# Tigers on trails: occupancy modeling for cluster sampling

> Tigers walk forest trails and leave a string of tracks and scats along them, so signs found on one stretch of trail are not independent of the next. Using a large tiger survey in southwestern India, the authors show that standard occupancy models can underestimate a species’ range by about 30% on trail surveys, and offer a new model that turns this trail behaviour into an advantage.

Summary by Kersey Lawrence, Original Wisdom. Updated 9 November 2026. Page: https://www.originalwisdom.com/tracking-literature/2010-hines-nichols-tigers-trails-occupancy-modeling/

## Citation and access

- **Authors:** J. E. Hines, J. D. Nichols, J. A. Royle, D. I. MacKenzie, A. M. Gopalaswamy, N. Samba Kumar and K. U. Karanth
- **Year published:** 2010
- **Journal or source:** Ecological Applications
- **Full citation:** Hines, J. E., Nichols, J. D., Royle, J. A., MacKenzie, D. I., Gopalaswamy, A. M., Samba Kumar, N. and Karanth, K. U. 2010. Tigers on trails: occupancy modeling for cluster sampling. Ecological Applications 20(5): 1456-1466.
- **DOI:** 10.1890/09-0321.1
- **Access status:** Hosted (freely available elsewhere)
- **PDF:** https://www.originalwisdom.com/wp-content/uploads/2026/11/2010_Hines_and_Nichols_etal_tigers-on-trails-occupancy-modeling-for-cluster-sampling.pdf

## Tags

- **Species:** Mammals > Tiger
- **Region:** Asia > India
- **Tracking:** Scat; Sign surveys; Tracking; Tracks; Tracks and signs identification
- **Behaviour and ecology:** Behaviour; Movement
- **Methods and technology:** Akaike Information Criterion (AIC); Detectability; Markov chain Monte Carlo (MCMC); Occupancy models; Simulation
- **Data type:** Modelling or theoretical; New field data
- **Observer qualification:** Reported but not validated
- **Discipline:** Behaviour and ecology; Wildlife tracking
- **Year published:** 2010

## Detailed keywords

- **Species:** Studied: tiger (Panthera tigris). Other examples discussed: carnivores and elephants surveyed with the same design elsewhere, wolf packs, birds of the North American Breeding Bird Survey, breeding frogs and toads (anurans) and butterflies.
- **Location and habitat:** Malnad-Mysore Tiger Landscape (spelled “Malenad-Mysore” in the Discussion), about 22,000 km², Karnataka State, southwestern India. Forest roads and trails in tropical forest; cells with less than 10% forest cover were not surveyed.
- **Topics:** occupancy estimation; detection probability; imperfect detection; spatial replication; trail-based sign surveys; cluster sampling; spatial dependence between neighbouring trail segments; Markov process for segment occupancy; trap response model; sampling design and segment length; animal use of trails as travel routes; availability versus detection given availability; temporary emigration; hidden Markov chains; habitat and occupancy.
- **Statistics or models:** standard occupancy models [psi(.), p(.)] and [psi(.), p(t)]; a new spatial Markov process occupancy model [psi(.), theta(.), theta’(.), p(.)]; a new trap response occupancy model [psi(.), p(.), p’(.)]; occupancy as a linear-logistic function of habitat; maximum likelihood in program PRESENCE; a Markov chain Monte Carlo (MCMC) version with WinBUGS code (supplement); simulation study (1000 data sets of 200 sample units with 10 segments each); bias and relative bias; Akaike’s information criterion (AIC) and AIC weights. Inconsistencies in the paper: the expected maximum tiger home range is given as about 200 km² in one section and 150 km² in another; the grid cells are 188 km² in the methods and Table 3 but called “200-km² cell” in the results.
- **Data type:** Modelling or theoretical: new models and a computer simulation study. New field data: initial data from a tiger sign survey across 205 grid cells, February 2006 to June 2007, by co-authors (the full survey is cited as an unpublished manuscript by Karanth, Kumar, Gopalaswamy and Srinivas).

## Synopsis

### 1. What they did

Occupancy surveys ask what share of an area a species uses. Because animals and their signs are easy to miss, surveyors need repeat checks of each area to estimate how often they miss them. One way is to split each area into several smaller places, ideally chosen at random. In practice, many large-carnivore surveys instead walk trails and forest roads, because animals travel and mark along them, and divide the walk into segments. These segments are not random and lie next to each other: a tiger walking a trail may leave signs on several segments in a row.

The authors built two new occupancy models for this kind of trail survey:

- A spatial Markov model. It splits detection into two steps: whether a tiger used a segment, and whether surveyors found the signs if they did. The chance of use depends on whether the previous segment was used.

- A trap response model. A simpler approximation in which the chance of finding signs on a segment depends on whether signs were found on the previous one.

They tested both against standard occupancy models with computer simulations (1000 data sets of 200 survey areas, each with a 10-segment trail). They then fitted all the models to data from a large tiger survey in Karnataka State, southwestern India (February 2006 to June 2007). There, three trained surveyors walked trails through 205 grid cells of 188 km² each, up to 40 km per cell depending on how much tiger habitat it held, and recorded tiger tracks and scats on 1-km segments (4 to 42 segments per cell).

### 2. What they found

Simulations

- When neighbouring segments were not linked, the standard models worked well.

- When they were linked, as on a trail, the standard models underestimated occupancy by about 30%.

- The trap response model did better but still underestimated occupancy.

- The spatial Markov model gave almost unbiased estimates (relative bias under 0.02), although it failed to converge in about 15% of simulations (and in nearly a quarter when segments were not linked).

Tiger survey

- Tiger signs were found in 36% of cells.

- The spatial Markov models were strongly favoured. The best model (weight about 0.87) also included the amount of habitat in each cell: cells with more habitat were more likely to be used by tigers.

- That model estimated that about half of the cells (0.50) were used by tigers; without the habitat effect the estimate was 0.57. The standard models gave the lowest estimates (0.38 and 0.41).

- A segment next to a segment without tigers had only a 7% chance of being used, but a segment following a used segment had a 79% chance. Where tigers had used a segment, surveyors found their signs 42% of the time.

### 3. What they concluded

- Trail-based sign surveys produce linked detections on neighbouring segments, and standard occupancy models then underestimate how widely a species occurs. The new spatial Markov model handles this and was appropriate for the tiger data.

- The design has great potential: it models the process that leaves signs along trails (animal behaviour) and suits the practical needs of fieldwork in tropical forests, where teams move easily along trails and signs are easy to see on them.

- What first looked like a nuisance is an advantage: linked segments let researchers separate whether animals used a place from whether their signs were found.

- Segments should not be so short that most hold no detections, nor so long that most areas have fewer than three segments. Beyond that, segment length should be worked out case by case. The model can also test whether combining 1-km segments into longer ones makes them independent.

- The models could apply to other road and trail surveys (track surveys, scent stations), roadside bird counts, and repeat surveys over time, such as a tiger or wolf pack moving through a large range or seasonal pulses of frogs and butterflies.

### 4. How tracking was or was not used

The tiger data came from sign surveys: teams walked forest roads and trails looking for tiger tracks and scats, and the signs found were verified and recorded for each 1-km segment. The paper notes that tigers use forest roads and trails as travel routes and mark them intensively with tracks, scent and scats, so searching trails gives far higher detection than random searching. No trailing of individual animals is reported. The new model treats a tiger walking along a trail for several segments as the reason detections cluster.

### 5. Tracker or observer certification

Reported but not validated. The trails were walked by “three trained surveyors”, but their training, experience and the way signs were verified are not described in this paper (it refers to the unpublished survey manuscript).

### 6. Other technologies, skills or innovations

- The new models were built into program PRESENCE (Hines 2006, United States Geological Survey), which can handle missing observations.

- A Markov chain Monte Carlo (MCMC) version with WinBUGS code is given as a supplement.

- Because the survey started from a random point in the middle of a trail, the authors adjusted the model for the first segment.

### 7. Implications for conservation (from the paper)

- The survey shows that occupancy surveys can map the current range of secretive animals over large landscapes, and that non-detection must be accounted for: the simple share of cells with signs (0.36) fell well short of the model estimate (0.50).

- Cells with more habitat were more likely to hold tigers, giving stronger evidence than expert statements about which areas matter most and supporting protection of the suitable habitat that remains.

### 8. Implications for tracking (Original Wisdom's observations)

_These are Original Wisdom's observations, not the authors'._

- The model puts numbers on something trackers know: an animal using a trail leaves a run of signs, not scattered single signs. Here, a segment after one used by a tiger had a 79% chance of use, against 7% otherwise.

- Even where tigers had used a segment, their signs were found only 42% of the time, which shows how much depends on finding and recognising signs, and why surveyor skill matters for such surveys.

## Cite this summary

Lawrence, K. 2026. Summary of: Tigers on trails: occupancy modeling for cluster sampling (Hines and Nichols et al., 2010). Original Wisdom, Tracking Literature. https://www.originalwisdom.com/tracking-literature/2010-hines-nichols-tigers-trails-occupancy-modeling/