Mapping biodiversity with environmental niche modeling

Schematic of environmental niche modeling

Joshua Ladau
Gladstone Institutes, GICD

(Left: Figure 1. Schematic of the environmental niche modeling methodology. Environmental niche modeling uses a statistical model to combine community samples with maps of environmental variables to generate predicted diversity maps or range maps. In this example, most of the samples are from North America, so although a global prediction is shown, a prediction to just North America may be better warranted.)

Maps of biodiversity at global and continental scales plays a pivotal role in ecology, evolutionary biology, and conservation biology.  Diversity maps give insight into the action of ecological and evolutionary processes at large scales, and they can guide management decisions. The diversity patterns of many macro-organisms are known; for instance, the global ranges of almost all known mammals, birds, and amphibians are available at a resolution of approximately 100 km. However, until recently, the ranges of many micro-organisms — including soil bacteria — have remained almost entirely unknown. In this post, I will describe an increasingly useful approach to inferring the diversity patterns and ranges of soil bacteria and other micro-organisms.

The reasons that the ranges of soil bacteria have remained elusive are at least two-fold.

First, sampling of soil bacterial communities has historically been sparse: Range maps of many macro-organisms are often based on many thousands of observations and museum records. By contrast, until recently the number of soil bacterial communities that have been sampled has been orders of magnitude smaller. However, the availability of cheaper and faster sequencing is rapidly changing this picture, and today numerous soil bacterial communities have been sequenced with additional surveys underway.

The second obstacle to mapping the ranges of soil bacteria is incomplete censusing of bacterial communities. Unlike macro-organism communities, many of which can effectively be censused given sufficient effort, fully assessing the diversity of bacteria at a location has been impossible. Although completely censusing most soil bacterial communities continues to remain impractical, improvements to sequencing technology have made it possible to get an increasingly complete picture of the bacterial communities.

Against this backdrop, we recently sought to ask not just whether we could map the ranges and diversity patterns of soil bacteria, but whether we could map the ranges and diversity patterns of soil bacteria across a largely extirpated ecosystem, the North American tallgrass prairie (Fierer et al, 2013, Reconstructing the Microbial Diversity and Function of Pre-Agricultural Tallgrass Prairie Soils in the United States. Science, 342:621-624). Historically, this ecosystem covered over 65 million ha of the United States. Today, as much as 99% of the tallgrass prairie has been lost to agriculture and other land use modification. These modifications have profoundly altered the soil communities. However, the bacterial communities that presumably inhabited the tallgrass prairie can still be sampled from the few tallgrass prairie relics that remain undisturbed, for instance in nature preserves and old cemeteries. We used samples to form these communities to infer the likely distributions of soil bacteria across the original extent of the tallgrass prairie.

To infer the distributions across the original tallgrass prairie extent, we used environmental niche modeling, a powerful methodology developed for studying the distribution of macro-organisms. To generate range maps, niche modeling combines (i) a limited number of community samples with (ii) maps of climate conditions and other environmental variables.

The idea behind niche modeling is straightforward: using the community samples, one infers the niches of taxa — e.g., bacterial OTUs or phylotypes – for the variables for which maps are available. For instance, if we had maps of precipitation and soil pH, we would check how the occurrence of an OTU of Acidobacteria depends on these variables. If it depends on these variables, for example the OTU is found only in samples with high precipitation and high pH, then we can draw its range by referencing the maps of precipitation and soil pH, shading in only regions that meet the OTU's precipitation and pH requirements.

Although the general idea behind environmental niche modeling is straightforward, the details are complicated. The challenges include:  Out of the many possible environmental variables that might predict a distribution of an OTU, how do we choose the important ones? How do we model non-linear relationships between the occurrence of OTUs and environmental variables? How do we prevent overfitting? Can we reasonably assume niche conservatism through space? Do dispersal limitation and interspecific interactions substantially affect range boundaries? What regions of the inferred maps require excessive extrapolation in environmental space, and how can we control the amount of extrapolation? Can we map diversity patterns in addition to ranges?

A large body of ecological and statistical research has been devoted to addressing these and other questions, and sophisticated niche modeling methods have been developed. In many circumstances, these methods can allow the construction accurate range maps.

We applied these niche modeling methods to map the distributions of the tallgrass prairie bacteria.  To make these maps across the original extent of the tallgrass prairie, we noted that climate conditions have changed little across this ecosystem in the last 150 years. Thus, if we found that a particular set of climate variables predicted the distribution of a taxon across the existing prairie relics, this would suggest that we could predict the original distribution across the prairie from the spatial distribution of this climate variable across the original extent of the tallgrass prairie. We found that climate variables were indeed excellent predictors for some taxa, and that we could additionally predict the spatial distributions of diversity across these region using climate variables. The results indicated that the diversity patterns of bacteria are sufficiently associated with climate variables that range maps can be constructed, and that their diversity patterns were driven largely by the distribution of Verrucomicrobia, a relatively little-studied phylum of bacteria.

Environmental niche modeling promises to be an important addition to soil microbial ecologists' computational toolbox. The application of environmental niche modeling to mapping tallgrass soil bacterial communities is but one example of the utility of this methodology. For instance, it may be useful for identifying microbial interactions, predicting the effects of climate change on microbial communities, and identifying unsampled regions that are likely to harbor beneficial or harmful bacterial taxa. Recent advances in sequencing technology, ongoing sampling of additional bacterial communities, and the development of the appropriate statistical tools are combining to make environmental niche modeling an increasingly practical and useful methodology for mapping the distributions of soil microbial communities.