Outlier detection methods are still effective even using virtual species created with the probabilistic approach

Liu et al. (Journal of Biogeography, 2018, 45 :164–176) presented an approach to detect outliers in species distribution data by developing virtual species created using the threshold approach. Meynard et al. (Journal of biogeography, 2019, 46 :2141–2144) raised concerns about this approach stating that ‘using a probabilistic approach … may significantly change results’. Here we provide a new series of simulations using the two approaches and demonstrate that the outlier detection approach based on pseudo species distribution models was still effective when using the probabilistic approach, although the detection rate was lower than when using the threshold approach.     

In search of relevant predictors for marine species distribution modelling using the MarineSPEED benchmark dataset

Aim

Ideally, datasets for species distribution modelling (SDM) contain evenly sampled records covering the entire distribution of the species, confirmed absences and auxiliary ecophysiological data allowing informed decisions on relevant predictors. Unfortunately, these criteria are rarely met for marine organisms for which distributions are too often only scantly characterized and absences generally not recorded. Here, we investigate predictor relevance as a function of modelling algorithms and settings for a global dataset of marine species.

Location

Global marine.

Methods

We selected well‐studied and identifiable species from all major marine taxonomic groups. Distribution records were compiled from public sources (e.g., OBIS, GBIF, Reef Life Survey) and linked to environmental data from Bio‐ORACLE and MARSPEC. Using this dataset, predictor relevance was analysed under different variations of modelling algorithms, numbers of predictor variables, cross‐validation strategies, sampling bias mitigation methods, evaluation methods and ranking methods. SDMs for all combinations of predictors from eight correlation groups were fitted and ranked, from which the top five predictors were selected as the most relevant.

Results

We collected two million distribution records from 514 species across 18 phyla. Mean sea surface temperature and calcite are, respectively, the most relevant and irrelevant predictors. A less clear pattern was derived from the other predictors. The biggest differences in predictor relevance were induced by varying the number of predictors, the modelling algorithm and the sample selection bias correction. The distribution data and associated environmental data are made available through the R package marinespeed and at http://marinespeed.org .

Main conclusions

While temperature is a relevant predictor of global marine species distributions, considerable variation in predictor relevance is linked to the SDM set‐up. We promote the usage of a standardized benchmark dataset (MarineSPEED) for methodological SDM studies.

     

An Integrated Approach for Identifying Wrongly Labelled Samples When Performing Classification in Microarray Data

Background

Using hybrid approach for gene selection and classification is common as results obtained are generally better than performing the two tasks independently. Yet, for some microarray datasets, both classification accuracy and stability of gene sets obtained still have rooms for improvement. This may be due to the presence of samples with wrong class labels (i.e. outliers). Outlier detection algorithms proposed so far are either not suitable for microarray data, or only solve the outlier detection problem on their own.

Results

We tackle the outlier detection problem based on a previously proposed Multiple-Filter-Multiple-Wrapper (MFMW) model, which was demonstrated to yield promising results when compared to other hybrid approaches (Leung and Hung, 2010). To incorporate outlier detection and overcome limitations of the existing MFMW model, three new features are introduced in our proposed MFMW-outlier approach: 1) an unbiased external Leave-One-Out Cross-Validation framework is developed to replace internal cross-validation in the previous MFMW model; 2) wrongly labeled samples are identified within the MFMW-outlier model; and 3) a stable set of genes is selected using an L1-norm SVM that removes any redundant genes present. Six binary-class microarray datasets were tested. Comparing with outlier detection studies on the same datasets, MFMW-outlier could detect all the outliers found in the original paper (for which the data was provided for analysis), and the genes selected after outlier removal were proven to have biological relevance. We also compared MFMW-outlier with PRAPIV (Zhang et al., 2006) based on same synthetic datasets. MFMW-outlier gave better average precision and recall values on three different settings. Lastly, artificially flipped microarray datasets were created by removing our detected outliers and flipping some of the remaining samples'' labels. Almost all the ‘wrong’ (artificially flipped) samples were detected, suggesting that MFMW-outlier was sufficiently powerful to detect outliers in high-dimensional microarray datasets.     

Predicting distributions,habitat preferences and associated conservation implications for a genus of rare fishes,seahorses (Hippocampus spp.)

Aim

To identify useful sources of species data and appropriate habitat variables for species distribution modelling on rare species, with seahorses as an example, deriving ecological knowledge and spatially explicit maps to advance global seahorse conservation.

Location

The shallow seas.

Methods

We applied a typical species distribution model (SDM), maximum entropy, to examine the utility of (1) two versions of habitat variables (habitat occurrences vs. proximity to habitats) and (2) three sources of species data: quality research‐grade (RG) data, quality‐unknown citizen science (CS) and museum‐collection (MC) data. We used the best combinations of species data and habitat variables to predict distributions and estimate species–habitat relations and threatened status for seahorse species.

Results

We demonstrated that using “proximity to habitats” and integrating all species datasets (RG, CS and MC) derived models with the highest accuracies among all dataset variations. Based on this finding, we derived reliable models for 33 species. Our models suggested that only 0.4% of potential seahorse range was suitable to more than three species together; seahorse biogeographic epicentres were mainly in the Philippines; and proximity to sponges was an important habitat variable. We found that 12 “Data Deficient” species might be threatened based on our predictions according to IUCN criteria.

Main conclusions

We highlight that using proper habitat variables (e.g., proximity to habitats) is critical to determine distributions and key habitats for low‐mobility animals; collating and integrating quality‐unknown occurrences (e.g., CS and MC) with quality research data are meaningful for building SDMs for rare species. We encourage the application of SDMs to estimate area of occupancy for rare organisms to facilitate their conservation status assessment.

     

The uncertain nature of absences and their importance in species distribution modelling

Species distribution models (SDM) are commonly used to obtain hypotheses on either the realized or the potential distribution of species. The reliability and meaning of these hypotheses depends on the kind of absences included in the training data, the variables used as predictors and the methods employed to parameterize the models. Information about the absence of species from certain localities is usually lacking, so pseudo‐absences are often incorporated to the training data. We explore the effect of using different kinds of pseudo‐absences on SDM results. To do this, we use presence information on Aphodius bonvouloiri, a dung beetle species of well‐known distribution. We incorporate different types of pseudo‐absences to create different sets of training data that account for absences of methodological (i.e. false absences), contingent and environmental origin. We used these datasets to calibrate SDMs with GAMs as modelling technique and climatic variables as predictors, and compare these results with geographical representations of the potential and realized distribution of the species created independently. Our results confirm the importance of the kind of absences in determining the aspect of species distribution identified through SDM. Estimations of the potential distribution require absences located farther apart in the geographic and/or environmental space than estimations of the realized distribution. Methodological absences produce overall bad models, and absences that are too far from the presence points in either the environmental or the geographic space may not be informative, yielding important overestimations. GLMs and Artificial Neural Networks yielded similar results. Synthetic discrimination measures such as the Area Under the Receiver Characteristic Curve (AUC) must be interpreted with caution, as they can produce misleading comparative results. Instead, the joint examination of ommission and comission errors provides a better understanding of the reliability of SDM results.     

Without quality presence–absence data,discrimination metrics such as TSS can be misleading measures of model performance

The discriminating capacity (i.e. ability to correctly classify presences and absences) of species distribution models (SDMs) is commonly evaluated with metrics such as the area under the receiving operating characteristic curve (AUC), the Kappa statistic and the true skill statistic (TSS). AUC and Kappa have been repeatedly criticized, but TSS has fared relatively well since its introduction, mainly because it has been considered as independent of prevalence. In addition, discrimination metrics have been contested because they should be calculated on presence–absence data, but are often used on presence‐only or presence‐background data. Here, we investigate TSS and an alternative set of metrics—similarity indices, also known as F‐measures. We first show that even in ideal conditions (i.e. perfectly random presence–absence sampling), TSS can be misleading because of its dependence on prevalence, whereas similarity/F‐measures provide adequate estimations of model discrimination capacity. Second, we show that in real‐world situations where sample prevalence is different from true species prevalence (i.e. biased sampling or presence‐pseudoabsence), no discrimination capacity metric provides adequate estimation of model discrimination capacity, including metrics specifically designed for modelling with presence‐pseudoabsence data. Our conclusions are twofold. First, they unequivocally impel SDM users to understand the potential shortcomings of discrimination metrics when quality presence–absence data are lacking, and we recommend obtaining such data. Second, in the specific case of virtual species, which are increasingly used to develop and test SDM methodologies, we strongly recommend the use of similarity/F‐measures, which were not biased by prevalence, contrary to TSS.     

Patterns of species richness hotspots and estimates of their protection are sensitive to spatial resolution

Aim

Species richness is a measure of biodiversity often used in spatial conservation assessments and mapped by summing species distribution maps. Commission errors inherent those maps influence richness patterns and conservation assessments. We sought to further the understanding of the sensitivity of hotspot delineation methods and conservation assessments to commission errors, and choice of threshold for hotspot delineation.

Location

United States.

Methods

We created range maps and 30‐m and 1‐km resolution habitat maps for terrestrial vertebrates in the United States and generated species richness maps with each dataset. With the richness maps and the GAP Protected Areas Dataset, we created species richness hotspot maps and calculated the proportion of hotspots within protected areas; calculating protection under a range of thresholds for defining hotspots. Our method allowed us to identify the influence of commission errors by comparing hotspot maps.

Results

Commission errors from coarse spatial grain data and lack of porosity in the range data inflated richness estimates and altered their spatial patterns. Coincidence of hotspots from different data types was low. The 30‐m hotspots were spatially dispersed, and some were very long distances from the hotspots mapped with coarser data. Estimates of protection were low for each of the taxa. The relationship between estimates of hotspot protection and threshold choice was nonlinear and inconsistent among data types (habitat and range) and grain size (30‐m and 1‐km).

Main conclusions

Coarse mapping methods and grain sizes can introduce commission errors into species distribution data that could result in misidentifications of the regions where hotspots occur and affect estimates of hotspot protection. Hotspot conservation assessments are also sensitive to choice of threshold for hotspot delineation. There is value in developing species distribution maps with high resolution and low rates of commission error for conservation assessments.     

Causal inference and large-scale expert validation shed light on the drivers of SDM accuracy and variance

Aim

To develop a causal understanding of the drivers of Species distribution model (SDM) performance.

Location

United Kingdom (UK).

Methods

We measured the accuracy and variance of SDMs fitted for 518 species of invertebrate and plant in the UK. Our measure of variance reflects variation among replicate model fits, and taxon experts assessed model accuracy. Using directed acyclic graphs, we developed a causal model depicting plausible effects of explanatory variables (e.g. species' prevalence, sample size) on SDM accuracy and variance and quantified those effects using a multilevel piecewise path model.

Results

According to our model, sample size and niche completeness (proportion of a species' niche covered by sampling) directly affect SDM accuracy and variance. Prevalence and range completeness have indirect effects mediated by sample size. Challenging conventional wisdom, we found that the effect of prevalence on SDM accuracy is positive. This reflects the facts that sample size has a positive effect on accuracy and larger sample sizes are possible for widespread species. It is possible, however, that the omission of an unobserved confounder biased this effect. Previous studies, which reported negative correlations between prevalence and SDM accuracy, conditioned on sample size.

Main conclusions

Our model explicates the causal basis of previously reported correlations between SDM performance and species/data characteristics. It also suggests that niche completeness has similarly large effects on SDM accuracy and variance as sample size. Analysts should consider niche completeness, or proxies thereof, in addition to sample size when deciding whether modelling is worthwhile.     

Beyond climate control on species range: The importance of soil data to predict distribution of Amazonian plant species

Aim

To evaluate the relative importance of climatic versus soil data when predicting species distributions for Amazonian plants and to gain understanding of potential range shifts under climate change.

Location

Amazon rain forest.

Methods

We produced species distribution models (SDM) at 5‐km spatial resolution for 42 plant species (trees, palms, lianas, monocot herbs and ferns) using species occurrence data from herbarium records and plot‐based inventories. We modelled species distribution with Bayesian logistic regression using either climate data only, soil data only or climate and soil data together to estimate their relative predictive powers. For areas defined as unsuitable to species occurrence, we mapped the difference between the suitability predictions obtained with climate‐only versus soil‐only models to identify regions where climate and soil might restrict species ranges independently or jointly.

Results

For 40 out of the 42 species, the best models included both climate and soil predictors. The models including only soil predictors performed better than the models including only climate predictors, but we still detected a drought‐sensitive response for most of the species. Edaphic conditions were predicted to restrict species occurrence in the centre, the north‐west and in the north‐east of Amazonia, while the climatic conditions were identified as the restricting factor in the eastern Amazonia, at the border of Roraima and Venezuela and in the Andean foothills.

Main conclusions

Our results revealed that soil data are a more important predictor than climate of plant species range in Amazonia. The strong control of species ranges by edaphic features might reduce species’ abilities to track suitable climate conditions under a drought‐increase scenario. Future challenges are to improve the quality of soil data and couple them with process‐based models to better predict species range dynamics under climate change.     

Bio‐ORACLE: a global environmental dataset for marine species distribution modelling

Aim The oceans harbour a great diversity of organisms whose distribution and ecological preferences are often poorly understood. Species distribution modelling (SDM) could improve our knowledge and inform marine ecosystem management and conservation. Although marine environmental data are available from various sources, there are currently no user‐friendly, high‐resolution global datasets designed for SDM applications. This study aims to fill this gap by assembling a comprehensive, uniform, high‐resolution and readily usable package of global environmental rasters. Location Global, marine. Methods We compiled global coverage data, e.g. satellite‐based and in situ measured data, representing various aspects of the marine environment relevant for species distributions. Rasters were assembled at a resolution of 5 arcmin (c. 9.2 km) and a uniform landmask was applied. The utility of the dataset was evaluated by maximum entropy SDM of the invasive seaweed Codium fragile ssp. fragile. Results We present Bio‐ORACLE (ocean rasters for analysis of climate and environment), a global dataset consisting of 23 geophysical, biotic and climate rasters. This user‐friendly data package for marine species distribution modelling is available for download at http://www.bio‐oracle.ugent.be . The high predictive power of the distribution model of C. fragile ssp. fragile clearly illustrates the potential of the data package for SDM of shallow‐water marine organisms. Main conclusions The availability of this global environmental data package has the potential to stimulate marine SDM. The high predictive success of the presence‐only model of a notorious invasive seaweed shows that the information contained in Bio‐ORACLE can be informative about marine distributions and permits building highly accurate species distribution models.     

Testing the utility of species distribution modelling using Random Forests for a species in decline

Habitat suitability estimates derived from species distribution models (SDMs) are increasingly used to guide management of threatened species. Poorly estimating species’ ranges can lead to underestimation of threatened status, undervaluing of remaining habitat and misdirection of conservation funding. We aimed to evaluate the utility of a SDM, similar to the models used to inform government regulation of habitat in our study region, in estimating the contemporary distribution of a threatened and declining species. We developed a presence‐only SDM for the endangered New Holland Mouse (Pseudomys novaehollandiae) across Victoria, Australia. We conducted extensive camera trap surveys across model‐predicted and expert‐selected areas to generate an independent data set for use in evaluating the model, determining confidence in absence data from non‐detection sites with occupancy and detectability modelling. We assessed the predictive capacity of the model at thresholds based on (1) sum of sensitivity and specificity (SSS), and (2) the lowest presence threshold (LPT; i.e. the lowest non‐zero model‐predicted habitat suitability value at which we detected the species). We detected P. novaehollandiae at 40 of 472 surveyed sites, with strong support for the species’ probable absence from non‐detection sites. Based on our post hoc optimised SSS threshold of the SDM, 25% of our detection sites were falsely predicted as non‐suitable habitat and 75% of sites predicted as suitable habitat did not contain the species at the time of our survey. One occupied site had a model‐predicted suitability value of zero, and at the LPT, 88% of sites predicted as suitable habitat did not contain the species at the time of our survey. Our findings demonstrate that application of generic SDMs in both regulatory and investment contexts should be tempered by considering their limitations and currency. Further, we recommend engaging species experts in the extrapolation and application of SDM outputs.     

Improved spatial estimates of climate predict patchier species distributions

Aim

Correlative species distribution models (SDMs) combined with spatial layers of climate and species' localities represent a frequently utilized and rapid method for generating spatial estimates of species distributions. However, an SDM is only as accurate as the inputs upon which it is based. Current best‐practice climate layers commonly utilized in SDM (e.g. ANUCLIM) are frequently inaccurate and biased spatially. Here, we statistically downscale 30 years of existing spatial weather estimates against empirical weather data and spatial layers of topography and vegetation to produce highly accurate spatial layers of weather. We proceed to demonstrate the effect of inaccurately quantified spatial data on SDM outcomes.

Location

The Australian Wet Tropics.

Methods

We use Boosted Regression Trees (BRTs) to generate 30 years of spatial estimates of daily maximum and minimum temperature for the study region and aggregate the resultant weather layers into ‘accuCLIM’ climate summaries, comparable with those generated by current best‐practice climate layers. We proceed to generate for seven species of rainforest skink comparable SDMs within species; one model based on ANUCLIM climate estimates and another based on accuCLIM climate estimates.

Results

Boosted Regression Trees weather layers are more accurate with respect to empirically measured temperature, particularly for maximum temperature, when compared to current best‐practice weather layers. ANUCLIM climate layers are least accurate in heavily forested upland regions, frequently over‐predicting empirical mean maximum temperature by as much as 7°. Distributions of the focal species as predicted by accuCLIM were more fragmented and contained less core distributional area.

Conclusion

Combined these results reveal a source of bias in climate‐based SDMs and indicate a solution in the form of statistical downscaling. This technique will allow researchers to produce fine‐grained, ground‐truthed spatial estimates of weather based on existing estimates, which can be aggregated in novel ways, and applied to correlative or process‐based modelling techniques.

     

Assessing habitat loss and fragmentation and their effects on population viability of forest specialist birds: Linking biogeographical and population approaches

Aim

Biogeographic approaches usually have been developed apart from population ecology, resulting in predictive models without key parameters needed to account for reproductive and behavioural limitations on dispersal. Our aim was to incorporate fully spatially explicit population traits into a classic species distribution model (SDM) using Geographic Information Systems (GIS), aiming at conservation purposes.

Location

Southern South America.

Methods

Our analysis incorporates the effects of habitat loss and fragmentation on population viability and therefore provides insights into how much spatially explicit population traits can improve the SDM prediction of habitable habitat. We utilized a well‐studied focal endemic bird of South American temperate rainforests (Scelorchilus rubecula). First, at a large scale, we assessed the historical extent habitat based on climate envelopes in an SDM. Second, we used a land cover change analysis at a regional scale to account for recent habitat loss and fragmentation. Third, we used empirically derived criteria to predict population responses to fragmented forest landscapes to identify actual losses of habitat and population. Then we selected three sites of high conservation value in southern Chile and applied our population model. Finally, we discuss the degree to which spatially explicit population traits can improve the SDM output without intervening in the modelling process itself.

Results

We found a historical habitat loss of 39.12% and an additional forest cover loss of 3.03% during 2000–2014; the latter occurred with a high degree of fragmentation, reducing the overall estimation of (1) carrying capacity by ?82.4%, ?33.1% and ?45.1% and (2) estimated number of pairs on viable populations by ?84.1%, ?33.0% and ?54.6% on the three selected sites.

Main conclusion

We conclude that our approach sharpened the SDM prediction on environmental suitability by 54.4%, adjusting the habitable area by adding population parameters through GIS, and allowing to incorporate other phenomena as fragmentation and habitat loss.

     

Nondetection sampling bias in marked presence‐only data

相似文献   

Modelling invasive alien species distributions from digital biodiversity atlases. Model upscaling as a means of reconciling data at different scales

Aim

There is a wealth of information on species occurrences in biodiversity data banks, albeit presence‐only, biased and scarce at fine resolutions. Moreover, fine‐resolution species maps are required in biodiversity conservation. New techniques for dealing with this kind of data have been reported to perform well. These fine‐resolution maps would be more robust if they could explain data at coarser resolutions at which species distributions are well represented. We present a new methodology for testing this hypothesis and apply it to invasive alien species (IAS).

Location

Catalonia, Spain.

Methods

We used species presence records from the Biodiversity data bank of Catalonia to model the distribution of ten IAS which, according to some recent studies, achieve their maximum distribution in the study area. To overcome problems inherent with the data, we prepared different correction treatments: three for dealing with bias and five for autocorrelation. We used the MaxEnt algorithm to generate models at 1‐km resolution for each species and treatment. Acceptable models were upscaled to 10 km and validated against independent 10 km occurrence data.

Results

Of a total of 150 models, 20 gave acceptable results at 1‐km resolution and 12 passed the cross‐scale validation test. No apparent pattern emerged, which could serve as a guide on modelling. Only four species gave models that also explained the distribution at the coarser scale.

Main conclusions

Although some techniques may apparently deliver good distribution maps for species with scarce and biased data, they need to be taken with caution. When good independent data at a coarser scale are available, cross‐scale validation can help to produce more reliable and robust maps. When no independent data are available for validation, however, new data gathering field surveys may be the only option if reliable fine‐scale resolution maps are needed.     

Species interactions weakly modify climate‐induced tree co‐occurrence patterns

Aims

Species distributions are hypothesized to be underlain by a complex association of processes that span multiple spatial scales including biotic interactions, dispersal limitation, fine‐scale resource gradients and climate. Species disequilibrium with climate may reflect the effects of non‐climatic processes on species distributions, yet distribution models have rarely directly considered non‐climatic processes. Here, we use a Joint Species Distribution Model (JSDM) to investigate the influence of non‐climatic factors on species co‐occurrence patterns and to directly quantify the relative influences of climate and alternative processes that may generate correlated responses in species distributions, such as species interactions, on tree co‐occurrence patterns.

Location

US Rocky Mountains.

Methods

We apply a Bayesian JSDM to simultaneously model the co‐occurrence patterns of ten dominant tree species across the Rocky Mountains, and evaluate climatic and residual correlations from the fitted model to determine the relative contribution of each component to observed co‐occurrence patterns. We also evaluate predictions generated from the fitted model relative to a single‐species modelling approach.

Results

For most species, correlation due to climate covariates exceeded residual correlation, indicating an overriding influence of broad‐scale climate on co‐occurrence patterns. Accounting for covariance among species did not significantly improve predictions relative to a single‐species approach, providing limited evidence for a strong independent influence of species interactions on distribution patterns.

Conclusions

Overall, our findings indicate that climate is an important driver of regional biodiversity patterns and that interactions between dominant tree species contribute little to explain species co‐occurrence patterns among Rocky Mountain trees.     

Species distributions models may predict accurately future distributions but poorly how distributions change: A critical perspective on model validation

Aim

Species distribution models (SDMs) are widely used to make predictions on how species distributions may change as a response to climatic change. To assess the reliability of those predictions, they need to be critically validated with respect to what they are used for. While ecologists are typically interested in how and where distributions will change, we argue that SDMs have seldom been evaluated in terms of their capacity to predict such change. Instead, typical retrospective validation methods estimate model's ability to predict to only one static time in future. Here, we apply two validation methods, one that predicts and evaluates a static pattern, while the other measures change and compare their estimates of predictive performance.

Location

Fennoscandia.

Methods

We applied a joint SDM to model the distributions of 120 bird species in four model validation settings. We trained models with a dataset from 1975 to 1999 and predicted species' future occurrence and abundance in two ways: for one static time period (2013–2016, ‘static validation’) and for a change between two time periods (difference between 1996–1999 and 2013–2016, ‘change validation’). We then measured predictive performance using correlation between predicted and observed values. We also related predictive performance to species traits.

Results

Even though static validation method evaluated predictive performance as good, change method indicated very poor performance. Predictive performance was not strongly related to any trait.

Main Conclusions

Static validation method might overestimate predictive performance by not revealing the model's inability to predict change events. If species' distributions remain mostly stable, then even an unfit model can predict the near future well due to temporal autocorrelation. We urge caution when working with forecasts of changes in spatial patterns of species occupancy or abundance, even for SDMs that are based on time series datasets unless they are critically validated for forecasting such change.     

Environmental controls on the global distribution of shallow‐water coral reefs

Aim Elucidating the environmental limits of coral reefs is central to projecting future impacts of climate change on these ecosystems and their global distribution. Recent developments in species distribution modelling (SDM) and the availability of comprehensive global environmental datasets have provided an opportunity to reassess the environmental factors that control the distribution of coral reefs at the global scale as well as to compare the performance of different SDM techniques. Location Shallow waters world‐wide. Methods The SDM methods used were maximum entropy (Maxent) and two presence/absence methods: classification and regression trees (CART) and boosted regression trees (BRT). The predictive variables considered included sea surface temperature (SST), salinity, aragonite saturation state (Ω_Arag), nutrients, irradiance, water transparency, dust, current speed and intensity of cyclone activity. For many variables both mean and SD were considered, and at weekly, monthly and annually averaged time‐scales. All were transformed to a global 1° × 1° grid to generate coral reef probability maps for comparison with known locations. Model performance was compared in terms of receiver operating characteristic (ROC) curves and area under the curve (AUC) scores. Potential geographical bias was explored via misclassification maps of false positive and negative errors on test data. Results Boosted regression trees consistently outperformed other methods, although Maxent also performed acceptably. The dominant environmental predictors were the temperature variables (annual mean SST, and monthly and weekly minimum SST), followed by, and with their relative importance differing between regions, nutrients, light availability and Ω_Arag. No systematic bias in SDM performance was found between major coral provinces, but false negatives were more likely for cells containing ‘marginal’ non‐reef‐forming coral communities, e.g. Bermuda. Main conclusions Agreement between BRT and Maxent models gives predictive confidence for exploring the environmental limits of coral reef ecosystems at a spatial scale relevant to global climate models (c. 1° × 1°). Although SST‐related variables dominate the coral reef distribution models, contributions from nutrients, Ω_Arag and light availability were critical in developing models of reef presence in regions such as the Bahamas, South Pacific and Coral Triangle. The steep response in SST‐driven probabilities at low temperatures indicates that latitudinal expansion of coral reef habitat is very sensitive to global warming.     

Correcting for bias in distribution modelling for rare species using citizen science data

Aim

To improve the accuracy of inferences on habitat associations and distribution patterns of rare species by combining machine‐learning, spatial filtering and resampling to address class imbalance and spatial bias of large volumes of citizen science data.

Innovation

Modelling rare species’ distributions is a pressing challenge for conservation and applied research. Often, a large number of surveys are required before enough detections occur to model distributions of rare species accurately, resulting in a data set with a high proportion of non‐detections (i.e. class imbalance). Citizen science data can provide a cost‐effective source of surveys but likely suffer from class imbalance. Citizen science data also suffer from spatial bias, likely from preferential sampling. To correct for class imbalance and spatial bias, we used spatial filtering to under‐sample the majority class (non‐detection) while maintaining all of the limited information from the minority class (detection). We investigated the use of spatial under‐sampling with randomForest models and compared it to common approaches used for imbalanced data, the synthetic minority oversampling technique (SMOTE), weighted random forest and balanced random forest models. Model accuracy was assessed using kappa, Brier score and AUC. We demonstrate the method by evaluating habitat associations and seasonal distribution patterns using citizen science data for a rare species, the tricoloured blackbird (Agelaius tricolor).

Main Conclusions

Spatial under‐sampling increased the accuracy of each model and outperformed the approach typically used to direct under‐sampling in the SMOTE algorithm. Our approach is the first to characterize winter distribution and movement of tricoloured blackbirds. Our results show that tricoloured blackbirds are positively associated with grassland, pasture and wetland habitats, and negatively associated with high elevations or evergreen forests during both winter and breeding seasons. The seasonal differences in distribution indicate that individuals move to the coast during the winter, as suggested by historical accounts.

     

Species distribution modelling for the people: unclassified landsat TM imagery predicts bird occurrence at fine resolutions

Aim

Assessing the influence of land cover in species distribution modelling is limited by the availability of fine‐resolution land‐cover data appropriate for most species responses. Remote‐sensing technology offers great potential for predicting species distributions at large scales, but the cost and required expertise are prohibitive for many applications. We test the usefulness of freely available raw remote‐sensing reflectance data in predicting species distributions of 40 commonly occurring bird species in western Oregon.

Location

Central Coast Range, Cascade and Klamath Mountains Oregon, USA.

Methods

Information on bird observations was collected from 4598 fixed‐radius point counts. Reflectance data were obtained using 30‐m resolution Landsat imagery summarized at scales of 150, 500, 1000 and 2000 m. We used boosted regression tree (BRT) models to analyse relationships between distributions of birds and reflectance values and evaluated prediction performance of the models using area under the receiver operating characteristic curve (AUC) values.

Results

Prediction success of models using all reflectance values was high (mean AUC = 0.79 ± 0.10 SD). Further, model performance using individual reflectance bands exceeded those that used only Normalized Difference Vegetation Index (NDVI). The relative influence of band 4 predictors was highest, indicating the importance of variables associated with vegetation biomass and photosynthetic activity. Across spatial scales, the average influence of predictors at the 2000 m scale was greatest.

Main Conclusions

We demonstrate that unclassified remote‐sensing imagery can be used to produce species distribution models with high prediction success. Our study is the first to identify general patterns in the usefulness of spectral reflectances for species distribution modelling of multiple species. We conclude that raw Landsat Thematic Mapper data will be particularly useful in species distribution models when high‐resolution predictions are required, including habitat change detection studies, identification of fine‐scale biodiversity hotspots and reserve design.