Performance evaluation of cetacean species distribution models developed using generalized additive models and boosted regression trees

Species distribution models (SDMs) are important management tools for highly mobile marine species because they provide spatially and temporally explicit information on animal distribution. Two prevalent modeling frameworks used to develop SDMs for marine species are generalized additive models (GAMs) and boosted regression trees (BRTs), but comparative studies have rarely been conducted; most rely on presence‐only data; and few have explored how features such as species distribution characteristics affect model performance. Since the majority of marine species BRTs have been used to predict habitat suitability, we first compared BRTs to GAMs that used presence/absence as the response variable. We then compared results from these habitat suitability models to GAMs that predict species density (animals per km²) because density models built with a subset of the data used here have previously received extensive validation. We compared both the explanatory power (i.e., model goodness of fit) and predictive power (i.e., performance on a novel dataset) of the GAMs and BRTs for a taxonomically diverse suite of cetacean species using a robust set of systematic survey data (1991–2014) within the California Current Ecosystem. Both BRTs and GAMs were successful at describing overall distribution patterns throughout the study area for the majority of species considered, but when predicting on novel data, the density GAMs exhibited substantially greater predictive power than both the presence/absence GAMs and BRTs, likely due to both the different response variables and fitting algorithms. Our results provide an improved understanding of some of the strengths and limitations of models developed using these two methods. These results can be used by modelers developing SDMs and resource managers tasked with the spatial management of marine species to determine the best modeling technique for their question of interest.     

基于集成模型的小黄鱼越冬群体适宜生境及其环境影响因素

小黄鱼是中韩渔业共同利用鱼种,其跨界洄游习性限制了对越冬场范围的调查和评估,导致对越冬群体适宜栖息地分布缺乏了解。本研究基于越冬期我国自然海域的物种分布点位数据和5个环境数据,运用8个物种分布模型(SDM)分析了小黄鱼越冬场分布范围,采用5折交叉验证,利用受试者工作特征曲线下面积(AUC)评价模型预测性能,并通过加权集成方法构建综合生境模型预测越冬场核心分布位置。结果表明: 出现/未出现数据模型预测准确度普遍高于仅出现模型;在出现/未出现数据模型中,机器学习方法预测准确度高于经典回归模型,支持向量模型(SVM)准确度最高(AUC=0.85),广义线性模型(GLM)准确度最低(AUC=0.73)。集成模型AUC较单一独立模型的准确度有所提升,表明集成模型能有效降低单一独立模型所带来的不确定性,提高模型预测准确度。变量重要性分析结果显示,盐度和温度是决定小黄鱼越冬场地理分布的重要因素,适宜分布区集中在黄海南部外海、东海北部外海和浙江省沿岸海域,而黄海南部沿岸海域和东海中南部外海为不适宜越冬区。研究结果为预测小黄鱼潜在越冬场提供了理论基础,可支撑越冬场渔业资源的空间规划和可持续利用。     

Uncertainty in ensemble forecasting of species distribution

Species distribution modelling has been widely applied in order to assess the potential impacts of climate change on biodiversity. Many methodological decisions, taken during the modelling process and forecasts, may, however, lead to a large variability in the assessment of future impacts. Using measures of species range change and turnover, the potential impacts of climate change on French stream fish species and assemblages were evaluated. Our main focus was to quantify the uncertainty in the projections of these impacts arising from four sources of uncertainty: initial datasets (Data), statistical methods [species distribution models (SDM)], general circulation models (GCM), and gas emission scenarios (GES). Several modalities of the aforementioned uncertainty sources were combined in an ensemble forecasting framework resulting in 8400 different projections. The variance explained by each source was then extracted from this whole ensemble of projections. Overall, SDM contributed to the largest variation in projections, followed by GCM, whose contribution increased over time equalling almost the proportion of variance explained by SDM in 2080. Data and GES had little influence on the variability in projections. Future projections of range change were more consistent for species with a large geographical extent (i.e., distribution along latitudinal or stream gradients) or with restricted environmental requirements (i.e., small thermal or elevation ranges). Variability in projections of turnover was spatially structured at the scale of France, indicating that certain particular geographical areas should be considered with care when projecting the potential impacts of climate change. The results of this study, therefore, emphasized that particular attention should be paid to the use of predictions ensembles resulting from the application of several statistical methods and climate models. Moreover, forecasted impacts of climate change should always be provided with an assessment of their uncertainty, so that management and conservation decisions can be taken in the full knowledge of their reliability.     

Constructing support vector machine ensembles for cancer classification based on proteomic profiling

In this study, we present a constructive algorithm for training cooperative support vector machine ensembles （CSVMEs）. CSVME combines ensemble architecture design with cooperative training for individual SVMs in ensembles. Unlike most previous studies on training ensembles, CSVME puts emphasis on both accuracy and collaboration among individual SVMs in an ensemble. A group of SVMs selected on the basis of recursive classifier elimination is used in CSVME, and the number of the individual SVMs selected to construct CSVME is determined by 10-fold cross-validation. This kind of SVME has been tested on two ovarian cancer datasets previously obtained by proteomic mass spectrometry. By combining several individual SVMs, the proposed method achieves better performance than the SVME of all base SVMs.     

Static species distribution models in dynamically changing systems: how good can predictions really be?

It is widely acknowledged that species respond to climate change by range shifts. Robust predictions of such changes in species’ distributions are pivotal for conservation planning and policy making, and are thus major challenges in ecological research. Statistical species distribution models (SDMs) have been widely applied in this context, though they remain subject to criticism as they implicitly assume equilibrium, and incorporate neither dispersal, demographic processes nor biotic interactions explicitly. In this study, the effects of transient dynamics and ecological properties and processes on the prediction accuracy of SDMs for climate change projections were tested. A spatially explicit multi‐species dynamic population model was built, incorporating species‐specific and interspecific ecological processes, environmental stochasticity and climate change. Species distributions were sampled in different scenarios, and SDMs were estimated by applying generalised linear models (GLMs) and boosted regression trees (BRTs). Resulting model performances were related to prevailing ecological processes and temporal dynamics. SDM performance varied for different range dynamics. Prediction accuracies decreased when abrupt range shifts occurred as species were outpaced by the rate of climate change, and increased again when a new equilibrium situation was realised. When ranges contracted, prediction accuracies increased as the absences were predicted well. Far‐dispersing species were faster in tracking climate change, and were predicted more accurately by SDMs than short‐dispersing species. BRTs mostly outperformed GLMs. The presence of a predator, and the inclusion of its incidence as an environmental predictor, made BRTs and GLMs perform similarly. Results are discussed in light of other studies dealing with effects of ecological traits and processes on SDM performance. Perspectives are given on further advancements of SDMs and for possible interfaces with more mechanistic approaches in order to improve predictions under environmental change.     

Trade-offs in covariate selection for species distribution models: a methodological comparison

Species distribution models (SDMs) are a common approach to describing species’ space-use and spatially-explicit abundance. With a myriad of model types, methods and parameterization options available, it is challenging to make informed decisions about how to build robust SDMs appropriate for a given purpose. One key component of SDM development is the appropriate parameterization of covariates, such as the inclusion of covariates that reflect underlying processes (e.g. abiotic and biotic covariates) and covariates that act as proxies for unobserved processes (e.g. space and time covariates). It is unclear how different SDMs apportion variance among a suite of covariates, and how parameterization decisions influence model accuracy and performance. To examine trade-offs in covariation parameterization in SDMs, we explore the attribution of spatiotemporal and environmental variation across a suite of SDMs. We first used simulated species distributions with known environmental preferences to compare three types of SDM: a machine learning model (boosted regression tree), a semi-parametric model (generalized additive model) and a spatiotemporal mixed-effects model (vector autoregressive spatiotemporal model, VAST). We then applied the same comparative framework to a case study with three fish species (arrowtooth flounder, pacific cod and walleye pollock) in the eastern Bering Sea, USA. Model type and covariate parameterization both had significant effects on model accuracy and performance. We found that including either spatiotemporal or environmental covariates typically reproduced patterns of species distribution and abundance across the three models tested, but model accuracy and performance was maximized when including both spatiotemporal and environmental covariates in the same model framework. Our results reveal trade-offs in the current generation of SDM tools between accurately estimating species abundance, accurately estimating spatial patterns, and accurately quantifying underlying species–environment relationships. These comparisons between model types and parameterization options can help SDM users better understand sources of model bias and estimate error.     

Testing the ability of species distribution models to infer variable importance

Models of species’ distributions and niches are frequently used to infer the importance of range- and niche-defining variables. However, the degree to which these models can reliably identify important variables and quantify their influence remains unknown. Here we use a series of simulations to explore how well models can 1) discriminate between variables with different influence and 2) calibrate the magnitude of influence relative to an ‘omniscient’ model. To quantify variable importance, we trained generalized additive models (GAMs), Maxent and boosted regression trees (BRTs) on simulated data and tested their sensitivity to permutations in each predictor. Importance was inferred by calculating the correlation between permuted and unpermuted predictions, and by comparing predictive accuracy of permuted and unpermuted predictions using AUC and the continuous Boyce index. In scenarios with one influential and one uninfluential variable, models failed to discriminate reliably between variables when training occurrences were < 8–64, prevalence was > 0.5, spatial extent was small, environmental data had coarse resolution and spatial autocorrelation was low, or when pairwise correlation between environmental variables was |r| > 0.7. When two variables influenced the distribution equally, importance was underestimated when species had narrow or intermediate niche breadth. Interactions between variables in how they shaped the niche did not affect inferences about their importance. When variables acted unequally, the effect of the stronger variable was overestimated. GAMs and Maxent discriminated between variables more reliably than BRTs, but no algorithm was consistently well-calibrated vis-à-vis the omniscient model. Algorithm-specific measures of importance like Maxent's change-in-gain metric were less robust than the permutation test. Overall, high predictive accuracy did not connote robust inferential capacity. As a result, requirements for reliably measuring variable importance are likely more stringent than for creating models with high predictive accuracy.     

Background sampling for multi-scale ensemble habitat selection modeling: Does the number of points matter?

Ensemble habitat selection modeling is becoming a popular approach among ecologists to answer different questions. Since we are still in the early stages of development and application of ensemble modeling, there remain many questions regarding performance and parameterization. One important gap, which this paper addresses, is how the number of background points used to train models influences the performance of the ensemble model. We used an empirical presence-only dataset and three different selections of background points to train scale-optimized habitat selection models using six modeling algorithms (GLM, GAM, MARS, ANN, Random Forest, and MaxEnt). We tested four ensemble models using different combinations of the component models: (a) equal numbers of background points and presences, (b) background points equaled ten times the number of presences, (c) 10,000 background points, and (d) optimized background points for each component model. Among regression-based approaches, MARS performed best when built with 10,000 background points. Among machine learning models, RF performed the best when built with equal presences and background points. Among the four ensemble models, AUC indicated that the best performing model was the ensemble with each component model including the optimized number of background points, while TSS increased as the number of background points models increased. We found that an ensemble of models, each trained with an optimal number of background points, outperformed ensembles of models trained with the same number of background points, although differences in performance were slight. When using a single modeling method, RF with equal number of presences and background points can perform better than an ensemble model, but the performance fluctuates when the number of background points is not properly selected. On the other hand, ensemble modeling provides consistently high accuracy regardless of background point sampling approach. Further, optimizing the number of background points for each component model within an ensemble model can provide the best model improvement. We suggest evaluating more models across multiple species to investigate how background point selection might affect ensemble models in different scenarios.     

Predicting the potential distribution of wheatear birds using stacked generalization-based ensembles

Habitat suitability models, usually referred to as species distribution models (SDMs), are widely applied in ecology for many purposes, including species conservation, habitat discovery, and gain evolutionary insights by estimating the distribution of species. Machine learning algorithms as well as statistical models have been recently used to predict the distribution of species. However, they seemed to have some limitations due to the data and the models used. Therefore, this study proposes a novel approach for assessing habitat suitability based on ensemble learning techniques. Three heterogeneous ensembles were built using the stacked generalization method to model the distribution of four wheatear species (Oenanthe deserti, Oenanthe leucopyga, Oenanthe leucura, and Oenanthe oenanthe) located in Morocco. Initially, a set of base-learners were constructed by virtue of training for each specie's dataset six machine learning algorithms (Multi-Layer Perceptron (MLP), Support Vector Classifier (SVC), K-nearest neighbors (KNN), Decision Trees (DT), Gradient Boosting Classifier (GB), and Random Forest (RF)). Then, the predictions of these base learners were fed as training data to train three meta-learners (Logistic Regression (LR), SVC, and MLP). To evaluate and assess the performance of the proposed approaches, we used: (1) six performance criteria (accuracy, recall, precision, F1-score, AUC, and TSS), (2) Borda Count (BC) ranking method based on multiple criteria to rank the best-performing models, and (3) Scott Knott (SK) test to statistically compare the performance of the presented models. The results based on the six-evaluation metrics showed that stacked ensembles outperformed their singles in all species datasets, and the stacked model with SVC as a meta-learner outperformed the other two ensembles. The results showed the potential of using ensemble learning techniques to model species distribution and recommend the use of the stacked generalization technique as a combination strategy since it gave better results compared to single models in four wheatear species datasets. Moreover, to assess the impact of future climate changes on the distribution of the four wheatear species, the best-performing distribution model was selected and projected into the current and future climatic conditions. The distributions of the Moroccan wheatear birds were found to be slightly affected by future climate changes.     

Incorporating intraspecific variation into species distribution models improves distribution predictions,but cannot predict species traits for a wide-spread plant species

The most common approach to predicting how species ranges and ecological functions will shift with climate change is to construct correlative species distribution models (SDMs). These models use a species’ climatic distribution to determine currently suitable areas for the species and project its potential distribution under future climate scenarios. A core, rarely tested, assumption of SDMs is that all populations will respond equivalently to climate. Few studies have examined this assumption, and those that have rarely dissect the reasons for intraspecific differences. Focusing on the arctic-alpine cushion plant Silene acaulis, we compared predictive accuracy from SDMs constructed using the species’ full global distribution with composite predictions from separate SDMs constructed using subpopulations defined either by genetic or habitat differences. This is one of the first studies to compare multiple ways of constructing intraspecific-level SDMs with a species-level SDM. We also examine the contested relationship between relative probability of occurrence and species performance or ecological function, testing if SDM output can predict individual performance (plant size) and biotic interactions (facilitation). We found that both genetic- and habitat-informed SDMs are considerably more accurate than a species-level SDM, and that the genetic model substantially differs from and outperforms the habitat model. While SDMs have been used to infer population performance and possibly even biotic interactions, in our system these relationships were extremely weak. Our results indicate that individual subpopulations may respond differently to climate, although we discuss and explore several alternative explanations for the superior performance of intraspecific-level SDMs. We emphasize the need to carefully examine how to best define intraspecific-level SDMs as well as how potential genetic, environmental, or sampling variation within species ranges can critically affect SDM predictions. We urge caution in inferring population performance or biotic interactions from SDM predictions, as these often-assumed relationships are not supported in our study.     

Uncertainty of bioclimate envelope models based on the geographical distribution of species

Aim We explored the effects of prevalence, latitudinal range and spatial autocorrelation of species distribution patterns on the accuracy of bioclimate envelope models of butterflies. Location Finland, northern Europe. Methods The data of a national butterfly atlas survey (NAFI) carried out in 1991–2003 with a resolution of 10 × 10 km were used in the analyses. Generalized additive models (GAM) were constructed, for each of 98 species, to estimate the probability of occurrence as a function of climate variables. Model performance was measured using the area under the curve (AUC) of a receiver operating characteristic (ROC) plot. Observed differences in modelling accuracy among species were related to the species’ geographical attributes using multivariate GAM. Results Accuracies of the climate–butterfly models varied from low to very high (AUC values 0.59–0.99), with a mean of 0.79. The modelling performance was related negatively to the latitudinal range and prevalence, and positively to the spatial autocorrelation of the species distribution. These three factors accounted for 75.2% of the variation in the modelling accuracy. Species at the margin of their range or with low prevalence were better predicted than widespread species, and species with clumped distributions better than scattered dispersed species. Main conclusions The results from this study indicate that species’ geographical attributes highly influence the behaviour and uncertainty of species–climate models, which should be taken into account in biogeographical modelling studies and assessments of climate change impacts.     

Scaling of species distribution models across spatial resolutions and extents along a biogeographic gradient. The case of the Iberian mole Talpa occidentalis

Scaling is a key process in modelling approaches since it allows for translating information from one scale to another. However, the success of this procedure may depend on ‘source’ and ‘target’ scales, but also on the biogeographic/ecological context of the study area. We aimed to quantify the performance and success of scaling species distribution model (SDM) predictions across spatial resolution and extent along a biogeographic gradient using the Iberian mole as study case. We ran separate MaxEnt models at two extents (national and regional) using independent datasets (species locations and environmental predictors) collected at 10 km and 50 m resolutions respectively. Model performance and success of scaling SDMs were quantified on the basis of accuracy measures and spatial predictions. Complementarily, we calculated marginality and tolerance as indicators of habitat availability and niche truncation along the biogeographic gradient. Model performance increased with resolution and extent, as well as from north to south (mainly for high resolution models). When regional models were validated at different scales, their performance reduced severely, particularly in the case of coarse resolution models (some of them performed worse than random). However, when the 10 km‐national model was downscaled within regions, it performed better (AUC_test: 0.82, 0.85 and 0.55 respectively for Galicia, Madrid and Granada) than models specifically calibrated within each region at 10 km (0.47, 0.65, 0.44). Indeed, it also had a better accuracy when projected at 50 m (0.77, 0.91, 0.79) than models fitted at that resolution (0.62, 0.83, 0.96) in two of the three cases. The success of scaling model predictions decreased along the biogeographic gradient, being these differences associated to niche truncation. Models representing non‐truncated niches were more successfully scaled across resolutions and extents (particularly in areas not offering all possible habitats for species), which has important implications for SDM applications.     

Consensus Forecasting of Species Distributions: The Effects of Niche Model Performance and Niche Properties

Ensemble forecasting is advocated as a way of reducing uncertainty in species distribution modeling (SDM). This is because it is expected to balance accuracy and robustness of SDM models. However, there are little available data regarding the spatial similarity of the combined distribution maps generated by different consensus approaches. Here, using eight niche-based models, nine split-sample calibration bouts (or nine random model-training subsets), and nine climate change scenarios, the distributions of 32 forest tree species in China were simulated under current and future climate conditions. The forecasting ensembles were combined to determine final consensual prediction maps for target species using three simple consensus approaches (average, frequency, and median [PCA]). Species’ geographic ranges changed (area change and shifting distance) in response to climate change, but the three consensual projections did not differ significantly with respect to how much or in which direction, but they did differ with respect to the spatial similarity of the three consensual predictions. Incongruent areas were observed primarily at the edges of species’ ranges. Multiple stepwise regression models showed the three factors (niche marginality and specialization, and niche model accuracy) to be related to the observed variations in consensual prediction maps among consensus approaches. Spatial correspondence among prediction maps was the highest when niche model accuracy was high and marginality and specialization were low. The difference in spatial predictions suggested that more attention should be paid to the range of spatial uncertainty before any decisions regarding specialist species can be made based on map outputs. The niche properties and single-model predictive performance provide promising insights that may further understanding of uncertainties in SDM.     

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.

     

Strengthening forecasts of climate change impacts with multi‐model ensemble averaged projections using MAGICC/SCENGEN 5.3

Climate output from general circulation models (GCMs) is being used with increasing frequency to explore potential climate change impacts on species’ distributional range shifts and extinction probability. However, different GCMs do not perform equally well in their ability to hindcast the key climatic factors that potentially influence species distributions. Previous research has demonstrated that multi‐model ensemble forecasts perform better than any single GCM in simulating observed conditions at a global scale. MAGICC/SCENGEN 5.3 is a freeware climate model ‘emulator’ that generates multi‐model ensemble forecasts, conditional on regional and/or global performance, for up to twenty GCMs. In combination with a new application ‘M/SGridder’, this software can be used to produce down‐scaled ensemble forecasts, which minimize climate‐model‐related uncertainty, for a range of ecological problems.     

Improved prediction of Canada lynx distribution through regional model transferability and data efficiency

The application of species distribution models (SDMs) to areas outside of where a model was created allows informed decisions across large spatial scales, yet transferability remains a challenge in ecological modeling. We examined how regional variation in animal‐environment relationships influenced model transferability for Canada lynx (Lynx canadensis), with an additional conservation aim of modeling lynx habitat across the northwestern United States. Simultaneously, we explored the effect of sample size from GPS data on SDM model performance and transferability. We used data from three geographically distinct Canada lynx populations in Washington (n = 17 individuals), Montana (n = 66), and Wyoming (n = 10) from 1996 to 2015. We assessed regional variation in lynx‐environment relationships between these three populations using principal components analysis (PCA). We used ensemble modeling to develop SDMs for each population and all populations combined and assessed model prediction and transferability for each model scenario using withheld data and an extensive independent dataset (n = 650). Finally, we examined GPS data efficiency by testing models created with sample sizes of 5%–100% of the original datasets. PCA results indicated some differences in environmental characteristics between populations; models created from individual populations showed differential transferability based on the populations'' similarity in PCA space. Despite population differences, a single model created from all populations performed as well, or better, than each individual population. Model performance was mostly insensitive to GPS sample size, with a plateau in predictive ability reached at ~30% of the total GPS dataset when initial sample size was large. Based on these results, we generated well‐validated spatial predictions of Canada lynx distribution across a large portion of the species'' southern range, with precipitation and temperature the primary environmental predictors in the model. We also demonstrated substantial redundancy in our large GPS dataset, with predictive performance insensitive to sample sizes above 30% of the original.     

The recognition of multi-class protein folds by adding average chemical shifts of secondary structure elements

The recognition of protein folds is an important step in the prediction of protein structure and function. Recently, an increasing number of researchers have sought to improve the methods for protein fold recognition. Following the construction of a dataset consisting of 27 protein fold classes by Ding and Dubchak in 2001, prediction algorithms, parameters and the construction of new datasets have improved for the prediction of protein folds. In this study, we reorganized a dataset consisting of 76-fold classes constructed by Liu et al. and used the values of the increment of diversity, average chemical shifts of secondary structure elements and secondary structure motifs as feature parameters in the recognition of multi-class protein folds. With the combined feature vector as the input parameter for the Random Forests algorithm and ensemble classification strategy, we propose a novel method to identify the 76 protein fold classes. The overall accuracy of the test dataset using an independent test was 66.69%; when the training and test sets were combined, with 5-fold cross-validation, the overall accuracy was 73.43%. This method was further used to predict the test dataset and the corresponding structural classification of the first 27-protein fold class dataset, resulting in overall accuracies of 79.66% and 93.40%, respectively. Moreover, when the training set and test sets were combined, the accuracy using 5-fold cross-validation was 81.21%. Additionally, this approach resulted in improved prediction results using the 27-protein fold class dataset constructed by Ding and Dubchak.     

Genetic trials improve the transfer of Douglas‐fir distribution models across continents

Climate change is likely to result in novel conditions with no analogy to current climate. Therefore, the application of species distribution models (SDMs) based on the correlation between observed species’ occurrence and their environment is questionable and calls for a better understanding of the traits that determine species occurrence. Here, we compared two intraspecific, trait‐based SDMs with occurrence‐based SDMs, all developed from European data, and analyzed their transferability to the native range of Douglas‐fir in North America. With data from 50 provenance trials of Douglas‐fir in central Europe multivariate universal response functions (URFs) were developed for two functional traits (dominant tree height and basal area) which are good indicators of growth and vitality under given environmental conditions. These trials included 290 North American provenances of Douglas‐fir. The URFs combine genetic effects i.e. the climate of provenance origin and environmental effects, i.e. the climate of planting locations into an integrated model to predict the respective functional trait from climate data. The URFs were applied as SDMs (URF‐SDMs) by converting growth performances into occurrence. For comparison, an ensemble occurrence‐based SDM was developed and block cross validated with the BIOMOD2 modeling platform utilizing the observed occurrence of Douglas‐fir in Europe. The two trait based SDMs and the occurrence‐based SDM, all calibrated with data from Europe, were applied to predict the known distribution of Douglas‐fir in its introduced and native range in Europe and North America. Both models performed well within their calibration range in Europe, but model transfer to its native range in North America was superior in case of the URF‐SDMs showing similar predictive power as SDMs developed in North America itself. The high transferability of the URF‐SDMs is a testimony of their applicability under novel climatic conditions highlighting the role of intraspecific trait variation for adaptation planning in climate change.     

Shaping up model transferability and generality of species distribution modeling for predicting invasions: implications from a study on Bythotrephes longimanus

When using species distribution models to predict distributions of invasive species, we are faced with the trade-off between model realism, generality, and precision. Models are most applicable to specific conditions on which they are developed, but typically not readily transferred to other situations. To better assist management of biological invasions, it is critical to know how to validate and improve model generality while maintaining good model precision and realism. We examined this issue with Bythotrephes longimanus, to determine the importance of different models and datasets in providing insights into understanding and predicting invasions. We developed models (linear discriminant analysis, multiple logistic regression, random forests, and artificial neural networks) on datasets with different sample sizes (315 or 179 lakes) and predictor information (environmental with or without fish data), and evaluated them by cross-validation and several independent datasets. In cross-validation, models developed on 315-lake environmental dataset performed better than those developed on 179-lake environmental and fish dataset. The advantage of a larger dataset disappeared when models were tested on independent datasets. Predictions of the models were more diverse when developed on environmental conditions alone, whereas they were more consistent when including fish (especially diversity) data. Random forests had relatively good and more stable performance than the other approaches when tested on independent datasets. Given the improvement of model transferability in this study by including relevant species occurrence or diversity index, incorporating biotic information in addition to environmental predictors, may help develop more reliable models with better realism, generality, and precision.