Study area
The study area covers the total land area of Thailand. Where it is useful, we divided Thailand into six regions (Fig. 2a), the names and boundaries of which are widely used, although they have no official administrative status. We focused on the elements of Thailand’s protected area system that were concerned principally with the in-situ conservation of biodiversity: existing and proposed National Parks, Wildlife Sanctuaries, Non-hunting Areas, and Forest Parks, covering 111, 201 km2 or 21.7% of the country’s land area37 (Fig. 1).
Environmental data
A set of environmental variables that were expected to be directly or indirectly related to species distributions in Thailand was used to model suitable habitat in the present and future (Supplementary Material Table S1). These variables were chosen to encompass ecologically relevant variables and enable consistent comparison between species, regardless of species-specific preferences. GIS layers for the whole of the study area were compiled using a variety of data sources at 1-km2 resolution. For variables originally at higher than 1-km resolutions, we used the plus function in ArcMap to combine them with a mask of the study area to use the mask dimensions for all cells.
The physical variables, altitude, slope, aspect, and soil pH are widely used in species distribution modeling. Slope and aspect have biologically significant impacts on both temperature and rainfall at these latitudes8 and are particularly important at the poleward margins of species ranges where species may be confined to one aspect. Slope also affects soil maturity and depth. Soil pH is a consistently measured soil variable that broadly correlates with fertility in tropical soils8. Additional soil variables, particularly soil phosphorus, have been shown to be important filters of plant species distributions in the tropics38, but they are not available for Thailand with a useful accuracy and spatial resolution. Altitudes were downloaded from the CGIAR-Consortium for Spatial Information, CGIAR-CSI version 4.1. Slope and aspect were generated by using surface tools in ArcGIS. Soil pH was extracted from ISRIC-World Soil Information version 2.0.
Unlike the temperate zone, where tolerances of winter cold and requirements for summer warmth dominate plant and animal distributions, our understanding of how tropical climates filter species distributions is still weak38,39. In Thailand, as in most of the tropics, there are two major climatic gradients which correlate with changes in species composition: a rainfall gradient in the lowlands, along which total rainfall declines and the length of the dry season increases, and a gradient of steadily declining temperature with elevation7. There is no simple relationship between elevation, and thus temperature, and rainfall. An additional complication is that temperature seasonality may be significant in northern Thailand (north of c.18° N), where cooler winters reduce dry-season water stress and extreme low temperatures at high altitudes may exceed physiological tolerances. We therefore chose 8 bioclimatic variables (Supplementary Material Table S1) related to precipitation and temperature, and their seasonality, all of which have previously been used in species distribution modelling in this region9,40. These are available at a resolution of 30 arc sec (approximately 1 km at the equator) from WorldClim ver. 1.4 based on averages of 1970–1990. These variables are available from the same source (and downscaled using the same methods) for the future climate projections.
Vegetation structure is an additional major influence on plant and animal distributions in the tropics, both in intact natural vegetation38,39 and when the original vegetation has been degraded or cleared8. Vegetation structure was represented through the inclusion of two continuous variables, percentage forest cover and tree density, as most of the modelled species are known to be sensitive to both the presence of forest and the degree of intactness of the tree cover9. Mean tree density per km2 was extracted from Crowther et al.41 version 2 and percentage coverage of forest per km2 was extracted from the European Space Agency (ESA) GlobCover Version 2.3.
Note that the mechanistic basis of the correlations between all these variables and the current distributions of tropical plants and animals are rarely known. Temperature has a direct physiological impact on all organisms, and water supply may be seasonally limiting for plants and some amphibians, but indirect links through biotic interactions are expected to be more important in the tropics, including pest pressure on plants38 and food supply for animals39. Competition is probably also important in shaping local species assemblies. For future projections, we assumed that temperature and precipitation were changing, and that other variables (topography, soil, and vegetation) were stable, so our analysis represents the impacts of climate alone. For 2070, we used the same variables projected by three CMIP5 Earth System Models, CNRM-CM5, GFDL-CM3 and HadGEM2-ES, which have been previously used in Southeast Asia9,42 and in Thailand7. We used two Representative Concentration Pathways, RCP2.6 and RCP8.5, representing low and high greenhouse-gas concentration scenarios, respectively, and thus the potential range of radiative forcing by the end of the century43. RCP2.6 is consistent with meeting the Paris Agreement’s 2 °C global warming target.
Species occurrence data
Many locality records for vertebrates were supplied by the Department of National Parks, Wildlife and Plant Conservation (DNP). Trained DNP staff walked along trails throughout the protected areas in Thailand during 2017–2018. They recorded 271,695 locations for 70 mammal species, 18 locations for 3 amphibian species, 318 locations for 18 reptile species, and 43,057 locations for 65 bird species44. We supplemented this with data downloaded from the Global Biodiversity Information Facility (GBIF, https://www.gbif.org/) for 1960–2019 for amphibians (2063 localities for 86 species)45, reptiles (1722 localities from 196 species)46, mammals (2508 localities from 191 species)47, and birds (1,559,222 localities from 884 species)48. More than 95% of the bird records from GBIF were identified as coming from eBird49, which is popular among birders in Thailand. For plants, we used occurrence data from the DNP’s forest resource inventory project from 221 plots, including 24,605 localities for 363 species, the DNP’s Forest Herbarium, including 227 localities for 141 species, and locations for 12 rare and endangered forest species collected from all over Thailand. We also downloaded data from the Botanical Information and Ecology Network (BIEN, https://bien.nceas.ucsb.edu/bien/), including 7209 localities for 1422 species.
We removed suspect records (coordinate issues, name problems, etc.), duplicates from the same locality (i.e., more than one individual of the same species recorded in a cell), and species with < 10 localities. High levels of duplication reduced the sizes of the datasets from GBIF and BIEN, so only 687 localities for 29 amphibian species, 1409 localities for 36 reptile species, 2050 localities for 52 mammal species, 282,026 localities for 697 bird species, and 1978 localities for 223 plant species were added to the data from the DNP. The total number of records used for analysis was therefore 705 localities for 31 amphibian species, 1697 localities for 53 reptile species, 243,423 localities for 80 mammal species, 325,008 localities for 702 bird species, and 34,440 localities for 591 plant species.
Species distribution modeling
Species distribution models (SDMs) assume the relationship between a pattern of interest (e.g., species abundance or species occurrence) and a set of factors assumed to control it can be quantified. Here we used Maxent, based on the maximum-entropy approach, to predict suitable habitats for each species, then summed them to look at changing diversity patterns. Maxent parameters, unless otherwise indicated, were set to the defaults, as these are broadly optimized across species. The same background area—the total land area of Thailand—was used for all species. For each species with more than 10 unique occurrence records, three replicates were run and the average of the three models used for further analysis. The Maxent output was then converted into binary presence-absence maps using the 10% cumulative logistic threshold, because it has previously been found to be the most accurate and conservative threshold for delineating suitable from unsuitable areas9. The predictive performance of Maxent was evaluated using the continuous Boyce Index, a presence-only and threshold-independent metric which measures how much the predictions of the SDMs differ from a random distribution of occurrences along the prediction gradient50. We considered values > 0.5 as adequate, but since only five SDMs out of the 1457 generated in this study had values lower than this (0.3–0.5), we retained all the models.
Assessment of climate change impacts
The estimated current distribution for each species from Maxent was used as the baseline for comparison with projected distributions of suitable habitat for these species by 2070, under the two emission scenarios and three ESMs, and with and without unlimited dispersal into newly available habitat. We then assessed the impacts of climate change, both on the spatial distribution of individual species and on the pattern of species richness. To generate a species richness map, the binary habitat suitability maps for all species were stacked to produce a consolidated map, which showed the number of species for each 1 km grid cell, and then classified them into five classes (lowest, low, moderate, high, and highest), using the mean ± standard deviation as a break class40.
Current and future maps were then compared for each species to calculate the change in species richness, and contingency tables were generated containing the numbers of cells (each of 1 km2) in each richness class. Suitable habitat areas were calculated for the current climate and projected for the future climate. For each species we estimated gained habitat as the areas that will become suitable for a species in future under that scenario, lost habitat as the areas currently predicted as suitable now but projected to become unsuitable under future climatic change, and stable habitat as the areas predicted as suitable now which will remain suitable into the future.
We then assessed the vulnerability of each species by estimating the projected change in its range over the next 50 years and using a criteria-based approach, which combined the mean of the suitable habitat area (interpreted as equivalent to extent of occurrence) in the three models and a simplified version of the IUCN Red List criteria51. For 2070, we modified criterion A3(c) as follows; Extinct (Ex) species are projected to lose 100% of suitable habitat by 2070, Critically Endangered (CR) species are projected to lose over 80%, Endangered (EN) species are projected to lose 50–80%; Vulnerable (VU) species are projected to lose 30–50%, Near Threatened (NT) species are projected to lose < 30%, and Least Concern (LC) show no projected loss. For the current status, we used only the modeled extent of suitable habitat: CR < 100 km2 of suitable habitat, EN < 5000 km2, VU < 20,000 km2, NT < 30,000 km2, and LC > 30,000 km2. In practice, other, non-climatic, factors will also influence extinction risk and may interact in unpredictable ways with changes in climate, so these assessments for individual species based only on climate change projections should not be used in isolation in conservation planning.
The 386 existing and 30 proposed PAs were overlaid on the predicted current and projected future species distributions to assess the impact of climate change on their ability to protect the modeled species. First, we assessed the difference between current and future predicted richness. We then assessed the ability of the protected area system as whole to support the conservation of species effectively, by calculating the change in the percentage of suitable habitat and identified which PAs complexes are likely to be most vulnerable.
Source: Ecology - nature.com
