changes at the grid-cell level, ecological niche models do as well, or better, than more complex models at predicting the magnitude of relative change in range size. Therefore, ecological niche models can provide a reasonable first approximation of the magnitude of species' potential range shifts, especially when more detailed data are lacking on dispersal dynamics, demographic processes underpinning population performance, and change in land cover.

Climate change could drive marine food web collapse through altered trophic flows and cyanobacterial proliferation

Global warming and ocean acidification are forecast to exert significant impacts on marine ecosystems worldwide. However, most of these projections are based on ecological proxies or experiments on single species or simplified food webs. How energy fluxes are likely to change in marine food webs in response to future climates remains unclear, hampering forecasts of ecosystem functioning. Using a sophisticated mesocosm experiment, we model energy flows through a species-rich multilevel food web, with live habitats, natural abiotic variability, and the potential for intra- and intergenerational adaptation. We show experimentally that the combined stress of acidification and warming reduced energy flows from the first trophic level (primary producers and detritus) to the second (herbivores), and from the second to the third trophic level (carnivores). Warming in isolation also reduced the energy flow from herbivores to carnivores, the efficiency of energy transfer from primary producers and detritus to herbivores and detritivores, and the living biomass of detritivores, herbivores, and carnivores. Whilst warming and acidification jointly boosted primary producer biomass through an expansion of cyanobacteria, this biomass was converted to detritus rather than to biomass at higher trophic levels—i.e., production was constrained to the base of the food web. In contrast, ocean acidification affected the food web positively by enhancing trophic flow from detritus and primary producers to herbivores, and by increasing the biomass of carnivores. Our results show how future climate change can potentially weaken marine food webs through reduced energy flow to higher trophic levels and a shift towards a more detritus-based system, leading to food web simplification and altered producer–consumer dynamics, both of which have important implications for the structuring of benthic communities.

Healthy marine ecosystems are crucial for people’s livelihoods and food production. Global climate stressors, such as warming and ocean acidification, can drastically impact the structure and function of marine food webs, diminishing the production of goods and services. Our ability to predict how future food webs will respond to a changing environment is limited by our understanding of species responses to climate change, which are often tested in isolation or in simplified experimental designs. More realistic predictions of the impacts of climate change on ecosystems requires consideration of entire species communities, including the species interactions that can buffer or exacerbate these impacts. We experimentally tested the effects of warming and acidification, both individually and in combination, on a benthic marine food web in a near-natural ecological setting. Energy flow from the first trophic level (primary producers and detritus) to the second (herbivores), and from the second to the third trophic level (carnivores) was quantified under these different regimes. We show that warming, either alone or in combination with acidification, can constrain productivity to the bottom of the food web by enhancing cyanobacterial biomass and reducing energy flow to higher trophic levels, thus lowering energy transfer efficiency between producers and consumers. In contrast, increased ocean acidification alone showed a positive effect on herbivores and carnivores. Our finding is important because it demonstrates that future warming could drive marine food web collapses to potentially simplified and less productive coastal systems.

On the Power of Uncertainties in Microbial System Modeling: No Need To Hide Them Anymore

For decades, microbiologists have considered uncertainties as an undesired side effect of experimental protocols. As a consequence, standard microbial system modeling strives to hide uncertainties for the sake of deterministic understanding.

For decades, microbiologists have considered uncertainties as an undesired side effect of experimental protocols. As a consequence, standard microbial system modeling strives to hide uncertainties for the sake of deterministic understanding. However, recent studies have highlighted greater experimental variability than expected and emphasized uncertainties not as a weakness but as a necessary feature of complex microbial systems. We therefore advocate that biological uncertainties need to be considered foundational facets that must be incorporated in models. Not only will understanding these uncertainties improve our understanding and identification of microbial traits, it will also provide fundamental insights on microbial systems as a whole. Taking into account uncertainties within microbial models calls for new validation techniques. Formal verification already overcomes this shortcoming by proposing modeling frameworks and validation techniques dedicated to probabilistic models. However, further work remains to extract the full potential of such techniques in the context of microbial models. Herein, we demonstrate how statistical model checking can enhance the development of microbial models by building confidence in the estimation of critical parameters and through improved sensitivity analyses.

Improving the forecast for biodiversity under climate change

New biological models are incorporating the realistic processes underlying biological responses to climate change and other human-caused disturbances. However, these more realistic models require detailed information, which is lacking for most species on Earth. Current monitoring efforts mainly document changes in biodiversity, rather than collecting the mechanistic data needed to predict future changes. We describe and prioritize the biological information needed to inform more realistic projections of species' responses to climate change. We also highlight how trait-based approaches and adaptive modeling can leverage sparse data to make broader predictions. We outline a global effort to collect the data necessary to better understand, anticipate, and reduce the damaging effects of climate change on biodiversity.

A Methodology to Evaluate Ecological Resources and Risk Using Two Case Studies at the Department of Energy's Hanford Site

An assessment of the potential risks to ecological resources from remediation activities or other perturbations should involve a quantitative evaluation of resources on the remediation site and in the surrounding environment. We developed a risk methodology to rapidly evaluate potential impact on ecological resources for the U.S. Department of Energy's Hanford Site in southcentral Washington State. We describe the application of the risk evaluation for two case studies to illustrate its applicability. The ecological assessment involves examining previous sources of information for the site, defining different resource levels from 0 to 5. We also developed a risk rating scale from non-discernable to very high. Field assessment is the critical step to determine resource levels or to determine if current conditions are the same as previously evaluated. We provide a rapid assessment method for current ecological conditions that can be compared to previous site-specific data, or that can be used to assess resource value on other sites where ecological information is not generally available. The method is applicable to other Department of Energy's sites, where its development may involve a range of state regulators, resource trustees, Tribes and other stakeholders. Achieving consistency across Department of Energy's sites for valuation of ecological resources on remediation sites will assure Congress and the public that funds and personnel are being deployed appropriately.

Elevated nonlinearity as an indicator of shifts in the dynamics of populations under stress

Populations occasionally experience abrupt changes, such as local extinctions, strong declines in abundance or transitions from stable dynamics to strongly irregular fluctuations. Although most of these changes have important ecological and at times economic implications, they remain notoriously difficult to detect in advance. Here, we study changes in the stability of populations under stress across a variety of transitions. Using a Ricker-type model, we simulate shifts from stable point equilibrium dynamics to cyclic and irregular boom-bust oscillations as well as abrupt shifts between alternative attractors. Our aim is to infer the loss of population stability before such shifts based on changes in nonlinearity of population dynamics. We measure nonlinearity by comparing forecast performance between linear and nonlinear models fitted on reconstructed attractors directly from observed time series. We compare nonlinearity to other suggested leading indicators of instability (variance and autocorrelation). We find that nonlinearity and variance increase in a similar way prior to the shifts. By contrast, autocorrelation is strongly affected by oscillations. Finally, we test these theoretical patterns in datasets of fisheries populations. Our results suggest that elevated nonlinearity could be used as an additional indicator to infer changes in the dynamics of populations under stress.

Predicting evolution in response to climate change: the example of sprouting probability in three dormancy-prone orchid species

Although many ecological properties of species respond to climate change, their evolutionary responses are poorly understood. Here, we use data from long-term demographic studies to predict evolutionary responses of three herbaceous perennial orchid species, Cypripedium parviflorum, C. candidum and Ophrys sphegodes, to predicted climate changes in the habitats they occupy. We focus on the evolution of sprouting probability, because all three species exhibit long-term vegetative dormancy, i.e. individual plants may not emerge above-ground, potentially for several consecutive years. The drivers of all major vital rates for populations of the species were analysed with general linear mixed models (GLMMs). High-dimensionality function-based matrix projection models were then developed to serve as core elements of deterministic and stochastic adaptive dynamics models used to analyse the adaptive context of sprouting in all populations. We then used regional climate forecasts, derived from high-resolution general atmospheric circulation models, of increased mean annual temperatures and spring precipitation at the occupied sites, to predict evolutionary trends in sprouting. The models predicted that C. parviflorum and O. sphegodes will evolve higher and lower probabilities of sprouting, respectively, by the end of the twenty-first century, whereas, after considerable variation, the probability of sprouting in C. candidum will return to its current level. These trends appear to be driven by relationships between mortality and size: in C. parviflorum and C. candidum, mortality is negatively related to size in the current year but positively related to growth since the previous year, whereas in O. sphegodes, mortality is positively related to size.

Partitioning prediction uncertainty in climate-dependent population models

The science of complex systems is increasingly asked to forecast the consequences of climate change. As a result, scientists are now engaged in making predictions about an uncertain future, which entails the efficient communication of this uncertainty. Here we show the benefits of hierarchically decomposing the uncertainty in predicted changes in animal population size into its components due to structural uncertainty in climate scenarios (greenhouse gas emissions and global circulation models), structural uncertainty in the demographic model, climatic stochasticity, environmental stochasticity unexplained by climate-demographic trait relationships, and sampling variance in demographic parameter estimates. We quantify components of uncertainty surrounding the future abundance of a migratory bird, the greater snow goose (Chen caeruslescens atlantica), using a process-based demographic model covering their full annual cycle. Our model predicts a slow population increase but with a large prediction uncertainty. As expected from theoretical variance decomposition rules, the contribution of sampling variance to prediction uncertainty rapidly overcomes that of process variance and dominates. Among the sources of process variance, uncertainty in the climate scenarios contributed less than 3% of the total prediction variance over a 40-year period, much less than environmental stochasticity. Our study exemplifies opportunities to improve the forecasting of complex systems using long-term studies and the challenges inherent to predicting the future of stochastic systems.

Evaluating citizen science data for forecasting species responses to national forest management

The extensive spatial and temporal coverage of many citizen science datasets (CSD) makes them appealing for use in species distribution modeling and forecasting. However, a frequent limitation is the inability to validate results. Here, we aim to assess the reliability of CSD for forecasting species occurrence in response to national forest management projections (representing 160,366 km²) by comparison against forecasts from a model based on systematically collected colonization–extinction data. We fitted species distribution models using citizen science observations of an old‐forest indicator fungus Phellinus ferrugineofuscus. We applied five modeling approaches (generalized linear model, Poisson process model, Bayesian occupancy model, and two MaxEnt models). Models were used to forecast changes in occurrence in response to national forest management for 2020‐2110. Forecasts of species occurrence from models based on CSD were congruent with forecasts made using the colonization–extinction model based on systematically collected data, although different modeling methods indicated different levels of change. All models projected increased occurrence in set‐aside forest from 2020 to 2110: the projected increase varied between 125% and 195% among models based on CSD, in comparison with an increase of 129% according to the colonization–extinction model. All but one model based on CSD projected a decline in production forest, which varied between 11% and 49%, compared to a decline of 41% using the colonization–extinction model. All models thus highlighted the importance of protected old forest for P. ferrugineofuscus persistence. We conclude that models based on CSD can reproduce forecasts from models based on systematically collected colonization–extinction data and so lead to the same forest management conclusions. Our results show that the use of a suite of models allows CSD to be reliably applied to land management and conservation decision making, demonstrating that widely available CSD can be a valuable forecasting resource.

Functional remediation components: A conceptual method of evaluating the effects of remediation on risks to ecological receptors

Governmental agencies, regulators, health professionals, tribal leaders, and the public are faced with understanding and evaluating the effects of cleanup activities on species, populations, and ecosystems. While engineers and managers understand the processes involved in different remediation types such as capping, pump and treat, and natural attenuation, there is often a disconnect between (1) how ecologists view the influence of different types of remediation, (2) how the public perceives them, and (3) how engineers understand them. The overall goal of the present investigation was to define the components of remediation types (= functional remediation). Objectives were to (1) define and describe functional components of remediation, regardless of the remediation type, (2) provide examples of each functional remediation component, and (3) explore potential effects of functional remediation components in the post-cleanup phase that may involve continued monitoring and assessment. Functional remediation components include types, numbers, and intensity of people, trucks, heavy equipment, pipes, and drill holes, among others. Several components may be involved in each remediation type, and each results in ecological effects, ranging from trampling of plants, to spreading invasive species, to disturbing rare species, and to creating fragmented habitats. In some cases remediation may exert a greater effect on ecological receptors than leaving the limited contamination in place. A goal of this conceptualization is to break down functional components of remediation such that managers, regulators, and the public might assess the effects of timing, extent, and duration of different remediation options on ecological systems.

Forecasting Epidemiological and Evolutionary Dynamics of Infectious Diseases

Mathematical models have been powerful tools in developing mechanistic understanding of infectious diseases. Furthermore, they have allowed detailed forecasting of epidemiological phenomena such as outbreak size, which is of considerable public-health relevance. The short generation time of pathogens and the strong selection they are subjected to (by host immunity, vaccines, chemotherapy, etc.) mean that evolution is also a key driver of infectious disease dynamics. Accurate forecasting of pathogen dynamics therefore calls for the integration of epidemiological and evolutionary processes, yet this integration remains relatively rare. We review previous attempts to model and predict infectious disease dynamics with or without evolution and discuss major challenges facing the development of the emerging science of epidemic forecasting.

Predicting stochastic community dynamics in grasslands under the assumption of competitive symmetry

Community dynamics is influenced by multiple ecological processes such as environmental spatiotemporal variation, competition between individuals and demographic stochasticity. Quantifying the respective influence of these various processes and making predictions on community dynamics require the use of a dynamical framework encompassing these various components. We here demonstrate how to adapt the framework of stochastic community dynamics to the peculiarities of herbaceous communities, by using a short temporal resolution adapted to the time scale of competition between herbaceous plants, and by taking into account the seasonal drops in plant aerial biomass following winter, harvesting or consumption by herbivores. We develop a hybrid inference method for this novel modelling framework that both uses numerical simulations and likelihood computations. Applying this methodology to empirical data from the Jena biodiversity experiment, we find that environmental stochasticity has a larger effect on community dynamics than demographic stochasticity, and that both effects are generally smaller than observation errors at the plot scale. We further evidence that plant intrinsic growth rates and carrying capacities are moderately predictable from plant vegetative height, specific leaf area and leaf dry matter content. We do not find any trade-off between demographical components, since species with larger intrinsic growth rates tend to also have lower demographic and environmental variances. Finally, we find that our model is able to make relatively good predictions of multi-specific community dynamics based on the assumption of competitive symmetry.

Predicting biotic interactions and their variability in a changing environment

Global environmental change is altering the patterns of biodiversity worldwide. Observation and theory suggest that species' distributions and abundances depend on a suite of processes, notably abiotic filtering and biotic interactions, both of which are constrained by species' phylogenetic history. Models predicting species distribution have historically mostly considered abiotic filtering and are only starting to integrate biotic interaction. However, using information on present interactions to forecast the future of biodiversity supposes that biotic interactions will not change when species are confronted with new environments. Using bacterial microcosms, we illustrate how biotic interactions can vary along an environmental gradient and how this variability can depend on the phylogenetic distance between interacting species.