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

Ecological drivers of movement for two sympatric marine predators in the California current large marine ecosystem

BACKGROUND

An animal's movement reflects behavioral decisions made to address ecological needs; specifically, that movement will become less directional in regions with high prey availability, indicating foraging behavior. In the marine realm, animal behavior occurs below the sea surface and is difficult to observe. We used an extensive satellite tagging dataset to explore how physical and biological habitat characteristics influence blue (Balaenoptera musculus) and fin (B. physalus) whale movement and foraging behavior in the California Current Ecosystem across four known bioregions.

METHODS

We fitted movement models to 14 years of blue whale satellite tracking data and 13 years of fin whale data to characterize their movement persistence, with higher move persistence values representing more directional movement and lower move persistence values representing less directional movement. Models were evaluated against a range of physical and biological environmental predictors to identify significant correlates of low move persistence (i.e., presumed intensified foraging behavior). We then used data from a subset of sensor-equipped tags that monitored vertical behavior (e.g., dive and feeding), in addition to movement, to test the relationship between vertical behavior and movement persistence.

RESULTS

Low move persistence was strongly correlated with shallower water depth and sea surface height for both species, with additional effects of chlorophyll-a concentration, vorticity and marine nekton biomass for blue whales. Data from sensor-equipped tags additionally showed that low move persistence occurred when whales made more numerous feeding dives. Temporal patterns of bioregion occupancy coincided with seasonal peaks in productivity. Most blue whale low-move-persistence movements occurred in the northern, nearshore bioregion with a late-season peak in productivity and were evenly distributed across all bioregions for fin whales.

CONCLUSIONS

We demonstrated that low move persistence is indicative of increased feeding behavior for both blue and fin whales. The environmental drivers of low move persistence were similar to those previously identified for survey-based species distribution models, linking environmental metrics to subsurface behavior. Occupancy and movement behavior patterns across bioregions indicate both species moved to exploit seasonal and spatial variability in productivity, with blue whales especially focusing on the bioregion of highest productivity during late summer and fall.

Evaluating Three Modelling Frameworks for Assessing Changes in Fin Whale Distribution in the Mediterranean Sea

Understanding the habitat of highly migratory species is aided by using species distribution models to identify species-habitat relationships and to inform conservation and management plans. While Generalized Additive Models (GAMs) are commonly used in ecology, and particularly the habitat modeling of marine mammals, there remains a debate between modeling habitat (presence/absence) versus density (# individuals). Our study assesses the performance and predictive capabilities of GAMs compared to boosted regression trees (BRTs) for modeling both fin whale density and habitat suitability alongside Hurdle Models treating presence/absence and density as a two-stage process to address the challenge of zero-inflated data. Fin whale data were collected from 2008 to 2022 along fixed transects crossing the NW Mediterranean Sea during the summer period. Data were analyzed using traditional line transect methodology, obtaining the Effective Area monitored. Based on existing literature, we select various covariates, either static in nature, such as bathymetry and slope, or variable in time, for example, SST, MLD, Chl concentration, EKE, and FSLE. We compared both the explanatory power and predictive skill of the different modeling techniques. Our results show that all models performed well in distinguishing presences and absences but, while density and presence patterns for the fin whale were similar, their dependencies on environmental factors can vary depending on the chosen model. Bathymetry was the most important variable in all models, followed by SST and the chlorophyll recorded 2 months before the sighting. This study underscores the role SDMs can play in marine mammal conservation efforts and emphasizes the importance of selecting appropriate modeling techniques. It also quantifies the relationship between environmental variables and fin whale distribution in an understudied area, providing a solid foundation for informed decision making and spatial management.

Comparative Analysis of MaxEnt and Deep Learning Approaches for Modeling Humpback Whale Distribution in North Iceland

In this study, we compared the established MaxEnt and a more novel deep learning approach for modeling the distribution of humpback whales (Megaptera novaeangliae) in north Iceland. We examined the mechanisms, structures, and optimization techniques of both approaches, highlighting their differences and similarities. Monthly distribution models for Skjálfandi Bay were created, from 2018 until 2021, using presence-only sighting data and satellite remote sensing data. Search efforts and boat tracklines were utilized to create pseudo-absence points for both models. Additionally, the trained models were used to create distribution projections for the year 2022, solely based on the available environmental data. We compared the results using the established area under the curve value. The findings indicate that both approaches have their limitations and advantages. MaxEnt does not allow continuous updating within a time series, yet it mitigates the risk of overfitting by employing the maximum entropy principle. The deep learning model is more likely to overfit, but the larger weight network increases the model's capability to capture complex relationships and patterns. Ultimately, the results show that the deep learning model had a higher predictive performance in modeling both current and future humpback whale distributions. Both modeling approaches have inherent limitations, such as the low resolution of the input data, spatial biases, and the inability to fully capture the entire complexity of natural processes. Despite this, deep learning showed promising results in modeling the distribution of humpback whales and prompts further research in different study areas and applications for other mobile animal species.

Will climate change cause Sargassum beds in temperate waters to expand or contract? Evidence from the range shift pattern of Sargassum

Understanding the range shift patterns of foundation species (e.g., macroalgae) under future climatic conditions is critical for biodiversity conservation in coastal ecosystems. These predictions are typically made using species distribution models (SDMs), and severe habitat loss has been predicted for most brown algal forests. Nevertheless, some models showed that local adaptation within species can reduce range loss projections. In this study, we used the brown algae Sargassum fusiforme and Sargassum thunbergii, which are distributed in the Northwest Pacific, to determine whether climate change will cause the Sargassum beds in Northwest Pacific temperate waters to expand or contract. We divided S. fusiforme and S. thunbergii into northern and southern lineages, considering the temperature gradients and phylogeographic structures. We quantified the realized niches of the two lineages using an n-dimensional hypervolume. Significant niche differentiation was detected between lineages for both species, suggesting the existence of local adaptation. Based on these results, lineage-level SDMs were constructed for both species. The prediction results showed the different responses of different lineages to climate change. The suitable distribution area for both species was predicted to move northward, retaining part of the suitable habitat at low latitudes (along the East China Sea). Unfortunately, this expansion could not compensate for losing middle-low latitude areas. Our results have important implications for the future management and protection of macroalgae and emphasize the importance of incorporating intraspecific variation into species distribution predictions.

Modeling Tick Populations: An Ecological Test Case for Gradient Boosted Trees

General linear models have been the foundational statistical framework used to discover the ecological processes that explain the distribution and abundance of natural populations. Analyses of the rapidly expanding cache of environmental and ecological data, however, require advanced statistical methods to contend with complexities inherent to extremely large natural data sets. Modern machine learning frameworks such as gradient boosted trees efficiently identify complex ecological relationships in massive data sets, which are expected to result in accurate predictions of the distribution and abundance of organisms in nature. However, rigorous assessments of the theoretical advantages of these methodologies on natural data sets are rare. Here we compare the abilities of gradient boosted and linear models to identify environmental features that explain observed variations in the distribution and abundance of blacklegged tick (Ixodes scapularis) populations in a data set collected across New York State over a ten-year period. The gradient boosted and linear models use similar environmental features to explain tick demography, although the gradient boosted models found non-linear relationships and interactions that are difficult to anticipate and often impractical to identify with a linear modeling framework. Further, the gradient boosted models predicted the distribution and abundance of ticks in years and areas beyond the training data with much greater accuracy than their linear model counterparts. The flexible gradient boosting framework also permitted additional model types that provide practical advantages for tick surveillance and public health. The results highlight the potential of gradient boosted models to discover novel ecological phenomena affecting pathogen demography and as a powerful public health tool to mitigate disease risks.

Predicting Carcass Weight of Grass-Fed Beef Cattle before Slaughter Using Statistical Modelling

Gaining insights into the utilization of farm-level data for decision-making within the beef industry is vital for improving production and profitability. In this study, we present a statistical model to predict the carcass weight (CW) of grass-fed beef cattle at different stages before slaughter using historical cattle data. Models were developed using two approaches: boosted regression trees and multiple linear regression. A sample of 2995 grass-fed beef cattle from 3 major properties in Northern Australia was used in the modeling. Four timespans prior to the slaughter, i.e., 1 month, 3 months, 9-10 months, and at weaning, were considered in the predictive modelling. Seven predictors, i.e., weaning weight, weight gain since weaning to each stage before slaughter, time since weaning to each stage before slaughter, breed, sex, weaning season (wet and dry), and property, were used as the potential predictors of the CW. To assess the predictive performance in each scenario, a test set which was not used to train the models was utilized. The results showed that the CW of the cattle was strongly associated with the animal's body weight at each stage before slaughter. The results showed that the CW can be predicted with a mean absolute percentage error (MAPE) of 4% (~12-16 kg) at three months before slaughter. The predictive error increased gradually when moving away from the slaughter date, e.g., the prediction error at weaning was ~8% (~20-25 kg). The overall predictive performances of the two statistical approaches was approximately similar, and neither of the models substantially outperformed each other. Predicting the CW in advance of slaughter may allow farmers to adequately prepare for forthcoming needs at the farm level, such as changing husbandry practices, control inventory, and estimate price return, thus allowing them to maximize the profitability of the industry.

Bottlenose dolphin distribution in a Mediterranean area exposed to intensive trawling

The Adriatic Sea is one of the areas most exposed to trawling, worldwide. We used four years (2018-2021) and 19,887 km of survey data to investigate factors influencing daylight dolphin distribution in its north-western sector, where common bottlenose dolphins Tursiops truncatus routinely follow fishing trawlers. We validated Automatic Identification System information on the position, type and activity of three types of trawlers based on observations from boats, and incorporated this information in a GAM-GEE modelling framework, together with physiographic, biological and anthropogenic variables. Along with bottom depth, trawlers (particularly otter and midwater trawlers) appeared to be important drivers of dolphin distribution, with dolphins foraging and scavenging behind trawlers during 39.3% of total observation time in trawling days. The spatial dimension of dolphin adaptations to intensive trawling, including distribution shifts between days with and without trawling, sheds light on the magnitude of ecological change driven by the trawl fishery.

Environmental variables and machine learning models to predict cetacean abundance in the Central-eastern Mediterranean Sea

Although the Mediterranean Sea is a crucial hotspot in marine biodiversity, it has been threatened by numerous anthropogenic pressures. As flagship species, Cetaceans are exposed to those anthropogenic impacts and global changes. Assessing their conservation status becomes strategic to set effective management plans. The aim of this paper is to understand the habitat requirements of cetaceans, exploiting the advantages of a machine-learning framework. To this end, 28 physical and biogeochemical variables were identified as environmental predictors related to the abundance of three odontocete species in the Northern Ionian Sea (Central-eastern Mediterranean Sea). In fact, habitat models were built using sighting data collected for striped dolphins Stenella coeruleoalba, common bottlenose dolphins Tursiops truncatus, and Risso's dolphins Grampus griseus between July 2009 and October 2021. Random Forest was a suitable machine learning algorithm for the cetacean abundance estimation. Nitrate, phytoplankton carbon biomass, temperature, and salinity were the most common influential predictors, followed by latitude, 3D-chlorophyll and density. The habitat models proposed here were validated using sighting data acquired during 2022 in the study area, confirming the good performance of the strategy. This study provides valuable information to support management decisions and conservation measures in the EU marine spatial planning context.

Association of Leopard Cat Occurrence with Environmental Factors in Chungnam Province, South Korea

This study was conducted to investigate the association of leopard cat (Prionailurus bengalensis) occurrences and environmental factors in Chungnam Province, South Korea, using two different analytical approaches for binomial responses: boosted regression trees and logistic regression. The extensive field survey data collected through the Chungnam Biotope Project were used to model construction and analysis. Five major influential factors identified by the boosted regression tree analysis were elevation, distance to road, distance to water channel/body, slope and population density. Logistic regression analysis indicated that distance to forest, population density, distance to water, and diameter class of the forest were the significant explanatory variables. The results showed that the leopard cats prefer the areas with higher accessibility of food resources (e.g., abundance and catchability) and avoid the areas adjacent to human-populated areas. The results also implied that boosted regression and logistic regression models could be used in a complementary manner for evaluating wildlife distribution and management.

Seascape genomics of common dolphins (Delphinus delphis) reveals adaptive diversity linked to regional and local oceanography

Background

High levels of standing genomic variation in wide-ranging marine species may enhance prospects for their long-term persistence. Patterns of connectivity and adaptation in such species are often thought to be influenced by spatial factors, environmental heterogeneity, and oceanographic and geomorphological features. Population-level studies that analytically integrate genome-wide data with environmental information (i.e., seascape genomics) have the potential to inform the spatial distribution of adaptive diversity in wide-ranging marine species, such as many marine mammals. We assessed genotype-environment associations (GEAs) in 214 common dolphins (Delphinus delphis) along > 3000 km of the southern coast of Australia.

Results

We identified 747 candidate adaptive SNPs out of a filtered panel of 17,327 SNPs, and five putatively locally-adapted populations with high levels of standing genomic variation were disclosed along environmentally heterogeneous coasts. Current velocity, sea surface temperature, salinity, and primary productivity were the key environmental variables associated with genomic variation. These environmental variables are in turn related to three main oceanographic phenomena that are likely affecting the dispersal of common dolphins: (1) regional oceanographic circulation, (2) localised and seasonal upwellings, and (3) seasonal on-shelf circulation in protected coastal habitats. Signals of selection at exonic gene regions suggest that adaptive divergence is related to important metabolic traits.

Conclusion

To the best of our knowledge, this represents the first seascape genomics study for common dolphins (genus Delphinus). Information from the associations between populations and their environment can assist population management in forecasting the adaptive capacity of common dolphins to climate change and other anthropogenic impacts.

Predicting the habitat suitability of the invasive white mango scale, Aulacaspis tubercularis; Newstead, 1906 (Hemiptera: Diaspididae) using bioclimatic variables

BACKGROUND

The white mango scale, Aulacaspis tubercularis (Hemiptera: Diaspididae), is an invasive pest that threatens the production of several crops of commercial value including mango. Though it is an important pest, little is known about its biology and ecology. Specifically, information on habitat suitability of A. tubercularis occurrence and potential distribution under climate change is largely unknown. In this study, we used four ecological niche models, namely maximum entropy, random forest, generalized additive models, and classification and regression trees to predict the habitat suitability of A. tubercularis under current and future [representative concentration pathways (RCPs): RCP4.5 and RCP8.5 of the year 2070] climatic scenarios, using bioclimatic variables. Models' performance was evaluated using the true skill statistic (TSS), the area under the curve (AUC), correlation (COR), and the deviance.

RESULTS

All models sufficiently predicted the occurrence of A. tubercularis with high accuracy (AUC ≥ 0.93, TSS ≥ 0.81 and COR ≥ 0.77). The random forest algorithm had the highest accuracy among the four models (AUC = 0.99, TSS = 0.93, COR = 0.90, deviance = 0.26). Temperature seasonality (Bio4), mean temperature of the driest quarter (Bio9), and precipitation seasonality (Bio15) were the most important variables influencing A. tubercularis occurrence. Models' predictions showed that countries in east, south, and west Africa are highly suitable for A. tubercularis establishment under current conditions. Similarly, Mexico, Brazil, India, Myanmar, Bangladesh, Thailand, Laos, Vietnam, and Cambodia are also highly suitable for the pest to thrive. Under future conditions, the suitable areas might slightly decrease in many countries of sub-Saharan Africa under both RCPs. However, the range of expansion of A. tubercularis is projected to be higher in Australia, Brazil, Spain, Italy, and Portugal under the future climatic scenarios.

CONCLUSION

The results reported here will be useful for guiding decision-making, developing an effective management strategy, and serving as an early warning tool to prevent further spread toward new areas. © 2022 Society of Chemical Industry.

Deep ocean drivers better explain habitat preferences of sperm whales Physeter macrocephalus than beaked whales in the Bay of Biscay

Species Distribution Models are commonly used with surface dynamic environmental variables as proxies for prey distribution to characterise marine top predator habitats. For oceanic species that spend lot of time at depth, surface variables might not be relevant to predict deep-dwelling prey distributions. We hypothesised that descriptors of deep-water layers would better predict the deep-diving cetacean distributions than surface variables. We combined static variables and dynamic variables integrated over different depth classes of the water column into Generalised Additive Models to predict the distribution of sperm whales Physeter macrocephalus and beaked whales Ziphiidae in the Bay of Biscay, eastern North Atlantic. We identified which variables best predicted their distribution. Although the highest densities of both taxa were predicted near the continental slope and canyons, the most important variables for beaked whales appeared to be static variables and surface to subsurface dynamic variables, while for sperm whales only surface and deep-water variables were selected. This could suggest differences in foraging strategies and in the prey targeted between the two taxa. Increasing the use of variables describing the deep-water layers would provide a better understanding of the oceanic species distribution and better assist in the planning of human activities in these habitats.

Towards a better characterisation of deep-diving whales' distributions by using prey distribution model outputs?

In habitat modelling, environmental variables are assumed to be proxies of lower trophic levels distribution and by extension, of marine top predator distributions. More proximal variables, such as potential prey fields, could refine relationships between top predator distributions and their environment. In situ data on prey distributions are not available over large spatial scales but, a numerical model, the Spatial Ecosystem And POpulation DYnamics Model (SEAPODYM), provides simulations of the biomass and production of zooplankton and six functional groups of micronekton at the global scale. Here, we explored whether generalised additive models fitted to simulated prey distribution data better predicted deep-diver densities (here beaked whales Ziphiidae and sperm whales Physeter macrocephalus) than models fitted to environmental variables. We assessed whether the combination of environmental and prey distribution data would further improve model fit by comparing their explanatory power. For both taxa, results were suggestive of a preference for habitats associated with topographic features and thermal fronts but also for habitats with an extended euphotic zone and with large prey of the lower mesopelagic layer. For beaked whales, no SEAPODYM variable was selected in the best model that combined the two types of variables, possibly because SEAPODYM does not accurately simulate the organisms on which beaked whales feed on. For sperm whales, the increase model performance was only marginal. SEAPODYM outputs were at best weakly correlated with sightings of deep-diving cetaceans, suggesting SEAPODYM may not accurately predict the prey fields of these taxa. This study was a first investigation and mostly highlighted the importance of the physiographic variables to understand mechanisms that influence the distribution of deep-diving cetaceans. A more systematic use of SEAPODYM could allow to better define the limits of its use and a development of the model that would simulate larger prey beyond 1,000 m would probably better characterise the prey of deep-diving cetaceans.

Combining Regional Habitat Selection Models for Large-Scale Prediction: Circumpolar Habitat Selection of Southern Ocean Humpback Whales

Machine learning algorithms are often used to model and predict animal habitat selection—the relationships between animal occurrences and habitat characteristics. For broadly distributed species, habitat selection often varies among populations and regions; thus, it would seem preferable to fit region- or population-specific models of habitat selection for more accurate inference and prediction, rather than fitting large-scale models using pooled data. However, where the aim is to make range-wide predictions, including areas for which there are no existing data or models of habitat selection, how can regional models best be combined? We propose that ensemble approaches commonly used to combine different algorithms for a single region can be reframed, treating regional habitat selection models as the candidate models. By doing so, we can incorporate regional variation when fitting predictive models of animal habitat selection across large ranges. We test this approach using satellite telemetry data from 168 humpback whales across five geographic regions in the Southern Ocean. Using random forests, we fitted a large-scale model relating humpback whale locations, versus background locations, to 10 environmental covariates, and made a circumpolar prediction of humpback whale habitat selection. We also fitted five regional models, the predictions of which we used as input features for four ensemble approaches: an unweighted ensemble, an ensemble weighted by environmental similarity in each cell, stacked generalization, and a hybrid approach wherein the environmental covariates and regional predictions were used as input features in a new model. We tested the predictive performance of these approaches on an independent validation dataset of humpback whale sightings and whaling catches. These multiregional ensemble approaches resulted in models with higher predictive performance than the circumpolar naive model. These approaches can be used to incorporate regional variation in animal habitat selection when fitting range-wide predictive models using machine learning algorithms. This can yield more accurate predictions across regions or populations of animals that may show variation in habitat selection.

Where did they not go? Considerations for generating pseudo-absences for telemetry-based habitat models

BACKGROUND

Habitat suitability models give insight into the ecological drivers of species distributions and are increasingly common in management and conservation planning. Telemetry data can be used in habitat models to describe where animals were present, however this requires the use of presence-only modeling approaches or the generation of 'pseudo-absences' to simulate locations where animals did not go. To highlight considerations for generating pseudo-absences for telemetry-based habitat models, we explored how different methods of pseudo-absence generation affect model performance across species' movement strategies, model types, and environments.

METHODS

We built habitat models for marine and terrestrial case studies, Northeast Pacific blue whales (Balaenoptera musculus) and African elephants (Loxodonta africana). We tested four pseudo-absence generation methods commonly used in telemetry-based habitat models: (1) background sampling; (2) sampling within a buffer zone around presence locations; (3) correlated random walks beginning at the tag release location; (4) reverse correlated random walks beginning at the last tag location. Habitat models were built using generalised linear mixed models, generalised additive mixed models, and boosted regression trees.

RESULTS

We found that the separation in environmental niche space between presences and pseudo-absences was the single most important driver of model explanatory power and predictive skill. This result was consistent across marine and terrestrial habitats, two species with vastly different movement syndromes, and three different model types. The best-performing pseudo-absence method depended on which created the greatest environmental separation: background sampling for blue whales and reverse correlated random walks for elephants. However, despite the fact that models with greater environmental separation performed better according to traditional predictive skill metrics, they did not always produce biologically realistic spatial predictions relative to known distributions.

CONCLUSIONS

Habitat model performance may be positively biased in cases where pseudo-absences are sampled from environments that are dissimilar to presences. This emphasizes the need to carefully consider spatial extent of the sampling domain and environmental heterogeneity of pseudo-absence samples when developing habitat models, and highlights the importance of scrutinizing spatial predictions to ensure that habitat models are biologically realistic and fit for modeling objectives.

Identifying priority habitat for conservation and management of Australian humpback dolphins within a marine protected area

Increasing human activity along the coast has amplified the extinction risk of inshore delphinids. Informed selection and prioritisation of areas for the conservation of inshore delphinids requires a comprehensive understanding of their distribution and habitat use. In this study, we applied an ensemble species distribution modelling approach, combining results of six modelling algorithms to identify areas of high probability of occurrence of the globally Vulnerable Australian humpback dolphin in northern Ningaloo Marine Park (NMP), north-western Australia. Model outputs were based on sighting data collected during systematic, boat-based surveys between 2013 and 2015, and in relation to various ecogeographic variables. Water depth and distance to coast were identified as the most important variables influencing dolphin presence, with dolphins showing a preference for shallow waters (5-15 m) less than 2 km from the coast. Areas of high probability (> 0.6) of dolphin occurrence were primarily (90%) in multiple use areas where extractive human activities are permitted, and were poorly represented in sanctuary (no-take) zones. This spatial mismatch emphasises the need to reassess for future spatial planning and marine park management plan reviews for NMP. Shallow, coastal waters identified here should be considered priority areas for the conservation of this Vulnerable species.