Predicting Species Distributions Research Articles

Multispecies deep learning using citizen science data produces more informative plant community models

In the age of big data, scientific progress is fundamentally limited by our capacity to extract critical information. Here, we map fine-grained spatiotemporal distributions for thousands of species, using deep neural networks (DNNs) and ubiquitous citizen science data. Based on 6.7 M observations, we jointly model the distributions of 2477 plant species and species aggregates across Switzerland with an ensemble of DNNs built with different cost functions. We find that, compared to commonly-used approaches, multispecies DNNs predict species distributions and especially community composition more accurately. Moreover, their design allows investigation of understudied aspects of ecology. Including seasonal variations of observation probability explicitly allows approximating flowering phenology; reweighting predictions to mirror cover-abundance allows mapping potentially canopy-dominant tree species nationwide; and projecting DNNs into the future allows assessing how distributions, phenology, and dominance may change. Given their skill and their versatility, multispecies DNNs can refine our understanding of the distribution of plants and well-sampled taxa in general.

A novel hybrid model for species distribution prediction using neural networks and Grey Wolf Optimizer algorithm

Neural networks are frequently employed to model species distribution through backpropagation methods, known as backpropagation neural networks (BPNN). However, the complex structure of BPNN introduces parameter settings challenges, such as the determination of connection weights, which can affect the accuracy of model simulation. In this paper, we integrated the Grey Wolf Optimizer (GWO) algorithm, renowned for its excellent global search capacity and rapid convergence, to enhance the performance of BPNN. Then we obtained a novel hybrid algorithm, the Grey Wolf Optimizer algorithm optimized backpropagation neural networks algorithm (GNNA), designed for predicting species’ potential distribution. We also compared the GNNA with four prevalent species distribution models (SDMs), namely the generalized boosting model (GBM), generalized linear model (GLM), maximum entropy (MaxEnt), and random forest (RF). These models were evaluated using three evaluation metrics: the area under the receiver operating characteristic curve, Cohen’s kappa, and the true skill statistic, across 23 varied species. Additionally, we examined the predictive accuracy concerning spatial distribution. The results showed that the predictive performance of GNNA was significantly improved compared to BPNN, was significantly better than that of GLM and GBM, and was even comparable to that of MaxEnt and RF in predicting species distributions with small sample sizes. Furthermore, the GNNA demonstrates exceptional powers in forecasting the potential non-native distribution of invasive plant species.

Predicting suitable habitat for the endangered tree Ormosia microphylla in China

Climate change has significantly influenced the growth and distribution of plant species, particularly those with a narrow ecological niche. Understanding climate change impacts on the distribution and spatial pattern of endangered species can improve conservation strategies. The MaxEnt model is widely applied to predict species distribution and environmental tolerance based on occurrence data. This study investigated the suitable habitats of the endangered Ormosia microphylla in China and evaluated the importance of bioclimatic factors in shaping its distribution. Occurrence data and environmental variables were gleaned to construct the MaxEnt model, and the resulting suitable habitat maps were evaluated for accuracy. The results showed that the MaxEnt model had an excellent simulation quality (AUC = 0.962). The major environmental factors predicting the current distribution of O. microphylla were the mean diurnal range (bio2) and precipitation of the driest month (bio14). The current core potential distribution areas were concentrated in Guangxi, Fujian, Guizhou, Guangdong, and Hunan provinces in south China, demonstrating significant differences in their distribution areas. Our findings contribute to developing effective conservation and management measures for O. microphylla, addressing the critical need for reliable prediction of unfavorable impacts on the potential suitable habitats of the endangered species.

Seasonal variation in marine bird distribution in the northeast Pacific Ocean

Human activities have profound influences on marine ecosystems. Marine birds are particularly sensitive to these impacts and, given their ease of observation and diverse life histories, often represent good indicators of ecosystem health. Conserving marine birds and their ecosystems requires robust predictions of species distribution to help mitigate human disturbance in areas where large aggregations of diverse species occur. We modelled variation in marine bird species diversity (Shannon-Wiener Index) and taxonomic family level probability of occurrence to map the intensity and extent of highly diverse ‘hotspots’ in Canada’s Pacific Exclusive Economic Zone. To do so, we paired 20 yr of survey data from the North Pacific Pelagic Seabird Database (1997-2017) and remote sensing data describing marine conditions and local geography (sea surface temperature, chlorophyll a, bathymetry, distance to shore, and benthic substrate type). These data were used to illustrate how seasonality within years and the El Niño-Southern Oscillation (ENSO) across years influenced spatial patterns in diversity. Hotspots were most persistent in Hecate Strait, off the west coast of Vancouver Island, and surrounding the Scott Islands in most seasons. Changes in hotspot locations and intensity were observed across seasons and within season under varying ENSO conditions. Our results provide a template for mapping marine species distribution for the purpose of identifying hotspots of diversity, and thereby facilitate planning to minimize harmful impacts in highly diverse and dynamic systems.

Bias correction in species distribution models based on geographic and environmental characteristics

Correcting sampling bias in species distribution models (SDMs) is challenging. The difficulty lies in accurately identifying and quantifying bias and the scarcity of samples, which greatly impedes the implementation of bias correction. Current methods often adjust the distribution of presence or background points within geographic or environmental spaces to correct the sampling bias in probability estimation within SDMs. However, these methods may lead to information loss, rely on subjective assumptions, and often separate geography and environment when correcting for bias. This study proposes a novel and easily implementable method termed “aggregation background.” This method selects background data based on the aggregation degree of presence points in the geographic and environmental feature space, thereby approximating the representation and correction of sampling bias in the presence samples. We compared this new method with other prevalent sampling bias correction methods in the existing literature by analyzing ecological authenticity. Under varying biases and sample sizes, the aggregation background and geographic filtering methods achieved more accurate species distribution predictions compared to the target group background and other methods. Notably, when the sample size was small (≤70), the aggregation background was superior to that obtained using the geographic filtering method. These findings underscore the effectiveness of the aggregation background in improving bias correction using limited available presence sample data, without relying on assumptions about sampling bias. Our method provides a new approach for correcting complex unknown biases in SDMs.

Estimating species distribution from camera trap by‐catch data, using jaguarundi (Herpailurus yagouaroundi) as an example

AbstractAimPlanning conservation action requires accurate estimates of abundance and distribution of the target species. For many mammals, particularly those inhabiting tropical forests, there are insufficient data to assess their conservation status. We present a framework for predicting species distribution using jaguarundi (Herpailurus yagouaroundi), a poorly known felid for which basic information on abundance and distribution is lacking.LocationMesoamerica and South America.Time PeriodFrom 2003 to 2021.Taxa Herpailurus yagouaroundi. MethodsWe combined camera‐trap data from multiple sites and used an occupancy modelling framework accounting for imperfect detection to identify habitat associations and predict the range‐wide distribution of jaguarundis.ResultsOur model predicted that the probability of jaguarundi occupancy is positively associated with rugged terrain, herbaceous cover, and human night‐time light intensity. Jaguarundi occupancy was predicted to be higher where precipitation was less seasonal, and at intermediate levels of diurnal temperature range. Our camera data also revealed additional detections of jaguarundis beyond the current International Union for Conservation of Nature (IUCN) range distribution, including the Andean foothills of Colombia and Bolivia.Main ConclusionOccupancy was predicted to be low throughout much of Amazonian lowlands, a vast area at the centre of jaguarundi known range. Further work is required to investigate whether this area represents sub‐optimal conditions for the species. Overall, we estimate a crude global jaguarundi population of 35,000 to 230,000 individuals, covering 4,453,406 km2 of Meso‐ and South America at the 0.5 probability level of occupancy. Our current framework allows for an initially detailed, well‐informed species distribution that should be challenged and refined with improved habitat layers and additional records of jaguarundi detection. We encourage similar studies of lesser‐known mammals, pooling existing by‐catch data from the growing bank of camera‐trap surveys around the world.

Predicting the effects of climate and land use changes on small rattlesnakes in central Mexico: Insights for conservation planning

Anthropogenic changes pose a significant threat to global biodiversity, especially in reptiles. Mexico, renowned for its diverse reptile population, is experiencing habitat disturbance and fragmentation, endangering rattlesnakes in particular due to their specific habitat requirements and limited mobility. The loss of structural connectivity further increases the risk of extinction for reptiles. Our study focuses on predicting the distribution and connectivity of nine Crotalus species in the Trans-Mexican Volcanic Belt (TMVB), using environmental modeling. We used climate models (CNRM-CM5 and MPI-ESM-LR) for the year 2050 and Representative Concentration Pathways (RCP 85) alongside changes in vegetation cover and land use. The Maxent modeling method in ENMeval helped assess distribution patterns and to identify key environmental variables influencing these species’ distributions and niche overlap. We also conducted an analysis of the structural connectivity of the different Crotalus species. Our findings indicate reductions in suitable habitats due to changes in agriculture, urbanization, and forest cover. The Maxent models showed high accuracy in predicting species distributions. The most influential variables varied among species and included forest types and climatic factors. We observed limited connectivity among small Crotalus species, although some species exhibited greater connectivity than others. Future models suggest potential distribution reductions for all species, with C. armstrongi facing the greatest reduction (78%) according to the cn85 global climate model. Protected areas do not encompass the majority of potential distribution for these small rattlesnake species. These findings highlight the urgency of conservation efforts and the need to mitigate the impacts of future environmental changes on reptile populations.

Dealing with area‐to‐point spatial misalignment in species distribution models

Species distribution models (SDMs) are extensively used to estimate species–environment relationships (SERs) and predict species distribution across space and time. For this purpose, it is key to choose relevant spatial grains for predictor and response variables at the onset of the modelling process. However, environmental variables are often derived from large‐scale climate models at a grain that can be coarser than the one of the response variable. Such area‐to‐point spatial misalignment can bias estimates of SER and jeopardise the robustness of predictions. We used a virtual species approach, running simulations across different levels of area‐to‐point spatial misalignment to seek statistical solutions to this problem. We specifically compared accuracy of SER estimates and predictive performances, assessed across different degrees of spatial heterogeneity in environmental conditions, of three SDMs: a GLM, a spatial GLM and a Berkson error model (BEM) that accounts for fine‐grain environmental heterogeneity within coarse‐grain cells. Only the BEM accurately estimates SER from relatively coarse‐grain environmental data (up to 50 times coarser than the response grain), while the two GLMs provide flattened SER. However, all three models perform poorly when predicting from coarse‐grain data, particularly in environments that are more heterogeneous than the training conditions. Conversely, decreasing environmental heterogeneity relative to the training dataset reduces the predictive biases. Because predictions are made from covariate‐grain data, the BEM displays lower predictive performance than the two GLMs. Thus, standard model selection methods would fail to select the model that best estimates SERs (here, the BEM), which could lead to false interpretations about the environmental drivers of species distributions. Overall, we conclude that the BEM, because it can robustly estimate SER at the response grain, holds great promise to overcome area‐to‐point misalignment.

Ground‐dwelling mammal and bird diversity in the southern Annamites: Exploring complex habitat associations and the ghost of past hunting pressure

AbstractThe Langbian Plateau, a biodiversity hotspot in the southern Annamites of Viet Nam, has undergone extensive hunting pressure. However, the limited information on the effects of overexploitation on the current status and community composition of wildlife hinders effective conservation efforts, including the implementation of targeted patrols to reduce snaring. In this study, we conducted a camera‐trapping survey across the Langbian Plateau, consisting of a broadleaf evergreen and coniferous habitat mosaic. We recorded 46 ground‐dwelling mammals and birds, including several threatened Annamite endemics. Using multi‐species Royle‐Nichols model and landscape covariates, we found higher richness in broadleaf evergreen forest located in more remote and less rugged areas. We then used species responses to covariates to predict species distribution and identify high‐priority areas for conservation. Furthermore, we constructed diversity profiles that indicated higher biodiversity in broadleaf evergreen forest compared to the coniferous forest. Finally, we used a dissimilarity index to assess the level of defaunation, revealing 16% of the community had been lost, with higher levels of defaunation for threatened and larger‐sized species. Our findings provide insights into the status, distribution, and occurrence of the ground‐dwelling mammal and bird communities in the Langbian Plateau, and can help stakeholders design more effective conservation strategies to protect existing populations.

Modelling heterogeneity in the classification process in multi-species distribution models can improve predictive performance.

Species distribution models and maps from large-scale biodiversity data are necessary for conservation management. One current issue is that biodiversity data are prone to taxonomic misclassifications. Methods to account for these misclassifications in multi-species distribution models have assumed that the classification probabilities are constant throughout the study. In reality, classification probabilities are likely to vary with several covariates. Failure to account for such heterogeneity can lead to biased prediction of species distributions. Here, we present a general multi-species distribution model that accounts for heterogeneity in the classification process. The proposed model assumes a multinomial generalised linear model for the classification confusion matrix. We compare the performance of the heterogeneous classification model to that of the homogeneous classification model by assessing how well they estimate the parameters in the model and their predictive performance on hold-out samples. We applied the model to gull data from Norway, Denmark and Finland, obtained from the Global Biodiversity Information Facility. Our simulation study showed that accounting for heterogeneity in the classification process increased the precision of true species' identity predictions by 30% and accuracy and recall by 6%. Since all the models in this study accounted for misclassification of some sort, there was no significant effect of accounting for heterogeneity in the classification process on the inference about the ecological process. Applying the model framework to the gull dataset did not improve the predictive performance between the homogeneous and heterogeneous models (with parametric distributions) due to the smaller misclassified sample sizes. However, when machine learning predictive scores were used as weights to inform the species distribution models about the classification process, the precision increased by 70%. We recommend multiple multinomial regression to be used to model the variation in the classification process when the data contains relatively larger misclassified samples. Machine learning prediction scores should be used when the data contains relatively smaller misclassified samples.

Modelling complex spatial–temporal drivers of habitat suitability for an imperilled stream fish

Fish populations rely on complex environmental conditions involving physical, chemical, and biological factors. Understanding the factors that control population persistence and productivity is essential for species management. We assessed the distribution and associated habitat features of a species at risk in Canada, Silver Shiner (Notropis photogenis), within Sixteen Mile Creek, a tributary of Lake Ontario. Using random forest models, we quantified a range of ecological factors (n = 25) to estimate habitat associations for sampled populations and life stages (juvenile, adult). A complex set of ecological factors were informative predictors of Silver Shiner distribution, including physical (stream morphology, water velocity, substrate type), and biological (aquatic and riparian vegetation) conditions. Juveniles were less responsive to habitat conditions but exhibited high seasonal variability in occurrence. Adults were most common in stream sections with greater than 0.5 m depth and stream velocity less than 0.6 m/s, and areas without silt substrate. Broadly, the models predicted Silver Shiner distribution with 68–92% accuracy in non-training data. Our findings describe the habitat conditions that Silver Shiner currently occupies in an urban drainage, which may serve as a point of reference for habitat protection and restoration. Further, predictive species distribution models can serve to identify habitat for further monitoring and restoration.

Evaluating trade-offs in spatial versus temporal replication when estimating avian community composition and predicting species distributions

Evaluating trade-offs in spatial versus temporal replication when estimating avian community composition and predicting species distributions

Context matters in ecological forecasting: Lessons in predicting species distributions.

Forecasting the future state of a species is a tricky process, as there are numerous hidden factors that influence species trajectories in addition to the obvious unknowns about the future state of the planet. We echo the guidance of Clare et al. (2024) to use near‐term and long‐term forecasting in complementary ways. Near‐term forecasts can be used to guide specific management and conservation actions, which can be updated as new data and evidence are collected. Long‐term forecasts can be used to characterize uncertainty further into the future, which can help guide longstanding conservation planning and legislative actions that are based on such uncertainty in possible future outcomes.

Is the GBIF appropriate for use as input in models of predicting species distributions? Study from the Czech Republic

Questions concerning species diversity have attracted ecologists and biogeographers for over a century, mainly because the diversity of life on Earth is in rapid decline, which is expected to continue in the future. One of the most important current database on species distribution data is the Global Biodiversity Information Facility (GBIF), which contains more than 2 billion occurrences for all organisms, and this number is continuously increasing with the addition of new data and by combining with other applications. Such data also exist in several national databases, most of which are unfortunately often not freely available and not included in GBIF. We suspected that the national databases, mostly professionally maintained by governmental organisations, may be more comprehensive than GBIF, which is not centrally organised and therefore the national databases may give more accurate predictions than GBIF. To test our assumptions, we have compared: (i) the amount of data included in the Czech database called Nálezová databáze ochrany přírody (NDOP, Discovery database of nature protection) with the amount of data in GBIF after its restriction to the Czech Republic, and (ii) the overlap of the predictions of species distributions for the Czech Republic, based on these two databases. We have used the family Orchidaceae as a model group. We found that: (i) there is a significantly larger number of records per studied region (Czech Republic) in NDOP, compared with GBIF, and (ii) the predictions of Maxent based on orchid records in NDOP are overlapping to a great degree with the predictions based on data based on orchid records in GBIF. Bearing in mind these results, we suggest that if only one database is available for the region studied, we must use this one. If more databases are available for the region studied, we should use the database containing most locations (usually some of the local ones, like NDOP), because using more locations implies larger significance of predictions of species distributions.

Incorporating eco-evolutionary information into species distribution models provides comprehensive predictions of species range shifts under climate change

As climate changes increasingly influence species distributions, ecosystem functions, and biodiversity, the urgency to understand how species' ranges shift under those changes is great. Species distribution models (SDMs) are vital approaches that can predict species distributions under changing climates. However, SDMs based on the species' current occurrences may underestimate the species' climatic tolerances. Integrating species' realized niches at different periods, also known as multi-temporal calibration, can provide an estimation closer to its fundamental niche. Based on this, we further proposed an integrated framework that combines eco-evolutionary data and SDMs (phylogenetically-informed SDMs) to provide comprehensive predictions of species range shifts under climate change. To evaluate our approach's performance, we applied it to a group of related species, the Chrysanthemum zawadskii species complex (Anthemidae, Asteracee). First, we investigated the niche differentiation between species and intraspecific lineages of the complex and estimated their rates of niche evolution. Next, using both standard SDMs and our phylogenetically-informed SDMs, we generated predictions of suitability areas for all species and lineages and compared the results. Finally, we reconstructed the historical range dynamics for the species of this complex. Our results showed that the species and intraspecific lineages of the complex had varying degrees of niche differentiation and different rates of niche evolution. Lineage-level SDMs can provide more realistic predictions for species with intraspecific differentiation than species-level models can. The phylogenetically-informed SDMs provided more complete environmental envelopes and predicted broader potential distributions for all species than the standard SDMs did. Range dynamics varied among the species that have different rates of niche evolution. Our framework integrating eco-evolutionary data and SDMs contributes to a better understanding of the species' responses to climate change and can help to make more targeted conservation efforts for the target species under climate change, particularly for rare species.

XGBOOST IN ENVIRONMENTAL ECOLOGY: A POWERFUL TOOL FOR SUSTAINABLE INSIGHTS

Environmental ecology stands at the forefront of understanding and addressing the challenges posed by a rapidly changing world. In this context, machine learning, particularly the XGBoost algorithm, has emerged as a pivotal tool, offering unparalleled accuracy and adaptability. This article delves into the origins and workings of XGBoost, highlighting its applications in predicting species distributions, assessing habitat suitability, and modeling climate change impacts. While the benefits of XGBoost, such as high predictive power and robustness to noisy data, are emphasized, the article also sheds light on potential challenges like overfitting and interpretability. The conclusion underscores the importance of a holistic approach, combining domain knowledge with algorithmic prowess, to harness the full potential of XGBoost in environmental ecology.

Predicting invasive species distributions using incremental ensemble-based pseudo-labeling

Control strategies for preventing the spread of invasive species require their accurate geographical distribution. Species distribution models (SDMs) that can predict potential habitats of invasive species and thereby derive habitat suitability maps have become a valuable tool for supporting regulatory strategies. To date, machine learning (ML)-based approaches have outperformed profile and statistical-based approaches in terms of species distribution prediction accuracy. However, ML-based approaches often suffer from poor predictive performance when there is insufficient labeled data. Recently, pseudo-labeling (PL), a semi-supervised learning method, has been proposed to alleviate this problem, but pseudo-labels generated using a single-teacher model with very few labeled data points are generally biased and not suitable for training SDMs. In this paper, we propose a novel prediction scheme for invasive species distributions using incremental ensemble-based PL (SDP-EPL). We first build an ensemble-based teacher using multiple conventional SDMs and then incrementally construct a training set, starting with very few labeled data points, by repeating the following process: (i) generating pseudo-labels for unlabeled data using the teacher model, (ii) appending the pseudo-labeled data representing high or low habitat suitability to the training set, and (iii) training the teacher model using the updated training set. We then train a student SDM using the training set. Based on extensive experiments using citizen science datasets for three species, we show that the proposed scheme outperforms other commonly used SDMs in terms of diverse evaluation metrics and achieves performance improvements of up to 14.61% and 5.45% compared to the baseline and state-of-the-art models, respectively. We also demonstrate the effectiveness of the ensemble teacher model and incremental labeling in terms of predictive accuracy.

Habitat selection of resident and non-resident gray wolves: implications for habitat connectivity

Habitat selection studies facilitate assessing and predicting species distributions and habitat connectivity, but habitat selection can vary temporally and among individuals, which is often ignored. We used GPS telemetry data from 96 Gray wolves (Canis lupus) in the western Great Lakes region of the USA to assess differences in habitat selection while wolves exhibited resident (territorial) or non-resident (dispersing or floating) movements and discuss implications for habitat connectivity. We used a step-selection function (SSF) to assess habitat selection by wolves exhibiting resident or non-resident movements, and modeled circuit connectivity throughout the western Great Lakes region. Wolves selected for natural land cover and against areas with high road densities, with no differences in selection among wolves when resident, dispersing, or floating. Similar habitat selection between resident and non-resident wolves may be due to similarity in environmental conditions, when non-resident movements occur largely within established wolf range rather than near the periphery or beyond the species range. Alternatively, non-resident wolves may travel through occupied territories because higher food availability or lower human disturbance outweighs risks posed by conspecifics. Finally, an absence of differences in habitat selection between resident and non-resident wolf movements may be due to other unknown reasons. We recommend considering context-dependency when evaluating differences in movements and habitat use between resident and non-resident individuals. Our results also provide independent validation of a previous species distribution model and connectivity analysis suggesting most potential wolf habitat in the western Great Lakes region is occupied, with limited connectivity to unoccupied habitat.

Predicting the spatial expansion of an animal population with presence-only data.

Predictive models can improve the efficiency of wildlife management by guiding actions at the local, landscape and regional scales. In recent decades, a vast range of modelling techniques have been developed to predict species distributions and patterns of population spread. However, data limitations often constrain the precision and biological realism of models, which make them less useful for supporting decision-making. Complex models can also be challenging to evaluate, and the results are often difficult to interpret for wildlife management practitioners. There is therefore a need to develop techniques that are appropriately robust, but also accessible to a range of end users. We developed a hybrid species distribution model that utilises commonly available presence-only distribution data and minimal demographic information to predict the spread of roe deer (Capreolus caprelous) in Great Britain. We take a novel approach to representing the environment in the model by constraining the size of habitat patches to the home-range area of an individual. Population dynamics are then simplified to a set of generic rules describing patch occupancy. The model is constructed and evaluated using data from a populated region (England and Scotland) and applied to predict regional-scale patterns of spread in a novel region (Wales). It is used to forecast the relative timing of colonisation events and identify important areas for targeted surveillance and management. The study demonstrates the utility of presence-only data for predicting the spread of animal species and describes a method of reducing model complexity while retaining important environmental detail and biological realism. Our modelling approach provides a much-needed opportunity for users without specialist expertise in computer coding to leverage limited data and make robust, easily interpretable predictions of spread to inform proactive population management.

The carbon balance of plants: economics, optimization, and trait spectra in a historical perspective.

Over fifty years have passed since the publication of Harold Mooney's formative paper, "The Carbon Balance of Plants" on pages 315-346 of Volume 3 (1972) of Annual Review of Ecology and Systematics. Arguably, the conceptual framework presented in that paper, and the work by Mooney and his students leading up to the paper, provided the foundational principles from which core disciplines emerged in plant economic theory, functional trait theory and, more generally, plant physiological ecology. Here, we revisit the primary impacts of those early discoveries to understand how researchers constructed major concepts in our understanding of plant adaptations, and where those concepts are likely to take us in the near future. The discipline of functional trait ecology, which is rooted in the principles of evolutionary and economic optimization, has captured the imagination of the plant physiological ecology research community, though its emphasis has shifted toward predicting species distributions and ecological roles across resource gradients. In the face of 'big-data' research pursuits that are revealing trait expression patterns at the cellular level and mass and energy exchange patterns at the planetary scale, an opportunity exists to reconnect the principles of plant carbon balance and evolutionary optimization with trait origins at the genetic and cellular scales and trait impacts at the global scale.