Species distribution modelling

Credit: Michele Menegon

Species distribution modelling (SDM) is commonly used to predict spatial patterns of biodiversity across sets of taxa with sufficient distributional records, while omitting narrow-ranging species due to statistical constraints. We investigate the implications of this dichotomy for conservation priority setting in Africa, now and in the future.

There are various ways to determine the spatial distribution pattern of species. Most attempt to predict species’ geographic ranges from occurrence records and environmental data from the same sites. Two types of output are very common: binary classifications of given sites as being within or outside the distribution, and probabilistic results often used to develop maps of predicted future distributions. 

Most species distribution models work with with basic ecological assumptions. Some of these assumptions relate to ecological processes that determine species’ distributions and abundances, whereas others are methodological and concern the way data are treated in species distribution modelling. In recent years, e.g. modeled impact of climate change on species distributions has become an additional consideration for conservation priority setting. 

A considerable body of literature exists on species distribution modelling and a couple of pitfalls have been identified. There are some known methodological issues and researchers try to address those by adapting either the algorithms used or the sampling regime:

- Incomplete sampling of niche space

  Distributional data are often collected unevenly across a species’ range with respect to space, time, and environment. The collection of biological data is time-consuming and expensive. As a result many studies have to rely on limited amounts of data and researchers need to make the best out of those. This is one of the reasons for global initiatives such as GBIF or OBIS to aggregate as much data from all possible sources to fill the gaps.

- Cause and correlation

  That is a dangerous one. Strong correlation does not automatically imply causation. For example climate data can be associated with many ecological phenomena for reasons other than causality. Strong correlations may allow climate data to predict species occurrences, even if climate has little to do with it. This is also true for the opposite scenario. Climate change might be the main cause for changes is species distributions but no significant correlations are obtained. 

- Scale mismatch

  Some available data are more tightly parameterized than others (e.g., precipitation patterns), and their accuracy and resolution can have important ramifications any predictions.  The scale mismatch issue brings us back to the opening paragraph. Many species require specific small-scale habitat attributes that are likely to be overlooked by common species distribution models. 

A new study demonstrates that the majority of threatened species are 'invisible' when using species distribution modelling to predict species distributions under climate change. Using African amphibians as a case study, the researchers found that more than 90 per cent of the species listed as threatened on the IUCN Red List of Threatened Species are omitted by most popular modelling tools. The researchers examined data on 733 African amphibians in Sub-Saharan Africa. They found that 400 have too few records to be used in species distribution modelling at continental scales, including 92% of those listed as Vulnerable, Endangered or Critically Endangered.

These results show that unless we use appropriate analysis for the impacts of climate change on species such as amphibians, we risk leaving many rare species under-represented in conservation plans, with the potential to misguide conservation efforts on the ground.

Effective biodiversity conservation, both now and in the future, relies on our ability to assess patterns of threat across all species, but particularly those close to extinction. There are ways around the problem, such as combining simple measures of exposure to climate change with knowledge of species' ability to disperse or adapt -- methods less reliant on sophisticated modelling tools, which are impractical for many of the rarest species.