The Effect of Species Geographical Distribution Estimation Methods on Richness and Phylogenetic Diversity Estimates. Amboni, M. P. M. & Laffan, S. W. 26(11):2097–2109.
The Effect of Species Geographical Distribution Estimation Methods on Richness and Phylogenetic Diversity Estimates [link]Paper  doi  abstract   bibtex   
Diversity assessments are widely used in various fields of knowledge and rely on good estimates of species distribution. There are several methods available to estimate species distribution and the effect of using them is not clearly understood. In this research, we assess the effect of species distributions derived from four geographical distribution estimation methods on derived species richness and phylogenetic diversity (PD). We used the following four most common approaches to determine species geographical distributions: (1) range-wide occurrences are records of presence from databases and museum collections; (2) marginal occurrences are generally expert drawn distributional maps; (3) species distribution modeling (SDM) combines geographic records and environmental data to predict species occurrence; and (4) a combined approach that constrains the statistical modeling predictions by the marginal occurrence distributions. Using these approaches, we estimated richness and PD and their correspondent geographic hotspots for three different analysis resolutions using non-overlapping square windows of 2°, 4°, and 6° across Australia. We also assessed the differences of the resultant geographical distributions for three different spatial resolutions. We found markedly different hotspots when using range-wide occurrences and statistical modeling approaches. Range-wide occurrences resulted in low values of species richness and PD and statistical modeling resulted in high values. The combined and marginal occurrences approaches both had intermediate values, with the combined approach showing a finer level of detail than the marginal occurrences. There is a tendency for species range sizes between methods to converge with decreasing spatial resolution. Even for a relatively well-sampled group such as the Australian marsupials, the range-wide occurrences approach is likely to underestimate the presence of species. Conversely, because SDMs usually do not account for dispersal abilities or biotic interactions, it is likely that species geographical distributions using this approach are overestimated. Depending on the method used, the resultant estimates of diversity may be completely different. Caution is needed when choosing the method to estimate species distribution. [Excerpt: Combined approach] In the fourth approach, we used a combination of both IUCN and museum data sets. For each species, we eliminated all museum records that were more than 100 km outside the boundary of the respective IUCN polygon. We then modeled the species distribution using the same parameters used for the statistical modeling approach. In order to minimize commission error, we also used the buffered IUCN distributions to exclude the predicted presences that were outside the IUCN species distribution. The 100 km buffer distance was selected to be small enough to eliminate false presence in cases where other factors may limit species distribution (e.g., dispersal limits, interaction with other species), but also large enough to account for areas where a species might occur but where it has not yet been recorded (e.g., due to insufficient sampling effort or sampling bias). It is important to acknowledge that this 100 km buffer zone is an assumed value and that species boundaries might change through time [...]. The effect of different distances needs to be explored in further research. [] [...] [Discussion] Our results show that, depending on the method used to estimate species distribution, the outcome can be dramatically different. Of note, the SDM approach results in completely different hotspots from the range-wide occurrences approach. [] Range-wide occurrences had the lowest estimates, and it is likely that species distribution is severely underestimated using this method. [...] [] The statistical modeling approach attempts to compensate for the lack of sampling by predicting the species geographical distributions. However, if an SDM does not account for biotic or evolutionary factors that might limit species distributions then it is likely to predict the fundamental niche instead of the realized niche [...], leading to an overprediction of species richness. Some researchers have shown that incorporating interspecific interactions can provide a more accurate model [...]. Even at macroecological scales, for which it was believed that biotic factors would be diluted [...], it has been shown to improve the model [...]. However, incorporating this parameter into the model requires knowledge that is usually not available. It is therefore important to acknowledge that different statistical modeling inputs [...], different algorithms [...], or different thresholds [...] could possibly produce very different results. This is a topic that needs further study. [] Species geographical distributions estimated from statistical modeling tend to be broader than those from the combined method, because we did not include biotic or evolutionary factors in our analysis. However, the environmental relationships tend to be similar between these two approaches with only slight changes in the relationships for wide-ranged species [...]. [] Similar hotspots were obtained using the marginal occurrences and combined approaches. These two methods show a correspondence in the estimates; however, some caution is needed when using the marginal occurrences approach. Hurlbert and Jetz (2007) argue that marginal occurrences need a minimum resolution of 1° for good quality mapping and lower resolution distribution maps may result in overprediction. The great majority of global extent studies use marginal occurrences [...], usually at a 1° resolution. [...] [] [...]
@article{amboniEffectSpeciesGeographical2012,
  title = {The Effect of Species Geographical Distribution Estimation Methods on Richness and Phylogenetic Diversity Estimates},
  author = {Amboni, Mayra P. M. and Laffan, Shawn W.},
  date = {2012-11},
  journaltitle = {International Journal of Geographical Information Science},
  volume = {26},
  pages = {2097--2109},
  issn = {1362-3087},
  doi = {10.1080/13658816.2012.717627},
  url = {https://doi.org/10.1080/13658816.2012.717627},
  abstract = {Diversity assessments are widely used in various fields of knowledge and rely on good estimates of species distribution. There are several methods available to estimate species distribution and the effect of using them is not clearly understood. In this research, we assess the effect of species distributions derived from four geographical distribution estimation methods on derived species richness and phylogenetic diversity (PD). We used the following four most common approaches to determine species geographical distributions: (1) range-wide occurrences are records of presence from databases and museum collections; (2) marginal occurrences are generally expert drawn distributional maps; (3) species distribution modeling (SDM) combines geographic records and environmental data to predict species occurrence; and (4) a combined approach that constrains the statistical modeling predictions by the marginal occurrence distributions. Using these approaches, we estimated richness and PD and their correspondent geographic hotspots for three different analysis resolutions using non-overlapping square windows of 2°, 4°, and 6° across Australia. We also assessed the differences of the resultant geographical distributions for three different spatial resolutions. We found markedly different hotspots when using range-wide occurrences and statistical modeling approaches. Range-wide occurrences resulted in low values of species richness and PD and statistical modeling resulted in high values. The combined and marginal occurrences approaches both had intermediate values, with the combined approach showing a finer level of detail than the marginal occurrences. There is a tendency for species range sizes between methods to converge with decreasing spatial resolution. Even for a relatively well-sampled group such as the Australian marsupials, the range-wide occurrences approach is likely to underestimate the presence of species. Conversely, because SDMs usually do not account for dispersal abilities or biotic interactions, it is likely that species geographical distributions using this approach are overestimated. Depending on the method used, the resultant estimates of diversity may be completely different. Caution is needed when choosing the method to estimate species distribution.

[Excerpt: Combined approach] In the fourth approach, we used a combination of both IUCN and museum data sets. For each species, we eliminated all museum records that were more than 100 km outside the boundary of the respective IUCN polygon. We then modeled the species distribution using the same parameters used for the statistical modeling approach. In order to minimize commission error, we also used the buffered IUCN distributions to exclude the predicted presences that were outside the IUCN species distribution. The 100 km buffer distance was selected to be small enough to eliminate false presence in cases where other factors may limit species distribution (e.g., dispersal limits, interaction with other species), but also large enough to account for areas where a species might occur but where it has not yet been recorded (e.g., due to insufficient sampling effort or sampling bias). It is important to acknowledge that this 100 km buffer zone is an assumed value and that species boundaries might change through time [...]. The effect of different distances needs to be explored in further research.

[] [...]

[Discussion]

Our results show that, depending on the method used to estimate species distribution, the outcome can be dramatically different. Of note, the SDM approach results in completely different hotspots from the range-wide occurrences approach.

[] Range-wide occurrences had the lowest estimates, and it is likely that species distribution is severely underestimated using this method. [...]

[] The statistical modeling approach attempts to compensate for the lack of sampling by predicting the species geographical distributions. However, if an SDM does not account for biotic or evolutionary factors that might limit species distributions then it is likely to predict the fundamental niche instead of the realized niche [...], leading to an overprediction of species richness. Some researchers have shown that incorporating interspecific interactions can provide a more accurate model [...]. Even at macroecological scales, for which it was believed that biotic factors would be diluted [...], it has been shown to improve the model [...]. However, incorporating this parameter into the model requires knowledge that is usually not available. It is therefore important to acknowledge that different statistical modeling inputs [...], different algorithms [...], or different thresholds [...] could possibly produce very different results. This is a topic that needs further study.

[] Species geographical distributions estimated from statistical modeling tend to be broader than those from the combined method, because we did not include biotic or evolutionary factors in our analysis. However, the environmental relationships tend to be similar between these two approaches with only slight changes in the relationships for wide-ranged species [...].

[] Similar hotspots were obtained using the marginal occurrences and combined approaches. These two methods show a correspondence in the estimates; however, some caution is needed when using the marginal occurrences approach. Hurlbert and Jetz (2007) argue that marginal occurrences need a minimum resolution of 1° for good quality mapping and lower resolution distribution maps may result in overprediction. The great majority of global extent studies use marginal occurrences [...], usually at a 1° resolution. [...]

[] [...]},
  keywords = {*imported-from-citeulike-INRMM,~INRMM-MiD:c-14076787,~to-add-doi-URL,australia,comparison,csmfa,diversity,habitat-suitability,integration-techniques,model-comparison,semantic-constraints,species-distribution,species-richness},
  number = {11}
}

Downloads: 0