Establishment of Spatial-time GAMs
We obtained data regarding the historical distribution of Chinese pangolins from the compendium of The Distributions and Changes of Rare Wild Animals in China, in which the occurrence times and locations of the species were recorded from standard histories and local gazetteers, as well as physical remains discovered in 1700-2000 (Wen 2009). The approach to the inclusion of historical literature records in the compendium was conservative; records lacking other supporting observations were excluded, and only confirmed records were included (Wen 2009). We extracted information on Chinese pangolin occurrences (year and location) from the compendium and then reconstructed the longitude and latitude of those locations (Liang 1980).
To track the historical extinction events of the Chinese pangolin, we divided China into 4345 square grids (50 km2 ×50 km2) and the full study period into ten 30-year periods (analyses based on 50-year periods were also conducted, but no significant results were obtained). For each grid, only presence and absence for each sampling year of were populated; specifically, if the Chinese pangolin was present in one 30-year period in a coded grid, we recorded this as event as 0 (presence) and if the occurrence of the species was not detected again in this grid, the next 30-year period would be recorded as 1 (absence). The above process represents a complete extinction event. After data verification, we detected 314 local extinction events of Chinese pangolins across China in 1700-2000.
Five anthropogenic variables, two climatic variables and their coordinates were used to establish GAMs. The anthropogenic factors used in the analysis were as follows: population density, defined as the number of persons per square kilometer per grid cell in a period from HYDE (version 3.2.1); cropland, defined as the proportion of cropland coverage in each grid; population counts, defined as the number of inhabitants in each grid; grazing, defined as the land use for grazing in each grid; and population, defined as the total population in China (download fromhttps://dataportaal.pbl.nl/downloads/HYDE/HYDE3.2/)(Goldewijk et al. 2017). Temperature was represented by oxygen isotopes (δ18O) of ice cores in the Tibetan Plateau was used as a proxy for holistic temperature fluctuation in China (download fromhttp://www.tpdc.ac.cn/zh-hans/) (Zhao et al., 2014). Regional temperature (5°×5° resolution) was represented by the average temperature during the Asian summer season (June to August) based on 357 publicly available proxy climate datasets (mainly tree ring sequence) from the World Data Center for Paleoclimatology archives. These data were used as the thermal proxy for regional temperature (download fromhttps://www.ncdc.noaa.gov/paleo-search/study/18635 ) (Shi et al., 2015). From 1700 AD to 2000 AD, the HYDE and climate data had a 10-year resolution. We used the average to represent the entire 30-year period.
The GAM algorithm was used to model the associations of population density, cropland coverage, population count, grazing, population, temperature and regional temperature with local extinctions of the Chinese pangolin. Based on the results of the Pearson correlation test, we found that temperature and population were highly correlated (r>0.7), and those two variables both exhibited a high correlation with local extinctions. However, significant effects of population density, cropland, grazing and regional temperature on local extinctions of Chinese pangolin were not detected with this method. Therefore, to avoid information loss, we established two categories of GAMs. Temperature and population were taken as the main variables of the regression models.