Effects of global warming on ancient mammalian communities and their environments

Mitigating carbon footprint for knowledge distillation based deep learning model compression

Deep learning techniques have recently demonstrated remarkable success in numerous domains. Typically, the success of these deep learning models is measured in terms of performance metrics such as accuracy and mean average precision (mAP). Generally, a model's high performance is highly valued, but it frequently comes at the expense of substantial energy costs and carbon footprint emissions during the model building step. Massive emission of CO2 has a deleterious impact on life on earth in general and is a serious ethical concern that is largely ignored in deep learning research. In this article, we mainly focus on environmental costs and the means of mitigating carbon footprints in deep learning models, with a particular focus on models created using knowledge distillation (KD). Deep learning models typically contain a large number of parameters, resulting in a 'heavy' model. A heavy model scores high on performance metrics but is incompatible with mobile and edge computing devices. Model compression techniques such as knowledge distillation enable the creation of lightweight, deployable models for these low-resource devices. KD generates lighter models and typically performs with slightly less accuracy than the heavier teacher model (model accuracy by the teacher model on CIFAR 10, CIFAR 100, and TinyImageNet is 95.04%, 76.03%, and 63.39%; model accuracy by KD is 91.78%, 69.7%, and 60.49%). Although the distillation process makes models deployable on low-resource devices, they were found to consume an exorbitant amount of energy and have a substantial carbon footprint (15.8, 17.9, and 13.5 times more carbon compared to the corresponding teacher model). The enormous environmental cost is primarily attributable to the tuning of the hyperparameter, Temperature (τ). In this article, we propose measuring the environmental costs of deep learning work (in terms of GFLOPS in millions, energy consumption in kWh, and CO2 equivalent in grams). In order to create lightweight models with low environmental costs, we propose a straightforward yet effective method for selecting a hyperparameter (τ) using a stochastic approach for each training batch fed into the models. We applied knowledge distillation (including its data-free variant) to problems involving image classification and object detection. To evaluate the robustness of our method, we ran experiments on various datasets (CIFAR 10, CIFAR 100, Tiny ImageNet, and PASCAL VOC) and models (ResNet18, MobileNetV2, Wrn-40-2). Our novel approach reduces the environmental costs by a large margin by eliminating the requirement of expensive hyperparameter tuning without sacrificing performance. Empirical results on the CIFAR 10 dataset show that the stochastic technique achieves an accuracy of 91.67%, whereas tuning achieves an accuracy of 91.78%-however, the stochastic approach reduces the energy consumption and CO2 equivalent each by a factor of 19. Similar results have been obtained with CIFAR 100 and TinyImageNet dataset. This pattern is also observed in object detection classification on the PASCAL VOC dataset, where the tuning technique performs similarly to the stochastic technique, with a difference of 0.03% mAP favoring the stochastic technique while reducing the energy consumptions and CO2 emission each by a factor of 18.5.

On the relationship between collagen- and carbonate-derived carbon isotopes with implications for the inference of carnivore dietary behavior

Studies of Rancho La Brea predators have yielded disparate dietary interpretations when analyzing bone collagen vs. enamel carbonate—requiring a better understanding of the relationship between stable carbon isotopes in these tissues. Stable carbon isotope spacing between collagen and carbonate (Δca-co) has also been used as a proxy for inferring the trophic level of mammals, with higher Δca-co values indicative of high carbohydrate consumption. To clarify the stable isotope ecology of carnivorans, past and present, we analyzed bone collagen (carbon and nitrogen) and enamel carbonate (carbon) of extinct and extant North American felids and canids, including dire wolves, sabertooth cats, coyotes, and pumas, supplementing these with data from African wild dogs and African lions. Our results reveal that Δca-co values are positively related to enamel carbonate values in secondary consumers and are less predictive of trophic level. Results indicate that the foraging habitat and diet of prey affects Δca-co in carnivores, like herbivores. Average Δca-co values in Pleistocene canids (8.7+/−1‰) and felids (7.0+/−0.7‰) overlap with previously documented extant herbivore Δca-co values suggesting that trophic level estimates may be relative to herbivore Δca-co values in each ecosystem and not directly comparable between disparate ecosystems. Physiological differences between felids and canids, ontogenetic dietary differences, and diagenesis at Rancho La Brea do not appear to be primary drivers of Δca-co offsets. Environmental influences affecting protein and fat consumption in prey and subsequently by predators, and nutrient routing to tissues may instead be driving Δca-co offsets in extant and extinct mammals.

Global long-term stability of individual dietary specialization in herbivorous mammals

Dietary variation within species has important ecological and evolutionary implications. While theoreticians have debated the consequences of trait variance (including dietary specialization), empirical studies have yet to examine intraspecific dietary variability across the globe and through time. Here, we use new and published serial sampled δ¹³C_enamel values of herbivorous mammals from the Miocene to the present (318 individuals summarized, 4134 samples) to examine how dietary strategy (i.e. browser, mixed-feeder, grazer) affects individual isotopic variation. We find that almost all herbivores, regardless of dietary strategy, are composed of individual specialists. For example, Cormohipparion emsliei (Equidae) from the Pliocene of Florida (approx. 5 Ma) exhibits a δ¹³C_enamel range of 13.4‰, but all individuals sampled have δ¹³C_enamel ranges of less than or equal to 2‰ (mean = 1.1‰). Most notably, this pattern holds globally and through time, with almost all herbivorous mammal individuals exhibiting narrow δ¹³C_enamel ranges (less than or equal to 3‰), demonstrating that individuals are specialized and less representative of their overall species' dietary breadth. Individual specialization probably reduces intraspecific competition, increases carrying capacities, and may have stabilizing effects on species and communities over time. Individual specialization among species with both narrow and broad dietary niches is common over space and time-a phenomenon not previously well recognized or documented empirically.

Pronghorn (Antilocapra americana) enamel phosphate δ18O values reflect climate seasonality: Implications for paleoclimate reconstruction

Stable oxygen isotope (δ18O) compositions from vertebrate tooth enamel are widely used as biogeochemical proxies for paleoclimate. However, the utility of enamel oxygen isotope values for environmental reconstruction varies among species. Herein, we evaluate the use of stable oxygen isotope compositions from pronghorn (Antilocapra americana Gray, 1866) enamel for reconstructing paleoclimate seasonality, an elusive but important parameter for understanding past ecosystems. We serially sampled the lower third molars of recent adult pronghorn from Wyoming for δ18O in phosphate (δ18OPO4) and compared patterns to interpolated and measured yearly variation in environmental waters as well as from sagebrush leaves, lakes, and rivers (δ18Ow). As expected, the oxygen isotope compositions of phosphate from pronghorn enamel are enriched in 18O relative to environmental waters. For a more direct comparison, we converted δ18Ow values into expected δ18OPO4* values (δ18OW-PO4*). Pronghorn δ18OPO4 values from tooth enamel record nearly the full amplitude of seasonal variation from Wyoming δ18OW-PO4* values. Furthermore, pronghorn enamel δ18OPO4 values are more similar to modeled δ18OW-PO4* values from plant leaf waters than meteoric waters, suggesting that they obtain much of their water from evaporated plant waters. Collectively, our findings establish that seasonality in source water is reliably reflected in pronghorn enamel, providing the basis for exploring changes in the amplitude of seasonality of ancient climates. As a preliminary test, we sampled historical pronghorn specimens (1720 ± 100 AD), which show a mean decrease (a shift to lower values) of 1-2‰ in δ18OPO4 compared to the modern specimens. They also exhibit an increase in the δ18O amplitude, representing an increase in seasonality. We suggest that the cooler mean annual and summer temperatures typical of the 18th century, as well as enhanced periods of drought, drove differences among the modern and historical pronghorn, further establishing pronghorn enamel as excellent sources of paleoclimate proxy data.

Dietary ecology of the scimitar-toothed cat Homotherium serum

The scimitar-toothed cat Homotherium was one of the most cosmopolitan cats of the Pleistocene, present throughout Eurasia, Africa, and the Americas until at least ~28 thousand years ago.^1-3 Friesenhahn Cave (Bexar County, Texas) contains some of the best-preserved specimens of Homotherium serum alongside an abundance of juvenile mammoths, leading some to argue that H. serum preferentially hunted juvenile mammoths.^1,4 Dietary data of Homotherium are rare, with their ecology inferred from morphological, taphonomic, and genetic data.^1,3-10 Here, we use a multi-proxy approach to clarify the dietary ecology of H. serum as compared to extinct and extant cats and their relatives. Dental microwear texture analysis (DMTA) reveals that H. serum consumed soft and tough foods, similar to the extant cheetah, which actively avoids bone,^11,12 but in stark contrast to extant lions and hyenas, which are observed to engage in durophagy (i.e., bone processing).^11-14 DMTA data are consistent with taphonomic evidence of bone defleshing and the absence of bone-crunching behavior in H. serum. Stable carbon isotope values of H. serum indicate a clear preference for C₄ grazers including juvenile mammoths, in agreement with taphonomic evidence suggestive of a "Homotherium den"^1,4 and morphological data indicative of a relatively cursorial lifestyle.^6-8 Notably, the inferred diet of H. serum contrasts with the extinct dirk-tooth sabertooth cat Smilodon fatalis, which preferred forest/woodland prey and engaged in bone processing.^15-19Homotherium serum exhibited a novel combination of morphological adaptations for acquiring open-country prey, consuming their soft and tough flesh-including the tough flesh of juvenile mammoths. VIDEO ABSTRACT.

Overkill, glacial history, and the extinction of North America's Ice Age megafauna

The end of the Pleistocene in North America saw the extinction of 38 genera of mostly large mammals. As their disappearance seemingly coincided with the arrival of people in the Americas, their extinction is often attributed to human overkill, notwithstanding a dearth of archaeological evidence of human predation. Moreover, this period saw the extinction of other species, along with significant changes in many surviving taxa, suggesting a broader cause, notably, the ecological upheaval that occurred as Earth shifted from a glacial to an interglacial climate. But, overkill advocates ask, if extinctions were due to climate changes, why did these large mammals survive previous glacial-interglacial transitions, only to vanish at the one when human hunters were present? This question rests on two assumptions: that previous glacial-interglacial transitions were similar to the end of the Pleistocene, and that the large mammal genera survived unchanged over multiple such cycles. Neither is demonstrably correct. Resolving the cause of large mammal extinctions requires greater knowledge of individual species' histories and their adaptive tolerances, a fuller understanding of how past climatic and ecological changes impacted those animals and their biotic communities, and what changes occurred at the Pleistocene-Holocene boundary that might have led to those genera going extinct at that time. Then we will be able to ascertain whether the sole ecologically significant difference between previous glacial-interglacial transitions and the very last one was a human presence.

The Great American Biotic Interchange revisited: a new perspective from the stable isotope record of Argentine Pampas fossil mammals

This study aims at assessing resource and habitat use, niche occupation and trophic interactions from a stable isotope perspective on fossil mammals from the Argentine Pampas during the Great American Biotic Interchange (GABI). We present stable isotope data of more than 400 samples belonging to 10 mammalian orders and spanning a temporal range from ~9.5 Ma to ~12 ky. Rodents, notoungulates and pilosians record an increase in the consumption of C₄ plants, whereas litopterns and cingulates show δ¹³C values that remain mostly within a C₃-dominated diet. Our stable isotope data indicates that the expansion of C₄ vegetation opened up new niche opportunities, probably alleviating resource competition among endemic taxa. Gomphothere, equid and camelid δ¹³C records show a broad variability pointing to consumption of C₃ and mixed C₃-C₄ vegetation. This flexible dietary behavior may have facilitated the successful settlement of immigrant groups in South America. In the case of carnivorous taxa, Late Miocene pre-GABI endemic sparassodonts consumed prey from C₃ environments, whereas immigrant carnivorans preferred prey from mixed C₃-C₄ areas. Our research contributes to the study of the GABI from a different perspective as stable isotope records permit to characterize, from a (semi)quantitative standpoint, ecological traits within extinct fauna.

A year in the life of a giant ground sloth during the Last Glacial Maximum in Belize

Stable isotope analysis of the first fossilized Eremotherium laurillardi remains from Belize offers valuable insights into the conditions within which this individual lived and its ability to adapt to the increasing aridity of the Last Glacial Maximum (LGM). Cathodoluminescence (CL) microscopy was used to identify chemical alteration of the tooth during fossilization. Results demonstrate that the inner orthodentin resists diagenesis, yielding potentially unaltered values. Using an intensive "vacuum milling" technique, the inner orthodentin produced an accelerator mass spectrometry (AMS) date of 26,975 ± 120 calibrated years before the present. The stable carbon and oxygen isotope analysis of this layer shows that the tooth recorded two wet seasons separated by one longer dry season and that this sloth was able to adapt its diet to the marked seasonality of the LGM. This study offers new insights into obtaining reliable isotope data from fossilized remains and suggests that this individual adapted to climate shifts, contributing to the conversation surrounding megafauna extinction.

Phylogenetic signal in tooth wear dietary niche proxies

In the absence of independent observational data, ecologists and paleoecologists use proxies for the Eltonian niches of species (i.e., the resource or dietary axes of the niche). Some dietary proxies exploit the fact that mammalian teeth experience wear during mastication, due to both tooth-on-tooth and food-on-tooth interactions. The distribution and types of wear detectible at micro- and macroscales are highly correlated with the resource preferences of individuals and, in turn, species. Because methods that quantify the distribution of tooth wear (i.e., analytical tooth wear methods) do so by direct observation of facets and marks on the teeth of individual animals, dietary inferences derived from them are thought to be independent of the clade to which individuals belong. However, an assumption of clade or phylogenetic independence when making species-level dietary inferences may be misleading if phylogenetic niche conservatism is widespread among mammals. Herein, we test for phylogenetic signal in data from numerous analytical tooth wear studies, incorporating macrowear (i.e., mesowear) and microwear (i.e., low-magnification microwear and dental microwear texture analysis). Using two measures of phylogenetic signal, heritability (H2) and Pagel's λ, we find that analytical tooth wear data are not independent of phylogeny and failing to account for such nonindependence leads to overestimation of discriminability among species with different dietary preferences. We suggest that morphological traits inherited from ancestral clades (e.g., tooth shape) influence the ways in which the teeth wear during mastication and constrain the foods individuals of a species can effectively exploit. We do not suggest that tooth wear is simply phylogeny in disguise; the tooth wear of individuals and species likely varies within some range that is set by morphological constraints. We therefore recommend the use of phylogenetic comparative methods in studies of mammalian tooth wear, whenever possible.

Cheek tooth morphology and ancient mitochondrial DNA of late Pleistocene horses from the western interior of North America: Implications for the taxonomy of North American Late Pleistocene Equus

Horses were a dominant component of North American Pleistocene land mammal communities and their remains are well represented in the fossil record. Despite the abundant material available for study, there is still considerable disagreement over the number of species of Equus that inhabited the different regions of the continent and on their taxonomic nomenclature. In this study, we investigated cheek tooth morphology and ancient mtDNA of late Pleistocene Equus specimens from the Western Interior of North America, with the objective of clarifying the species that lived in this region prior to the end-Pleistocene extinction. Based on the morphological and molecular data analyzed, a caballine (Equus ferus) and a non-caballine (E. conversidens) species were identified from different localities across most of the Western Interior. A second non-caballine species (E. cedralensis) was recognized from southern localities based exclusively on the morphological analyses of the cheek teeth. Notably the separation into caballine and non-caballine species was observed in the Bayesian phylogenetic analysis of ancient mtDNA as well as in the geometric morphometric analyses of the upper and lower premolars. Teeth morphologically identified as E. conversidens that yielded ancient mtDNA fall within the New World stilt-legged clade recognized in previous studies and this is the name we apply to this group. Geographic variation in morphology in the caballine species is indicated by statistically different occlusal enamel patterns in the specimens from Bluefish Caves, Yukon Territory, relative to the specimens from the other geographic regions. Whether this represents ecomorphological variation and/or a certain degree of geographic and genetic isolation of these Arctic populations requires further study.

Stable isotope ecology of the koala (Phascolarctos cinereus)

Australia has undergone significant climate change, both today and in the past. Koalas, due to their restricted diet of predominantly eucalyptus leaves and limited drinking behaviour may serve as model organisms for assessing past climate change via stable isotopes of tooth enamel. Here, we assess whether stable carbon and oxygen isotopes from tooth enamel record known climate variables, including proxies of relative aridity (e.g. mean annual precipitation, mean annual maximum temperature, and relative humidity). The results demonstrate significant negative relationships between oxygen isotope values and both relative humidity and mean annual precipitation, proxies for relative aridity. The best model for predicting enamel oxygen isotope values incorporates mean annual precipitation and modelled oxygen isotope values of local precipitation. These data and the absence of any relationship between modelled oxygen isotope precipitation values, independently, suggest that koalas do not track local precipitation values but instead record relative aridity. The lack of significant relationships between carbon isotopes and climate variables suggests that koalas may instead be tracking the density of forests and/or their location in the canopy. Collectively, these data suggest that koalas are model organisms for assessing relative aridity over time – much like kangaroos.

Exploring the Potential of Laser Ablation Carbon Isotope Analysis for Examining Ecology during the Ontogeny of Middle Pleistocene Hominins from Sima de los Huesos (Northern Spain)

Laser ablation of tooth enamel was used to analyze stable carbon isotope compositions of teeth of hominins, red deer, and bears from middle Pleistocene sites in the Sierra de Atapuerca in northern Spain, to investigate the possibility that this technique could be used as an additional tool to identify periods of physiological change that are not detectable as changes in tooth morphology. Most of the specimens were found to have minimal intra-tooth variation in carbon isotopes (< 2.3‰), suggesting isotopically uniform diets through time and revealing no obvious periods of physiological change. However, one of the two sampled hominin teeth displayed a temporal carbon isotope shift (3.2‰) that was significantly greater than observed for co-occurring specimens. The δ¹³C value of this individual averaged about -16‰ early in life, and -13‰ later in life. This isotopic change occurred on the canine crown about 4.2 mm from the root, which corresponds to an approximate age of two to four years old in modern humans. Our dataset is perforce small owing to the precious nature of hominid teeth, but it demonstrates the potential utility of the intra-tooth isotope profile method for extracting ontogenetic histories of human ancestors.

Relict endemism of extant Rhineuridae (Amphisbaenia): testing for phylogenetic niche conservatism in the fossil record

Rhineurid amphisbaenians are represented by a rich Cenozoic fossil record in North America, but today conisist of a single living species restricted to the Florida Peninsula. Such relict endemism may be the result of phylogenetic niche conservatism (PNC), the retention of ancestral traits preventing expansion into new environments. Most tests of PNC derive ancestral niche preferences from species' extant ecologies, while ignoring valuable paleontological information. To test if PNC contributes to the restricted distribution of modern Rhineura floridana, we compare the species' current environmental preferences (temperature, precipitation and soil) to paleoenvironmental data from the rhineurid fossil record. We find no evidence of PNC in modern R. floridana, as it also occurred in Florida during drier glacial periods. Ancient rhineurids also exhibit tolerance to changing climates, having undergone a shift from subtropical-humid to semi-arid savanna conditions during the Eocene-Oligocene transition. However, rhineurids nearly disappear from North America after the middle Miocene, potentially due to the onset of prolonged freezing temperatures following the mid-Miocene Climatic Optimum. This physiological limit of environmental tolerances could be interpreted as PNC for the entire family, but also characterizes much of Amphisbaenia, emphasizing the relevance of the temporal as well as phylogenetic scale at which PNC is investigated.

Mammalian niche conservation through deep time

Climate change alters species distributions, causing plants and animals to move north or to higher elevations with current warming. Bioclimatic models predict species distributions based on extant realized niches and assume niche conservation. Here, we evaluate if proxies for niches (i.e., range areas) are conserved at the family level through deep time, from the Eocene to the Pleistocene. We analyze the occurrence of all mammalian families in the continental USA, calculating range area, percent range area occupied, range area rank, and range polygon centroids during each epoch. Percent range area occupied significantly increases from the Oligocene to the Miocene and again from the Pliocene to the Pleistocene; however, mammalian families maintain statistical concordance between rank orders across time. Families with greater taxonomic diversity occupy a greater percent of available range area during each epoch and net changes in taxonomic diversity are significantly positively related to changes in percent range area occupied from the Eocene to the Pleistocene. Furthermore, gains and losses in generic and species diversity are remarkably consistent with ∼2.3 species gained per generic increase. Centroids demonstrate southeastern shifts from the Eocene through the Pleistocene that may correspond to major environmental events and/or climate changes during the Cenozoic. These results demonstrate range conservation at the family level and support the idea that niche conservation at higher taxonomic levels operates over deep time and may be controlled by life history traits. Furthermore, families containing megafauna and/or terminal Pleistocene extinction victims do not incur significantly greater declines in range area rank than families containing only smaller taxa and/or only survivors, from the Pliocene to Pleistocene. Collectively, these data evince the resilience of families to climate and/or environmental change in deep time, the absence of terminal Pleistocene “extinction prone” families, and provide valuable insights to understanding mammalian responses to current climate change.

Evolution of climate niches in European mammals?

Our ability to predict consequences of climate change is severely impaired by the lack of knowledge on the ability of species to adapt to changing environmental conditions. We used distribution data for 140 mammal species in Europe, together with data on climate, land cover and topography, to derive a statistical description of their realized climate niche. We then compared climate niche overlap of pairs of species, selected on the basis of phylogenetic information. In contrast to expectations, related species were not similar in their climate niche. Rather, even species pairs that had a common ancestor less than 1 Ma already display very high climate niche distances. We interpret our finding as a strong interspecific competitive constraint on the realized niche, rather than a rapid evolution of the fundamental niche. If correct, our results imply a very limited usefulness of climate niche models for the prediction of future mammal distributions.