Tradeoffs of different types of species occurrence data for use in systematic conservation planning

Data on the occurrence of species are widely used to inform the design of reserve networks. These data contain commission errors (when a species is mistakenly thought to be present) and omission errors (when a species is mistakenly thought to be absent), and the rates of the two types of error are inversely related. Point locality data can minimize commission errors, but those obtained from museum collections are generally sparse, suffer from substantial spatial bias and contain large omission errors. Geographic ranges generate large commission errors because they assume homogenous species distributions. Predicted distribution data make explicit inferences on species occurrence and their commission and omission errors depend on model structure, on the omission of variables that determine species distribution and on data resolution. Omission errors lead to identifying networks of areas for conservation action that are smaller than required and centred on known species occurrences, thus affecting the comprehensiveness, representativeness and efficiency of selected areas. Commission errors lead to selecting areas not relevant to conservation, thus affecting the representativeness and adequacy of reserve networks. Conservation plans should include an estimation of commission and omission errors in underlying species data and explicitly use this information to influence conservation planning outcomes.     

SPATIOTEMPORAL PREDICTION MODELS OF CETACEAN HABITATS IN THE MID-WESTERN NORTH ATLANTIC OCEAN (FROM CAPE HATTERAS, NORTH CAROLINA, U.S.A. TO NOVA SCOTIA, CANADA)

Habitat prediction models were developed for 13 cetacean species of the mid-western North Atlantic Ocean: beaked whale, fin whale, humpback whale, minke whale, pilot whale, sperm whale, bottlenose dolphin, common dolphin, Risso's dolphin, spotted dolphin, whitesided dolphin, and harbor porpoise. Using the multiple logistic regression, sightings of cetaceans during the 1990–1996 summer (June-September) surveys were modeled with oceanographic (sea surface temperature, monthly probability of front occurrence) and topographic (depth, slope) variables for the same period. Predicted habitat maps for June and August were created for each species using a Geographical Information System. The predicted habitat locations matched with current and historic cetacean sighting locations. The model also predicted habitat shifts for some species associated with oceanographic changes. The correct classification rate of the prediction models with 1997–1998 summer survey data ranged from 44% to 70%, of which most of the misclassifications were caused by false positives ( i.e. , absence of sightings at locations where the models predicted).     

Ranking Mahalanobis Distance Models for Predictions of Occupancy From Presence-Only Data

ABSTRACT The Mahalanobis distance statistic (D²) has emerged as an effective tool to identify suitable habitat from presence data alone, but there has been no mechanism to select among potential habitat covariates. We propose that the best combination of explanatory variables for a D² model can be identified by ranking potential models based on the proportion of the entire study area that is classified as potentially suitable habitat given that a predetermined proportion of occupied locations are correctly classified. In effect, our approach seeks to minimize errors of commission, or maximize specificity, while holding the omission error rate constant. We used this approach to identify potentially suitable habitat for the Olympic marmot (Marmota olympus), a declining species endemic to Olympic National Park, Washington, USA. We compared models built with all combinations of 11 habitat variables. A 7-variable model identified 21,143 ha within the park as potentially suitable for marmots, correctly classifying 80% of occupied locations. Additional refinements to the 7-variable model (e.g., eliminating small patches) further reduced the predicted area to 18,579 ha with little reduction in predictive power. Although we sought a model that would allow field workers to find 80% of Olympic marmot locations, in fact, <3% of 376 occupied locations and <9% of abandoned locations were >100 m from habitat predicted by the final model, suggesting that >90% of occupied marmot habitat could be found by observant workers surveying predicted habitat. The model comparison procedure allowed us to identify the suite of covariates that maximized specificity of our model and, thus, limited the amount of less favorable habitat included in the final prediction area. We expect that by maximizing specificity of models built from presence-only data, our model comparison procedure will be useful to conservation practitioners planning reintroductions, searching for rare species, or identifying habitat for protection.     

Spatio-temporal changes in habitat potential of endangered freshwater fish in Japan

In recent years, biodiversity conservation and ecosystem restoration have been key issues of watershed management in many countries. To maintain or restore the environmental quality of watersheds, we need to assess the impact of anthropogenic changes on stream ecosystems with accuracy. In addition, watershed conservation planners have to make strategic plans and determine priorities of each conservation activity.

A new monitoring methodology to evaluate the change of habitat condition for freshwater fish based on a predictive habitat model using logistic regression was developed and applied to the whole of Japan. The main contributions of our approach were 1) the construction of a Geographical Information System (GIS) database that integrates many types of data, including freshwater fish species, water quality, habitat fragmentation by damming, geology, and climate; 2) spatial analysis for quantitative assessment and predictive habitat modeling using logistic regression to combine fish survey data and environmental habitat factors to determine critical and major habitat variables for each target fish; and 3) digital mapping and changes detection of fish habitat potential for targeted endangered fish species to show habitat distribution and spatio-temporal changes of habitat potential over a 25-year period (from 1977 to 2002). We found that predicted suitable habitat and actual fish habitat showed high overlap, and that habitat conditions and distribution patterns of target freshwater fish had been affected by major habitat variables to target species respectively.     

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.     

ROC曲线形状在生态位模型评价中的重要性——以美国白蛾为例

【目的】生态位模型在生物地理学、入侵生物学和保护生物学中具有广泛的应用,被越来越多地用于预测物种潜在分布和现实分布的研究中。本文以美国白蛾为例介绍pROC方案在生态位模型评价中的应用及其注意事项,以期对物种潜在分布预测进行合理的评价,促进生态位模型在我国的合理运用和发展。【方法】介绍ROC曲线和AUC值基本原理,总结其在生态位模型评价中的应用,从物种存在分布点和不存在分布点的可信度出发,分析AUC值用于模型评价的优点和不足,最后介绍局部受试者工作特征曲线的线下面积方案(pROC方案)来弥补传统AUC值的不足。【结果】AUC值虽独立于阈值,但因其综合灵敏度和特异度,而屏蔽这2个指标各自的特征,不能分别评估预测结果的灵敏度和特异度,同时对遗漏率和记账错率不能进行权衡,会误导使用者对模型的评价。与AUC值相比,ROC曲线的形状更具有价值,蕴含丰富的模型评价信息。【结论】模型评价需要将灵敏度和特异度区别对待,ROC曲线形状比AUC值在生态位模型评价中更为重要,pROC方案相对于传统AUC值具有优势,但容易对过度模拟做出不当判断。模型评价与作者研究目的密切相关:当以预测物种潜在分布为目的时(如入侵物种潜在分布、气候变化对物种分布的影响和谱系生物地理学),模型评价应当给予灵敏度(或者遗漏率)更多的权重;当以预测物种现实分布为目的时(如保护区界定和濒危物种引入),模型评价应当给予灵敏度和特异度同等的权重。     

Effects of habitat configuration and quality on species richness and distribution in fragmented forest patches near Rome

Question: How important are habitat configuration, quality, history and anthropic disturbance in determining nemoral plant species richness and distribution of fragmented forest patches in a Mediterranean region? Location: Agricultural landscape north of Rome, Italy. Methods: Sixty‐nine woodland patches, identified through a stratified random sampling, were sampled for nemoral plant species. The homogeneity of woodlands was tested through a hierarchical classification of the floristic data and a Mann‐Whitney test of dependent and independent variables. The importance of habitat configuration (area, isolation, shape), quality (soil properties, forest structure, anthropic disturbance) and history (age of woodland) in determining species richness was estimated through a Poisson regression model. Presence‐absence of each species was analysed by logistic regression. Differences among plant life‐trait types (life span, dispersal mode, habitat preference) were analysed by comparing their median β‐values through ANOVA models. Results: Through hierarchical classification, two woodland types were identified that differed in species composition, habitat quality and spatial configuration. Poisson regression showed that habitat configuration and history influenced species richness. Multiple logistic regression resulted in significant fits for 88 species/variable combinations: 38 are habitat quality variables, 25 are habitat configuration variables, and 13 are anthropic factors. Dispersal strategies varied significantly with respect to area, isolation and age, while generalist and specialist species differed according to age of the woodland. Conclusion: Our results show that habitat history and configuration are the key factors determining species richness of woodland. Together with habitat configuration, habitat quality (mainly soil acidity) appeared to influence species composition.     

Use of simulated data from a process‐based habitat model to evaluate methods for predicting species occurrence

Much research has centered on determining which habitat model best predicts species occurrence. However, previous work typically used data sets that are inherently biased for evaluation. The use of simulated data provides a way of testing model performance using un‐biased data where the true relationships between species occurrence and population processes are predefined using sound ecological theory. We used a process‐based habitat model to generate simulated occurrence data to evaluate presence–absence and presence–only methods: generalized linear and generalized additive models (GLM, GAM), maximum entropy model (Maxent), and discrete choice models (DCM). This is the first study to use a DCM for predicting species distributions. We varied the effect that habitat quality had on fecundity and reported the model responses to these changes. When the effect of habitat quality on fecundity was weak, model performance was no better than random for all methods, however, performance increased as the habitat/fecundity relationship became stronger. For each level of habitat quality effect, there was little variation in performance between the presence–absence and presence–only methods. The use of a process‐based habitat model to generate occurrence data for evaluating model performance has a distinct advantage over other testing methods, because no errors are made during sampling and the true ecological relationships between population process and species occurrence are known. This leads to un‐biased results and increased confidence in assessing model performance and making management recommendations.     

Are All Global Positioning System Collars Created Equal? Correcting Habitat-Induced Bias Using Three Brands in the Central Canadian Rockies

Abstract: Global positioning system (GPS) collars are changing the face of wildlife research, yet they still possess biases such as habitat-induced fix-rate bias, which is a serious concern for habitat selection studies. We studied GPS bias in the Central Canadian Rockies, a critical area for wildlife conservation, to provide a statistical approach to correct GPS habitat bias for habitat selection studies using GPS collars. To model GPS habitat bias we deployed 11 different collars from 3 brands of GPS collars (Advanced Telemetry Systems [ATS], Asanti, MN; LOTEK Engineering Ltd., Newmarket, ON, Canada; and Televilt, Lindesberg, Sweden) in a random-stratified design at 86 sites across habitat and topographic conditions. We modeled the probability of obtaining a successful location, P_FIX, as a function of habitat, topography, and collar brand using mixed-effects logistic regression in an information theoretic approach. For LOTEK collars, we also investigated the effect of 8 and 12 GPS channels on fix rate. The ATS collars had the highest overall fix rates (97.4%), followed by LOTEK 12 channel (94.5%), LOTEK 8 channel (85.6%), and Televilt (82.3%). Sufficient model selection uncertainty existed to warrant model averaging for logistic regression P_FIX models. Collar brand influenced fix rate in all P_FIX models: fix rates for ATS and LOTEK 12 channel were not statistically different, whereas LOTEK 8 channel receivers had intermediate fix rates, and Televilt had the lowest. Fix rate was reduced in aspen stands, closed coniferous stands, and sites in narrow mountainous valleys but was higher on upper mountain slopes. Slight discrepancies between fix rates from field trials and observed species fix rates (wolf [Canis lupus] and elk [Cervus elaphus]) suggest uncorrected behavioral or movement-induced bias similar to other recent studies. Regardless, the strong habitat-induced bias in GPS fix rates confirms that in our study area habitat effects are critical, especially for poorer performance brands. Based on previous studies of effects of the amount of bias on inferences, our results suggest correction for GPS bias should be mandatory for Televilt collars in the Canadian Rockies, optional for LOTEK (dependent on the no. of channels), and unnecessary for ATS. Thus, our GPS bias model will be useful to researchers using GPS collars on a variety of species throughout the Rocky Mountain cordillera.     

Developing landscape habitat models for rare amphibians with small geographic ranges: a case study of Siskiyou Mountains salamanders in the western USA

To advance the development of conservation planning for rare species with small geographic ranges, we determined habitat associations of Siskiyou Mountains salamanders (Plethodon stormi) and developed habitat suitability models at fine (10 ha), medium (40 ha), and broad (202 ha) spatial scales using available Geographic Information Systems data and logistic regression analysis with an information theoretic approach. Across spatial scales, there was very little support for models with structural habitat features, such as tree canopy cover and conifer diameter. Model-averaged 95% confidence intervals for regression coefficients and associated odds ratios indicated that the occurrence of Siskiyou Mountains salamanders was positively associated with rocky soils and Pacific madrone (Abutus menziesii) and negatively associated with elevation and white fir (Abies concolor); these associations were consistent across 3 spatial scales. The occurrence of this species also was positively associated with hardwood density at the medium spatial scale. Odds ratios projected that a 10% decrease in white fir abundance would increase the odds of salamander occurrence 3.02–4.47 times, depending on spatial scale. We selected the model with rocky soils, white fir, and Oregon white oak (Quercus garryana) as the best model across 3 spatial scales and created habitat suitability maps for Siskiyou Mountains salamanders by projecting habitat suitability scores across the landscape. Our habitat suitability models and maps are applicable to selection of priority conservation areas for Siskiyou Mountains salamanders, and our approach can be easily adapted to conservation of other rare species in any geographical location.     

基于逻辑斯蒂回归模型的鹭科水鸟栖息地适宜性评价

近年来湿地生态系统遭到不同程度破坏,湿地水鸟及其生存空间日益受到威胁。以香港米埔-后海湾湿地为例,收集2003年1月份与鹭科水鸟密切相关的15个自变量和鹭科水鸟实测数据作为因变量构建逻辑斯蒂回归模型,通过筛选获取9个变量因子,分别为土地利用,NDVI,坡度,降雨,TM4纹理,TM3纹理,道路密度,道路距离,人居密度。经Nagelkerke R2检验模型精度达到0.743,拟合度较高。利用模型结果快速聚类,对栖息地进行适宜性分级,分级结果与同期鹭科水鸟实测数据做拟合,精度达到77.4%。最后采集2009年1月份各变量因子数据对回归方程进行时间尺度检验,与同期实测鹭科水鸟数据拟合精度同样达到75.8%,模型具有较好的通用性。     

Building Hierarchical Models of Avian Distributions for the State of Georgia

Abstract To predict the distributions of breeding birds in the state of Georgia, USA, we built hierarchical models consisting of 4 levels of nested mapping units of decreasing area: 90,000 ha, 3,600 ha, 144 ha, and 5.76 ha. We used the Partners in Flight database of point counts to generate presence and absence data at locations across the state of Georgia for 9 avian species: Acadian flycatcher (Empidonax virescens), brown-headed nuthatch (Sitta pusilla), Carolina wren (Thryothorus ludovicianus), indigo bunting (Passerina cyanea), northern cardinal (Cardinalis cardinalis), prairie warbler (Dendroica discolor), yellow-billed cuckoo (Coccyzus americanus), white-eyed vireo (Vireo griseus), and wood thrush (Hylocichla mustelina). At each location, we estimated hierarchical-level-specific habitat measurements using the Georgia GAP Analysis18 class land cover and other Geographic Information System sources. We created candidate, species-specific occupancy models based on previously reported relationships, and fit these using Markov chain Monte Carlo procedures implemented in OpenBugs. We then created a confidence model set for each species based on Akaike's Information Criterion. We found hierarchical habitat relationships for all species. Three-fold cross-validation estimates of model accuracy indicated an average overall correct classification rate of 60.5%. Comparisons with existing Georgia GAP Analysis models indicated that our models were more accurate overall. Our results provide guidance to wildlife scientists and managers seeking predict avian occurrence as a function of local and landscape-level habitat attributes.     

Spatio-temporal co-occurrence of cougars (Felis concolor), wolves (Canis lupus) and their prey during winter: a comparison of two analytical methods

Aims To examine the spatio‐temporal co‐occurrence of cougars (Felis concolor), wolves (Canis lupus), and their prey during winter using monthly (November–March) species–environment relationship models. In addition, to contrast predictions across two methods: logistic regression and Geographic Information System (GIS) image correlation. Location The eastern front ranges of the Canadian Rocky Mountains (south‐central Alberta), approximately 100 km west of Calgary, including portions of Banff National Park and Kananaskis Country. Methods Snow‐tracking data were collected simultaneously for cougars, wolves, elk (Cervus elaphus), and deer (Odocoileus virginianus and O. hemionus) between November and March, 1997–2000. Track data were synthesized in a GIS. Logistic regression and Akaike's information criterion (AIC) were used to select optimal environmental relationship models for each species. We first examined co‐occurrence by iteratively using each species as a dependent variable (presence/absence) in a logistic regression analysis and using all other species track‐density estimates as independent variables. We built predictive surfaces in a GIS using the exponent form of the logistic regression models, and assessed model accuracy with a receiver operating characteristic curve. We then re‐examined co‐occurrence using pairwise correlations of species probability surfaces by month. The correlation results were compared with logistic regression results to illuminate mechanisms of co‐occurrence and to investigate predictive consistency across the two methods. Results Cougars showed a trend in distribution from higher elevation and less rugged terrain in December, to lower elevation and more rugged terrain in March. This trend differed from that for wolves, which showed a more stable affinity for low elevation and less rugged valley bottoms across all months. The logistic regression models indicated variable positive and negative associations of cougars with wolves by month, and changes in prey associations over time. Notably, there was a shift in co‐occurrence for both predators from elk to deer in March. We found high predictive accuracy for all probability surfaces, except for the month of January. Our image comparison showed that spatial co‐occurrence amongst all species increased over winter, except that wolves and cougars were negatively correlated in February. Combining the results of each approach we found that cougars and wolves converged spatially over winter at the landscape scale (i.e. the valley), while showing more discrete use of that space over time and by habitat attributes (e.g. forest cover, topographic complexity, and prey track density). Main conclusions In the Rocky Mountains, the spatial distributions of cougars and wolves converged into the valley floor as winter progressed. Cougars were distinct from wolves and prey in the intensity of this shift. We determined that a comparison of predictive surfaces alone fails to explain species co‐occurrence. The surfaces must be coupled with investigation of respective species–environment models to account for temporal changes in associations. We suggest that the two approaches represent different ecological scales: image comparison may be best for landscape‐ (valley) level analysis, while logistic regression is best for site‐level analysis. Ultimately, both approaches were critical to our analysis. Finally, the variability observed over time suggested that annual and seasonal models may obscure important ecological patterns and processes, especially for cougars.     

Multiscale assessment of the influence of habitat structure and composition on bird assemblages in boreal forest

The relative contributions of habitat structure and composition to biodiversity are often scale-dependent. Although bird communities in boreal forest have been largely altered and threatened by forest harvesting, bird habitat selection in this ecosystem has not been fully understood. Our study aimed to assess the relative contributions of habitat structure and composition on the assemblages of boreal birds at multiple spatial scales characterized by radii ranging from 100 to 1,000?m. We recorded bird species occurrence at 96 stations located in an old-growth forest in the C?te-Nord region of Québec, Canada. We characterized habitat structure using the proportion of dense, open, and sparse stands, and habitat composition using the proportions of coniferous, mixedwood, and deciduous stands. We used partial canonical correspondence analyses and hierarchical variance partitioning to assess the relative contribution of habitat structure and composition on bird assemblage, and logistic regression to model the probability of occurrence for individual species in response to habitat variables. Our results revealed that habitat structure and composition explained similar proportions of the variance in bird assemblage (21.7 vs. 21.6?%), regardless of spatial scale. Whilst logistic regression yielded fair predictions in the occurrence of individual species (i.e., area under the receiver-operating characteristic curve >0.70 for 90?% of the species), it further confirmed our findings in community level analysis. Our study indicates that habitat structure and composition are both important in shaping bird assemblages, but spatial scale draws little influence on their relative contributions.     

Density and habitat requirements of introduced Eastern cottontail <Emphasis Type="Italic">Sylvilagus floridanus</Emphasis> in northern Italy

Following the introductions carried out in late 1960s, Eastern cottontail Sylvilagus floridanus Allen, 1890 rapidly colonized the Po Plain (northern Italy), following the Po River and its tributaries. We monitored a cottontail population using the line-transect method from autumn 2005 to spring 2009 in a 8.2-km2 study area located along the Po River, and we investigated species habitat requirements by assessing the presence/absence of faecal pellets in 200 randomly distributed plots from September 2006 to August 2007 and by Resource Selection Probability Function through logistic regression analyses and multi-model inference. The cottontail population varied dramatically over time in size, with a great drop at the end of the breeding period. Cottontails selected foraging habitats at the macro- and micro-scales, with some differences among seasons. Two macro-habitat variables differed significantly between used and unused plots through seasons: arboriculture stands were always greater in presence plots, whereas winter cereals were always greater in absence ones. On the macro-level, woody and herbaceous habitats, such as fallow fields, characterized presence plots. At the micro-habitat level, presence plots were associated with permanent dense cover except during summer. Several logistic regression models were built through seasons and ranked using the Akaike’s Information Criterion. Arboriculture stands enhanced cottontail presence mostly during the growing season contrary to crop fields. Hedgerows were used according to availability during feeding activity. Cottontail habitat selection varied according to seasonal changes in resource availability and suitability of the different habitat types.     

Logistic regression models for predicting occurrence of terrestrial molluscs in southern Sweden – importance of environmental data quality and model complexity

We compared the performance of four logistic regression models of different complexity with different environmental data quality, in predicting the occurrence of 49 terrestrial mollusc species in southern Sweden. Performance of models derived from an explanatory data set was evaluated on a confirmatory data set. The overall predictive success of our models (>80% for the three best model approaches), is as good as in other studies, despite the fact that we had to transform a text database into quantitative habitat variables. Simple models (no variable interactions), with forward selection, and detailed habitat data (from field visits) showed the best overall predictive success (mean=84.8%). From comparisons of model approaches, we conclude that data quality (map‐derived data vs habitat mapping) had a stronger impact than model complexity on model performance. However, most of these models showed relatively low values (mean=0.29) for Kappa (statistic for model evaluation), suggesting that the models need to be improved before they would be applied. Predictive success was strongly associated with species incidence but also Kappa was positively correlated with species incidence in univariate tests. Predictive success for true absences was negatively correlated with predictive success for true presences (R²=0.69) and most models failed to give a good prediction of both categories. Models for species with a high incidence in “Open dry sites” or “Mesic interior forests” had a better performance than expected, suggesting that occurrences of species with preference for “narrow” habitats are most easy to predict. Tree layer variables (openness and species abundance) were included in 48 of the 49 final predictive models, suggesting that these variables were good “indicators” of habitat conditions for ground‐living molluscs. Twenty‐four species models included distance to coast and altitude, and we interpret these associations as partly being related to differences in climate. In the final models, true presences (36.9% correctly classified) were much more difficult to predict than true absences (89.7% correct). Possible explanations might be that important habitat variables (e.g. chemical variables and site history) were not included. On the other hand, all suitable sites would not be expected to be occupied due to dynamics in local extinctions (meta‐population theory).     

Limited transferability of stream‐fish distribution models among river catchments: reasons and implications

相似文献   

Insights into the area under the receiver operating characteristic curve (AUC) as a discrimination measure in species distribution modelling

Aim The area under the receiver operating characteristic (ROC) curve (AUC) is a widely used statistic for assessing the discriminatory capacity of species distribution models. Here, I used simulated data to examine the interdependence of the AUC and classical discrimination measures (sensitivity and specificity) derived for the application of a threshold. I shall further exemplify with simulated data the implications of using the AUC to evaluate potential versus realized distribution models. Innovation After applying the threshold that makes sensitivity and specificity equal, a strong relationship between the AUC and these two measures was found. This result is corroborated with real data. On the other hand, the AUC penalizes the models that estimate potential distributions (the regions where the species could survive and reproduce due to the existence of suitable environmental conditions), and favours those that estimate realized distributions (the regions where the species actually lives). Main conclusions Firstly, the independence of the AUC from the threshold selection may be irrelevant in practice. This result also emphasizes the fact that the AUC assumes nothing about the relative costs of errors of omission and commission. However, in most real situations this premise may not be optimal. Measures derived from a contingency table for different cost ratio scenarios, together with the ROC curve, may be more informative than reporting just a single AUC value. Secondly, the AUC is only truly informative when there are true instances of absence available and the objective is the estimation of the realized distribution. When the potential distribution is the goal of the research, the AUC is not an appropriate performance measure because the weight of commission errors is much lower than that of omission errors.     

Classification trees: An alternative non‐parametric approach for predicting species distributions

Abstract. The use of Generalized Linear Models (GLM) in vegetation analysis has been advocated to accommodate complex species response curves. This paper investigates the potential advantages of using classification and regression trees (CART), a recursive partitioning method that is free of distributional assumptions. We used multiple logistic regression (a form of GLM) and CART to predict the distribution of three major oak species in California. We compared two types of model: polynomial logistic regression models optimized to account for non‐linearity and factor interactions, and simple CART‐models. Each type of model was developed using learning data sets of 2085 and 410 sample cases, and assessed on test sets containing 2016 and 3691 cases respectively. The responses of the three species to environmental gradients were varied and often non‐homogeneous or context dependent. We tested the methods for predictive accuracy: CART‐models performed significantly better than our polynomial logistic regression models in four of the six cases considered, and as well in the two remaining cases. CART also showed a superior ability to detect factor interactions. Insight gained from CART‐models then helped develop improved parametric models. Although the probabilistic form of logistic regression results is more adapted to test theories about species responses to environmental gradients, we found that CART‐models are intuitive, easy to develop and interpret, and constitute a valuable tool for modeling species distributions.     

Integrating individual habitat choices and regional distribution of a biodiversity indicator and top predator

Aim Habitat selection studies have mainly focused on behavioural choices of individuals or on the habitat‐related regional distribution of a population, with little integration of the two approaches. This is despite the fact that traditional biogeography theory sees the geographical distribution of a species as the collective outcome of the adaptive habitat choices of individuals. Here, we integrate individual habitat choices with regional distribution through a bottom‐up Geographical Information System (GIS)‐based approach, by using a 9‐year data set on a large avian predator, the eagle owl (Bubo bubo L.). We further examine the potential population level and biodiversity consequences of this approach. Location The study was conducted in the Trento Region (central‐eastern Italian Alps) and in six other areas of the nearby Lombardia Region in the central Alps. Methods We used stepwise logistic regression to build a habitat suitability model discriminating between eagle owl territories and an equal number of random locations. The model was applied to the whole Trento region by means of a GIS so as to predict suitable habitat patches. The predicted regional distribution (presence–absence in 10‐km grid quadrats) was then compared with the observed one. Furthermore, we compared estimates of biodiversity in quadrats with and without eagle owls, so as to test whether the presence of this top predator may signal macro‐areas of high biodiversity. Results The logistic habitat suitability model showed that, compared with a random distribution, eagle owls selected low‐elevation breeding sites with high availability of prey‐rich habitats in their surroundings. Breeding performance increased with the availability of prey‐rich habitats, confirming the adaptiveness of the detected habitat choices. We applied the habitat suitability model to the 6200 km² study region by means of a GIS and found a close fit between the observed and predicted regional distribution. Furthermore, population abundance was positively related to the availability of habitat defined as suitable by the above analyses. Finally, high biodiversity levels were associated with owl presence and with the amount of suitable owl habitat, demonstrating that modelling habitat suitability of a properly chosen indicator species may provide key conservation information at the wider ecosystem level. Main conclusions Our bottom‐up modelling approach may increase the conservation‐value of habitat selection models, by (1) predicting local and regional distribution, (2) estimating regional population size, (3) stimulating further hypothesis testing, (4) forecasting the population effects of future habitat loss and degradation and (5) aiding in the identification and prioritization of high‐biodiversity areas.