Skip to main content
Dryad logo

Hidden treasure of the Gobi: understanding how water limits range use of khulan in the Mongolian Gobi

Citation

Payne, John C. et al. (2020), Hidden treasure of the Gobi: understanding how water limits range use of khulan in the Mongolian Gobi, Dryad, Dataset, https://doi.org/10.5061/dryad.pg4f4qrjw

Abstract

Most large herbivores in arid landscapes need to drink which constrains their movements and makes them vulnerable to disturbance. Asiatic wild ass or khulan (Equus hemionus) were widespread and abundant throughout the arid landscapes of Central Asia and Mongolia, but have undergone dramatic population declines and range constrictions; denying khulan access to water is believed to have played a major role. Mongolia’s South Gobi Region now houses the world largest remaining khulan population, but is undergoing rapid land use changes. Khulan water use is poorly understood, largely due to the difficulty of mapping waterpoints used by khulan throughout their exceptionally large ranges, prone to high variations in precipitation. We used the special movement path characteristics of GPS tagged khulan to show us where water is located. We identified 367 waterpoints, 53 of which were of population importance, characterized the seasonal and circadian use, and identified snow cover as the most important variable predicting khulan visits during the non-growing season, and vegetation greenness during the growing season. Our results provide a data layer to help guide a regional khulan conservation strategy, allow predictions for other part of the global khulan range, and illustrates the overall importance of waterpoints for dryland herbivores.

Methods

Water point characterization

To identify unique waterpoints with similar patterns of use, we applied  ahierarchical cluster analysis (Ward algorithm) to data on waterpoint use (the number of total visits, the number of unique individuals of the visits, and the number of years the waterpoint was visited). We plotted the relationship between the number of clusters and the within-cluster variation to assess the appropriate number of clusters.

The influence of environmental variables on waterpoint use and visitation rates

Environmental variables

To test which environmental variables influence waterpoint visitation rates, we obtained values for precipitation, snow cover, vegetation greenness (as a proxy for plant water content), and ambient temperature from the following sources:

  1. Precipitation from the Global Precipitation Climatology Centre’s GPCC First Guess Daily product75 available on a 1-degree grid (roughly 110km longitude x 82 km latitude, at the latitude of the study area) and available open source at: https://opendata.dwd.de/climate_environment/GPCC/html/gpcc_firstguess_daily_doi_download.html. We had previously found good agreement of this landscape-scale precipitation data with localized rain information from a weather station at the Oyu Tolgoi mine site (JP unpubl. data) and evidence of rain from camera collar images56. For each khulan, we calculated the daily rainfall as the mean of the grid values which were intersected by its hourly GPS positions.
  2. Vegetation greenness based on the Normalized Vegetation Index (NDVI) from the 250m Grid, 16-day composite MODIS/Terra Vegetation Indices product (Version 6, MOD13Q1, open source available from the SGS website at: https://e4ftl01.cr.usgs.gov/MOLT/MOD13Q1.006/. We mosaiced and re-projected the tiles (from Sin to Geographic), extracted the NDVI index and NASA’s pixel reliability layer76. We simplified the pixel reliability layer by converting all categories except “Good data” to missing-data values, then used the resulting layer as a mask to exclude questionable pixels from the NDVI data. We subsequently loaded the layers into PostgreSQL database tables, using a custom R function and NASA’s MODIS Reprojection Tool available at: https://lpdaac.usgs.gov/tools/modis_reprojection_tool. For each khulan we calculated the daily mean NDVI based on 9 pixels around each hourly GPS position.
  3. Snow cover from the MODIS/Terra Snow Cover Daily L3 Global 500m Grid, Version 6 (MOD10A1; available at: http://nsidc.org/data/mod10a1). Satellite images of the Gobi in winter are often obscured by cloud, sometimes over very large areas for days on end. Matching khulan track locations with individual pixels from the daily snow cover images resulted in time series with many missing values. We therefore calculated a 5-day running mean of the proportion of snow cover for the entire khulan range (100% minimum convex polygon of all track positions, Fig. 1) for each day, and the same daily snow cover values were assigned to each animal.
  4. Ambient temperature from the hourly temperatures measured by each khulan collar (as a rough proxy for the temperatures experienced by the animals). For each khulan we calculated the daily mean of the collar temperature and assigned individual daily means to each khulan.

GAM modeling

To test the effect of abiotic variables on water use, we modelled the probability of a waterpoint visit by a khulan on a given day as a binary variable in a generalized additive mixed effects model77. We tested the effects of climate variables snow cover, temperature, rainfall and NDVI as fixed effects in two separate models because snow and rain / NDVI are more or less mutually exclusive. For the growing season (May to September) we tested for the effect of temperature, rainfall and NDVI and for the non-growing season (October – April) for the effect of temperature and snow cover. These variables were standardized to units of standard deviations so that their effect sizes represented their relative importance. We included a random intercept for individual ID and random slopes for each individual for the effect of each abiotic variable. To account for spatial autocorrelation, we also included a Duchon spline term for the interaction of latitude and longitude of the centroid of each individual’s daily movement track (Fig. S2 in SOM). Because the response was a binary variable, we specified the family argument as binomial. These models were fitted with the R package gamm477 (R script provided in the SOM).

Usage Notes

Water point characterization: given the population level importance of these waterpoints we did remove the GPS coordinates

The influence of environmental variables on waterpoint use and visitation rates: the GPS coordinates represent the centroid of each individual’s daily movement track

Funding

Research Council of Norway, Award: 251112