Understanding the distribution of biodiversity and its determinants, particularly that of ecologically sensitive ones, has long been intriguing to the science community and will help formulate conservation strategies under future climate changes. To this end, we conducted extensive field surveys on the distribution of orchid flora in the Beipan River Basin in Guizhou Province, which is one of the biodiversity conservation priorities in China. The data we acquired, together with those published previously, were converted into orchid species richness for each of the 3km × 3km grid cells covering the study region. Redundancy analysis (RDA) and Geographically Weighted Regression (GWR) were then applied to determine which of the 30 environmental factors are potentially critical for the spatial distribution of orchid flora we have observed. Despite a moderate spatial extent, we found that the Beipan River Basin harbors about 249 native orchid species belonging to 74 genera, equivalent to 14.5% of orchid flora of China. Orchid species richness in this area follows a descending gradient from the southeast to the northwest, 70.41% of its variation among grid cells can be explained by environmental factors and spatial variables, and spatial variables accounted for 63.90% of the spatial variation of orchid distribution, indicating that spatial variables played a dominant role in the distribution of wild orchidaceae species richness. In addition, the main environmental driver is the mean temperature of the wettest quarter. Our study provides a good example for revealing the main drivers of orchid distribution characteristics, and has a certain reference value for the development of orchid conservation strategies.

We first downloaded a total of 30 environmental factors categorized into four types: energy, water, habitat heterogeneity, and human activities, covering most of those known to have large impacts on the distribution of terrestrial plant diversity.

Energy variables: Altogether 13 energy variables were selected, including mean annual temperature (MAT), mean diurnal range (MDR), isothermality (ISO), temperature seasonality (TS), maximal temperature of the warmest month (MTWM), minimal temperature of the coldest month (MTCM), temperature annual range (TAR), mean temperature of the wettest quarter (MTWetQ), mean temperature of the driest quarter (MTDQ), mean temperature of the coldest quarter (MTCQ), mean temperature of the warmest quarter (MTWarmQ), and potential evapo-transpiration (PET). All these twelve variables, except PET (from International Agricultural Database, https://cgiarcsi. community), were downloaded from the Worldclim Database (http://www.worldclim.org) and with a resolution of 30 arc seconds. Additionally, water deficit (WD) was computed as WD=PET－MAP.

Water variables: We selected 10 water variables, including mean annual precipitation (MAP), precipitation of the wettest month (PWM), precipitation of the driest month (PDM), precipitation seasonality (PS), precipitation of the wettest quarter (PWetQ), precipitation of the driest quarter (PDQ), precipitation of the warmest quarter (PWarmQ), precipitation of the coldest quarter (PCQ), and actual evapo-transpiration (AET). All these nine variables, except AET , were downloaded from the Wordclim Database (http://www. worldclim.org) and with the resolution of 30 arc seconds . Additionally, moisture index (MI) was computed as MI＝(MAP/PET－1)×100.

Habitat heterogeneity variables: We selected 4 variables that quantify habitat heterogeneity, including elevational range (ER), the number of vegetation formations (NVF), ranges of MAT and MAP (RMAT and RMAP, respectively) for each grid. ER was extracted from a 12.5 m digital elevation model (DEM) (https://search. asf.alaska.edu/), and NVF with the resolution of 1km2 from the Center for Resources and Environmental Science and Data, Chinese Academy of Sciences (http://resdc.cn/).

Human activities variables: We selected 3 variables as proxies of human activity intensity, including human population density (HPD) in 2017 with the resolution of 30 arc seconds downloaded from the Global Demographic Dynamics Statistical Analysis Database (https:// landscan.ornl.gov), gross domestic product (GDP) in 2015 with the resolution of 1km2 from the Center for Resources and Environmental Science and Data, Chinese Academy of Sciences (http://resdc.cn/), and the area of cropland (AOC) in 2015 with the resolution of 1km2 from the National Qinghai-Tibet Plateau Scientific Data Center (http://data.tpdc.ac.cn/zh-hans/).