Deep learning models map rapid plant species changes from citizen science and remote sensing data.

Anthropogenic habitat destruction and climate change are reshaping the geographic distribution of plants worldwide. However, we are still unable to map species shifts at high spatial, temporal, and taxonomic resolution. Here, we develop a deep learning model trained using remote sensing images from California paired with half a million citizen science observations that can map the distribution of over 2,000 plant species. Our model-Deepbiosphere-not only outperforms many common species distribution modeling approaches (AUC 0.95 vs. 0.88) but can map species at up to a few meters resolution and finely delineate plant communities with high accuracy, including the pristine and clear-cut forests of Redwood National Park. These fine-scale predictions can further be used to map the intensity of habitat fragmentation and sharp ecosystem transitions across human-altered landscapes. In addition, from frequent collections of remote sensing data, Deepbiosphere can detect the rapid effects of severe wildfire on plant community composition across a 2-y time period. These findings demonstrate that integrating public earth observations and citizen science with deep learning can pave the way toward automated systems for monitoring biodiversity change in real-time worldwide.

Genomic evidence of an ancient inland temperate rainforest in the Pacific Northwest of North America.

The disjunct temperate rainforests of the Pacific Northwest of North America (PNW) are characterized by late-successional dominant tree species Thuja plicata (western redcedar) and Tsuga heterophylla (western hemlock). The demographic histories of these species, along with the PNW rainforest ecosystem in its entirety, have been heavily impacted by geological and climatic changes the PNW has experienced over the last 5 million years, including mountain orogeny and repeated Pleistocene glaciations. These environmental events have ultimately shaped the history of these species, with inland populations potentially being extirpated during the Pleistocene glaciations. Here, we collect genomic data for both species across their ranges to test multiple demographic models, each reflecting a different phylogeographical hypothesis on how the ecosystem-dominating species may have responded to dramatic climatic change. Our results indicate that inland and coastal populations in both species diverged ~2.5 million years ago in the early Pleistocene and experienced decreases in population size during glacial cycles, with subsequent population expansion. Importantly, we found evidence for gene flow between coastal and inland populations during the mid-Holocene. It is likely that intermittent migration in these species during this time has prevented allopatric speciation via genetic drift alone. In conclusion, our results from combining genomic data and demographic inference procedures establish that populations of the ecosystem dominants Thuja plicata and Tsuga heterophylla persisted in refugia located in both the coastal and inland regions of the PNW throughout the Pleistocene, with populations expanding and contracting in response to glacial cycles with occasional gene flow.

Integrating life history traits into predictive phylogeography.

Predictive phylogeography seeks to aggregate genetic, environmental and taxonomic data from multiple species in order to make predictions about unsampled taxa using machine-learning techniques such as Random Forests. To date, organismal trait data have infrequently been incorporated into predictive frameworks due to difficulties inherent to the scoring of trait data across a taxonomically broad set of taxa. We refine predictive frameworks from two North American systems, the inland temperate rainforests of the Pacific Northwest and the Southwestern Arid Lands (SWAL), by incorporating a number of organismal trait variables. Our results indicate that incorporating life history traits as predictor variables improves the performance of the supervised machine-learning approach to predictive phylogeography, especially for the SWAL system, in which predictions made from only taxonomic and climate variables meets only moderate success. In particular, traits related to reproduction (e.g., reproductive mode; clutch size) and trophic level appear to be particularly informative to the predictive framework. Predictive frameworks offer an important mechanism for integration of organismal trait, environmental data, and genetic data in phylogeographic studies.

Combining allele frequency and tree-based approaches improves phylogeographic inference from natural history collections.

Model selection approaches in phylogeography have allowed researchers to evaluate the support for competing demographic histories, which provides a mode of inference and a measure of uncertainty in understanding climatic and spatial influences on intraspecific diversity. Here, to rank all models in the comparison set and determine what proportion of the total support the top-ranked model garners, we conduct model selection using two analytical approaches-allele frequency-based, implemented in fastsimcoal2, and gene tree-based, implemented in phrapl. We then expand this model selection framework by including an assessment of absolute fit of the models to the data. For this, we utilize DNA isolated from existing natural history collections that span the distribution of red alder (Alnus rubra) in the Pacific Northwest of North America to generate genomic data for the evaluation of 13 demographic scenarios. The quality of DNA recovered from herbarium specimen leaf tissue was assessed for its utility and effectiveness in demographic model selection, specifically in the two approaches mentioned. We present strong support for the use of herbarium tissue in the generation of genomic DNA, albeit with the inclusion of additional quality control checks prior to library preparation and analyses with multiple approaches that incorporate various data. Analyses with allele frequency spectra and gene trees predominantly support A. rubra having experienced an ancient vicariance event with intermittent and frequent gene flow between the disjunct populations. Additionally, the data consistently fit the most frequently selected model, corroborating the model selection techniques. Finally, these results suggest that the A. rubra disjunct populations do not represent separate species.

Demographic model selection using random forests and the site frequency spectrum.

Phylogeographic data sets have grown from tens to thousands of loci in recent years, but extant statistical methods do not take full advantage of these large data sets. For example, approximate Bayesian computation (ABC) is a commonly used method for the explicit comparison of alternate demographic histories, but it is limited by the "curse of dimensionality" and issues related to the simulation and summarization of data when applied to next-generation sequencing (NGS) data sets. We implement here several improvements to overcome these difficulties. We use a Random Forest (RF) classifier for model selection to circumvent the curse of dimensionality and apply a binned representation of the multidimensional site frequency spectrum (mSFS) to address issues related to the simulation and summarization of large SNP data sets. We evaluate the performance of these improvements using simulation and find low overall error rates (~7%). We then apply the approach to data from Haplotrema vancouverense, a land snail endemic to the Pacific Northwest of North America. Fifteen demographic models were compared, and our results support a model of recent dispersal from coastal to inland rainforests. Our results demonstrate that binning is an effective strategy for the construction of a mSFS and imply that the statistical power of RF when applied to demographic model selection is at least comparable to traditional ABC algorithms. Importantly, by combining these strategies, large sets of models with differing numbers of populations can be evaluated.

Identifying cryptic diversity with predictive phylogeography.

Identifying units of biological diversity is a major goal of organismal biology. An increasing literature has focused on the importance of cryptic diversity, defined as the presence of deeply diverged lineages within a single species. While most discoveries of cryptic lineages proceed on a taxon-by-taxon basis, rapid assessments of biodiversity are needed to inform conservation policy and decision-making. Here, we introduce a predictive framework for phylogeography that allows rapidly identifying cryptic diversity. Our approach proceeds by collecting environmental, taxonomic and genetic data from codistributed taxa with known phylogeographic histories. We define these taxa as a reference set, and categorize them as either harbouring or lacking cryptic diversity. We then build a random forest classifier that allows us to predict which other taxa endemic to the same biome are likely to contain cryptic diversity. We apply this framework to data from two sets of disjunct ecosystems known to harbour taxa with cryptic diversity: the mesic temperate forests of the Pacific Northwest of North America and the arid lands of Southwestern North America. The predictive approach presented here is accurate, with prediction accuracies placed between 65% and 98.79% depending of the ecosystem. This seems to indicate that our method can be successfully used to address ecosystem-level questions about cryptic diversity. Further, our application for the prediction of the cryptic/non-cryptic nature of unknown species is easily applicable and provides results that agree with recent discoveries from those systems. Our results demonstrate that the transition of phylogeography from a descriptive to a predictive discipline is possible and effective.

Detecting the ecological footprint of selection.

The structure of communities is influenced by many ecological and evolutionary processes, but the way these manifest in classic biodiversity patterns often remains unclear. Here we aim to distinguish the ecological footprint of selection-through competition or environmental filtering-from that of neutral processes that are invariant to species identity. We build on existing Massive Eco-evolutionary Synthesis Simulations (MESS), which uses information from three biodiversity axes-species abundances, genetic diversity, and trait variation-to distinguish between mechanistic processes. To correctly detect and characterise competition, we add a new and more realistic form of competition that explicitly compares the traits of each pair of individuals. Our results are qualitatively different to those of previous work in which competition is based on the distance of each individual's trait to the community mean. We find that our new form of competition is easier to identify in empirical data compared to the alternatives. This is especially true when trait data are available and used in the inference procedure. Our findings hint that signatures in empirical data previously attributed to neutrality may in fact be the result of pairwise-acting selective forces. We conclude that gathering more different types of data, together with more advanced mechanistic models and inference as done here, could be the key to unravelling the mechanisms of community assembly and question the relative roles of neutral and selective processes.

From coral reefs to Joshua trees: What ecological interactions teach us about the adaptive capacity of biodiversity in the Anthropocene.

The pervasive loss of biodiversity in the Anthropocene necessitates rapid assessments of ecosystems to understand how they will respond to anthropogenic environmental change. Many studies have sought to describe the adaptive capacity (AC) of individual species, a measure that encompasses a species' ability to respond and adapt to change. Only those adaptive mechanisms that can be used over the next few decades (e.g. via novel interactions, behavioural changes, hybridization, migration, etc.) are relevant to the timescale set by the rapid changes of the Anthropocene. The impacts of species loss cascade through ecosystems, yet few studies integrate the capacity of ecological networks to adapt to change with the ACs of its species. Here, we discuss three ecosystems and how their ecological networks impact the AC of species and vice versa. A more holistic perspective that considers the AC of species with respect to their ecological interactions and functions will provide more predictive power and a deeper understanding of what factors are most important to a species' survival. We contend that the AC of a species, combined with its role in ecosystem function and stability, must guide decisions in assigning 'risk' and triaging biodiversity loss in the Anthropocene. This article is part of the theme issue 'Ecological complexity and the biosphere: the next 30 years'.

Genetic diversity loss in the Anthropocene.

Anthropogenic habitat loss and climate change are reducing species' geographic ranges, increasing extinction risk and losses of species' genetic diversity. Although preserving genetic diversity is key to maintaining species' adaptability, we lack predictive tools and global estimates of genetic diversity loss across ecosystems. We introduce a mathematical framework that bridges biodiversity theory and population genetics to understand the loss of naturally occurring DNA mutations with decreasing habitat. By analyzing genomic variation of 10,095 georeferenced individuals from 20 plant and animal species, we show that genome-wide diversity follows a mutations-area relationship power law with geographic area, which can predict genetic diversity loss from local population extinctions. We estimate that more than 10% of genetic diversity may already be lost for many threatened and nonthreatened species, surpassing the United Nations' post-2020 targets for genetic preservation.

Phylogenetic diversity and community assembly in a naturally fragmented system.

We sought to assess effects of fragmentation and quantify the contribution of ecological processes to community assembly by measuring species richness, phylogenetic, and phenotypic diversity of species found in local and regional plant communities. Specifically, our fragmented system is Craters of the Moon National Monument and Preserve, Idaho, USA. CRMO is characterized by vegetated islands, kipukas, that are isolated in a matrix of lava. We used floristic surveys of vascular plants in 19 kipukas to create a local species list to compare traditional dispersion metrics, mean pairwise distance, and mean nearest taxon distance (MPD and MNTD), to a regional species list with phenotypic and phylogenetic data. We combined phylogenetic and functional trait data in a novel machine-learning model selection approach, Community Assembly Model Inference (CAMI), to infer probability associated with different models of community assembly given the data. Finally, we used linear regression to explore whether the geography of kipukas explained estimated support for community assembly models. Using traditional metrics of MPD and MNTD neutral processes received the most support when comparing kipuka species to regional species. Individually no kipukas showed significant support for overdispersion. Rather, five kipukas showed significant support for phylogenetic clustering using MPD and two kipukas using MNTD. Using CAMI, we inferred neutral and filtering models structured the kipuka plant community for our trait of interest. Finally, we found as species richness in kipukas increases, model support for competition decreases and lower elevation kipukas show more support for habitat filtering models. While traditional phylogenetic community approaches suggest neutral assembly dynamics, recently developed approaches utilizing machine learning and model choice revealed joint influences of assembly processes to form the kipuka plant communities. Understanding ecological processes at play in naturally fragmented systems will aid in guiding our understanding of how fragmentation impacts future changes in landscapes.

A unified model of species abundance, genetic diversity, and functional diversity reveals the mechanisms structuring ecological communities.

Biodiversity accumulates hierarchically by means of ecological and evolutionary processes and feedbacks. Within ecological communities drift, dispersal, speciation, and selection operate simultaneously to shape patterns of biodiversity. Reconciling the relative importance of these is hindered by current models and inference methods, which tend to focus on a subset of processes and their resulting predictions. Here we introduce massive ecoevolutionary synthesis simulations (MESS), a unified mechanistic model of community assembly, rooted in classic island biogeography theory, which makes temporally explicit joint predictions across three biodiversity data axes: (i) species richness and abundances, (ii) population genetic diversities, and (iii) trait variation in a phylogenetic context. Using simulations we demonstrate that each data axis captures information at different timescales, and that integrating these axes enables discriminating among previously unidentifiable community assembly models. MESS is unique in generating predictions of community-scale genetic diversity, and in characterizing joint patterns of genetic diversity, abundance, and trait values. MESS unlocks the full potential for investigation of biodiversity processes using multidimensional community data including a genetic component, such as might be produced by contemporary eDNA or metabarcoding studies. We combine MESS with supervised machine learning to fit the parameters of the model to real data and infer processes underlying how biodiversity accumulates, using communities of tropical trees, arthropods, and gastropods as case studies that span a range of data availability scenarios, and spatial and taxonomic scales.

Identifying models of trait-mediated community assembly using random forests and approximate Bayesian computation.

Ecologists often use dispersion metrics and statistical hypothesis testing to infer processes of community formation such as environmental filtering, competitive exclusion, and neutral species assembly. These metrics have limited power in inferring assembly models because they rely on often-violated assumptions. Here, we adapt a model of phenotypic similarity and repulsion to simulate the process of community assembly via environmental filtering and competitive exclusion, all while parameterizing the strength of the respective ecological processes. We then use random forests and approximate Bayesian computation to distinguish between these models given the simulated data. We find that our approach is more accurate than using dispersion metrics and accounts for uncertainty in model selection. We also demonstrate that the parameter determining the strength of the assembly processes can be accurately estimated. This approach is available in the R package CAMI; Community Assembly Model Inference. We demonstrate the effectiveness of CAMI using an example of plant communities living on lava flow islands.