Sampling bias in snow leopard population estimation studies

1. What they did

To judge how threatened a species is, conservationists need to know how many animals there are and where they live. For snow leopards, published global population figures range from 3,920 to 8,745, but they rest mostly on expert opinion, interviews, questionnaires and counts of snow leopard signs, which do not give reliable numbers. The authors point to two more reliable methods: camera traps, where individual snow leopards are recognized by their spot patterns, and genetic analysis of scat (faeces), where individuals are identified from DNA. Both use statistical models that allow for animals being missed.

The authors asked whether the existing studies are good enough to estimate the world’s snow leopard population:

• Review. They collected every peer-reviewed study that estimated snow leopard numbers with camera traps or scat genetics, and recorded each study’s location, years, methods, area sampled, number of snow leopards, density and whether it was inside a protected area. Where a study gave only a number of animals, they worked out a rough density by dividing by the area sampled.
• Model. They plotted density against the size of the area sampled and fitted a curve to see whether smaller study areas gave higher densities.
• Simulations. Using camera trap data from a large study site (1,684 km²) in Tost Nature Reserve, Mongolia, they re-estimated density for smaller and smaller pieces of the site. In one set the pieces all started from the centre, the better snow leopard habitat; in the other they were placed at random.

2. What they found

The published studies

• 17 studies were found. Three were left out (one caught no snow leopards on camera, one was a laboratory study of genetic markers, and one used pugmarks and radio telemetry).
• The 17 studies covered 24 sites in only 6 of the 12 range countries. The remaining 14 studies gave 32 estimates, mostly India (8 estimates), Mongolia (8), China (5) and Kyrgyzstan (4). The Tost Mountains of Mongolia alone were sampled eight times.
• Together they covered 9,596 km², only 0.3 to 0.9% of the snow leopard’s presumed range. Study areas ranged from 60 to 1,684 km².
• Only 4 of the 32 estimates used spatially explicit capture-recapture, the method the authors regard as most robust.

Study area size

• Density fell steeply as study area size grew, levelling off at about 0.79 snow leopards per 100 km² (0.78 in the fitted equation). The curve explained 79.2% of the variation.
• The model suggests study areas need to be at least 481 km² to give a density close to that of a very large area. 55% (17) of the estimates came from smaller areas; the authors note that 55% of study sites were smaller than the home ranges of just two adult male snow leopards.
• Density did not differ between protected areas (mean 2.13 per 100 km²) and unprotected areas (2.30).

Simulations

• When the sampled area was centred on the best habitat, the estimated density dropped as the area grew, from 1.63 per 100 km² in the smallest area (81 km²) to 0.347 in the full area. The number of snow leopards identified rose from 8 to 18.
• When the sampled areas were placed at random, density did not fall with area size (mean 0.9 per 100 km²).

3. What they concluded

• The existing studies cannot give a reliable global estimate of snow leopard numbers: they cover too few countries and mountain ranges, too little of the range, and often too small an area.
• Researchers tend to put their study areas where snow leopards are most common, to get the most camera captures. Combined with small study areas, this may have overestimated densities by up to five times.
• Small study areas are less of a problem if the choice of site is not biased towards the best habitat. Global assessments should not use estimates from small areas centred on good habitat.
• Since densities inside and outside protected areas were similar, snow leopard conservation needs to work across whole landscapes, well beyond protected area boundaries.
• The authors call for: (a) better sampling and analysis protocols for snow leopard population estimation; (b) agreement among researchers and governments across the range to use the same standardized protocols; and (c) sampling of larger and under-represented areas. They welcome the 12 range countries’ agreement on the Population Assessment of the World’s Snow Leopards (PAWS).

6. Other technologies, skills or innovations

• Camera trapping with individual identification from spot patterns. The authors note that the assumption that individual snow leopards can be identified accurately from photographs has not been tested.
• Identification of individuals from DNA in scat samples.
• Spatially explicit capture-recapture analysis in the R statistics package “secr”.
• Simulation by sub-sampling an existing camera trap grid to test how study area size and site choice affect density estimates.

7. Implications for conservation (from the paper)

• Current global snow leopard population figures, and the existing peer-reviewed estimates, are not a sound basis for judging the species’ status or trends.
• Studies that must sample small areas should take special care not to concentrate on the best habitat, especially if their results will be used for larger landscapes or global assessments.
• Coordinated surveys over large landscapes, analysed with spatially explicit capture-recapture and combining camera trap, scat genetics and telemetry data, could improve estimates.
• Snow leopards occur outside protected areas at similar densities, so conservation needs a landscape-level, land-sharing approach.