Estimating Lion Abundance using N-mixture Models for Social Species

1. What they did

Lion numbers are falling across Africa, and managing them well depends on accurate counts. Lions are often counted directly, by call-in surveys, or by track counts, but few studies repeat their surveys in a way that allows for lions (or their tracks) that are present but missed. The authors tested N-mixture models, statistical models that estimate abundance from repeated counts at many sites, on lions in southeastern Serengeti National Park, Tanzania.

Over 7 weeks from mid-September 2015 they ran two surveys at the same time:

• Call-in survey. At 39 sites about 6 km apart, two crews broadcast a recording of a female lion roar, a wildebeest in distress and a spotted hyena whoop from four roof-mounted speakers, for 70 minutes starting at 7 pm. Each site was visited once a week. Lions that came in were counted through the roof hatch with a red-filtered spotlight and a thermal (forward-looking infrared) monocular. Because fewer lions came as the weeks went by, buffalo distress calls replaced wildebeest calls at some sites in weeks 6 and 7.
• Track survey. Ten road routes of about 25 km each (253 km in total) were swept with a tire drag each evening and checked the next morning from about 7 am, with an experienced tracker sitting on the hood of the vehicle. Each route was surveyed once a week.

Because lions live in groups, a whole pride may answer a call or not. The authors built this into the model: first, does a group respond in a given week; then, if it does, is each lion seen. They also included moonlight, land cover, rivers, roads, kopjes and which crew did the survey.

2. What they found

Call-in survey

• The model estimated 270 lions in the sampled area (median 242; 95% credible interval 170 to 551), or 14.4 lions per 100 km². Sites held between 2.0 and 17.9 lions.
• The chance of lions responding to a call fell sharply, from 0.93 in week 1 to 0.11 in week 5, then rose to 0.50 in week 7. The authors read this as rapid habituation to the calls.
• Once a group responded, the chance of seeing each lion was steadier (0.74 to 0.92 across weeks).
• Brighter moonlight lowered the chance of detecting individual lions. Land cover stayed in the model, but its effect on abundance could not be determined.

Track survey

• The crews recorded 456 lion track occurrences: 0 to 129 per route and 52 to 91 per week. Predicted density was 5.0 to 12.9 tracks per km.
• Land cover stayed in the track model too, with no clear effect.
• Predicted track density had a weak negative relationship with predicted lion numbers from the call-ins (R² = 0.113), so the authors could not estimate lion abundance from tracks.

Comparison with earlier counts

Their estimate was close to earlier distance-sampling estimates for larger areas that included theirs (314 lions in 2002; 247 in 2005) and a little higher than a camera-trap reference count of known females, which suggested to them that it is reasonable.

3. What they concluded

• This was the first estimate of lion abundance with N-mixture models, and the first model built to allow for the way social animals respond and are detected as groups.
• The estimate may be slightly high if the same lions came to neighbouring call-in sites, though earlier work suggests lions rarely come from more than 3 km away.
• Lions may move less on bright nights, when prey can see them more easily, which could make them less likely to approach calls.
• Changing the call sequence or site order between weeks, or leaving longer gaps between surveys, could reduce habituation and should be tested.
• Tracks did not track lion numbers well. Possible reasons are the small grid cells, missing covariates, and, more practically, not being able to tell individual lions apart from their tracks. Like an earlier study by Midlane and colleagues in Kafue National Park, they found call-ins more suitable than track surveys for estimating lion abundance.
• Call-ins work in both savanna and forest, can be done quickly, and with N-mixture models allow for observer differences and changing conditions such as moonlight.
• They suggest N-mixture models with a group-level observation process for other social, herding and grouping species, such as ungulates, birds and fish.

4. How tracking was or was not used

Track surveys ran alongside the call-ins. Roads were dragged clean the evening before, and each morning an experienced tracker on the vehicle hood looked for lion tracks while the driver drove at up to 10 km per hour, finishing by about noon to avoid harsh sunlight on the tracks. For each set of lion tracks the team:

• estimated the number of lions from track size, how the tracks lay in relation to each other, and direction of travel;
• measured the length and width of one representative track per lion and photographed it;
• counted tracks further along the route as new lions unless the measurements and photos showed they were the same animals.

The occasional leopard track was told apart from lion by size, pad shape, group size (leopards usually travel alone) and location (leopards mostly keep to wooded riverbanks). Any track that could not reliably be identified as lion was discarded. No trailing was reported.

The authors note that they did not measure all four paws, as recommended for identifying individuals, because of uneven road surfaces and varying substrate, and that overlapping measurements made telling lions apart hard. Lion numbers from tracks may therefore have been both over- and underestimated.

5. Tracker or observer certification

Reported but not validated. Each track crew had “an experienced tracker”, but no name, certification or test of skill is given. The model included a random observer effect for both surveys, and crews swapped routes and sites each week to allow for differences between them. Six people are thanked for field assistance.

6. Other technologies, skills or innovations

• A two-level (hierarchical) detection model for social species: group response, then individual detection.
• A commercial game-calling system (Foxpro) broadcasting at up to 116 decibels from four speakers, alternating between speaker pairs to spare the vehicle battery.
• A red-filtered spotlight (lions showed some aversion to unfiltered light) and a forward-looking infrared (FLIR) monocular for counting lions at night.
• Moonlight values for each night calculated with the R package lunar; landscape layers from the Serengeti-Mara database in a Geographic Information System (GIS).
• Tire drags to clear road routes before each track survey.

7. Implications for conservation (from the paper)

• Accurate estimates of large carnivore numbers are needed for harvest rules, conservation status assessments and understanding change.
• Allowing for animals that are present but missed, and for observer and moonlight effects, can improve abundance estimates.
• Reducing habituation to broadcast calls would improve the precision of call-in estimates.
• Better landscape measures tied to lion resource use, and monitoring of prey, could improve the models.