Ensemble Of Small Models Research Articles

Intraspecific differentiation: Implications for niche and distribution modelling

AbstractAimMounting evidence suggests that failure of species distribution models to integrate local adaptation hinders our ability to predict distribution ranges, raising the question of whether modelling should be performed at the level of species (clade models) or intraspecific lineages (subclade models), characterized by the restricted availability of occurrence points. While Ensembles of Small Models (ESMs) offer an attractive framework for small datasets, their evaluation remains critical. We address these issues in the case of very small datasets inherent to subclade models and discuss which modelling strategy should be applied based on niche overlap among lineages.LocationSweden.TaxonMosses.MethodsEnsembles of Small Models were evaluated by null models built from randomly sampled presence points. We compared the extent of suitable area predicted by the projections of clade and subclade models. Niche overlap was quantified using Schoener's D and Hellinger'sImetrics, and the significance of these metrics in terms of niche conservatism or divergence was assessed by similarity tests.ResultsWe introduced a simple procedure for evaluating ESMs based on the pooling of the statistics used to assess model accuracy from the replicates. Despite fairly high AUC and TSS values, 2 of the 23 subclade models did not perform better than null models and should be discarded. Combined predictions from subclade models contributed, on average, five times more than clade models to the total suitable area predicted by the combination of subclade and clade models. The D and I metrics averaged 0.45 and 0.71, with evidence for niche conservatism in half of the species and no signal for niche divergence.Main conclusionsIn addition to the assessment of ESM accuracy based on the simple procedure described here, we recommend that ESMs should be systematically evaluated against null models. Lumping or splitting occurrence data at the intraspecific level substantially impacted model projections. Given the poor performance of models based on small datasets, even when employing ESMs, we pragmatically suggest that, in the absence of evidence for niche divergence during diversification of closely related intraspecific lineages, SDMs should be based on all available occurrence data at the species level.

Modeling the geographic distribution of pawpaw (Asimina triloba [L.] Dunal) in a portion of its northern range limits, western New York State, USA

Pawpaw (Asimina triloba [L.] Dunal) is a small deciduous tree in eastern North America notable for its edible fruit. Its northern range extends into New York State (NYS) where it is listed as threatened, with only approximately 20 sites known to contain pawpaw. This research models the geographic distribution of pawpaw in western NYS (land area = 27,263 km2) for four reasons: to understand factors shaping its distribution, to assess the degree to which it is protected, to locate potential introduction or reintroduction sites, and to locate potential sites with undiscovered pawpaw. This study uses an ensemble of small models (ESM) approach devised for modeling the distribution of rare species, which averages predictions from simple (i.e., bivariate) models weighted by model predictive performance. The ESM in this study, constructed from boosted regression tree (BRT) models, demonstrated a high ability to discern pawpaw presences from pseudo-absences. The area under the receiver operating characteristic curve (AUC) value was 0.986 when calculated using training data, and BRT models comprising the ESM exhibited AUC values calculated using cross-validation data as high as 0.957. The ESM suggests that pawpaw is found in the warmest (mean annual temperature > 9.0 °C,) and driest (May–September precipitation 110 cm) soils. Locations predicted to be suitable for pawpaw are overwhelmingly found on unprotected lands; forested areas in PAs with “high” or “highest” suitability for pawpaw only comprise 0.1–0.2% of the study area.

Toward community predictions: Multi-scale modelling of mountain breeding birds' habitat suitability, landscape preferences, and environmental drivers.

Across a large mountain area of the western Swiss Alps, we used occurrence data (presence‐only points) of bird species to find suitable modelling solutions and build reliable distribution maps to deal with biodiversity and conservation necessities of bird species at finer scales. We have performed a multi‐scale method of modelling, which uses distance, climatic, and focal variables at different scales (neighboring window sizes), to estimate the efficient scale of each environmental predictor and enhance our knowledge on how birds interact with their complex environment. To identify the best radius for each focal variable and the most efficient impact scale of each predictor, we have fitted univariate models per species. In the last step, the final set of variables were subsequently employed to build ensemble of small models (ESMs) at a fine spatial resolution of 100 m and generate species distribution maps as tools of conservation. We could build useful habitat suitability models for the three groups of species in the national red list. Our results indicate that, in general, the most important variables were in the group of bioclimatic variables including “Bio11” (Mean Temperature of Coldest Quarter), and “Bio 4” (Temperature Seasonality), then in the focal variables including “Forest”, “Orchard”, and “Agriculture area” as potential foraging, feeding and nesting sites. Our distribution maps are useful for identifying the most threatened species and their habitat and also for improving conservation effort to locate bird hotspots. It is a powerful strategy to improve the ecological understanding of the distribution of bird species in a dynamic heterogeneous environment.

Dispersal ability of threatened species affects future distributions

To track future climate space, seed dispersal will be essential for plants, but dispersal ability is rarely measured or incorporated into species distribution models. Species distribution models (SDMs) can rank habitat suitability at a local scale, and they may be a valuable conservation planning tool for rare, patchily distributed species. Dispersal is essential for species to survive the threats of habitat destruction and climate change. Combining dispersal ability and habitat suitability at the landscape scale is important to understand and predict species spatial responses to environmental change. This study analyzed future species distributions based on habitat suitability and dispersal scenarios for 7 endangered and vulnerable taxa, using SDM and dispersal ability to predict the range of species distribution across central Italy in the near future 2050. Species distribution models (Ensembles of Small Models, ESMs) were carried out using 19 bioclimatic environmental variables downloaded from WorldClim 1.4. The study identified the most suitable area for this species in the central Apennines. The innovation of this study is the combination of SDMs with dispersal ability based on measured traits of local populations of rare species and subspecies. The main results show a combination of climate change with dispersal ability strongly affects the future potential distribution of the rare species. Thus, obtaining suitable and accessible areas in the near future possible is possible to identify the high suitable sites for the reinforcement of the natural population, ensuring habitat connectivity.

Disentangling the effects of host resources, local, and landscape variables on the occurrence pattern of the dusky large blue butterfly (Phengaris nausithous) in upland grasslands

Determining the effects of local and landscape drivers on endangered species and predicting potential suitable habitats for their persistence is crucial for effective conservation management. Here, we applied a multi-scale approach to disentangle the effects of host resources, local, and landscape variables on the occurrence pattern of Phengaris (= Maculinea) nausithous in semi-natural upland grasslands. Our approach comprised the assessment of host parameters (plant cover, density, height, flower heads density, ant nest density, ant colony size), local grassland management (pasture, meadow), site conditions (area, shape, terrain attributes), and landscape variables (landscape composition, connectivity). We used ensemble of small models based on bivariate generalized linear models for explaining and predicting the butterfly occurrence pattern. Bivariate models revealed that host ant nest density, plant cover and height, local grassland management type (pasture), slope and eastness, landscape forest cover and grassland connectivity had a positive effect on the occurrence of P. nausithous (average explained deviance 20.5%). Host ant density, host plant cover, and local grassland management were the most influential factors on the ensemble predictions. The presence of P. nausithous in upland grasslands is not only determined by host resources, but also by local and landscape factors. Such factors proved to be relevant for identifying and predicting suitable grassland sites for this endangered species. Consequently, we recommend that conservation actions should include a landscape perspective to promote connectivity by facilitating coherent grazing networks enabling dispersal between semi-natural upland grasslands and thus species persistence.

Ensemble of small models for estimating potential abundance of Caucasian grouse (Lyrurus mlokosiewiczi) in Iran

The need to obtain information on population size to inform management actions for conservation is imperative. Despite this, reliable data on Caucasian grouse abundance is scarce in Iran. The goal of this study was to explore the potential distribution of Caucasian grouse using an ensemble of small models with outstanding performance for modelling rare species’ distributions to estimate the potential population size in Iran. We fitted an ensemble of small models with generalized boosted model (GBM) and maximum entropy (MaxEnt), and then built a final ensemble prediction by averaging across these two ensembles of small models.

Predicting climate change impacts on the threatened Quercus arbutifolia in montane cloud forests in southern China and Vietnam: Conservation implications

Trees of montane cloud forests (MCFs) often have small and isolated populations and face numerous threats. Due to limited conservation resources, management decisions are often based on scarce biological information. This study investigated the current status of populations of the critically endangered oak species Quercus arbutifolia in southern China, including its distribution, threats, population structure, and area of occupancy (AOO). Additionally, by using ensembles of small models (ESMs), the present and future (2050) distributions of climatically suitable habitats were predicted throughout south China and Vietnam. The present distribution of Q. arbutifolia is extremely fragmented, with only eight confirmed populations and a very small number of individuals (ca. 1200 in total). The results presented here show that Q. arbutifolia populations suffer from habitat destruction and fragmentation, small population size, unbalanced population structure, and pressure from strong competitor species, even though all known occurrences of Q. arbutifolia are in already established nature reserves. Based on the utilized models, the current potential distribution is limited to MCFs, and 17 new areas were predicted to have complete habitats suitability for Q. arbutifolia. However, only a small area in Fujian province will remain suitable for Q. arbutifolia in the future. The current AOO of Q. arbutifolia is very small (8.49 km2), with one-third of all populations predicted to be extinct by 2050, even under the minimum emission assumption. Finally, various actions and conservation measures, such as search for new unknown populations as well as ex situ and in situ conservation, are introduced and discussed in this paper.

A habitat suitability analysis at multi-spatial scale of two sympatric flying fox species reveals the urgent need for conservation action

In this study we used a multi-spatial scale approach to investigate habitat suitability, roosting characteristics, and ecological niche in two flying fox species on the Comoros Islands—Pteropus livingstonii and Pteropus seychellensis comorensis. At a broad scale, we assessed the ecological niche and habitat suitability for both species using the Species Distribution Modeling method based on the recent ensembles of small models (ESM) approach. At a fine scale, Ecological Niche Factor Analysis (ENFA) was applied to assess habitat selection by each species. Direct observation was used at each roost to estimate the total number of individuals and to identify the roost characteristics. At both broad and fine scales, the analyses highlighted clear niche partitioning by the two species. We found that P. livingstonii has a very limited distribution, restricted to steep, high-elevation slopes of the islands’ remaining natural forests, and the patterns were the same for roosting, foraging sites and the entire habitat. By contrast, P. s. comorensis has a relatively large geographic range that extends over low-elevation farmlands and villages and it was negatively correlated to natural forest across the entire area and all roosting sites, but its foraging areas were positively correlated to natural forest and high elevation areas. Both species selected large, tall trees for roosting. The total number of individuals in the studied area was estimated to be 1243 P. livingstonii and 11,898 P. s. comorensis. The results of our study demonstrated that these two species use different habitat types and ensure different ecosystem services in pollination and seed dispersion and thus are both critical for maintaining overall ecosystem dynamics. However, the currently high level of hunting pressure and roost disturbance makes them vulnerable to extinction. To ensure the viability of both species, conservation measures need to be taken by the Comoros government.

Ecospat: an R package to support spatial analyses and modeling of species niches and distributions

The aim of the ecospat package is to make available novel tools and methods to support spatial analyses and modeling of species niches and distributions in a coherent workflow. The package is written in the R language (R Development Core Team) and contains several features, unique in their implementation, that are complementary to other existing R packages. Pre‐modeling analyses include species niche quantifications and comparisons between distinct ranges or time periods, measures of phylogenetic diversity, and other data exploration functionalities (e.g. extrapolation detection, ExDet). Core modeling brings together the new approach of ensemble of small models (ESM) and various implementations of the spatially‐explicit modeling of species assemblages (SESAM) framework. Post‐modeling analyses include evaluation of species predictions based on presence‐only data (Boyce index) and of community predictions, phylogenetic diversity and environmentally‐constrained species co‐occurrences analyses. The ecospat package also provides some functions to supplement the ‘biomod2’ package (e.g. data preparation, permutation tests and cross‐validation of model predictive power). With this novel package, we intend to stimulate the use of comprehensive approaches in spatial modelling of species and community distributions.

Overcoming limitations of modelling rare species by using ensembles of small models

SummarySpecies distribution models (SDMs) have become a standard tool in ecology and applied conservation biology. Modelling rare and threatened species is particularly important for conservation purposes. However, modelling rare species is difficult because the combination of few occurrences and many predictor variables easily leads to model overfitting. A new strategy using ensembles of small models was recently developed in an attempt to overcome this limitation of rare species modelling and has been tested successfully for only a single species so far. Here, we aim to test the approach more comprehensively on a large number of species including a transferability assessment.For each species, numerous small (here bivariate) models were calibrated, evaluated and averaged to an ensemble weighted byAUCscores. These ‘ensembles of small models’ (ESMs) were compared to standardSDMs using three commonly used modelling techniques (GLM,GBMand Maxent) and their ensemble prediction. We tested 107 rare and under‐sampled plant species of conservation concern in Switzerland.We show thatESMs performed significantly better than standardSDMs. The rarer the species, the more pronounced the effects were.ESMs were also superior to standardSDMs and their ensemble when they were evaluated using a transferability assessment.By averaging simple small models to an ensemble,ESMs avoid overfitting without losing explanatory power through reducing the number of predictor variables. They further improve the reliability of species distribution models, especially for rare species, and thus help to overcome limitations of modelling rare species.