Article

Predicting the distribution potential of an invasive frog using remotely sensed data in Hawaii

Wiley
Diversity and Distributions
Authors:
To read the full-text of this research, you can request a copy directly from the authors.

Abstract

Aim Eleutherodactylus coqui (commonly known as the coqui) is a frog species native to Puerto Rico and non-native in Hawaii. Despite its ecological and economic impacts, its potential range in Hawaii is unknown, making control and management efforts difficult. Here, we predicted the distribution potential of the coqui on the island of Hawaii. Location Puerto Rico and Hawaii. Methods We predicted its potential distribution in Hawaii using five biophysical variables derived from Moderate Resolution Imaging Spectroradiometer (MODIS) as predictors, presence/absence data collected from Puerto Rico and Hawaii and three classification methods – Classification Trees (CT), Random Forests (RF) and Support Vector Machines (SVM). Results Models developed separately using data from the native range and the invaded range predicted potential coqui habitats in Hawaii with high performance. Across the three classification methods, mean area under the ROC curve (AUC) was 0.75 for models trained using the native range data and 0.88 for models trained using the invaded range data. We achieved the highest AUC value of 0.90 using RF for models trained with invaded range data. Main conclusions Our results showed that the potential distribution of coquis on the island of Hawaii is much larger than its current distribution, with RF predicting up to 49% of the island as suitable coqui habitat. Predictions also show that most areas with an elevation between 0 and 2000 m are suitable coqui habitats, whereas the cool and dry high elevation areas beyond 2000 m elevation are unsuitable. Results show that MODIS-derived biophysical variables are capable of characterizing coqui habitats in Hawaii.

No full-text available

Request Full-text Paper PDF

To read the full-text of this research,
you can request a copy directly from the authors.

... In the three decades since it established, coqui have spread across much of the Island of Hawai'i, but mainly in lowland areas according to the most current survey of their distribution in 2007 (Bisrat et al., 2012). More recently, it is thought that they have begun to spread into high-elevation habitats, although this has not been systematically documented. ...
... A presence-absence survey of the coqui frog was conducted in Hawai'i, USA from 3 to 15 May 2021. Survey points were created by overlaying a 1-km grid over a map of Hawai'i and selecting every other pixel of the grid intersecting with a major road on the island (Bisrat et al., 2012;Olson et al., 2012). Because the mating call of coqui frogs can be heard over 1000 m away (Olson et al., 2012), a distance of 2 km between survey sites allowed us to assume separate individuals were heard at each site. ...
... Roads were used to survey coqui because vehicular transport is a primary mode of spread for coqui frogs (Peacock et al., 2009), and because multiple sites can easily be visited within a short window of time. A total of 476 established survey points from a previous study (Bisrat et al., 2012), and 26 new survey points generated for this study were sampled (n = 502; Figure 1a). ...
Article
The coqui frog (Eleutherodactylus coqui) was introduced to the island of Hawai'i in the 1980s, and has spread across much of the island. There is concern they will invade higher elevation areas where negative impacts on native species are expected. It is not known if coqui change behavior and baseline physiology in ways that allow them to invade higher elevations. We investigated where coqui are found across the island and whether that includes recent invasion into higher elevations. We also investigated whether elevation is related to coqui's microhabitat use, including substrate use and height off the forest floor, and physiological metrics, including plasma osmolality, oxidative status, glucose, free glycerol, and triglycerides, that might be associated with invading higher elevations. We found coqui have increased the area they occupy along roads from 31% to 50% and have moved into more high-elevation locations (16% vs. 1%) compared to where they were found 14 years ago. We also found frogs at high elevation on different substrates and closer to the forest floor than frogs at lower elevations-perhaps in response to air temperatures which tended to be warmer close to the forest floor. We observed that blood glucose and triglycerides increase in frogs with elevation. An increase in glucose is likely an acclimation response to cold temperatures while triglycerides may also help frogs cope with the energetic demands of suboptimal temperatures. Finally, we found that female coqui have higher plasma osmolality, reactive oxygen metabolites (dROMs), free glycerol, and triglycerides than males. Our study suggests coqui behavior and physiology in Hawai'i may be influenced by elevation in ways that allow them to cope with lower temperatures and invade higher elevations.
... Random forests (RF; Breiman, 2001) has been widely applied in ecology for its advantages including (1) its very high accuracy, (2) its ability to determine variable importance, (3) its ability to model complex interactions between predictor variables, (4) the flexibility to perform several types of statistical data analysis including regression, classification and unsupervised learning, and (5) its ability to handle missing values (Cutler et al., 2007). Its high predictive capability has been supported by previous comparative studies with other machine learning techniques (Bisrat et al., 2012;Fukuda et al., 2013;Kampichler et al., 2010;Pino-Mejías et al., 2010). We used the randomForest package (Liaw and Wiener, 2002) of the R software (R Development Core Team, 2011), in which default settings were applied. ...
... As such, local optima do not exist and efficient optimization procedures can be used to find the global optimum. Given this merit and high predictive performance, SVMs have been widely applied to SDM studies (Bisrat et al., 2012;Fukuda et al., 2013;Kampichler et al., 2010;Pino-Mejías et al., 2010;Pouteau et al., 2012). ...
... In this study, all the models could achieve a very high accuracy (Tables 3 and 4). Among these models, RF and SVMs outperformed FHSMs, of which the high accuracy concurred with previous comparative studies (Bisrat et al., 2012;Fukuda et al., 2013;Kampichler et al., 2010;Peters et al., 2007;Pino-Mejías et al., 2010;Slabbinck et al., 2009), with RF being more accurate in most cases except Pouteau et al. (2012). This further supports the applicability of RF for species distribution modeling. ...
Article
The study of species' response is a key to understand the ecology of a species (e.g. critical habitat requirement and biological invasion processes) and design better conservation and management plans (e.g. problem identification, priority assessment and risk analysis). Predictive machine learning methods can be used as a tool for modelling species distributions as well as for describing important variables and specific habitat conditions required for a target species. This study aims (1) to demonstrate how habitat information such as species response curves can be retrieved from a species distribution model (SDM), (2) to assess the effects of data prevalence on model accuracy and habitat information retrieved from SDMs, and (3) to illustrate the differences between three data-driven methods, namely a fuzzy habitat suitability model (FHSM), random forests (RF) and support vector machines (SVMs). Nineteen sets of virtual species data with different data prevalences were generated using field-observed habitat conditions and hypothetical habitat suitability curves under four interaction scenarios governing the species-environment relationship for a virtual species. The effects of data prevalence on species distribution modelling were evaluated based on model accuracy and habitat information such as species response curves. Data prevalence affected both model accuracy and the assessment of species' response, with a stronger influence on the latter. The effects of data prevalence on model accuracy were less pronounced in the case of RF and SVMs which showed a higher performance. While the response curves were similar among the three models, data prevalence markedly affected the shapes of the response curves. Specifically, response curves obtained from a data set with higher prevalence showed higher tolerance to unsuitable habitat conditions, emphasizing the importance of accounting for data prevalence in the assessment of species-environment relationships. In a practical implementation of an SDM, data prevalence should be taken into account when interpreting the model results.
... Logistic regression is commonly used in species distribution modeling [20,21,37]. Ecologists have recently started using the random forest method as well, because it does not require any assumptions about the distribution of the data [38,39]. Random forest is a machine learning classification method that extends classification and regression tree (CART) approaches, which work by recursive binary partitioning of the data space into increasingly homogenous regions [39,40]. ...
... Ecologists have recently started using the random forest method as well, because it does not require any assumptions about the distribution of the data [38,39]. Random forest is a machine learning classification method that extends classification and regression tree (CART) approaches, which work by recursive binary partitioning of the data space into increasingly homogenous regions [39,40]. Random forest works by fitting and combining many CARTs to create a more accurate prediction [36,39]. ...
... Random forest is a machine learning classification method that extends classification and regression tree (CART) approaches, which work by recursive binary partitioning of the data space into increasingly homogenous regions [39,40]. Random forest works by fitting and combining many CARTs to create a more accurate prediction [36,39]. Both methods were used separately on the two datasets (one from the 10 by 10 km area and the other from the 15 monthly ground surveys in Aduoyo-Miyare and Nguka). ...
Conference Paper
Full-text available
The spatial distribution of malaria prevalence is often heterogeneous across a landscape, differing among households within a community. This is especially true in areas where community-wide malaria interventions are implemented and reduce malaria prevalence in most households. Understanding the spatial heterogeneity of malaria prevalence is especially important in this context of public health interventions in order to identify the factors that limit an intervention’s effectiveness. In the case of insecticide-treated bed nets (ITNs) and long lasting impregnated nets (LLINs), it is vital to understand the factors that influence the spatial distribution of the vectors. Using data from indoor resting mosquito collections and larval habitat surveys, we quantified the relative contributions of larval habitat spatial distribution and ITN/LLIN use on the spatial distributions of the malaria vector populations in a holoendemic region of western Kenya. We sampled 526 houses and collected 227 An. gambiae s.l. and 125 An. funestus females. Ownership and use of LLINs was high, but at least one person did not sleep under a bed net in 21% of the houses sampled. Larval Anopheles habitats were more likely to be found in areas of agricultural land use, closer to streams, and where topography favored the pooling of runoff water. Houses where at least one person did not sleep under an LLIN had more Anopheles females than houses where everyone slept under a bed net. Additionally, the number of An. gambiae s.l. females increased with the number of larval habitats within 50m of a house. While ITNs and LLINs have been shown to be effective at reducing malaria vector populations across broad scales, our results highlight fine scale factors influencing where vectors persist.
... Logistic regression is commonly used in species distribution modeling [20,21,37]. Ecologists have recently started using the random forest method as well, because it does not require any assumptions about the distribution of the data [38,39]. Random forest is a machine learning classification method that extends classification and regression tree (CART) approaches, which work by recursive binary partitioning of the data space into increasingly homogenous regions [39,40]. ...
... Ecologists have recently started using the random forest method as well, because it does not require any assumptions about the distribution of the data [38,39]. Random forest is a machine learning classification method that extends classification and regression tree (CART) approaches, which work by recursive binary partitioning of the data space into increasingly homogenous regions [39,40]. Random forest works by fitting and combining many CARTs to create a more accurate prediction [36,39]. ...
... Random forest is a machine learning classification method that extends classification and regression tree (CART) approaches, which work by recursive binary partitioning of the data space into increasingly homogenous regions [39,40]. Random forest works by fitting and combining many CARTs to create a more accurate prediction [36,39]. Both methods were used separately on the two datasets (one from the 10 by 10 km area and the other from the 15 monthly ground surveys in Aduoyo-Miyare and Nguka). ...
Article
Full-text available
Background Predictive models of malaria vector larval habitat locations may provide a basis for understanding the spatial determinants of malaria transmission. Methods We used four landscape variables (topographic wetness index [TWI], soil type, land use-land cover, and distance to stream) and accumulated precipitation to model larval habitat locations in a region of western Kenya through two methods: logistic regression and random forest. Additionally, we used two separate data sets to account for variation in habitat locations across space and over time. Results Larval habitats were more likely to be present in locations with a lower slope to contributing area ratio (i.e. TWI), closer to streams, with agricultural land use relative to nonagricultural land use, and in friable clay/sandy clay loam soil and firm, silty clay/clay soil relative to friable clay soil. The probability of larval habitat presence increased with increasing accumulated precipitation. The random forest models were more accurate than the logistic regression models, especially when accumulated precipitation was included to account for seasonal differences in precipitation. The most accurate models for the two data sets had area under the curve (AUC) values of 0.864 and 0.871, respectively. TWI, distance to the nearest stream, and precipitation had the greatest mean decrease in Gini impurity criteria in these models. Conclusions This study demonstrates the usefulness of random forest models for larval malaria vector habitat modeling. TWI and distance to the nearest stream were the two most important landscape variables in these models. Including accumulated precipitation in our models improved the accuracy of larval habitat location predictions by accounting for seasonal variation in the precipitation. Finally, the sampling strategy employed here for model parameterization could serve as a framework for creating predictive larval habitat models to assist in larval control efforts.
... Species distribution model uses associated animal presence data to predict the presence probability in other specific areas or the same area at a different time. It has been widely used to analyze the relationship between species and environment variables, especially the distribution patterns under the influence of climate change [2], the expanding area of invasive species [3], or shrinking habitat of endangered species [4]. Several existing species distribution methods such as BIOCLIM [5], DOMAIN, and maximum entropy method (MaxEnt) [2] can model potential distributions with presence-only data along with environmental information for the whole study area. ...
... Several existing species distribution methods such as BIOCLIM [5], DOMAIN, and maximum entropy method (MaxEnt) [2] can model potential distributions with presence-only data along with environmental information for the whole study area. For presence/absence data, logistic regression methods (LR), Support Vector Machines (SVM), and artificial neural network (ANN) are the most commonly used statistical procedures [3,6,7]. Indices derived from remote-sensing data such as the Normalized Difference Vegetation Index (NDVI) have been used extensively in species distribution models [4,6,[8][9][10][11]. ...
Article
Full-text available
With the application of various data acquisition devices, a large number of animal movement data can be used to label presence data in remote sensing images and predict species distribution. In this paper, a two-stage classification approach for combining movement data and moderate-resolution remote sensing images was proposed. First, we introduced a new density-based clustering method to identify stopovers from migratory birds’ movement data and generated classification samples based on the clustering result. We split the remote sensing images into 16 × 16 patches and labeled them as positive samples if they have overlap with stopovers. Second, a multi-convolution neural network model is proposed for extracting the features from temperature data and remote sensing images, respectively. Then a Support Vector Machines (SVM) model was used to combine the features together and predict classification results eventually. The experimental analysis was carried out on public Landsat 5 TM images and a GPS dataset was collected on 29 birds over three years. The results indicated that our proposed method outperforms the existing baseline methods and was able to achieve good performance in habitat suitability prediction.
... Because the Hawaiian bird community has so many rare and endemic species, understanding the effect of the coqui invasion on birds is important for guiding management decisions. For example, on the island of Hawaii, where the coqui is most widespread, 6 of 13 native bird species found in forests are listed as endangered (Banko and Banko 2009b), and could overlap with and be affected by the coqui, currently or in the future , Bisrat et al. 2012. The overall goal of our study was to determine, at the landscape scale, whether there are measurable differences in bird communities in areas where the coqui has invaded. ...
... We also measured habitat variables in our study plots, and no habitat variables that we measured, expect elevation, were different across the invasion fronts. The elevational difference that we found likely occurred because coquis first established in lowlands and tend to move upslope (Bisrat et al. 2012, Olson et al. 2012); however, it should be noted that the difference in elevation between plots with and without coquis was small (mean difference of 20 m) and probably not relevant for the bird species that we studied. We acknowledge that our design cannot completely rule out the possibility that coquis and birds are both responding to a variable that we did not measure. ...
Article
The global spread of invasive species has created significant challenges for avian conservation. Introduced predators and pathogens have long been recognized for their direct negative effects on birds, but introduced amphibians can reach high densities on islands with no native amphibians, where they interact with native species. The coqui frog (Eleutherodactylus coqui), introduced to the Hawaiian Islands in the late 1980s, could have significant impacts on birds because it is fully terrestrial and achieves high densities. Coquis have been hypothesized to compete with native birds for invertebrate prey, but could also serve as a novel food resource for birds that consume small vertebrates. To test whether coquis measurably affect bird abundance, we conducted point counts of birds in coqui-invaded and adjacent uninvaded plots across 15 sites on the island of Hawaii, USA. We used N-mixture models to estimate the effect of coqui presence and density on the abundances of both native and nonnative birds, while controlling for possible habitat differences between plots with and without coquis. We found that coquis were associated with ∼35% higher abundance of nonnative birds in general, and more specifically generalist birds that sometimes consume small vertebrates. We suggest that generalist birds increase in abundance with coquis primarily because coquis serve as an abundant food resource. While 4 native bird species co-occurred with coquis, native bird abundance (20% of our total observations) did not show a difference across coqui-invaded and uninvaded plots. Coquis do not appear to be important competitors with native birds in Hawaii, but the frogs are associated with increased abundances of some nonnative birds, which could induce undesirable ecosystem impacts.
... Coquis are now widespread on the island of Hawaii and have colonized many moist habitats, while they have been controlled or restricted on the other Hawaiian Islands (Beard et al. 2009;Bisrat et al. 2012;. They reproduce through direct development (Stewart and Woolbright 1996), and are terrestrial throughout all life stages. ...
... Manuka is one of the Fig. 4 Mean isotope values (±SE bars) for discriminationcorrected coqui, Hawaii amakihi, Japanese white-eye, red-billed leiothrix, and Hawaiian hoary bats plotted with invertebrates (gray) and plants (black) few mid-elevation areas where native birds are still abundant on the island of Hawaii. In many cases, native Hawaiian birds are restricted to elevations above 1500 m (Camp et al. 2009), where the coqui has not yet invaded or may be unable to invade because of colder temperatures (Bisrat et al. 2012;. ...
Article
Full-text available
Non-native amphibians often compete with native amphibians in their introduced range, but their competitive effects on other vertebrates are less well known. The Puerto Rican coqui frog (Eleutherodactylus coqui) has colonized the island of Hawaii, and has been hypothesized to compete with insectivorous birds and bats. To address if the coqui could compete with these vertebrates, we used stable isotope analyses to compare the trophic position and isotopic niche overlap between the coqui, three insectivorous bird species, and the Hawaiian hoary bat. Coquis shared similar trophic position to Hawaii amakihi, Japanese white-eye, and red-billed leiothrix. Coquis were about 3 % less enriched in d15N than the Hawaiian hoary bat, suggesting the bats feed at a higher trophic level than coquis. Analyses of potential diet sources between coquis and each of the three bird species indicate that there was more dietary overlap between bird species than any of the birds and the coqui. Results suggest that Acari, Amphipoda, and Blattodea made up[90% of coqui diet, while Araneae made up only 2% of coqui diet, but approximately 25% of amakihi and white-eye diet. The three bird species shared similar proportions of Lepidoptera larvae, which were *25% of their diet. Results suggest that coquis share few food resources with insectivorous birds, but occupy a similar trophic position, which could indicate weak competition. However, resource competition may not be the only way coquis impact insectivorous birds, and future research should examine whether coqui invasions are associated with changes in bird abundance.
... We also compare the results of dif-matic feature presented a barrier ferent niche modeling approaches. Such comparisons have been from occupying otherwise suitable performed on many taxa (Sehgal et al. 2011, Bisrat et al. 2012, gorical representation of the season Oppel et al. 2012), but have been rare for Bd. Using multiple sampled, niche modeling techniques allows us to distinguish between cli matic variables that have a consistent correlation with Bd infec-Variable selection: maximum relevance, minimum tion rates across approaches, and those that appear significant redundancy.-To ...
... Batracbochytrium Because it is extremely difficult dendrobatidis infection rates were lowest in areas with less than assumptions that best fit the distri 60 mm rainfall in the driest month, and increased with précipita-we used multiple modeling app tion above that threshold (Fig. 4). cess in extrapolating to areas not included in the training data Previous modeling studies on Behave identified precipitation (Bisrat et al. 2012, Oppel et al. 2012. To do so, we jackvariables as important to Bd distribution. ...
... The use of both the BRT and RF techniques is presently becoming increasingly popular (Elith et al., 2006;Williams et al., 2009;Bisrat et al., 2012). These machine-learning methods provide a number of advantages over the more traditional GLM approach, including: robust parameter estimates; model structure learned from data; and easy implementation of complex interactions. ...
... Whereas, RF modelling method is able to fit complex non-linear surfaces from high-dimensional input data (Cutler et al., 2007). The use of both the BRT and RF techniques is presently becoming increasingly popular (Elith et al., 2006;Williams et al., 2009;Bisrat et al., 2012). ...
... We also compare the results of different niche modeling approaches. Such comparisons have been performed on many taxa (Sehgal et al. 2011, Bisrat et al. 2012, Oppel et al. 2012, but have been rare for Bd. Using multiple niche modeling techniques allows us to distinguish between climatic variables that have a consistent correlation with Bd infection rates across approaches, and those that appear significant only as a result of the assumptions of one approach. ...
... Because it is extremely difficult, a priori, to identify the set of assumptions that best fit the distribution of Bd in our study area, we used multiple modeling approaches and compared their success in extrapolating to areas not included in the training data (Bisrat et al. 2012, Oppel et al. 2012. To do so, we jack-knifed over our collection locations and found the MSE, R 2 , and Pearson correlation coefficient of the predictions. ...
Article
We model Batrachochytrium dendrobatidis (Bd) infection rates in Jamaican frogs—one of the most threatened amphibian fauna in the world. The majority of species we surveyed were terrestrial direct-developing frogs or frogs that breed in tank bromeliads, rather than those that use permanent water bodies to breed. Thus, we were able to investigate the climatic correlates of Bd infection in a frog assemblage that does not rely on permanent water bodies. We sampled frogs for Bd across all of the major habitat types on the island, used machine learning algorithms to identify climatic variables that are correlated with infection rates, and extrapolated infection rates across the island. We compared the effectiveness of the machine learning algorithms for species distribution modeling in the context of our study, and found that infection rate rose quickly with precipitation in the driest month. Infection rates also increased with mean temperature in the warmest quarter until 22 °C, and remained relatively level thereafter. Both of these results are in accordance with previous studies of the physiology of Bd. Based on our environmental results, we suggest that frogs occupying high-precipitation habitats with cool rainy-season temperatures, though zcurrently experiencing low frequencies of infection, may experience an increase in infection rates as global warming increases temperatures in their habitat.
... Martinuzzi et al. 2009;Mochizuki & Murakami 2011). Similar to RF, applications of classification trees are diverse ranging from biogeographical, evolutionary to ecological issues (Bisrat et al. 2012;O'Connor et al. 1996;Olden et al. 2008). ...
... Similarly, Peters et al. (2007) concluded that on the whole RF leads to better predictive models than Multiple Logistic Regression (MLR) although both models showed high accuracy measures. Bisrat et al. (2012) determined RF and Support Vector Machines (SVM) to be superior to CT, in stability as well as in prediction accuracy. According to that, Garzón et al. (2006) found RF to perform better than Neural networks (NN) and Classification and regression trees (CART) showing the least accurate predictive model. ...
Article
Loss and deterioration of habitats are major threats for Tetrao urogallus in central Europe, where forests are highly fragmented and forest practices have distinctly changed during the last decades. Habitat models are important tools for conservation planning, often relying on presence-absence data. We mapped indirect signs of Tetrao urogallus presence as well as habitat variables over a series of seven study areas in the Austrian Alps, situated on limestone and on silicate rock. We modelled habitat use of Tetrao urogallus with one parametric approach (binary logistic regression) and two machine learning classification algorithms (classification trees and random forests) for both geological substrata separately. All three modelling approaches performed equally well in terms of accuracy or predictive power, but differed in model calibration. Three variables significantly contributed to all three habitat models on limestone and on silicate substrate, respectively, i.e. the cover of field-layer, the cover of dwarf shrubs and the proportion of deciduous trees in forest stands on limestone and the cover of field-layer, the canopy cover and the occurrence of Abies alba and/or Pinus sylvestris in forest stands on silicate rock. Some variables like the cover of Rubus sp. appeared in several models, which are not frequently mentioned in other studies. There have been some explanatory variables, which would have been missed, when applying just one single modelling approach, for example the occurrence of forest edges, the availability of canopy gaps and the supply of ant hills. Our results suggest fairly differing habitat management strategies on limestone and on silicate rock. Considering the large spatial requirements of Tetrao urogallus the necessity of active habitat management for Tetrao urogallus becomes obvious.
... Despite their utility, remotely sensed predictor variables remain underutilized in SDMs, possibly because the literature offers little guidance on appropriate datasets (Buermann et al., 2008) and interpretation of results obtained from remotely sensed data (Turner et al., 2003). Since the scale at which organisms perceive and interact with their environment is often much smaller than the scale at which many remotely sensed variables are obtained, concerns have also been raised as to whether remotely sensed data can be used to detect environmental variation at scales relevant to SDMs (Bistrat et al., 2011;Laurent et al., 2005). The accelerating availability of diverse, remotely sensed products has generated questions about which and how many parameters to incorporate into model building. ...
... Further research is required to confirm whether our approach, which is highly flexible, is applicable to other species, spatial scales, modeling algorithms, and predictor variables. Our study confirms that current remote sensors are able to provide environmental predictor variables relevant to SDMs (Laurent et al., 2005;Bradley and Fleishman, 2008;Bistrat et al., 2011). Despite challenges in interpreting remote sensing-based SDM output (Bradley and Fleishman, 2008;Cord and Rödder, 2011), remotely sensed datasets provide unbiased, high-resolution environmental data over larger areas through logistically and economically more efficient means than traditional field-based methods (Gillespie et al., 2008). ...
Article
Habitat assessments for biodiversity conservation are often complicated by the lack of detailed knowledge of a study species’ distribution. As an alternative to resource-intensive field-based methods to obtain such information, remotely sensed products can be utilized in species distribution models to infer a species’ distribution and ecological needs. Here we demonstrate how to arbitrate among a variety of remotely sensed predictor variables to estimate the distribution and ecological needs of an endangered butterfly species occurring mainly in inaccessible areas. We classified 19 continuous environmental predictor variables into three conceptually independent predictor classes, terrain, land cover, and vertical vegetation structure, and compared the accuracy of competing Maxent habitat models consisting of different combinations of each class. Each class contributed, though disproportionately, to our most reliable model that considered all 19 variables. We confirm that variables obtained from remote sensors can effectively estimate the distribution and ecological needs of a relatively unknown imperiled species occurring in inaccessible locations. Importantly, increasing the variety of predictor classes through multi-sensor fusion resulted in greater model accuracy than increasing the absolute number of predictor variables.
... A more extreme thermal acclimation might also be informative in assessing the limits of the coqui's acclimation ability, which could provide insight into the coldest possible environmental temperatures the coqui could tolerate. Relating these limits to the daily and annual minimum nocturnal temperatures across the island would be informative as to the potential spatial distribution of the coqui in Hawai'i (Bisrat et al., 2012). ...
Article
The coqui frog (Eleutherodactylus coqui) was introduced to the island of Hawai'i in the 1980s and has spread across much of the island. Concern remains that this frog will continue to expand its range and invade higher elevation habitats where much of the island's endemic species are found. We determined whether coqui thermal tolerance and physiology change along Hawai'i’s elevational gradients. We measured physiological responses using a short-term experiment to determine baseline tolerance and physiology by elevation, and a long-term experiment to determine the coqui's ability to acclimate to different temperatures. We collected frogs from low, medium, and high elevations. After both the short and long-term experiments, we measured critical thermal minimum (CTmin), blood glucose, oxidative stress, and corticosterone levels. CTmin was lower in high elevation frogs than low elevation frogs after the short acclimation experiment, signifying that they acclimate to local conditions. After the extended acclimation, CTmin was lower in frogs acclimated to cold temperatures compared to warm-acclimated frogs and no longer varied by elevation. Blood glucose levels were positively correlated with elevation even after the extended acclimation, suggesting glucose may also be related to lower temperatures. Oxidative stress was higher in females than males, and corticosterone was not significantly related to any predictor variables. The extended acclimation experiment showed that coquis can adjust their thermal tolerance to different temperatures over a 3-week period, suggesting the expansion of coqui into higher elevation habitats may still be possible, and they may not be as restricted by cold temperatures as previously thought.
... A more extreme thermal acclimation might also be informative in assessing the limits of the coqui's acclimation ability, which could provide insight into the coldest possible environmental temperatures the coqui could tolerate. Relating these limits to the daily and annual minimum nocturnal temperatures across the island would be informative as to the potential spatial distribution of the coqui in Hawai'i (Bisrat et al., 2012). ...
Article
The coqui frog (Eleutherodactylus coqui) was introduced to the island of Hawai‘i in the 1980s and has spread across much of the island. Concern remains that this invasive frog will continue to expand its range and invade higher elevation habitats where much of the island’s endemic species are found. We determined whether coqui thermal tolerance and physiology change along Hawai‘i’s elevational gradients. We measured physiological responses using a short-term experiment to determine baseline tolerance and physiology by elevation, and a long-term experiment to determine the coqui’s ability to acclimate to different temperatures. We collected frogs from low, medium, and high elevations. After both the short and long-term experiments, we measured critical thermal minimum (CTmin) and tested blood glucose, oxidative stress, and corticosterone levels. CTmin was lower in high elevation frogs than low elevation frogs after the short acclimation experiment. After the extended acclimation, CTmin was lower in frogs acclimated to cold temperatures compared to warm-acclimated frogs and no longer varied by elevation. Blood glucose levels had a positive correlation with elevation, suggesting that coqui frogs use glucose to cope with low temperatures. Oxidative stress was higher in females than males, and corticosterone was not significantly related to any predictor variables. Thermal tolerance to low temperatures was greater in high elevation-collected than low elevation-collected coqui, suggesting they acclimate to local temperatures. The extended acclimation experiment showed that coquis can adjust their thermal tolerance to different temperatures over a few weeks. However, glucose levels were higher with increasing elevation, even after the extended acclimation, suggesting inherent physiological differences among populations at different elevations. Because coqui can readily acclimate to lower temperatures after a 3-week period, this suggests expansion of coqui into higher elevation habitats is still possible, and they may not be as restricted by cold temperatures as previously thought.
... However, many studies of metapopulation dynamics that seek to understand for the factors that determine whether a species will exist at a location [7][8][9]. Large-scale monitoring programs for amphibian species [10][11][12] often relied on remotely sensed data (data on amphibians that has been gathered using biophysical variables derived from moderate resolution imaging spectroradiometers or from remote-sensing instruments on satellites) to depict spatial models in habitat occupancy [13][14][15]. Ignoring detectability may lead to biased estimations of site occupancy [16][17][18] and studies of habitat occupancy are often hampered by imperfect detectability for the species [1,[19][20][21][22]. ...
Chapter
Full-text available
We conducted a large-scale assessment at 35 primary forest sites and 42 secondary forest sites in Bach Ma National Park, central Vietnam, using the detection/non-detection data for each site over multiple visits, to quantify the site proportions that were occupied by granular spiny frogs (Quasipaa verrucospinosa). We additionally investigated the effect of site covariates (primary versus secondary forests) and sample covariates (temperature, humidity, and precipitation) to examine the environmental needs that may be incorporated for conserving rain forest amphibians in Vietnam. From the best model among all candidate models, We estimated a site occupancy probability of 0.632 that was higher than the naïve occupancy estimate of 0.403 and a 57% increase over the proportion of sites at which frogs were actually observed. The primary forest variable was an important determinant of site occupancy, whereas occupancy was not associated with the variable of secondary forest. In a combined AIC model weight, the detection model p (temperature, humidity, precipitation) included 90.9% of the total weight, providing clear evidence that environmental conditions were important sample covariates in modeling detection probabilities of granular spiny frogs. Our results substantiate the importance of incorporating occupancy and detection probabilities into studies of habitat relationships and suggest that the primary forest factor associated with environmental conditions influence the occupancy of granular spiny frogs.
... Although the terms species distribution modelling (SDM) and ENM are often used to describe the same concept, Feng et al. [55] suggested that the difference was that SDMs tend to focus more on the geographic distribution of species, while ENMs are more strongly geared to analyzing components of the fundamental ecological niche. These models have been used successfully to map species distributions for other snake species, including western rattlesnakes in California and Oregon, U.S.A. [56], midget faded rattlesnakes (Crotalus oreganus concolor) in Wyoming [57], eastern hog-nosed snakes (Heterodon platirhinos) in Ontario, Canada [58], invasive frogs in Hawaii [59] and amphibians in Europe [60]. Specifically, Spear et al. [57] used MaxEnt to determine factors relating to abundance of rattlesnakes, including distance to urban areas. ...
Article
Full-text available
Modelling the distribution and abundance of species at risk is extremely important for their conservation and management. We used ecological niche models (ENMs) to predict the occurrence of western rattlesnakes (Crotalus oreganus) in British Columbia (BC), Canada. We applied this to existing population estimates to support a threshold of occurrence for management and conservation. We also identified predictors influencing rattlesnake distribution and abundance in this region. Using a Geographic Information Systems platform, we incorporated ENMs, capture–mark–recapture (CMR) and radio-telemetry results, province-wide observations, Landsat imagery and provincial databases for agricultural land use to produce quantitative, spatially explicit, population estimates across BC. Using available western rattlesnake habitat estimated at 183.9 km2 and averaging estimates calculated from densities in three study populations, we generated a mean adult population size of 9722 (±SD 3009; 0.8 relative index of occurrence [RIO] threshold). Only a small area (21.6 km2) of suitable land cover was located within protected areas, potentially protecting an estimated 1144 (±354) adults. Most suitable land cover was within 500 m of roads (170.6 km2), representing potential habitat being used by an estimated 9017 (±2791) adults. At the threshold RIO value chosen (0.8), only a very small area of farmland provided suitable land cover. Our results highlight the possibility of high mortality rates for western rattlesnakes near roads and the fact that protected areas do not provide sufficient coverage to conserve the population. Given that this species has relatively low mobility and high site fidelity to home ranges, our population estimate for BC provides a useful reference for the northern part of the species’ range. It also fulfills a need to estimate population size within political jurisdictions where conservation management decisions are made, as well as presenting a method that can be applied to other parts of the range, including the southern United States. Our study provides an important benchmark for future monitoring of western rattlesnakes in BC using a repeatable and transparent approach. Similar applications can be extrapolated and applied for other threatened species to identify and quantify population distributions and threats, further supporting conservation prioritization tools to be used to maximize the effectiveness of conservation strategies under financial constraints.
... Additionally, remotely sensed variables are continuous observations without interpolation and geographical bias, and therefore with less uncertainty (He et al., 2015). Recent studies have revealed how remotely sensed LST data could improve species distribution modelling studies (e.g., Buermann et al., 2008;Bisrat et al., 2012;Still et al., 2013;Bobrowski et al., 2018). As time series data of vegetation characteristics (i.e., phenological metrics) are becoming more and more readily available, changing habitat suitability can be estimated and incorporated into model approaches. ...
Article
Full-text available
Modelling species across vast distributions in remote, high mountain regions like the Himalayas remains a challenging task. Challenges include, first and foremost, large-scale sampling of species occurrences and acquisition of sufficient high quality, fine-scale environmental parameters. We compiled a review of 157 Himalayan studies published between 2010 and 2021, aiming at identifying their main modelling objective in relation to the conceptualization of their methodological framework, evaluating origin of species occurrence data, taxonomic groups, spatial and temporal scale, selection of predictor variables and applied modelling algorithms. The majority of the analysed studies (40%) attempted to answer questions about potential range changes under future or past climatic conditions. The most studied organisms were trees (27%), followed by mammals (22%), herbaceous plants (20%), and birds (9%). For almost all studies we noted that a critical investigation and evaluation of input parameters and their ability to account for the species ecological requirements were neglected. Over 87% of all studies used Worldclim climate data as predictor variables, while around 50% of these studies solely relied on Worldclim climate data. Climate data from other sources were incorporated in only 7% and an additional 6% solely used remotely sensed predictors. Only around 2% of all studies attempted to compare the influence of different climate data sources on model performance. By far, Maxent was the most used modelling algorithm with 66%, followed by ensemble approaches (16%), whereas statistical modelling techniques lagged far behind (9%). Surprisingly, we found in 37% of the studies no interpretation on the relationship between the species and the predictor variables, while 27% of all studies included brief information, and 36% provided an elaborate, detailed interpretation on species ecological needs reflected in the final model. With this review we highlight the necessity to identify and reduce biases and uncertainty associated with species’ occurrence records and environmental data a priori. Since flawed input parameters produce misleading models without ecological causality, their implementation may have detrimental consequences when the best possible adaptation to future climatic conditions is at stake.
... Large-scale monitoring programs for amphibian species (Seber, 1982;Williams et al., 2002;Kaiser, 2008) of-ten relied on remotely sensed data (data on amphibians that has been gathered using biophysical variables derived from remote-sensing instruments on satellites or from moderate resolution imaging spectroradiometers) to depict spatial models in habitat occupancy (Carey et al., 2001;Shive et al., 2010;Bisrat et al., 2012). Ignoring detectability may lead to biased estimations of site occupancy (Tyre et al., 2003;Gu and Swihart, 2004;Weir et al., 2005) and studies of habitat occupancy are often hampered by imperfect detectability for the target species (MacKenzie et al., 2003(MacKenzie et al., , 2004Dorazio et al., 2006;Bailey et al., 2007;Nichols et al., 2007). ...
Article
Amphibian species are rarely detected with perfect accuracy, regardless of the method employed. A large-scale assessment forQuasipaa verrucospinosaoccupancy was conducted at 35 sites in the primary forest and 42 sites in the secondary forest of Bach Ma National Park, central Vietnam. Based on the detection data for each site, the distribution of Q. verrucospinosawas estimated in different habitat types using occupancy models. From the best model among all performed models, we estimated a site occupancy probability of 0.576 that was higher than the naive occupancy estimate of 0.403 and a 43.1% increase over the site proportion at whichQ. verrucospinosawas actually observed. The site covariate of the primary forest was an important determinant of site occupancy, which was not associated with the variable of secondary forest. In a combined AIC model weight: the p(temperature), p(humidity), and p(precipitation) models have 47.3, 67.1, and 90.9% of the total, respectively; providing evidence that aforementioned environmental conditions were important sample covariates in modelling detection probabilities ofQ. verrucospinosa. Our results substantiate the importance of incorporating detection and occupancy probabilities into studies of habitat relationships and suggest that the primary forests associated with weather conditions influence the site occupancy ofQ. verrucospinosain Bach Ma National Park, central Vietnam.
... Large-scale monitoring programs for amphibian species (Seber, 1982;Williams et al., 2002;Kaiser, 2008) of-ten relied on remotely sensed data (data on amphibians that has been gathered using biophysical variables derived from remote-sensing instruments on satellites or from moderate resolution imaging spectroradiometers) to depict spatial models in habitat occupancy (Carey et al., 2001;Shive et al., 2010;Bisrat et al., 2012). Ignoring detectability may lead to biased estimations of site occupancy (Tyre et al., 2003;Gu and Swihart, 2004;Weir et al., 2005) and studies of habitat occupancy are often hampered by imperfect detectability for the target species (MacKenzie et al., 2003(MacKenzie et al., , 2004Dorazio et al., 2006;Bailey et al., 2007;Nichols et al., 2007). ...
Article
Full-text available
Amphibian species are rarely detected with perfect accuracy, regardless of the method employed. A large-scale assessment for Quasipaa verrucospinosa occupancy was conducted at 35 sites in the primary forest and 42 sites in the secondary forest of Bach Ma National Park, central Vietnam. Based on the detection data for each site, the distribution of Q. verrucospinosa was estimated in different habitat types using occupancy models. From the best model among all performed models, we estimated a site occupancy probability of 0.576 that was higher than the naive occupancy estimate of 0.403 and a 43.1% increase over the site proportion at which Q. verrucospinosa was actually observed. The site covariate of the primary forest was an important determinant of site occupancy, which was not associated with the variable of secondary forest. In a combined AIC model weight: the p(temperature), p(humidity), and p(precipitation) models have 47.3, 67.1, and 90.9% of the total, respectively; providing evidence that aforementioned environmental conditions were important sample covariates in modelling detection probabilities of Q. verrucospinosa. Our results substantiate the importance of incorporating detection and occupancy probabilities into studies of habitat relationships and suggest that the primary forests associated with weather conditions influence the site occupancy of Q. verrucospinosa in Bach Ma National Park, central Vietnam.
... For comparison to the integrated PPM, we also fit SDMs using Random Forests (RF; Breiman, 2001) and Maxent (Phillips et al., 2006), machine learning algorithms that perform well at predicting species distributions (Bisrat et al., 2012;Roura-Pascual et al., 2009). The RF model was fit using presence/non-detection data for western spadefoots from the 273 ponds surveyed from 1990 to 1992. ...
Article
Determining the spatial scale at which landscape features influence population persistence is an important task for conservation planning. One challenge is that sampling biases confound factors that influence species occurrence and survey effort. Recent developments in Point Process Models (PPMs) enable researchers to disentangle the sampling process from ecological drivers of species' distributions. Land-cover change is a driver of decline for the western spadefoot (Spea hammondii), which has been extirpated from much of its range in California. Assessing this species' status requires information on the current distribution of suitable habitat within its historical range, but little is known about the effect of the landscape surrounding breeding ponds on spadefoot occurrence. Critically, surveys for western spadefoots often occur along roads, potentially biasing data used to fit species distribution models. We created PPMs integrating historical presence/non-detection and presence-only data for western spadefoots and land-cover data at multiple spatial scales to model the distribution of this species while removing the influence of sampling bias. There was spatial sampling bias in presence-only data; records were more likely to be reported near roads and urban centers and PPMs that removed sampling bias outperformed models that ignored sampling bias. The occurrence of western spadefoots was positively related to the proportion of grassland within a 2000 m buffer. The remaining habitat for western spadefoots is largely found in the foothills surrounding California's Central Valley. Our study illustrates how PPMs can improve projections of habitat suitability and our understanding of the drivers of species' distributions.
... These extremes were smoothed out, however, when using the EuroLST temperature averages over a 10-year period. Although the use of satellite-based LSTs for SDMs has been largely underexplored until now, our study adds to the growing list of recent studies indicating the potential of these untapped data resources for accurately predicting species distributions (see e.g., Bisrat, White, Beard, & Richard Cutler, 2012;Cord & Rödder, 2011;Neteler et al., 2013). We expect that LST timeseries with an even higher spatial resolution, such as Landsat (Cook, 2014), will turn out to be the crucial link between local-scale temperature measurements and global climate models. ...
Article
Full-text available
Aim: While species distribution models (SDMs) traditionally link species occurrences to free-air temperature data at coarse spatiotemporal resolution, the distribution of organisms might rather be driven by temperatures more proximal to their habitats. Several solutions are currently available, such as downscaled or interpolated coarse-grained free-air temperatures, satellite-measured land surface temperatures (LST) or in-situ measured soil temperatures. A comprehensive comparison of temperature data sources and their performance in SDMs is however currently lacking. Location: Northern Scandinavia Time period: 1970 - 2017 Major taxa studied: Higher plants Methods: We evaluated different sources of temperature data (WorldClim, CHELSA, MODIS, E-OBS, topoclimate and soil temperature from miniature data loggers), differing in spatial resolution (1’’ to 0.1°), measurement focus (free-air, ground-surface or soil temperature) and temporal extent (year-long vs. long-term averages), and use them to fit SDMs for 50 plant species with different growth forms in a high-latitudinal mountain region. Results: Differences between these temperature data sources originating from measurement focus and temporal extent overshadow the effects of temporal climatic differences and spatiotemporal resolution, with elevational lapse rates ranging from -0.6 °C per 100 m for long-term free-air temperature data to -0.2 °C per 100 m for in-situ soil temperatures. Most importantly, we found that the performance of the temperature data in SDMs depended on species’ growth forms. The use of in-situ soil temperatures improved the explanatory power of our SDMS (R² on average +16%), especially for forbs and graminoids (R²: +24% and +21% on average, respectively) compared to the other data sources. Main conclusions: We suggest future studies using SDMs to use the temperature dataset that best reflects the species’ ecology, rather than automatically using coarse-grained data from WorldClim or CHELSA.
... At such densities, their predation on invertebrates is great enough to influence herbivory, plant growth, and litter decomposition in ways that may confer a competitive advantage to invasive plants, further altering natural environments (Sin et al. 2008). On the island of Hawaii, where complete eradication is no longer believed possible (Beard et al. 2009;Bisrat et al. 2012), government action against the coqui focuses on preventing further spread. Control efforts have eradicated the coqui from the islands of Oahu and Kauai, and reduced the species to one population on Maui. ...
Article
Full-text available
The increasing worldwide spread of non-native species is both a component and a consequence of environmental change, and islands are especially vulnerable to negative effects. Efforts to control non-native species often include public education intended to promote behaviors designed to reduce or reverse their spread. To inform the use of information strategies to control the invasive, non-native frog Eleutherodactylus coqui in Hawaii, USA, we surveyed over 700 property owners about their attitudes and behaviors regarding the species. Included were residents of the island of Hawaii, where the species is common and management emphasizes prevention of further spread, and three other islands where the species is largely absent and management emphasizes detection and eradication. Where frogs are present, 61% of respondents reported taking actions to reduce their population, typically clearing vegetation or hand-capturing individual frogs. For these individuals, intentions to engage in future control activities were not significantly related to reports of past behavior. Intentions to participate in future control efforts on the island of Hawaii were best predicted by attitudes toward practices. On the other islands, behavioral intentions were best predicted by subjective norms (i.e., beliefs about others’ expectations that they should manage frogs). Thus, intentions to engage in non-native species management behaviors appear to be influenced by prior exposure to, and experience with, that species. Understanding the predictors of behavioral intentions at different stages of invasion have implications for the design of information strategies that can promote participation in control activities.
... However, the outcomes have so far only been occasionally used 212 as microclimatic data input in SDMs (e.g. Bisrat et al. 2012, Neteler et al. 2013, as IR images are 213 limited to surface temperatures, and suffer from either temporal extent or spatial resolution 214 limitations when using airborne or satellite-borne sensors, respectively (Potter et al. 2013). 215 ...
Article
Full-text available
Species distribution models (SDMs) are widely used to make predictions and assess questions regarding the spatial distribution and redistribution of species under environmental changes. Current SDMs are, however, often based on free‐air or synoptic temperature conditions with a coarse resolution, and thus may fail to capture apparent temperature (cf. microclimate) experienced by living organisms within their habitats. Microclimate is nevertheless crucial in habitats characterized by a vertical component (e.g. forests, mountains, or cities) or by horizontal variation in surface cover. The mismatch between how we usually express climate (cf. coarse‐grained free‐air conditions) and the apparent microclimatic conditions that living organisms experience has only recently been acknowledged in SDMs, yet several studies have already made considerable progress in tackling this problem from different angles. In this review, we summarize the currently available methods to obtain meaningful microclimatic data for use in distribution modelling. We discuss the issue of extent and resolution, and propose an integrated framework using a selection of appropriately‐placed sensors in combination with both detailed measurements of the habitat 3D structure, for example derived from digital elevation models or airborne laser scanning, and long‐term records of free‐air conditions from weather stations. As such, we can obtain microclimatic data with finer spatiotemporal resolution and of sufficient extent to model current and future species distributions. This article is protected by copyright. All rights reserved.
... In 2007, the population density of E. coqui was estimated to be 7780 AE 1708 adults and subadults/ha in Nanawale Forest Reserve and 7713 AE 1115 adults and subadults/ha in Lava Trees State Monument, with no significant differ- ence between M. polymorpha and F. moluccana stands (Table 1; McGuire 2008). E. coqui popula- tions had spread into the study area by the early 2000s, but significant areas of forest were still unoccupied in 2006 even though they were con- sidered suitable for E. coqui invasion ( Bisrat et al. 2012). Control treatments to reduce E. coqui num- bers on Hawai'i Island did not start until 2007, a year after field work for this study was completed (Beard and Pitt 2012). ...
Article
Full-text available
Invasions of introduced species have homogenized ecological communities worldwide, leading to losses of native species and the services they provide. Some of these invaders substantially alter nutrient cycling, which changes conditions for all other organisms, but less is known about the potential influence of these species on nitrogen (N) trace gas emissions that affect atmospheric processes. We used a natural experiment to explore whether the establishment of an introduced nitrogen (N) fixing tree (Falcataria moluccana) and recent invasion of an amphibian predator, the Caribbean tree frog (Eleutherodactylus coqui), into native Hawaiian rainforests have affected soil emissions of nitrous oxide (N2O) and nitric oxide (NO), two atmospherically important trace gases produced by soil microorganisms. Soil N2O and NO emissions and rates of soil N cycling were significantly higher in F. moluccana‐dominated stands compared to native Metrosideros polymorpha (Ohi'a) stands. Additionally, invasion of E. coqui frogs moderately increased soil N2O emissions, primarily in non‐native F. moluccana forests where soil N availability was already elevated. N2O emissions were positively and significantly related to net potential N mineralization, and total N2O+NO fluxes increased with soil nitrate (NO3−) concentration and rates of nitrification. Previous work in these Hawaiian rainforest sites has shown that F. moluccana substantially increases N availability by increasing ecosystem N supply compared to uninvaded stands, and E. coqui accelerates N availability and litter decomposition, although moderately, due to enhanced fluxes of nutrient‐rich waste products. Here, we show that acceleration of nutrient cycling by introduced species can also alter biosphere–atmosphere exchange of N‐oxides.
... The invasion of the insectivorous frog, Eleutherodactylus coqui (Anura: Leptodactylidae; hereafter "coqui"), into the Hawaiian Islands in the 1980s has raised concerns about its impact on Hawaiian forest communities [13][14][15][16]. On Hawaii Island, coqui are now an abundant top consumer in low-to mid-elevation forest communities that lack an evolutionary history with amphibian predators. ...
Article
Full-text available
Invasive predators can profoundly impact native communities, especially in insular ecosystems where functionally equivalent predators were evolutionarily absent. Beyond direct consumption, predators can affect communities indirectly by creating or altering food web linkages among existing species. Where invasive predators consume prey from multiple distinct resource channels, novel links may couple the dynamics of disjunct modules and create indirect interactions between them. Our study focuses on invasive populations of Eleutherodactylus coqui (Anura: Leptodactylidae) on Hawaii Island. Coqui actively forage in the understory and lower canopy at night but return to the forest floor and belowground retreats by day. Recent dietary studies using gut contents and naturally occurring stable isotopes indicate higher than expected consumption of litter arthropods, which in these Hawaiian forests are primarily non-native species. We used laboratory studies to observe diurnal and nocturnal foraging behavior, and experimental field additions of C4 vegetation as a litter tracer to distinguish epigaeic sources from food web pools in the C3 canopy. Lab trials revealed that prey consumption during diurnal foraging was half that consumed during nocturnal foraging. Analysis of δ¹³C isotopes showed incorporation of C4 carbon into litter arthropods within one month, and Bayesian mixing models estimated that 15–25% of the carbon in coqui tissue was derived from litter sources. These results support recent findings that E. coqui are not quiescent diurnally but instead actively forage. Such activity by a mobile invasive predator may introduce a novel linkage that integrates detrital and foliar resource pools, potentially distributing influences of invasive litter arthropods through the broader system to amplify impacts on native species.
... Major advantages of LST-related variables include continuous observations without interpolation and geographical bias and therefore fewer uncertainties [23]. Recent studies showed how LST data could improve species modelling studies (e.g., [17,96,97]). These parameters offer numerous possibilities, such as tailored predictors in high resolution. ...
Article
Full-text available
Modelling ecological niches across vast distribution ranges in remote, high mountain regions like the Himalayas faces several data limitations, in particular nonavailability of species occurrence data and fine-scale environmental information of sufficiently high quality. Remotely sensed data provide key advantages such as frequent, complete, and long-term observations of land surface parameters with full spatial coverage. The objective of this study is to evaluate modelled climate data as well as remotely sensed data for modelling the ecological niche of Betula utilis in the subalpine and alpine belts of the Himalayan region covering the entire Himalayan arc. Using generalized linear models (GLM), we aim at testing factors controlling the species distribution under current climate conditions. We evaluate the additional predictive capacity of remotely sensed variables, namely remotely sensed topography and vegetation phenology data (phenological traits), as well as the capability to substitute bioclimatic variables from downscaled numerical models by remotely sensed annual land surface temperature parameters. The best performing model utilized bioclimatic variables, topography, and phenological traits, and explained over 69% of variance, while models exclusively based on remotely sensed data reached 65% of explained variance. In summary, models based on bioclimatic variables and topography combined with phenological traits led to a refined prediction of the current niche of B. utilis, whereas models using solely climate data consistently resulted in overpredictions. Our results suggest that remotely sensed phenological traits can be applied beneficially as supplements to improve model accuracy and to refine the prediction of the species niche. We conclude that the combination of remotely sensed land surface temperature parameters is promising, in particular in regions where sufficient fine-scale climate data are not available.
... For both animal as well as plant species, SRS-based species distribution models are a key source of information to identify areas that are vulnerable to invasion, both now and in the future. For instance, Bisrat et al. (2012) modelled the habitat of an invasive frog (Eleutherodactylus coqui) in its native Puerto Rico, and used this information to predict its potential distribution in Hawaii (where it is invasive. Similarly, Roura-Pascual et al. (2004) used SRS-based habitat modelling to identify areas vulnerable to invasion by Argentine fire ants (Linepithema humile) worldwide, and Clark et al. (2014) used remotely sensed information about vegetation phenology to identify areas along the Appalachian trail (USA) which may be vulnerable to the spread of tree-of-heaven (Ailanthus altissima), an invasive shrub, as part of a natural resource management support system. ...
Technical Report
Full-text available
This latest chapter in the Conservation Technology Series from WWF-UK looks at the opportunities, challenges and state-of-the-art of satellite remote sensing for conservation applications. This issue reviews available satellite imagery and derived datasets, a comprehensive guide to data sources, common processing workflows and case studies.
... To improve the ecological niche model (ENM) we used a combination of bioclimatic data and remote sensing data as environmental layers (see Buermann et al., 2008;Bisrat et al., 2012). For the bioclimatic data, we used 19 variables obtained from the WorldClim database (www.worldclim.org ...
Article
Full-text available
We reviewed the distribution of Dendropsophus columbianus and introduced three new localities for the Andean region in Department of Cauca: i) the upper Patía River basin; ii) the western slopes of the Central Cordillera (Quintana, Popayán municipality); and iii) the western slopes of the Western Cordillera in El Tambo municipality, Colombia. The Patía River record extends its distribution 40 km south from previously known records and constitutes the lowest confirmed elevation known for the species (700 m a.s.l.). The Quintana record extends the upper high altitudinal distribution by 160 meters. Our records also extend the distribution of the species to two additional river basins (Patía and San Joaquín rivers). We also reviewed extra-distributional records outside the Andean region available from the literature from two localities (Quimarí and Guapi) in the Caribbean and Pacific regions of Colombia. Specimens from Quimarí were re-identified as D. phlebodes; and the specimens from Guapi belong to D. columbianus but the locality is likely erroneous. Using the validated localities, we built a distribution model corroborating its high association to Andean ecoregions in the Western and Central cordilleras of Colombia. The potential distribution of the species extends from Nariño to Antioquia following the Cauca River basin. Although the model shows that the presence of D. columbianus in Antioquia (north) and Nariño (south) is likely, the northern and southern boundaries of the species distribution is unclear due to the absence of records. Future expeditions are necessary to verify the limits of the species distribution.
... Species distribution models are typically fit to measured or interpolated climate or environmental measurements of the terrestrial environment. This is because the most reliably available climate variables come from global datasets (Hijmans et al. 2005, Tyberghein et al. 2012) or remote sensing (Cord and R€ odder 2011,Bisrat et al. 2012) that primarily measures terrestrial climate variables. These data are often only an indirect indicator of freshwater environmental conditions, and several physical characteristics of water bodies can strongly influence the correlation between atmospheric and aquatic conditions (Mohseni and Stefan 1999). ...
Article
Full-text available
Efficient management and prevention of species invasions requires accurate prediction of where species of concern can arrive and persist. Species distribution models provide one way to identify potentially suitable habitat by developing the relationship between climate variables and species occurrence data. However, these models when applied to freshwater invasions are complicated by two factors. The first is that the range expansions that typically occur as part of the invasion process violate standard species distribution model assumptions of data stationarity. Second, predicting potential range of freshwater aquatic species is complicated by the reliance on terrestrial climate measurements to develop occurrence relationships for species that occur in aquatic environments. To overcome these obstacles, we combined a recently developed algorithm for species distribution modeling—range bagging—with newly available aquatic habitat-specific information from the North American Great Lakes region to predict suitable habitat for three potential invasive species: golden mussel, killer shrimp, and northern snakehead. Range bagging may more accurately predict relative suitability than other methods because it focuses on the limits of the species environmental tolerances rather than central tendency or “typical” cases. Overlaying the species distribution model output with aquatic habitat-specific data then allowed for more specific predictions of areas with high suitability. Our results indicate there is suitable habitat for northern snakehead in the Great Lakes, particularly shallow coastal habitats in the lower four Great Lakes where literature suggests they will favor areas of wetland and submerged aquatic vegetation. These coastal areas also offer the highest suitability for golden mussel, but our models suggest they are marginal habitats. Globally, the Great Lakes provide the closest match to the currently invaded range of killer shrimp, but they appear to pose an intermediate risk to the region. Range bagging provided reliable predictions when assessed either by a standard test set or by tests for spatial transferability, with golden mussel being the most difficult to accurately predict. Our approach illustrates the strength of combining multiple sources of data, while reiterating the need for increased measurement of freshwater habitat at high spatial resolutions to improve the ability to predict potential invasive species.
... The components for the first scenario (wind turbine development) are depicted spatially in Fig. 1; the benefit component and final output for the second scenario (biodiversity conservation) are depicted in Fig. 3 (Hijmans et al., 2014). The random forest algorithm has been shown to be quite robust and accurate for this type of spatial modeling (Bisrat et al., 2012;Iverson et al., 2004;Lawler et al., 2006). We tested the accuracy of the model by applying leave-one-out cross-validation for all turbines grouped into 100 groups based on their location (Fielding, 2002). ...
... Allele des ob bei der Vorhersage von Risikogebieten das gesamte Verbreitungsgebiet in die Modellierung einbezogen werden sollte oder entweder nur das heimische Verbrei- tungsgebiet (Abb. 10) (Broennimann & Guisann 2008, Peterson & Viglais 2001) oder nur nicht-heimische Areale ( Bisrat et al. 2012). Eng verbunden damit ist die Frage, ob eine In- vasion einer gebietsfremden Art mit einer Änderung der realisierten oder gar der fundamentalen Nische stattgefunden hat ( Broennimann et al. 2007). ...
Article
Full-text available
Understanding range limitations of species fascinates researchers for more than a hundred years and is the core of Biogeography. Since increasing anthropogenic pressures summarized as global change alters the earth this knowledge has become more important than ever. Like biodiversity in general, birds are suffering from these pressures but due to their enormous popularity they are among the best studied organisms in this regard. Therefore, this review aims to summarize the state of knowledge of avian range dynamics under global change. We first briefly introduce the general biogeographical concepts, history, and future prospects in a changing world from a bird’s eye. Further, we point to problems of overemphasizing climate change as single driver of distributional changes, while neglecting other factors such as changes in habitats or food availability that highly correlate and interact with climate change in time. To this end, we critically discuss the emergence and use of species distribution modeling for assessing and predicting bird ranges under various conditions. Finally, we review various key processes that, next to climate, affect distributions and illustrate them with examples from the ornithological literature. We call for an enhanced consideration and an improved integration of these various processes into a holistic perspective when assessing future responses of bird distributions to changing environmental conditions.
... for Europe (Metz et al. 2014). MODIS LST data are increasingly being used in SDMs to understand and predict ecological processes Bisrat et al. 2012;Neteler et al. 2013;Pau et al. 2013;Still et al. 2014). Recently, efforts have been made to use LST to facilitate interpolation of weather station data as weather station data have a long temporal span, which cannot be fully covered by remote sensing data (Parmentier et al. 2015). ...
Article
Full-text available
Two prominent limitations of species distribution models (SDMs) are spatial biases in existing occurrence data and a lack of spatially explicit predictor variables to fully capture habitat characteristics of species. Can existing and emerging remote sensing technologies meet these challenges and improve future SDMs? We believe so. Novel products derived from multispectral and hyperspectral sensors, as well as future Light Detection and Ranging (LiDAR) and RADAR missions, may play a key role in improving model performance. In this perspective piece, we demonstrate how modern sensors onboard satellites, planes and unmanned aerial vehicles are revolutionizing the way we can detect and monitor both plant and animal species in terrestrial and aquatic ecosystems as well as allowing the emergence of novel predictor variables appropriate for species distribution modeling. We hope this interdisciplinary perspective will motivate ecologists, remote sensing experts and modelers to work together for developing a more refined SDM framework in the near future.
... for Europe (Metz et al. 2014). MODIS LST data are increasingly being used in SDMs to understand and predict ecological processes Bisrat et al. 2012;Neteler et al. 2013;Pau et al. 2013;Still et al. 2014). Recently, efforts have been made to use LST to facilitate interpolation of weather station data as weather station data have a long temporal span, which cannot be fully covered by remote sensing data (Parmentier et al. 2015). ...
Article
Full-text available
Two prominent limitations of species distribution models (SDMs) are spatial biases in existing occurrence data and a lack of spatially explicit predictor variables to fully capture habitat characteristics of species. Can existing and emerging remote sensing technologies meet these challenges and improve future SDMs? We believe so. Novel products derived from multispectral and hyper-spectral sensors, as well as future Light Detection and Ranging (LiDAR) and RADAR missions, may play a key role in improving model performance. In this perspective piece, we demonstrate how modern sensors onboard satellites, planes and unmanned aerial vehicles are revolutionizing the way we can detect and monitor both plant and animal species in terrestrial and aquatic ecosystems as well as allowing the emergence of novel predictor variables appropriate for species distribution modeling. We hope this interdisciplinary perspective will motivate ecologists, remote sensing experts and modelers to work together for developing a more refined SDM framework in the near future.
... In this regard, analyzing niche relationships between species can provide more accurate species distribution maps. Species distribution models, also known as habitat suitability models, have been widely used in different fields of ecology, such as control of invasive species (Bisrat et al., 2012;Gallien et al., 2010;Václavík and Meentemeyer, 2009), effect of climate and environmental change on species distribution (Franklin, 2010;Tang and Beckage, 2010;Taylor and Kumar, 2013), design of biodiversity protective network (Wilson et al., 2005), and conservation biology (Nazeri et al., 2012(Nazeri et al., , 2014Rood et al., 2010). These models are based on Hutchinson's (1957) fundamental niche theory, which lies at the core of ecological research, arguing that each species has a unique, ndimensional array of ecological resources and environmental needs. ...
... island of Hawaii in the late 1980s, the coqui's range has increased (Kraus et al., 1999;Kraus and Campbell, 2002) and it is listed as one of the 100 ''world's worst'' invaders (ISSG, 2005) and is designated as a 'pest' and 'injurious wildlife' by the Hawaii Department of Agriculture . The coqui is expected to continue expanding its range on the island, and eradication is no longer believed to be possible, even though control efforts have been quite successful on Oahu and Kauai while the populations were still restricted in range (Beard et al., 2009;Bisrat et al., 2012). One reason that eradication on the island of Hawaii is viewed as no longer possible is because the coqui is often found on private properties, which means that landowners have to agree to and even participate in control operations. ...
... Despite the fact that correlative species distribution models (or ecological niche models; Peterson et al. 2011;Anderson 2012) are often developed at regional or continental extents, the number of studies employing remote sensing data is still relatively small; notable exceptions include Bradley & Fleishman (2008), Buermann et al. (2008), Saatchi et al. (2008), Tuanmu et al. (2010), Bisrat et al. (2012 and Papes ß, Peterson & Powell (2012). These pioneering studies have used remote sensing data to characterize land cover and vegetation indices, often in combination with WorldClim data, to improve prediction of species ranges (reviews by Gillespie et al. 2008;Boyd & Foody 2011;Cord et al. 2013). ...
Article
Remote sensing techniques offer an opportunity to improve biodiversity modelling and prediction world‐wide. Yet, to date, the weather station‐based WorldClim data set has been the primary source of temperature and precipitation information used in correlative species distribution models. WorldClim consists of grids interpolated from in situ station data recorded primarily from 1960 to 1990. Those data sets suffer from uneven geographic coverage, with many areas of Earth poorly represented. Here, we compare two remote sensing data sources for the purposes of biodiversity prediction: MERRA climate reanalysis data and AMSR ‐E, a pure remote sensing data source. We use these data to generate novel temperature‐based bioclimatic information and to model the distributions of 20 species of vertebrates endemic to four regions of South America: Amazonia, the Atlantic Forest, the Cerrado and Patagonia. We compare the bioclimatic data sets derived from MERRA and AMSR ‐E information with in situ station data and contrast species distribution models based on these two products to models built with WorldClim. Surface temperature estimates provided by MERRA and AMSR ‐E showed warm temperature biases relative to the in situ data fields, but the reliability of these data sets varied in geographic space. Species distribution models derived from the MERRA data performed equally well (in Cerrado, Amazonia and Patagonia) or better (Atlantic Forest) than models built with the WorldClim data. In contrast, the performance of models constructed with the AMSR ‐E data was similar to (Amazonia, Atlantic Forest, Cerrado) or worse than (Patagonia) that of models built with WorldClim data. Whereas this initial comparison assessed only temperature fields, efforts to estimate precipitation from remote sensing information hold great promise; furthermore, other environmental data sets with higher spatial and temporal fidelity may improve upon these results.
... Whereas in some studies only locations from the native range have been used (Peterson & Viglais 2001, other studies have used occurrences from the non-native range (Mau-Crimmins et al. 2006). In a recent study, Bisrat et al. (2012) found higher area under the curve (AUC) values for data from the non-native range when predicting another non-native range. In contrast, Broennimann and Guisan (2008) recommended that occurrences from both the native and the non-native range should be used to capture a more comprehensive view on the realized niche of the species (see also Jim enez-Valverde et al. 2011). ...
Article
Non-native species can have severe impacts on ecosystems. Therefore, predictions of potentially suitable areas that are at risk of the establishment of non-native populations are desirable. In recent years, species distribution models (SDMs) have been widely applied for this purpose. However, the appropriate selection of species records, whether from the native area alone or also from the introduced range, is still a matter of debate. We combined analyses of native and non-native realized climate niches to understand differences between models based on all locations, as well as on locations from the native range only. Our approach was applied to four estrildid finch species that have been introduced to many regions around the world. Our results showed that SDMs based on location data from native areas alone may underestimate the potential distribution of a given species. The climatic niches of species in their native ranges differed from those of their non-native ranges. Niche comparisons resulted in low overlap values, indicating considerable niche shifts, at least in the realized niches of these species. All four species have high potential to spread over many tropical and subtropical areas. However, transferring these results to temperate areas has a high degree of uncertainty, and we urge caution when assessing the potential spread of tropical species that have been introduced to higher latitudes.
... One such landscape based approach, species distribution modeling (SDM), has been used extensively to understand both single and multi-species (i.e. richness/diversity) distributions in both natural and invaded landscapes [25][26][27]. The multitude of SDM methodologies all have the ability (with varying accuracy) to both define and predict the theorized realized niche of an organism (based on biotic and abiotic variables), and project that habitat onto specific climate change scenarios [26,28,29]. ...
Article
Full-text available
Occupation of native ecosystems by invasive plant species alters their structure and/or function. In Hawaii, a subset of introduced plants is regarded as extremely harmful due to competitive ability, ecosystem modification, and biogeochemical habitat degradation. By controlling this subset of highly invasive ecosystem modifiers, conservation managers could significantly reduce native ecosystem degradation. To assess the invasibility of vulnerable native ecosystems, we selected a proxy subset of these invasive plants and developed robust ensemble species distribution models to define their respective potential distributions. The combinations of all species models using both binary and continuous habitat suitability projections resulted in estimates of species richness and diversity that were subsequently used to define an invasibility metric. The invasibility metric was defined from species distribution models with <0.7 niche overlap (Warrens I) and relatively discriminative distributions (Area Under the Curve >0.8; True Skill Statistic >0.75) as evaluated per species. Invasibility was further projected onto a 2100 Hawaii regional climate change scenario to assess the change in potential habitat degradation. The distribution defined by the invasibility metric delineates areas of known and potential invasibility under current climate conditions and, when projected into the future, estimates potential reductions in native ecosystem extent due to climate-driven invasive incursion. We have provided the code used to develop these metrics to facilitate their wider use (Code S1). This work will help determine the vulnerability of native-dominated ecosystems to the combined threats of climate change and invasive species, and thus help prioritize ecosystem and species management actions.
... Among various machine learning methods, Random Forests (RF; Breiman, 2001) has been widely applied to many disciplines because of the following advantages: (1) very high accuracy; (2) a novel method of determining variable importance; (3) ability to model complex interactions among predictor variables; (4) flexibility to perform several types of statistical data analyses including regression, classification, survival analysis and unsupervised learning; and (5) an algorithm for imputing missing values (Cutler et al., 2007). The highly predictive capability of RF has been supported by previous comparative studies with other machine learning techniques (Benito Garzón et al., 2006;Slabbinck et al., 2009;Kampichler et al., 2010;Pino-Mejías et al., 2010;Bisrat et al., 2012). However, so far no study has applied RF to estimate the yield of agricultural products (but see Vincenzi et al., 2011 for an RF application in clam fisheries). ...
Article
Mango (Mangifera indica L.) is one of the major tropical fruits exported through long supply chains to export markets. Production of high quality fruits and monitoring postharvest changes during storage and transport are thus primary concerns for exporters to ensure the premium value of fresh mango fruit after distribution. This study aims to demonstrate the applicability of Random Forests (RF) for estimating the internal qualities of mango based on peel colour. Two cultivars, namely Nam Dokmai and Irwin, having different fruit properties and grown in intensively managed orchards in Thailand and Japan, respectively, were used in this study. Postharvest changes in peel colour and fruit quality were observed under three storage conditions with respect to temperature. RF models were applied to establish a relationship between peel colour and fruit quality, and then tested the applicability based on model accuracy and variable importance computed by the RF. Specifically, this work demonstrates how the variable importance can be used to interpret the model results. The high accuracy and the information retrieved by the RF models suggest the applicability and practicality as a non-destructive assessment method for the quality of fresh mango fruit.
... Random Forest (Breiman, 2001) is a machine learning technique which has revealed itself as a very convenient tool in ecological studies (Cutler et al., 2007), including its use as a SDM technique (Sehgal et al., 2010;Bisrat et al., 2012;Tôrres et al., 2012). Random Forest fits many regression trees to a data set, and then combines the predictions from all the trees. ...
Article
The importance of parasitism for host populations depends on local parasite richness and prevalence: usually host individuals face higher infection risk in areas where parasites are most diverse, and host dispersal to or from these areas may have fitness consequences. Knowing how parasites are and will be distributed in space and time (in a context of global change) is thus crucial from both an ecological and a biological conservation perspective. Nevertheless, most research articles focus just on elaborating models of parasite distribution instead of parasite diversity. We produced distribution models of the areas where haemosporidian parasites are currently highly diverse (both at community and within-host levels) and prevalent among Iberian populations of a model passerine host: the blackcap Sylvia atricapilla; and how these areas are expected to vary according to three scenarios of climate change. Based on these models, we analysed whether variation among populations in parasite richness or prevalence are expected to remain the same or change in the future, thereby reshuffling the geographic mosaic of host-parasite interactions as we observe it today. Our models predict a rearrangement of areas of high prevalence and richness of parasites in the future, with Haemoproteus and Leucocytozoon parasites (today the most diverse genera in blackcaps) losing areas of high diversity and Plasmodium parasites (the most virulent ones) gaining them. Likewise, the prevalence of multiple infections and parasite infracommunity richness would be reduced. Importantly, differences among populations in the prevalence and richness of parasites are expected to decrease in the future, creating a more homogeneous parasitic landscape. This predicts an altered geographic mosaic of host-parasite relationships, which will modify the interaction arena in which parasite virulence evolves. This article is protected by copyright. All rights reserved.
... To complement the habitat preference information retrieved from the FHPM, the Random Forests (RF; Breiman, 2001) method was used, since in contrast to the FHPM, variable importance can be derived from RF. RF has been widely applied for its advantages including (1) its very high accuracy, (2) the ability to determine variable importance, (3) its ability to model complex interactions between predictor variables, (4) the flexibility to perform several types of statistical data analyses including regression, classification, survival analysis and unsupervised learning, and (5) its ability to handle missing values (Cutler et al., 2007). Its high predictive capability has been supported by previous comparative studies with other machine learning techniques (Benito Garzón et al., 2006;Slabbinck et al., 2009;Kampichler et al., 2010;Pino-Mejías et al., 2010;Mouton et al., 2011;Bisrat et al., 2012). ...
Article
Topmouth gudgeon ( Pseudorasbora parva ) is widely known as a highly invasive freshwater fish and has expanded from East Asia (native range) to Central Asia, Europe and Northern Africa (introduced range). Although the relationship between the occurrence of P . parva and its habitat conditions remains unclear, information on factors affecting its distribution, especially in its native range, is important for predicting its expansion. This study provides primary information on the distribution of P . parva in rivers and agricultural canals in northern Kyushu Island, Japan, where the fish is native. Fuzzy habitat preference models (FHPMs) and Random Forests (RF) were applied to link landscape features to the distribution of P . parva based on field observation data collected from two distinct ecoregions, the north‐western (NW) and north‐eastern (NE) parts of Kyushu Island. The results show a clear habitat preference of P . parva for areas with a lower elevation, a gentler slope and a smaller number of river‐to‐river connections as general landscape features across the ecoregions. Weak preferences are observed for sites with a higher number of river‐to‐canal connections, a higher canal network index, a larger area of paddy fields, a larger residential area, more crop fields and fewer forests and orchards. Of these site‐specific features, five landscape features – elevation, slope, canal network index, area of paddy fields, and presence of forests and orchards – are identified as the most important features for predicting the distribution of P . parva . The general and specific habitat preference information, as demonstrated in this study, may be important in biogeography and invasion ecology. Further research is needed to accumulate information for a better understanding of the invasion ecology and the design of improved management and control strategies against P . parva . Copyright © 2013 John Wiley & Sons, Ltd.
... To complement the habitat preference information retrieved from the FHPM, the Random Forests (RF; Breiman, 2001) method was used, since in contrast to the FHPM, variable importance can be derived from RF. RF has been widely applied for its advantages including (1) its very high accuracy, (2) the ability to determine variable importance, (3) its ability to model complex interactions between predictor variables, (4) the flexibility to perform several types of statistical data analyses including regression, classification, survival analysis and unsupervised learning, and (5) its ability to handle missing values (Cutler et al., 2007). Its high predictive capability has been supported by previous comparative studies with other machine learning techniques (Benito Garzón et al., 2006;Slabbinck et al., 2009;Kampichler et al., 2010;Pino-Mejías et al., 2010;Mouton et al., 2011;Bisrat et al., 2012). ...
Article
The comparison of independent random variables can be modeled by a set of dice and a reciprocal relation expressing the winning probability of one dice over another. It is well known that dice transitivity is a necessary 3-cycle condition for a reciprocal relation to be dice representable, i.e. to be the winning probability relation of a set of dice. Although this 3-cycle condition is sufficient for a rational-valued reciprocal relation on a set of three elements to be dice representable, it has been shown that this is no longer the case for sets consisting of four or more elements. In this contribution, we provide a necessary 4-cycle condition for dice representability of reciprocal relations. Moreover, we show that our condition is sufficient in the sense that a given rational-weighted 4-cycle and reciprocally weighted inverse cycle, both fulfilling the 4-cycle condition, can be extended to a winning probability graph representing a dice-representable reciprocal relation on four elements.
... Remotely sensed datasets are continuously observed and not modeled or interpolated, and they are often freely available. As demonstrated by Buermann et al. (2008), including remotely sensed data can improve model predictions for some species (see also Phillips et al., 2006; Bradley & Fleishman, 2008; Gillespie et al., 2008; Bisrat et al., 2011). One satellite dataset we use is a measure of the surface skin temperature , a quantity that has rarely been used for SDMs despite its obvious linkage to surface microclimates, and we demonstrate its potential importance here. ...
Article
A number of studies have demonstrated the ecological sorting of C3 and C4 grasses along temperature and moisture gradients. However, previous studies of C3 and C4 grass biogeography have often inadvertently compared species in different and relatively unrelated lineages, which are associated with different environmental settings and distinct adaptive traits. Such confounded comparisons of C3 and C4 grasses may bias our understanding of ecological sorting imposed strictly by photosynthetic pathway. Here, we used MaxEnt species distribution modeling in combination with satellite data to understand the functional diversity of C3 and C4 grasses by comparing both large clades and closely related sister taxa. Similar to previous work, we found that C4 grasses showed a preference for regions with higher temperatures and lower precipitation compared with grasses using the C3 pathway. However, air temperature differences were smaller (2 °C vs. 4 °C) and precipitation and % tree cover differences were larger (1783 mm vs. 755 mm, 21.3% vs. 7.7%, respectively) when comparing C3 and C4 grasses within the same clade vs. comparing all C4 and all C3 grasses (i.e., ignoring phylogenetic structure). These results were due to important differences in the environmental preferences of C3 BEP and PACMAD clades (the two main grass clades). Winter precipitation was found to be more important for understanding the distribution and environmental niche of C3 PACMADs in comparison with both C3 BEPs and C4 taxa, for which temperature was much more important. Results comparing closely related C3 -C4 sister taxa supported the patterns derived from our modeling of the larger clade groupings. Our findings, which are novel in comparing the distribution and niches of clades, demonstrate that the evolutionary history of taxa is important for understanding the functional diversity of C3 and C4 grasses, and should have implications for how grasslands will respond to global change.
... The RF algorithm is able to fit complex non-linear surfaces from high-dimensional input data (Cutler et al., 2007). The use of both the BRT and RF techniques is presently becoming increasingly popular (Bisrat et al., 2012;Elith et al., 2006;Williams et al., 2009). These machine-learning methods provide a number of advantages over the more traditional GLM approach, including: robust parameter estimates; model structure learned from data; and easy implementation of complex interactions. ...
Article
Full-text available
The spread of ecosystem modifying invasive plant (EMIP) species is one of the largest threats to native ecosystems in Hawaiʻi. However, differences in niche characteristics between Hawaiʻi’s isolated insular environment and the wider global distribution of these species have not been carefully examined. We used species distribution modeling (SDM) methods to assess similarities and differences in niche characteristics between global and regional scales for 17 EMIPs present in Hawaiʻi. With a clearer understanding of the global context of regional plant invasion, we combined two SDM methods to better understand the potential future regional spread: (1) a nested modeling approach to integrate global and regional invasive species distribution projections; and (2) integrating all available agency and citizen science data to minimize the effect of monitoring gaps and biases. Our results show there are multiple similarities in niche characteristics across regional and global scales for most species, such as similar sets of climatic determinants of distribution, similar responses along environmental gradients, and moderate to high niche overlap between global and regional models. However, some differences were apparent and likely due to several factors including incomplete regional spread, community assembly or diversity effects. Invaders that established earlier showed a higher degree of niche overlap and similar environmental gradient responses when comparing global and regional models. This pattern, coupled with the tendency for regionally-based projections to predict narrower distributions than global projections, indicates a potential for continued spread of several invasive species across the Hawaiian landscape. Our study has broader implications for understanding the distribution and spread of invasive species in other regions, as similar analyses and models, including a novel way to characterize environmental gradient response differences across regions or scales, can likely provide valuable information for conservation and management efforts.
Chapter
Mountains are fascinating habitats, characterized by steep ecological vertical gradients and corresponding altitudinal vegetation zonation. Alpine treelines as upper boundaries of more or less contiguous tree stands are the most conspicuous vegetation limits; they have always attracted great research interest. Globally, alpine treeline elevations in the mountains are caused by heat deficiency. At landscape and local scales, however, multiple interactions of influencing factors and mechanisms determine treeline position, spatial pattern and dynamics. In the course of climate change, it is postulated that treelines will shift to higher elevations. To be able to quantify potential shifts, an analysis of the underlying factors and a correct modelling of the treeline ecotone under current climatic conditions are of great importance. For this purpose, statistical models are used to calculate the ecological niche of species based on climatic factors. These models serve as a baseline for models that project the distribution under future climatic conditions. The Himalayas are the largest mountain range in the world, yet they are often under-represented in the scientific literature. This holds particularly true in relation to modelling studies. Modelling treeline species in remote high-altitude regions faces several challenges, especially the availability of occurrence data and high-quality environmental variables. This book chapter summarizes recent results modelling the ecological niche of the Himalayan birch (Betula utilis) under present climatic conditions in the Himalayan mountain system. B. utilis represents a favourable target species for modelling studies, since it is widespread as a treeline-forming species along the entire Himalayan arch. Due to less distinctive habitat requirements and high adaptation potential, it is gaining importance as a pioneer tree species for possible succession developments at treelines under future climate conditions. In a synergistic approach, a detailed study on comparing the underlying climatic, topographical and plant phenological factors was undertaken to model the potential and the actual distribution of the focal species. The present results provide a new starting point for further investigations aimed at modelling the distribution of the species under past or future climate scenarios. Simultaneously, the presented approaches can also be transferred to other treeline species in high mountains.
Article
The invasion of alien species in their non-native range has resulted in inevitable consequences. Thus, the potential distribution of alien species must be delineated to anticipate and reduce their negative effect on native ecosystems. The potential distribution can be predicted using invasive species distribution models (iSDMs). Thus far, few studies have investigated the human influence on the distribution of alien species when modelling their potential distribution. In the present study, we predict the potential distribution of Acacia farnesiana in the Himalayan hotspot using a popular iSDM. The effect of human influence was studied by comparing the potential distribution predicted using only bioclimatic variables and that using both bioclimatic and human footprint variables. We found that using both bioclimatic and human footprint variables, the potential distribution of target species could be 55.38% larger than that of using only bioclimatic variables. This proves the positive effect of human activities on distribution of invasive species. Among the six considered bioclimatic variables, the mean temperature of the coldest quarter, the precipitation of the coldest quarter, and temperature seasonality are the most influential factors in determining the potential distribution of A. farnesiana.
Article
Full-text available
The Greenhouse Frog (Eleutherodactylus planirostris) is one of the most widespread frog species in the world. Because of its high population densities, widespread distribution, and consumption of native invertebrates in some invaded sites, understanding its impacts in Hawaii is important. We analyzed stomach contents of 397 frogs from 10 study sites in Maui. Results suggest Greenhouse Frogs are active, ant-specialist predators in the leaf litter. Ants (Formicidae) were the dominant prey flushed from stomachs in both number and volume. Furthermore, only ants were consumed in a higher proportion than they were sampled in the environment. Because ants dominated their diets, and because all ants are nonnative to Hawaii, this means Greenhouse Frogs consumed primarily nonnative invertebrates (>80%) in the areas sampled. Although results suggest that most native taxa are not at risk from Greenhouse Frog predation, the only locations where we could currently find Greenhouse Frogs were in human-dominated lowlands, which have a lower proportion of native species. Greenhouse Frogs may consume more native species if they invade more native-dominated habitat. Alternatively, nonnative ants are known to impact negatively many native invertebrates in Hawaii, and their possible reduction through Greenhouse Frog predation could affect other species positively. Our research highlights the need to understand better the effects of Greenhouse Frog predation on both native and nonnative invertebrates in Hawaii.
Article
Predicting species geographic distributions in the future is an important yet exceptionally challenging endeavor. Overall, it requires a two-step process: (1) a niche model characterizing suitability, applied to projections of future conditions and linked to (2) a dispersal/demographic simulation estimating the species' future occupied distribution. Despite limitations, for the vast majority of species, correlative approaches are the most feasible avenue for building niche models. In addition to myriad technical issues regarding model building, researchers should follow critical principles for selecting predictor variables and occurrence data, demonstrating effective performance in prediction across space, and extrapolating into nonanalog conditions. Many of these principles relate directly to the niche space, dispersal/demographic noise, biotic noise, and human noise assumptions defined here. Issues requiring progress include modeling interactions between abiotic variables, integrating biotic variables, considering genetic heterogeneity, and quantifying uncertainty. Once built, the niche model identifying currently suitable conditions must be processed to approximate the areas that the species occupies. That estimate serves as a seed for the simulation of persistence, dispersal, and establishment in future suitable areas. The dispersal/demographic simulation also requires data regarding the species' dispersal ability and demography, scenarios for future land use, and the capability of considering multiple interacting species simultaneously.
Article
Full-text available
1. The stacking of species-distribution models (S-SDMs) is receiving attention by conservation researchers because this approach is capable of simultaneously predicting species richness and composition. However, the steps required to build S-SDMs implies at least two choices that influence its predictive performance which have not been extensively assessed: the selection of the modelling algorithm and the application of a threshold to transform the species-distribution models into binary maps to be added together to build the final S-SDM. Our goal was to provide guidelines concerning the best combinations of modelling algorithms and thresholds with which to build more accurate S-SDMs. 2. We generated 380 S-SDMs of 1224 tree species in Mesoamerica by combining 19 distribution modelling methods with 20 different thresholds using presence-only data from the Global Biodiversity Information Facility. We compared the predicted richness and composition with inventory data obtained from the BIOTREE-NET forest plot database. We designed two indicators of predictive performance that were based on the diversity factors used to measure species turnover: a (shared species between the observed and predicted compositions), b and c (the exclusive species of the predicted and observed compositions respectively) and compared them with the Sorensen and Beta-Simpson turnover measures. 3. Our proposed indexes and the Sorensen index proved suitable as indicators of predictive performance for S-SDMs, whereas the Beta-Simpson turnover measure presented issues that would prevent its application to evaluate S-SDMs. 4. Some modelling methods –especially machine learning and ensemble model forecasting methods performed significantly better than others in minimizing the error in predicted richness and composition. Our results also points out that restrictive thresholds (with high omission errors) lead to more accurate S-SDMs in terms of species richness and composition. Here, we demonstrate that particular combinations of modelling methods and thresholds provide results with higher predictive performance. 5. These results provide clear modelling guidelines that will help S-SDM modellers to select the appropriate combination of modelling methods and thresholds to build more accurate S-SDMs, and therefore will have a positive impact on the quality of the diversity models used to assist conservation planning.
Article
Full-text available
The Caribbean is one of the world’s centers of biodiversity and endemism. As in similar regions, many of its islands have complex topography, climate and soils, and ecological zones change over small areas. A segmented, supervised classification approach using Landsat TM imagery enabled us to develop the most detailed island-wide map of Puerto Rico’s extremely complex natural vegetation cover. Many Caribbean forest formations that are not spectrally distinct had distributions approximately separable using climatic zone, geology, elevation, and rainfall. Classification accuracy of 26 land cover and woody vegetation classes was 71 % overall and 83 % after combining forest successional stages within image mapping zones. In 1991-92, Puerto Rico had about 364,000 ha of closed forest, which covered 41.6 % of the main island. Unlike previous island-wide mapping, this map better identifies the spatial distributions of forest formations where certain groups of endemic species occur. Approximately 5 % of Puerto Rico’s forest area is under protection, but the reserve system grossly underrepresents lowland moist, seasonal evergreen forests.
Article
Full-text available
This article introduces yaImpute, an R package for nearest neighbor search and imputation. Although nearest neighbor imputation is used in a host of disciplines, the methods implemented in the yaImpute package are tailored to imputation-based forest attribute estimation and mapping. The impetus to writing the yaImpute is a growing interest in nearest neighbor imputation methods for spatially explicit forest inventory, and a need within this research community for software that facilitates comparison among different nearest neighbor search algorithms and subsequent imputation techniques, yaImpute provides directives for defining the search space, subsequent distance calculation, and imputation rules for a given number of nearest neighbors. Further, the package offers a suite of diagnostics for comparison among results generated from different imputation analyses and a set of functions for mapping imputation results.
Article
Full-text available
Adult American toads, Bufo americanus americanus Holbrook, may be exposed to conditions that result in considerable variation in both body temperature and hydration state during their daily activity periods. We measured the interactive effects of body temperature and hydration state on locomotor performance of toads. Performance, measured as distance moved in 10 min of forced locomotion, decreased at all body temperatures as animals were dehydrated. The magnitude of this decrease was affected by temperature. Similarly, the effects of temperature on performance were influenced by the hydration state of the toad. Fully hydrated and slightly dehydrated animals travelled farthest at high test temperatures, whereas more dehydrated animals moved farthest at intermediate temperatures. Body temperature and hydration state showed statistically significant interaction and individual toads varied in their tolerance of those parameters. Thus, the relative performance capacities of individual toads under natural conditions cannot be predicted from laboratory measurements at a single combination of body temperature and hydration.
Article
Full-text available
ABSTRACT Invading non-indigenous species in the United States cause major environmental damages,and losses adding up to more than $138 billion per year. There are approximately 50,000 foreign species and the number is increasing. About 42% of the species on the Threatened or Endangered species lists are at risk primarily because of non-indigenous species. In the history of the United States, approximately 50,000 non-indigenous (non-native)
Article
Full-text available
Classification and regression trees are ideally suited for the analysis of complex ecological data. For such data, we require flexible and robust analytical methods, which can deal with nonlinear relationships, high-order interactions, and missing values. Despite such difficulties, the methods should be simple to understand and give easily interpretable results. Trees explain variation of a single response variable by repeatedly splitting the data into more homogeneous groups, using combinations of explanatory variables that may be categorical and/or numeric. Each group is characterized by a typical value of the response variable, the number of observations in the group, and the values of the explanatory variables that define it. The tree is represented graphically, and this aids exploration and understanding. Trees can be used for interactive exploration and for description and prediction of patterns and processes. Advantages of trees include: (1) the flexibility to handle a broad range of response types, including numeric, categorical, ratings, and survival data; (2) invariance to monotonic transformations of the explanatory variables; (3) ease and robustness of construction; (4) case of interpretation; and (5) the ability to handle missing values in both response and explanatory variables. Thus, trees complement or represent an alternative to many traditional statistical techniques, including multiple regression, analysis of variance, logistic regression, log-linear models, linear discriminant analysis, and survival models. We use classification and regression trees to analyze survey data from the Australian central Great Barrier Reef, comprising abundances of soft coral taxa (Cnidaria: Octocorallia) and physical and spatial environmental information. Regression tree analyses showed that dense aggregations, typically formed by three taxa, were restricted to distinct habitat types, each of which was defined by combinations of 3-4 environmental variables. The habitat definitions were consistent with known experimental findings on the nutrition of these taxa. When used separately, physical and spatial variables were similarly strong predictors of abundances and lost little in comparison with their joint use. The spatial variables are thus effective surrogates for the physical variables in this extensive reef complex, where information on the physical environment is often not available. Finally, we compare the use of regression trees and linear models for the analysis of these data and show how linear models fail to find patterns uncovered by the trees.
Article
Full-text available
Predictive vegetation mapping can be defined as pedicting the geographic distribution of the vegetation composition across a landscape from mapped environmental variables. Computerized predictive vegetation mapping is made possibly by the availability of digital maps of topography and other environmental variables such as soils, geology and climate variables, and geographic information system software for manipulating these data. Especially important to predictive vegetation mapping are interpolated climatic variables related to physiological tolerances, and topographic variables, derived from digital elevation grids, related to site energy and moisture balance. Predictive vegetation mapping is founded in ecological niches theory and gradient analysis, and driven by the need to map vegetation patterns over large areas for resource conservation planning, and to predict the effects of environmental change on vegetation distributions. -from Author
Article
Full-text available
Aim Several studies have found that more accurate predictive models of species’ occurrences can be developed for rarer species; however, one recent study found the relationship between range size and model performance to be an artefact of sample prevalence, that is, the proportion of presence versus absence observations in the data used to train the model. We examined the effect of model type, species rarity class, species’ survey frequency, detectability and manipulated sample prevalence on the accuracy of distribution models developed for 30 reptile and amphibian species. Location Coastal southern California, USA. Methods Classification trees, generalized additive models and generalized linear models were developed using species presence and absence data from 420 locations. Model performance was measured using sensitivity, specificity and the area under the curve (AUC) of the receiver‐operating characteristic (ROC) plot based on twofold cross‐validation, or on bootstrapping. Predictors included climate, terrain, soil and vegetation variables. Species were assigned to rarity classes by experts. The data were sampled to generate subsets with varying ratios of presences and absences to test for the effect of sample prevalence. Join count statistics were used to characterize spatial dependence in the prediction errors. Results Species in classes with higher rarity were more accurately predicted than common species, and this effect was independent of sample prevalence. Although positive spatial autocorrelation remained in the prediction errors, it was weaker than was observed in the species occurrence data. The differences in accuracy among model types were slight. Main conclusions Using a variety of modelling methods, more accurate species distribution models were developed for rarer than for more common species. This was presumably because it is difficult to discriminate suitable from unsuitable habitat for habitat generalists, and not as an artefact of the effect of sample prevalence on model estimation.
Article
Full-text available
Aim In response to a recent paper suggesting the failure of ecological niche models to predict between native and introduced distributional areas of fire ants ( Solenopsis invicta ), we sought to assess methodological causes of this failure. Location Ecological niche models were developed on the species’ native distributional area in South America, and projected globally. Methods We developed ecological niche models based on six different environmental data sets, and compared their respective abilities to anticipate the North American invasive distributional area of the species. Results We show that models based on the ‘bioclimatic variables’ of the WorldClim data set indeed fail to predict the full invasive potential of the species, but that models based on four other data sets could predict this potential correctly. Main conclusions The difference in predictive abilities appears to centre on the complexity of the environmental variables involved. These results emphasize important influences of environmental data sets on the generality and ability of ecological niche models to anticipate novel phenomena, and offer a simpler explanation for the lack of predictive ability among native and invaded distributional areas than that of niche shifts.
Article
Full-text available
Transforming the results of species distribution modelling from probabilities of or suitabilities for species occurrence to presences/absences needs a specific threshold. Even though there are many approaches to determining thresholds, there is no comparative study. In this paper, twelve approaches were compared using two species in Europe and artificial neural networks, and the modelling results were assessed using four indices: sensitivity, specificity, overall prediction success and Cohen's kappa statistic. The results show that prevalence approach, average predicted probability/suitability approach, and three sensitivity-specificity-combined approaches, including sensitivity-specificity sum maximization approach, sensitivity-specificity equality approach and the approach based on the shortest distance to the top-left corner (0,1) in ROC plot, are the good ones. The commonly used kappa maximization approach is not as good as the afore-mentioned ones, and the fixed threshold approach is the worst one. We also recommend using datasets with prevalence of 50% to build models if possible since most optimization criteria might be satisfied or nearly satisfied at the same time, and therefore it's easier to find optimal thresholds in this situation.
Article
Full-text available
1. The potential invasive success of exotic plant species is thought to be associated with similarity in climate and biome between the original and novel range. We tested this assumption by quantifying the match between the realized climatic niches and biomes occupied in the exotic and native range of 26 plant species introduced to Australia. We then explored correlations between the propensity to shift climatic niche with residence time, invasion status, geographic range size, and species traits. 2. Occurrence data from the native and exotic range of 26 species introduced to Australia were obtained, and the overlap between native and exotic climate niches was calculated using between‐class analysis. Shifts into novel biomes were assessed using a Geographic Information System (GIS). Correlations between introduction, distribution and species traits and the degree of climate matching were examined using nonparametric statistical tests. 3. Exotic species frequently occurred in climatic conditions outside those occupied in their native range (20 of 26 species). Nineteen species inhabited biomes in Australia not occupied in the native range and in some instances this shift represented the establishment of populations in novel biomes not present in the native range. No single‐species trait, introduction or distributional characteristic was significantly associated with the degree of climatic niche shift. 4. Synthesis. Exotic species are able to occupy climate niches in the new range that differ substantially from those of the native range, and generally do not show biome conservatism between their native and introduced ranges. This implies that novel climatic conditions are not a major obstacle for exotic species establishing populations outside their native range. These results have important implications for the use and interpretation of ecological niche models used to predict the distribution of species in novel climates in time or space. The results also highlight the importance of alternate mechanisms, such as enemy release, phenotypic plasticity or rapid evolution, in the establishment of naturalized and invasive populations.
Article
Full-text available
The fundamental niche can be viewed as the set of conditions and resources that allow a given organism to survive and reproduce in the absence of biotic interactions. Quantitative descriptions of the environmental variables with which organisms are asso-ciated are becoming common with the advent of geographic information systems (GIS). Although such descriptive approaches to the niche are useful for interpolating species distributions, they implicitly incorporate biotic interactions and therefore do not represent the fundamental niche. A mechanistic understanding of the fundamental niche, when com-bined with GIS data, can provide us with greater insight into the causes of distribution and abundance, a solid foundation for exploring the role of biotic interactions, and greater confidence in extrapolating to novel circumstances such as climate change and species introductions. We apply such a mechanistic approach to study the climatic component of the fundamental niche of a nocturnal lizard, Heteronotia binoei, across an entire continent. We combine physiological measurements of this species (thermal requirements for egg development, thermal preferences and tolerances, metabolic and evaporative water loss rates), and high-resolution climatic data for the Australian continent (air temperature, cloud cover, wind speed, humidity, and radiation), with biophysical models to calculate the cli-matic component of the fundamental niche of this lizard and map it onto the Australian landscape at high resolution. We also use this approach to predict the effects of a mild global warming on the degree-days in the soil for egg development and the potential for aboveground activity of the study organism.
Article
Full-text available
The Caribbean is one of the world's centers of biodiversity and endemism. As in similar regions, many of its islands have complex topography, climate and soils, and ecological zones change over small areas. A segmented, supervised classification approach using Landsat TM imagery enabled us to develop the most detailed island-wide map of Puerto Rico's extremely complex natural vegetation cover. Many Caribbean forest formations that are not spectrally distinct had distributions approximately separable using climatic zone, geology, elevation, and rainfall. Classification accuracy of 26 land cover and woody vegetation classes was 71 % overall and 83 % after combining forest successional stages within image mapping zones. In 1991-92, Puerto Rico had about 364,000 ha of closed forest, which covered 41.6 % of the main island. Unlike previous island-wide mapping, this map better identifies the spatial distributions of forest formations where certain groups of endemic species occur. Approximately 5 % of Puerto Rico's forest area is under protection, but the reserve system grossly underrepresents lowland moist, seasonal evergreen forests.
Article
Full-text available
This paper presents an evaluation of the Earth Observing System (EOS) Moderate Resolution Imaging Spectroradiometer (MODIS) thermal infrared bands and the status of land surface temperature (LST) version-3 standard products retrieved from Terra MODIS data. The accuracy of daily MODIS LST products has been validated in more than 20 clear-sky cases with in situ measurement data collected in field campaigns in 2000–2002. The MODIS LST accuracy is better than 1C in the range from 210 to 50C. Refinements and improvements were made to the new version of MODIS LST product generation executive code. Using both Terra and Aqua MODIS data for LST retrieval improves the quality of the LST product and the diurnal feature in the product due to better temporal, spatial and angular coverage of clear-sky observations.
Article
Aim The use of species distribution models (SDMs) to predict biological invasions is a rapidly developing area of ecology. However, most studies investigating SDMs typically ignore prediction errors and instead focus on regions where native distributions correctly predict invaded ranges. We investigated the ecological significance of prediction errors using reciprocal comparisons between the predicted invaded and native range of the red imported fire ant (Solenopsis invicta) (hereafter called the fire ant). We questioned whether fire ants occupy similar environments in their native and introduced range, how the environments that fire ants occupy in their introduced range changed through time relative to their native range, and where fire ant propagules are likely to have originated. Location We developed models for South America and the conterminous United States (US) of America. Methods We developed models using the Genetic Algorithm for Rule-set Prediction (GARP) and 12 environmental layers. Occurrence data from the native range in South America were used to predict the introduced range in the US and vice versa. Further, time-series data recording the invasion of fire ants in the US were used to predict the native range. Results Native range occurrences under-predicted the invasive potential of fire ants, whereas occurrence data from the US over-predicted the southern boundary of the native range. Secondly, introduced fire ants initially established in environments similar to those in their native range, but subsequently invaded harsher environments. Time-series data suggest that fire ant propagules originated near the southern limit of their native range. Conclusions Our findings suggest that fire ants from a peripheral native population established in an environment similar to their native environment, and then ultimately expanded into environments in which they are not found in their native range. We argue that reciprocal comparisons between predicted native and invaded ranges will facilitate a better understanding of the biogeography of invasive and native species and of the role of SDMs in predicting future distributions.
Article
The paucity of empirical data on processes in species life cycles demands tools to extract insight from field observations. Such insights help inform policy on invasive species and on impacts of climate change at regional and local scales. We used the CLIMEX model to infer the response of the red imported fire ant, Solenopsis invicta Buren (Hymenoptera: Formicidae), to temperature and moisture from its range in the United States. We tested hypotheses on the mechanisms that limit the distribution of the ant and estimated the potential global area at risk from invasion. The ant can spread further in the United States, including north along the west coast, where patterns of infestation will differ from those in the east. We analyzed the risk of colonization in Australia and New Zealand, where the ant was recently discovered. The patterns of infestation of the ant in Oceania will differ from those in the eastern United States, with slower growth and less winter mortality. This study adds to earlier temperature-based models by incorporating a moisture response; by replacing arbitrary categories of colony size to predict overwintering success with a site-specific model based on the balance between annual growth and survival; and by comparing different hypotheses on low temperature-related mechanisms that limit the geographical distribution. It shows how the response of a species to climate can be synthesized from field observations to provide useful insights into its population dynamics. Such analyses provide a basis for making decisions on regional management of invasive species and an informative context for local studies.
Article
Bagging predictors is a method for generating multiple versions of a predictor and using these to get an aggregated predictor. The aggregation averages over the versions when predicting a numerical outcome and does a plurality vote when predicting a class. The multiple versions are formed by making bootstrap replicates of the learning set and using these as new learning sets. Tests on real and simulated data sets using classification and regression trees and subset selection in linear regression show that bagging can give substantial gains in accuracy. The vital element is the instability of the prediction method. If perturbing the learning set can cause significant changes in the predictor constructed, then bagging can improve accuracy.
Article
We studied the reproductive phenology and ecology, and patterns of male and female reproductive success of Eleutherodactylus coqui, a frog with terrestrial reproduction, direct development, and male parental care. Reproductive activity (calling and oviposition) occurred during every month of our two-year study, although it was distinctly seasonal, with the lowest activity during the dry season and the highest during the wet season. Male chorus density and female clutch deposition were temporally associated with temperature and rainfall. Female breeding was asynchronous, with at least one female mating on 63% of all wet season nights. The mean interclutch interval was about 8 weeks during the wet season. Clutch size was correlated with female body size, but also showed a significant seasonal effect with clutch size increasing in the wet season. In contrast, ovipositing female body size declined from dry to wet season. Male mating success was variable, with some males mating up to five times over one six-month period. Mating success was not correlated with male body size and we found no evidence of size-assortative mating. Mating success was correlated with calling effort measured as chorus attendance. Variance in hatching rate comprised most (89%) of the total variation in female reproductive success, measured as number of hatchlings per clutch. Our results suggest that female coquies should be strongly selected to exercise mate choice based on variables that correlate with hatching rate, i.e., male parental care quality and nest site quality.
Article
Random forests are a combination of tree predictors such that each tree depends on the values of a random vector sampled independently and with the same distribution for all trees in the forest. The generalization error for forests converges a.s. to a limit as the number of trees in the forest becomes large. The generalization error of a forest of tree classifiers depends on the strength of the individual trees in the forest and the correlation between them. Using a random selection of features to split each node yields error rates that compare favorably to Adaboost (Y. Freund & R. Schapire, Machine Learning: Proceedings of the Thirteenth International conference, ∗∗∗, 148–156), but are more robust with respect to noise. Internal estimates monitor error, strength, and correlation and these are used to show the response to increasing the number of features used in the splitting. Internal estimates are also used to measure variable importance. These ideas are also applicable to regression.
Article
Two species of neotropical frog, Eleutherodactylus coqui and E. planirostris, have been introduced into the State of Hawaii via the horticulture trade. E. coqui was introduced prior to 1988 and E. planirostris was first reported in 1994. Since these dates frog colonies have rapidly spread accidentally and intentionally and frog abundance within colonies has grown rapidly. Although these frogs were originally restricted to horticulture sites, they are now found in residential areas, resorts and hotels, and public lands. Due to the high potential biomass of introduced frogs there are realistic anthropogenic and ecological concerns associated with the spread of these frogs. Though there currently is a tool that can be used for localized control of frogs in limited circumstances, overall efforts by Federal, State, and County agencies to control the frog in Hawaii have been hampered by limited authorities and funds, disbelief in the threat, and the reluctance to act.