Eliciting expert knowledge in conservation science

Expert knowledge is used widely in the science and practice of conservation because of the complexity of problems, relative lack of data, and the imminent nature of many conservation decisions. Expert knowledge is substantive information on a particular topic that is not widely known by others. An expert is someone who holds this knowledge and who is often deferred to in its interpretation. We refer to predictions by experts of what may happen in a particular context as expert judgments. In general, an expert-elicitation approach consists of five steps: deciding how information will be used, determining what to elicit, designing the elicitation process, performing the elicitation, and translating the elicited information into quantitative statements that can be used in a model or directly to make decisions. This last step is known as encoding. Some of the considerations in eliciting expert knowledge include determining how to work with multiple experts and how to combine multiple judgments, minimizing bias in the elicited information, and verifying the accuracy of expert information. We highlight structured elicitation techniques that, if adopted, will improve the accuracy and information content of expert judgment and ensure uncertainty is captured accurately. We suggest four aspects of an expert elicitation exercise be examined to determine its comprehensiveness and effectiveness: study design and context, elicitation design, elicitation method, and elicitation output. Just as the reliability of empirical data depends on the rigor with which it was acquired so too does that of expert knowledge.     

Fuzzy experts on recreational vessels, a risk modelling approach for marine invasions

Understanding risks from the human-mediated spread of non-indigenous species (NIS) is a critical component of marine biosecurity management programmes. Recreational boating is well-recognised as a NIS pathway, especially at a regional scale. Assessment of risks from this pathway is therefore desirable for coastal environments where recreational boating occurs. However, formal or quantitative risk assessment for the recreational vessel pathway is often hampered by lack of data, hence often relies on expert opinion. The use of expert opinion itself is sometimes limited by its inherent vagueness, which can be an important source of uncertainty that reduces the validity and applicability of the assessment. Fuzzy logic, specifically interval type-2 fuzzy logic, is able to model and propagate this type of uncertainty, and is a useful technique in risk assessment where expert opinion is relied upon. The present paper describes the implementation of a NIS fuzzy expert system (FES) for assessing the risk of invasion in marine environments via recreational vessels. The FES was based on expert opinion gathered through systematic elicitation exercises, designed to acknowledge important uncertainty sources (e.g., underspecificity and ambiguity). The FES, using interval type-2 fuzzy logic, calculated an invasion risk value (integrating NIS infection and detection probabilities) for a range of invasion scenarios. These scenarios were defined by all possible combinations of two vessel types (moored and trailered), five vessel components (hull, deck, internal spaces, anchor, fishing gear), two infection modes (fouling, water/sediment retention) and six frequently visited marine habitats (marina, mooring, farm, ramp, wharf, anchorage). Although invasion risk values determined using the FES approach was scenario-specific, general patterns were identified. Moored vessels consistently showed higher invasion risk values than trailered vessels. Invasion risk values were higher for anchorages, moorings and wharves. Similarly, hull-fouling was revealed as the highest infection risk mode after pooling results across all habitats. The NIS fuzzy expert system presented here appears as a valuable prioritising and decision-making tool for NIS research, prevention and control activities. Its easy implementation and wide applicability should encourage the development and application of this type of system as an integral part of biosecurity, and other environmental management plans.     

Uncertainty in integrated coastal zone management

Uncertainty plays a major role in Integrated Coastal Zone Management (ICZM). A large part of this uncertainty is connected to our lack of knowledge of the integrated functioning of the coastal system and to the increasing need to act in a pro-active way. Increasingly, coastal managers are forced to take decisions based on information which is surrounded by uncertainties. Different types of uncertainty can be identified and the role of uncertainty in decision making, scientific uncertainty and model uncertainty in ICZM is discussed. The issue of spatial variability, which is believed to be extremely important in ICZM and represents a primary source of complexity and uncertainty, is also briefly introduced. Some principles for complex model building are described as an approach to handle, in a balanced way, the available data, information, knowledge and experience. The practical method of sensitivity analysis is then introduced as a method for a posterior evaluation of uncertainty in simulation models. We conclude by emphasising the need for the definition of an analysis plan in order to handle model uncertainty in a balanced way during the decision making process.     

Uncertainty characterization for emergy values

While statistical estimation of uncertainty has not typically accompanied published emergy values, as with any other quantitative model, uncertainty is embedded in these values, and lack of uncertainty characterization makes their accuracy not only opaque, it also prevents the use of emergy values in statistical tests of hypotheses. This paper first attempts to describe sources of uncertainty in unit emergy values (UEVs) and presents a framework for estimating this uncertainty with analytical and stochastic models, with model choices dependent upon on how the UEV is calculated and what kind of uncertainties are quantified. The analytical model can incorporate a broader spectrum of uncertainty types than the stochastic model, including model and scenario uncertainty, which may be significant in emergy models, but is only appropriate for the most basic of emergy calculations. Although less comprehensive in its incorporation of uncertainty, the proposed stochastic method is suitable for all types of UEVs. The distributions of unit emergy values approximate the lognormal distribution with variations depending on the types of uncertainty quantified as well as the way the UEVs are calculated. While both methods of estimating uncertainty in UEVs have their limitations in their presented stage of development, this paper provides methods for incorporating uncertainty into emergy, and demonstrates how this can be depicted and propagated so that it can be used in future emergy analyses and permit emergy to be more readily incorporated into other methods of environmental assessment, such as LCA.     

An integrated optimization and simulation approach for air pollution control under uncertainty in open-pit metal mine

Air Pollution Control model is developed for open-pit metal mines. Model will aid decision makers to select a cost-effective solution. Open-pit metal mines contribute toward air pollution and without effective control techniques manifests the risk of violation of environmental guidelines. This paper establishes a stochastic approach to conceptualize the air pollution control model to attain a sustainable solution. The model is formulated for decision makers to select the least costly treatment method using linear programming with a defined objective function and multi-constraints. Furthermore, an integrated fuzzy based risk assessment approach is applied to examine uncertainties and evaluate an ambient air quality systematically. The applicability of the optimized model is explored through an open-pit metal mine case study, in North America. This method also incorporates the meteorological data as input to accommodate the local conditions. The uncertainties in the inputs, and predicted concentration are accomplished by probabilistic analysis using Monte Carlo simulation method. The output results are obtained to select the cost-effective pollution control technologies for PM_2.5, PM₁₀, NO_x, SO₂ and greenhouse gases. The risk level is divided into three types (loose, medium and strict) using a triangular fuzzy membership approach based on different environmental guidelines. Fuzzy logic is then used to identify environmental risk through stochastic simulated cumulative distribution functions of pollutant concentration. Thus, an integrated modeling approach can be used as a decision tool for decision makers to select the cost-effective technology to control air pollution.     

Uncertainty analyses for calibrating a soil carbon balance model to agricultural field trial data in Sweden and Kenya

How do additional data of the same and/or different type contribute to reducing model parameter and predictive uncertainties? Most modeling applications of soil organic carbon (SOC) time series in agricultural field trial datasets have been conducted without accounting for model parameter uncertainty. There have been recent advances with Monte Carlo-based uncertainty analyses in the field of hydrological modeling that are applicable, relevant and potentially valuable in modeling the dynamics of SOC. Here we employed a Monte Carlo method with threshold screening known as Generalized Likelihood Uncertainty Estimation (GLUE) to calibrate the Introductory Carbon Balance Model (ICBM) to long-term field trail data from Ultuna, Sweden and Machang’a, Kenya. Calibration results are presented in terms of parameter distributions and credibility bands on time series simulations for a number of case studies. Using these methods, we demonstrate that widely uncertain model parameters, as well as strong covariance between inert pool size and rate constant parameters, exist when root mean square simulation errors were within uncertainties in input estimations and data observations. We show that even rough estimates of the inert pool (perhaps from chemical analysis) can be quite valuable to reduce uncertainties in model parameters. In fact, such estimates were more effective at reducing parameter and predictive uncertainty than an additional 16 years time series data at Ultuna. We also demonstrate an effective method to jointly, simultaneously and in principle more robustly calibrate model parameters to multiple datasets across different climatic regions within an uncertainty framework. These methods and approaches should have benefits for use with other SOC models and datasets as well.     

Predicting the distributions of marine organisms at the global scale

We present and evaluate AquaMaps, a presence-only species distribution modelling system that allows the incorporation of expert knowledge about habitat usage and was designed for maximum output of standardized species range maps at the global scale. In the marine environment there is a significant challenge to the production of range maps due to large biases in the amount and location of occurrence data for most species. AquaMaps is compared with traditional presence-only species distribution modelling methods to determine the quality of outputs under equivalently automated conditions. The effect of the inclusion of expert knowledge to AquaMaps is also investigated. Model outputs were tested internally, through data partitioning, and externally against independent survey data to determine the ability of models to predict presence versus absence. Models were also tested externally by assessing correlation with independent survey estimates of relative species abundance. AquaMaps outputs compare well to the existing methods tested, and inclusion of expert knowledge results in a general improvement in model outputs. The transparency, speed and adaptability of the AquaMaps system, as well as the existing online framework which allows expert review to compensate for sampling biases and thus improve model predictions are proposed as additional benefits for public and research use alike.     

Uncertainty analysis of a pharmacokinetic modeling for inorganic arsenic

This paper presents an uncertainty and sensitivity analysis of a pharmacokinetic modeling of inorganic arsenic deposition in rodents for a short‐term exposure. Efforts to develop the pharmacokinetic model are directed towards predicting the kinetic behavior of inorganic arsenic in the body, including tissue and blood concentrations, and especially, the urinary excretion of arsenic and its methylated metabolites. However, the use of the model raises an important question when fixed values of model parameters are used: how is the uncertainty in the model prediction based on the collective uncertainties in the model inputs? This study focuses on an “epistemic”; uncertainty in order to handle this problem. In this case, the uncertainty refers to an input that has a single value which cannot be known with precision due to a lack of knowledge about items or its measurement. The combination of the pharmacokinetic model and the uncertainty analysis would help understand the uncertainties in risk assessment associated with inorganic arsenic.     

Parameter uncertainties in the modelling of vegetation dynamics—Effects on tree community structure and ecosystem functioning in European forest biomes

Dynamic vegetation models are useful tools for analysing terrestrial ecosystem processes and their interactions with climate through variations in carbon and water exchange. Long-term changes in structure and composition (vegetation dynamics) caused by altered competitive strength between plant functional types (PFTs) are attracting increasing attention as controls on ecosystem functioning and potential feedbacks to climate. Imperfect process knowledge and limited observational data restrict the possibility to parameterise these processes adequately and potentially contribute to uncertainty in model results. This study addresses uncertainty among parameters scaling vegetation dynamic processes in a process-based ecosystem model, LPJ-GUESS, designed for regional-scale studies, with the objective to assess the extent to which this uncertainty propagates to additional uncertainty in the tree community structure (in terms of the tree functional types present and their relative abundance) and thus to ecosystem functioning (carbon storage and fluxes). The results clearly indicate that the uncertainties in parameterisation can lead to a shift in competitive balance, most strikingly among deciduous tree PFTs, with dominance of either shade-tolerant or shade-intolerant PFTs being possible, depending on the choice of plausible parameter values. Despite this uncertainty, our results indicate that the resulting effect on ecosystem functioning is low. Since the vegetation dynamics in LPJ-GUESS are representative for the more complex Earth system models now being applied within ecosystem and climate research, we assume that our findings will be of general relevance. We suggest that, in terms of carbon storage and fluxes, the heavier parameterisation requirement of the processes involved does not widen the overall uncertainty in model predictions.     

Implications of different population model structures for management of threatened plants

Population viability analysis (PVA) is a reliable tool for ranking management options for a range of species despite parameter uncertainty. No one has yet investigated whether this holds true for model uncertainty for species with complex life histories and for responses to multiple threats. We tested whether a range of model structures yielded similar rankings of management and threat scenarios for 2 plant species with complex postfire responses. We examined 2 contrasting species from different plant functional types: an obligate seeding shrub and a facultative resprouting shrub. We exposed each to altered fire regimes and an additional, species‐specific threat. Long‐term demographic data sets were used to construct an individual‐based model (IBM), a complex stage‐based model, and a simple matrix model that subsumes all life stages into 2 or 3 stages. Agreement across models was good under some scenarios and poor under others. Results from the simple and complex matrix models were more similar to each other than to the IBM. Results were robust across models when dominant threats are considered but were less so for smaller effects. Robustness also broke down as the scenarios deviated from baseline conditions, likely the result of a number of factors related to the complexity of the species’ life history and how it was represented in a model. Although PVA can be an invaluable tool for integrating data and understanding species’ responses to threats and management strategies, this is best achieved in the context of decision support for adaptive management alongside multiple lines of evidence and expert critique of model construction and output.     

Quantifying parameter uncertainty in a coral reef model using Metropolis-Coupled Markov Chain Monte Carlo

Coral reefs are threatened ecosystems, so it is important to have predictive models of their dynamics. Most current models of coral reefs fall into two categories. The first is simple heuristic models which provide an abstract understanding of the possible behaviour of reefs in general, but do not describe real reefs. The second is complex simulations whose parameters are obtained from a range of sources such as literature estimates. We cannot estimate the parameters of these models from a single data set, and we have little idea of the uncertainty in their predictions.We have developed a compromise between these two extremes, which is complex enough to describe real reef data, but simple enough that we can estimate parameters for a specific reef from a time series. In previous work, we fitted this model to a long-term data set from Heron Island, Australia, using maximum likelihood methods. To evaluate predictions from this model, we need estimates of the uncertainty in our parameters. Here, we obtain such estimates using Bayesian Metropolis-Coupled Markov Chain Monte Carlo. We do this for versions of the model in which corals are aggregated into a single state variable (the three-state model), and in which corals are separated into four state variables (the six-state model), in order to determine the appropriate level of aggregation. We also estimate the posterior distribution of predicted trajectories in each case.In both cases, the fitted trajectories were close to the observed data, but we had doubts about the biological plausibility of some parameter estimates. We suggest that informative prior distributions incorporating expert knowledge may resolve this problem. In the six-state model, the posterior distribution of state frequencies after 40 years contained two divergent community types, one dominated by free space and soft corals, and one dominated by acroporid, pocilloporid, and massive corals. The three-state model predicts only a single community type. We conclude that the three-state model hides too much biological heterogeneity, but we need more data if we are to obtain reliable predictions from the six-state model. It is likely that there will be similarly large, but currently unevaluated, uncertainty in the predictions of other coral reef models, many of which are much more complex and harder to fit to real data.     

Using global sensitivity analysis of demographic models for ecological impact assessment

Population viability analysis (PVA) is widely used to assess population‐level impacts of environmental changes on species. When combined with sensitivity analysis, PVA yields insights into the effects of parameter and model structure uncertainty. This helps researchers prioritize efforts for further data collection so that model improvements are efficient and helps managers prioritize conservation and management actions. Usually, sensitivity is analyzed by varying one input parameter at a time and observing the influence that variation has over model outcomes. This approach does not account for interactions among parameters. Global sensitivity analysis (GSA) overcomes this limitation by varying several model inputs simultaneously. Then, regression techniques allow measuring the importance of input‐parameter uncertainties. In many conservation applications, the goal of demographic modeling is to assess how different scenarios of impact or management cause changes in a population. This is challenging because the uncertainty of input‐parameter values can be confounded with the effect of impacts and management actions. We developed a GSA method that separates model outcome uncertainty resulting from parameter uncertainty from that resulting from projected ecological impacts or simulated management actions, effectively separating the 2 main questions that sensitivity analysis asks. We applied this method to assess the effects of predicted sea‐level rise on Snowy Plover (Charadrius nivosus). A relatively small number of replicate models (approximately 100) resulted in consistent measures of variable importance when not trying to separate the effects of ecological impacts from parameter uncertainty. However, many more replicate models (approximately 500) were required to separate these effects. These differences are important to consider when using demographic models to estimate ecological impacts of management actions.     

Multi-model approach and evaluation of the uncertainty of model results. Rationale and applications to predict the behaviour of contaminants in the abiotic components of the fresh water environment

This work aims at discussing some concepts pertaining to the theory and practice of environmental modelling in view of the results of several model validation exercises performed by the group “Model validation for radionuclide transport in the system watershed-river and in estuaries” of project EMRAS (Environmental Modelling for Radiation Safety) supported by the IAEA (International Atomic Energy Agency). The analyses here performed concern models applied to real scenarios of environmental contamination. In particular, the reasons for the uncertainty of the models and the EBUA (empirically based uncertainty analysis) methodology are discussed. The foundations of multi-model approach in environmental modelling are presented and motivated. An application of EBUA to the results of a multi-model exercise concerning three models aimed at predicting the wash-off of radionuclide deposits from the Pripyat floodplain (Ukraine) was described. Multi-model approach is, definitely, a tool for uncertainty analysis. EBUA offers the opportunity of an evaluation of the uncertainty levels of predictions in multi-model applications.     

Methods for estimating the uncertainty in emergy table-form models

Emergy studies have suffered criticism due to the lack of uncertainty analysis and this shortcoming may have directly hindered the wider application and acceptance of this methodology. Recently, to fill this gap, the sources of uncertainty in emergy analysis were described and analytical and stochastic methods were put forward to estimate the uncertainty in unit emergy values (UEVs). However, the most common method used to determine UEVs is the emergy table-form model, and only a stochastic method (i.e., the Monte Carlo method) was provided to estimate the uncertainty of values calculated in this way. To simplify the determination of uncertainties in emergy analysis using table-form calculations, we introduced two analytical methods provided by the Guide to the Expression of Uncertainty in Measurement (GUM), i.e., the Variance method and the Taylor method, to estimate the uncertainty of emergy table-form calculations for two different types of data, and compared them with the stochastic method in two case studies. The results showed that, when replicate data are available at the system level, i.e., the same data on inputs and output are measured repeatedly in several independent systems, the Variance method is the simplest and most reliable method for determining the uncertainty of the model output, since it considers the underlying covariance of the inputs and requires no assumptions about the probability distributions of the inputs. However, when replicate data are only available at the subsystem level, i.e., repeat samples are measured on subsystems without specific correspondence between an output and a certain suite of inputs, the Taylor method will be a better option for calculating uncertainty, since it requires less information and is easier to understand and perform than the Monte Carlo method.     

Sensitivity of ecological models to their climate drivers: statistical ensembles for forcing.

Global and regional numerical models for terrestrial ecosystem dynamics require fine spatial resolution and temporally complete historical climate fields as input variables. However, because climate observations are unevenly spaced and have incomplete records, such fields need to be estimated. In addition, uncertainty in these fields associated with their estimation are rarely assessed. Ecological models are usually driven with a geostatistical model's mean estimate (kriging) of these fields without accounting for this uncertainty, much less evaluating such errors in terms of their propagation in ecological simulations. We introduce a Bayesian statistical framework to model climate observations to create spatially uniform and temporally complete fields, taking into account correlation in time and space, spatial heterogeneity, lack of normality, and uncertainty about all these factors. A key benefit of the Bayesian model is that it generates uncertainty measures for the generated fields. To demonstrate this method, we reconstruct historical monthly precipitation fields (a driver for ecological models) on a fine resolution grid for a climatically heterogeneous region in the western United States. The main goal of this work is to evaluate the sensitivity of ecological models to the uncertainty associated with prediction of their climate drivers. To assess their numerical sensitivity to predicted input variables, we generate a set of ecological model simulations run using an ensemble of different versions of the reconstructed fields. We construct such an ensemble by sampling from the posterior predictive distribution of the climate field. We demonstrate that the estimated prediction error of the climate field can be very high. We evaluate the importance of such errors in ecological model experiments using an ensemble of historical precipitation time series in simulations of grassland biogeochemical dynamics with an ecological numerical model, Century. We show how uncertainty in predicted precipitation fields is propagated into ecological model results and that this propagation had different modes. Depending on output variable, the response of model dynamics to uncertainty in inputs ranged from uncertainty in outputs that matched that of inputs to those that were muted or that were biased, as well as uncertainty that was persistent in time after input errors dropped.     

Choosing ecosystem service investments that are robust to uncertainty across multiple parameters

Info-gap decision theory facilitates decision making for problems in which uncertainty is large and probability distributions of uncertain variables are unknown. The info-gap framework allows the decision maker to maximize robustness to failure in the presence of uncertainty, where uncertainty is in the parameters of the model and failure is defined as the model output falling below some minimally acceptable performance threshold. Info-gap theory has found particular application to problems in conservation biology and ecological economics. In this study, we applied info-gap theory to an ecosystem services tradeoff case study in which a decision maker aiming to maximize ecosystem service investment returns must choose between two alternative land uses: native vegetation conservation or the establishment of an exotic timber plantation. The uncertain variables are the carbon price and the water price. With a "no-information" uncertainty model that assumes equal relative uncertainty across both variables, info-gap theory identifies a minimally acceptable reward threshold above which conservation is preferred, but below which plantation establishment is preferred. However, with an uncertainty model that allows the carbon price to be substantially more uncertain than the water price, conservation of native vegetation becomes an economically more robust investment option than establishing alien pine plantations. We explored the sensitivity of the results to the use of alternative uncertainty models, including asymmetric uncertainty in individual variables. We emphasize the general finding that the results of info-gap analyses can be sensitive to the choice of uncertainty model and that, therefore, future applications to ecological problems should be careful to incorporate all available qualitative and quantitative information relating to uncertainties or should at least justify the no-information uncertainty model.     

An evaluation of environmental factors affecting species distributions

Many different models can be built to explain the distributions of species. Often there is no single model that is clearly better than the alternatives, and this leads to uncertainty over which environmental factors are limiting species’ distributions. We investigated the support for different environmental factors by determining the drop in model performance when selected predictors were excluded from the model building process. We used a paired t-test over 37 plant species so that an environmental factor was only deemed significant if it consistently improved the results for multiple species. Geology and winter minimum temperatures were found to be the environmental factors with the most support, with a significant drop in model performance when either of these factors was excluded. However, there was less support for summer maximum temperature, as other environmental factors could combine to produce similar model performance. Our method of evaluating environmental factors using multiple species will not be capable of detecting predictors that are only important for one or two species, but it is difficult to distinguish these from spurious correlations. The strength of the method is that it increases inference for factors that consistently affect the distributions of many species. We discourage the assessment of models against predefined benchmarks, such as an area under the curve (AUC) of more than 0.7, as many alternative models for the same species produce similar results. Therefore, the benchmarks do not provide any indication of how the performance of the selected model compares to alternative models, and they provide weak inference to accept any selected model.     

From data and theory to environmental models and indices formation

A prerequisite for environmental indices is that they represent environmental pressure, and the state of, and impact on environmental conditions. In other words, they should capture as much as possible of the cause-effect chains they represent and relate pressure and effect to criteria of environmental quality. The approach proposed in the article attempts to link the pressure–state–impact–response framework of indicators to the integrated environmental model, based on the method of response function (MRF). The MRF allows to construct purposeful, credible models from data and prior knowledge or information. The data are usually time series observations of system inputs and outputs, and sometimes of internal states. The output of such models is presented with highly aggregated environmental indices, reflecting the main pressure–state–impact–response cause-effect chains. The proposed approach is illustrated with the example of soil erosion indices.     

Incorporating Uncertainty of Management Costs in Sensitivity Analyses of Matrix Population Models

The importance of accounting for economic costs when making environmental‐management decisions subject to resource constraints has been increasingly recognized in recent years. In contrast, uncertainty associated with such costs has often been ignored. We developed a method, on the basis of economic theory, that accounts for the uncertainty in population‐management decisions. We considered the case where, rather than taking fixed values, model parameters are random variables that represent the situation when parameters are not precisely known. Hence, the outcome is not precisely known either. Instead of maximizing the expected outcome, we maximized the probability of obtaining an outcome above a threshold of acceptability. We derived explicit analytical expressions for the optimal allocation and its associated probability, as a function of the threshold of acceptability, where the model parameters were distributed according to normal and uniform distributions. To illustrate our approach we revisited a previous study that incorporated cost‐efficiency analyses in management decisions that were based on perturbation analyses of matrix population models. Incorporating derivations from this study into our framework, we extended the model to address potential uncertainties. We then applied these results to 2 case studies: management of a Koala (Phascolarctos cinereus) population and conservation of an olive ridley sea turtle (Lepidochelys olivacea) population. For low aspirations, that is, when the threshold of acceptability is relatively low, the optimal strategy was obtained by diversifying the allocation of funds. Conversely, for high aspirations, the budget was directed toward management actions with the highest potential effect on the population. The exact optimal allocation was sensitive to the choice of uncertainty model. Our results highlight the importance of accounting for uncertainty when making decisions and suggest that more effort should be placed on understanding the distributional characteristics of such uncertainty. Our approach provides a tool to improve decision making.     

Learning habitat models for the diatom community in Lake Prespa

Habitat suitability modelling studies the influence of abiotic factors on the abundance or diversity of a given taxonomic group of organisms. In this work, we investigate the effect of the environmental conditions of Lake Prespa (Republic of Macedonia) on diatom communities. The data contain measurements of physical and chemical properties of the environment as well as the relative abundances of 116 diatom taxa. In addition, we create a separate dataset that contains information only about the top 10 most abundant diatoms. We use two machine learning techniques to model the data: regression trees and multi-target regression trees. We learn a regression tree for each taxon separately (from the top 10 most abundant) to identify the environmental conditions that influence the abundance of the given diatom taxon. We learn two multi-target regression trees: one for modelling the complete community and the other for the top 10 most abundant diatoms. The multi-target regression trees approach is able to detect the conditions that affect the structure of a diatom community (as compared to other approaches that can model only a single target variable). We interpret and compare the obtained models. The models present knowledge about the influence of metallic ions and nutrients on the structure of the diatom community, which is consistent with, but further extends existing expert knowledge.