Track surveys do not provide accurate or precise lion density estimates in Serengeti

1. What they did

Track surveys are a cheap and repeatable way to estimate how many large carnivores live in an area, and they are widely used for African lions. But whether they give reliable numbers has long been questioned. In 2016, Winterbach and colleagues reanalysed earlier track data (Funston et al. 2010) and produced new, simpler regression models that convert the number of tracks found along a road into an estimate of carnivore density. They asked for independent data to test these models.

This study is the first independent test of one of those models, “lion and cheetah on clay through origin”, chosen because the study area has deep clay soils and the model was built partly on data from the same part of the Serengeti.

• Track surveys. The team set up 10 transects on dirt roads (about 25 km each, 253 km in total) in southeastern Serengeti National Park, Tanzania. Each transect was surveyed once a week for 7 weeks in 2015 (September to November) and 6 weeks in 2016 (September to October).
• Clearing the roads. The evening before each survey (usually between 17:00 and 18:30), they dragged a tyre behind a vehicle to wipe out old tracks, so every track counted was less than a day old.
• Reading the tracks. Each of two crews had a driver and an experienced tracker sitting on the bonnet. They drove at up to 10 km per hour, between about 07:00 and 12:00, before strong sunlight made tracks hard to see. For each set of lion tracks they worked out how many lions there were from track size, how the tracks lay in relation to each other, and direction of travel. They measured and photographed a representative track of each lion. Following Funston et al. (2010), similar tracks within 500 m of each other that could not be told apart were counted only once.
• Comparing with a reference. They converted track counts to lion densities with the model, for each week and each year, and compared the 2015 results with an independent estimate from a call-in survey in the same season (lions attracted by broadcast calls, counted repeatedly and analysed with N-mixture models, Belant et al. 2016).

2. What they found

Survey effort and tracks

• They drove 3,289 km of transects (1,771 km in 2015 and 1,518 km in 2016), 253 km in each of 13 weekly sessions.
• They found 22.22 lion tracks per 100 km in 2015 and 18.68 per 100 km in 2016.
• Weekly track counts ranged from 44 to 80 in 2015 and from 35 to 54 in 2016, always more than the 30 tracks thought to be enough for a precise estimate.

Density from tracks

• 2015: 41.15 lions per 100 km² (95% confidence limits 31.92 to 57.86), or about 773 lions in the study area.
• 2016: 34.59 lions per 100 km² (26.83 to 46.04), or about 650 lions, a 16% drop.
• Weekly estimates swung widely: from 31.86 to 57.55 lions per 100 km² in 2015, and from 27.22 to 40.40 in 2016. The authors report this as variation of up to 55.6% within 2015 and 39.0% within 2016, though the weekly confidence limits overlapped. This week-to-week variation was bigger than the change between years.

Compared with the call-in survey

• The call-in survey in 2015 gave 14.36 lions per 100 km² (9.04 to 29.31), or about 270 lions.
• Every annual and weekly track estimate was higher than this, and the 2015 track estimate’s range did not overlap the call-in range: the track surveys gave close to three times as many lions.
• The authors are confident in the call-in figure: it fits earlier research in the area and their own observations of prides. Combining the Funston et al. (2010) dry-season reference densities for short and long grass also gives a similar figure (13.87 lions per 100 km²).

Possible reasons

• Different methods. Funston et al. drove at 10 to 20 km per hour; in trials, this team felt less able to detect tracks reliably above 10 km per hour. Funston et al. counted only tracks judged to be under 24 hours old instead of clearing the roads; this team could not confidently tell a 23-hour-old track from a 25-hour-old one, and suggest estimated track age could lead to over- or undercounting, varying between observers. Teams could also differ in how they decided whether tracks within 500 m belonged to the same lion.
• Too little data behind the model. The Serengeti part of the model rests on only 4 data points each for lions and cheetahs, from surveys of 66.8 to 375.3 km. If about 30 tracks are needed for a reliable estimate, only 1 of the 8 Serengeti estimates in Funston et al. (2010) had enough.
• Errors in the original reference counts. The Funston et al. (2010) Serengeti densities were single values with no measure of precision and only a season given, but the authors think they were probably reasonable.

3. What they concluded

• Track surveys, as currently done, gave inaccurate and imprecise lion densities in the Serengeti. The likely causes are differences in method between studies, the subjectivity of assigning tracks to individuals and estimating track age, and too little data behind the original and revised regression models.
• For track-based estimates to work, track deposition would have to be steady across space and time, meaning lions would have to move consistently, or tracks would have to be laid down at random and sampled enough to cover the variation. The authors suggest none of this holds: how lions use roads changes with moonlight, landscape features that help them catch prey, and prey numbers. Their study area lies on the route of the great annual ungulate migration, which shifts where lions move.
• The large changes from week to week and year to year were not real changes in lion numbers.
• They recommend caution in using track surveys to estimate lion density, and favour designs that include direct observation of lions (transect or road surveys, call-in surveys), account for imperfect detection in a spatially explicit framework, and include factors such as prey and land cover that shape lion movements.

4. How tracking was or was not used

Tracking was the method being tested. Experienced trackers riding on the vehicle bonnet looked for lion tracks on dirt roads that had been cleared with a tyre drag the evening before. They identified tracks as lion, estimated the number of lions from track size, the arrangement of the tracks and direction of travel, measured the length and width of one track per lion and photographed it. They told the occasional leopard track apart from lion tracks by track size, pad shape, group size (leopards usually travel alone) and location (leopards mostly stay in wooded areas along rivers). Any track that could not reliably be identified as lion was discarded. No trailing was reported.

5. Tracker or observer certification

Reported but not validated. Each crew included “an experienced tracker”, but the trackers are not named, no certification is given and their skill was not tested. Field assistants are thanked by name in the acknowledgements, without their roles being given. The authors themselves raise observer differences: they suggest that estimating track age could vary within or between observers, and that teams may differ in how they assign nearby tracks to the same or different lions.

6. Other technologies, skills or innovations

• A tyre drag pulled behind a vehicle to clear the road surface the evening before each survey, so that all tracks were laid down in a known period of under a day, instead of estimating track age.
• A deliberately slow driving speed (up to 10 km per hour), chosen after trial surveys.
• Call-in surveys with broadcast lion vocalisations and N-mixture models, used as the independent reference.

7. Implications for conservation (from the paper)

• Lion numbers have reportedly fallen by 43% from 1993 to 2014, and lions now live in only 8% of their historic range, so accurate counts matter.
• Track surveys were one of the methods used to produce lion population estimates behind large-scale conservation recommendations, global conservation assessments and national and international policy (such as the 2015 United States listing of two lion subspecies). The authors suggest at least some of these estimates may be wrong, and encourage caution in using them, to keep conservation decisions scientifically rigorous.