Mapping National Plant Biodiversity Patterns in South Korea with the MARS Species Distribution Model

Abstract

Accurate information on the distribution of existing species is crucial to assess regional biodiversity. However, data inventories are insufficient in many areas. We examine the ability of Multivariate Adaptive Regression Splines (MARS) multi-response species distribution model to overcome species’ data limitations and portray plant species distribution patterns for 199 South Korean plant species. The study models species with two or more observations, examines their contribution to national patterns of species richness, provides a sensitivity analysis of different range threshold cutoff approaches for modeling species’ ranges, and presents considerations for species modeling at fine spatial resolution. We ran MARS models for each species and tested four threshold methods to transform occurrence probabilities into presence or absence range maps. Modeled occurrence probabilities were extracted at each species’ presence points, and the mean, median, and one standard deviation (SD) calculated to define data-driven thresholds. A maximum sum of sensitivity and specificity threshold was also calculated, and the range maps from the four cutoffs were tested using independent plant survey data. The single SD values were the best threshold tested for minimizing omission errors and limiting species ranges to areas where the associated occurrence data were correctly classed. Eight individual species range maps for rare plant species were identified that are potentially affected by resampling predictor variables to fine spatial scales. We portray spatial patterns of high species richness by assessing the combined range maps from three classes of species: all species, endangered and endemic species, and range-size rarity of all species, which could be used in conservation planning for South Korea. The MARS model is promising for addressing the common problem of few species occurrence records. However, projected species ranges are highly dependent on the threshold and scale criteria, which should be assessed on a per-project basis.

An assessment of human impacts on endangered red pandas (Ailurus fulgens) living in the Himalaya

“…The maximum sum of the sensitivity and specificity (MaxSSS) threshold is appropriate to convert the continuous probability map to a binary map when only presence data are available (Liu, Newell, & White, ; Liu, White, & Newell, ). This is a widely used threshold that has been used in similar studies (Bista et al, ; Choe, Thorne, & Seo, ; KC et al, ). In this study, we used the MaxSSS threshold to generate the final suitable habitat maps.…”

Integrating the Rabinowitz rarity framework with a National Plant Inventory in South Korea

“…However, for species richness in the rarity categories, we found that Gangwon province (northeastern South Korea) contains the highest species richness, especially for species with narrow geographic ranges. The primary mountain ranges of South Korea, including the highly biodiverse Baekdudaegan Mountains (Choe et al, ; Ministry of Environment, ), are mostly located in Gangwon province. Our study also found high species richness of rarity classes and of nationally listed endangered and endemic species in each rarity class in these mountains (Figure and Figure ), which indicates our simple approach can identify high biodiversity areas for rare species.…”

“…Uncertainties in SDM results arise from the quantity and quality of observation data, from predictor variables used, and from errors in the modeling (Elith & Leathwick, 2009). Model results are influenced significantly by the sample size (Hernandez, Graham, Master, & Albert, 2006), and in many cases, the number of records available to fit a reliable model is limited, particularly for many rare species and regions (Choe, Thorne, & Seo, 2016;Wisz et al, 2008). Furthermore, the testing of the model performance is not sufficient in many SDM studies to convince ecologists (Vaughan & Ormerod, 2005).…”

Integrating Climate Change and Land Use Impacts to Explore Forest Conservation Policy

“…In this study, we used the SDM results modeled using only climate parameters, but, in reality, the distributions of species are confined by interactions with other species [56]. We did not consider the capacity of species to reach the future climate suitable regions, and stacking species' ranges may tend to overestimate species richness [58,59]. Therefore, future species richness may be lower than projected, so uncertainties caused by using SDMs for predicting the impacts of climate change should be understood by policy makers.…”