Erin Saupe's research while affiliated with University of Oxford and other places

The fossil record is spatiotemporally heterogeneous: taxon occurrence data have patchy spatial distributions, and this patchiness varies through time. Inferences from large-scale quantitative paleobiology studies that fail to account for heterogeneous sampling coverage will be uninformative at best and confidently wrong at worst. Explicitly spatial methods of standardization are necessary for analyses of large-scale fossil datasets, because non-spatial sample standardization, such as diversity rarefaction, is insufficient to reduce the signal of varying spatial coverage through time or between environments and clades. Spatial standardization should control both geographic area and dispersion (spread) of fossil localities. In addition to spatial standardization, other factors may be standardized, including environmental heterogeneity or the number of publications or field collecting units that report taxon occurrences. Using a case study of published global Paleobiology Database occurrences, we demonstrate the strong signals of sampling that could be misinterpreted as biologically meaningful, and which spatial standardization accounts for successfully. We discuss practical issues of implementing spatial standardization via subsampling and present the new R package "divvy" to improve the accessibility of spatial analysis. The software provides three spatial subsampling approaches, as well as related tools to quantify spatial coverage. After reviewing the theory, practice, and history of equalizing spatial coverage between data comparison groups, we outline priority areas to improve related data collection, analysis, and reporting practices in paleobiology.

The latitudinal diversity gradient (LDG) describes the pattern of increasing numbers of species from the poles to the equator. Although recognized for over 200 years, the mechanisms responsible for the largest-scale and longest-known pattern in macroecology are still actively debated. I argue here that any explanation for the LDG must invoke differential rates of speciation, extinction, extirpation, or dispersal. These processes themselves may be governed by numerous abiotic or biotic factors. Hypotheses that claim not to invoke differential rates, such as 'age and area' or 'time for diversification', eschew focus from rate variation that is assumed by these explanations. There is still significant uncertainty in how rates of speciation, extinction, extirpation, and dispersal have varied regionally over Earth history. However, to better understand the development of LDGs, we need to better constrain this variation. Only then will the drivers of such rate variation - be they abiotic or biotic in nature - become clearer.

Ecological niche modelling is applied broadly in ecology to model a species’ niche and map suitable habitat. The approach links species’ occurrences with environmental predictors to statistically derive response curves. Although commonly applied to study extant taxa, ecological niche modelling is an emerging method in palaeobiology, providing opportunities to test ecological hypotheses regarding extinct taxa. However, the extent to which the approach can be applied to fossil data remains unconstrained. The fossil record is inherently incomplete and biased by heterogeneous spatial sampling. Consequently, the complete geographic distribution of a species, and its occupation of environmental space, is often unknown. These limitations can bias niche characterisations, leading to potentially erroneous conclusions about niche dynamics through time. Here, we use a virtual species approach to quantify information loss when using fossil data to estimate species’ climatic niches and geographic distributions through time. We focus on the Late Cretaceous fossil record to quantify the completeness of species’ niches after sampling virtual species by the ‘known’ spatial sampling window. Our results suggest niche characterisations are often incomplete and biased towards a limited range of climatic conditions. Consequently, statistically derived response curves can be misleading in some cases, resulting in erroneous predictions of suitable habitat.

The field of distributional ecology has seen considerable recent attention, particularly surrounding the theory, protocols, and tools for Ecological Niche Modeling (ENM) or Species Distribution Modeling (SDM). Such analyses have grown steadily over the past two decades—including a maturation of relevant theory and key concepts—but methodological consensus has yet to be reached. In response, and following an online course taught in Spanish in 2018, we designed a comprehensive English-language course covering much of the underlying theory and methods currently applied in this broad field. Here, we summarize that course, ENM2020, and provide links by which resources produced for it can be accessed into the future. ENM2020 lasted 43 weeks, with presentations from 52 instructors, who engaged with >2500 participants globally through >14,000 hours of viewing and >90,000 views of instructional video and question-and-answer sessions. Each major topic was introduced by an “Overview” talk, followed by more detailed lectures on subtopics. The hierarchical and modular format of the course permits updates, corrections, or alternative viewpoints, and generally facilitates revision and reuse, including the use of only the Overview lectures for introductory courses. All course materials are free and openly accessible (CC-BY license) to ensure these resources remain available to all interested in distributional ecology.