# Global Navigation Satellite System Wildlife Tracking Collar Positioning Accuracy

> How close is a GPS collar point to where the animal really stood? In her master's thesis, Kersey Lawrence (then Kirstin Lawrence-Apfel) tested wildlife collars against survey-grade control in Connecticut forests and the mountains of Patagonia: nearly half the positions were more than 15 m off, and a few were out by up to 2 km. She recommends using GPS for the big picture, then checking on the ground with camera traps and tracks and signs surveys.

Summary by Kersey Lawrence, Original Wisdom. Updated 9 November 2026. Page: https://www.originalwisdom.com/tracking-literature/2009-lawrence-apfel-gnss-wildlife-tracking-collar-accuracy/

## Citation and access

- **Authors:** Kirstin Lawrence-Apfel (now Dr. Kersey Lawrence)
- **Year published:** 2009
- **Journal or source:** Master of Science thesis, University of Connecticut
- **Full citation:** Lawrence-Apfel, K. 2009. Global Navigation Satellite System Wildlife Tracking Collar Positioning Accuracy. Master of Science thesis, University of Connecticut.
- **Access status:** Hosted with permission
- **PDF:** https://www.originalwisdom.com/wp-content/uploads/2026/11/2009_Lawrence-Apfel_global-navigation-satellite-system-wildlife-tracking-collar-positioning-accuracy.pdf

## Tags

- **Region:** South America > Chile; North America > United States
- **Behaviour and ecology:** Behaviour; Habitat selection; Home range; Movement
- **Methods and technology:** Analysis of variance (ANOVA); Descriptive statistics; Global Navigation Satellite System (GNSS); Global Positioning System (GPS); Global Positioning System (GPS) collars; Hemispherical photography; Linear regression; Literature review; Polynomial regression; Position dilution of precision (PDOP); Two-sample t-test
- **Data type:** New field data; Review
- **Observer qualification:** Not needed
- **Discipline:** Behaviour and ecology
- **Year published:** 2009

## Detailed keywords

- **Species:** None studied. The Televilt collars had been used earlier in a study of white-tailed deer (Odocoileus virginianus) (Williams et al. 2008). Trees at the Connecticut test sites included red oak (Quercus rubra), shagbark hickory (Carya ovata), red maple (Acer rubrum), American hornbeam (Carpinus caroliniana), spicebush (Lindera benzoin) and eastern hemlock (Tsuga canadensis). Species named only in the literature review: wolves, elk, grizzly bears, black bears, maned wolves and moose.
- **Location and habitat:** Connecticut, United States: a rooftop control point on the University of Connecticut campus in Storrs (open sky), and a forested 12-acre private property in North Franklin (oak and hickory broadleaf canopy). Torres del Paine National Park, Magallanes, southern Patagonia, Chile: open steppe and grassland (Camp Valley, Laguna Amarga sector); scrubland with groves of deciduous trees in a canyon (Vega Puma, Pehoe sector); mature forest below steep, narrow cliffs near a glacial ice field (Lago Grey sector). Control markers were in the Sarmiento Lake, Laguna Amarga and Pehoe sectors. Field conditions in December and January: high winds averaging 20 m/sec and average humidity of 48%; the thesis gives the average temperature as 49°C, which is a typo (probably 49°F).
- **Topics:** Author note: Kirstin Lawrence-Apfel is a previous name of Dr. Kersey Lawrence, founder of Original Wisdom. GPS collar positioning accuracy; fix-rate bias (scheduled positions not recorded); collar malfunction; canopy and sky obstruction; antenna angle; topography; satellite geometry; two-dimensional (2-D) versus three-dimensional (3-D) positions; data screening; error ellipses; the GPS error budget; multipath; geodetic control; Continuously Operating Reference Stations (CORS); differential GPS; manufacturer accuracy claims; implications for movement modelling, home range and habitat-use estimates; how animals' behaviour and habitat choice (sleeping under cover, digging, posture, use of dense forest and steep terrain) affect whether and how accurately a collar records a position; habitat misclassification; wildlife management decisions; ground-truthing.
- **Statistics or models:** Analysis of variance (ANOVA) in Mathematica. Statistical Analysis System (SAS) Trend Analysis: the Mixed procedure for polynomial terms, the Reg procedure for coefficients and the RSREG procedure for lack-of-fit tests, on natural-log transformed errors. Three predictive models of positioning error: Model 1 from canopy and antenna angle; Model 2 from cubic Position Dilution of Precision (PDOP); Model 3 from linear PDOP. All were significant (p<0.0001), with low R-squared values (0.15 to 0.18). Models were checked in Microsoft Excel. Folded-F tests and Satterthwaite two-sample t-tests between collars, and t-tests for directional bias. Descriptive statistics. Inconsistencies inside the thesis: (1) in Chapter 2, the text says 76 of 122 positions had errors over 100 m, but Table 2.6 shows 102 such positions (22 + 80); (2) in Chapter 3, the results text says the BlueSky and SirTrack collars were biased north to south but not east to west, while the p-values given and the conclusion indicate a bias in longitude (east to west).
- **Data type:** New field data: Connecticut stationary tests in 2007 and 2008; Patagonia tests in December and January of 2005 and 2006. Chapter 1 is a review of satellite positioning for wildlife research. Chapter 3 was later published separately as Lawrence-Apfel et al. (2012), Anales Instituto Patagonia, which has its own entry.

## Synopsis

### 1. What they did

Global Positioning System (GPS) collars let researchers follow animals without being in the field, but the places animals live (under trees, below cliffs, in canyons) are the places where satellite signals struggle. This thesis asked how accurate wildlife tracking collars really are in those conditions. Chapter 1 explains how Global Navigation Satellite Systems (GNSS) work and where their errors come from, as a critique for wildlife researchers.

In Chapter 2, six Televilt Simplex collars were mounted on a pipe at fixed points in Connecticut whose true positions had first been surveyed to within 4 cm with a dual-frequency receiver. The sites ranged from open sky on a rooftop to 40%, 65% and 85% of the sky blocked by forest canopy, measured with hemispherical photos. The collars recorded a position twice an hour while their antennas were turned through eight angles, from pointing straight up to pointing down, for at least 24 hours each, and the pipe was turned north to south and east to west. In 2008 the test was repeated with the one collar that had performed well.

In Chapter 3, collars from three manufacturers (SirTrack, Advanced Telemetry Systems (ATS) and BlueSky) were tested on surveyed control markers in Torres del Paine National Park, Chile. A SirTrack collar was then carried, together with a survey-grade rover receiver, along “simulated animal movement corridors” in three habitats of increasing difficulty: open steppe, a scrubby canyon, and steep forested cliffs near a glacier.

### 2. What they found

Connecticut, 2007

- Only 6,743 of 15,863 scheduled positions were recorded, a 43% success rate. One collar recorded nothing and another under 10% of its schedule.

- Of 6,740 positions (after three extreme outliers linked to a failing battery were removed), 54% were within 15 m of the true point, 29.6% were 15 to 30 m off, 15% were 30 to 100 m off, and a few were 100 m to just over 2 km off. No position was ever closer than 2 m, even in the best conditions.

- Mean error rose with canopy: 10 m under open sky, 19 m under 40% cover and 70 m under 65% cover. Antenna angle mattered less (19 to 26 m on average).

- 72% of positions were 2-dimensional (2-D), with a mean error of 27 m; 3-dimensional (3-D) positions averaged 13 m.

- Removing all 2-D positions would have discarded 72% of the data. Removing positions with a Position Dilution of Precision (PDOP) above 5 discarded only 8% and took out 76 of the largest errors; positions above that threshold averaged about 51 m more error.

Connecticut, 2008

- The best collar recorded 92% of its positions.

- Error and missed positions both increased with canopy, with the antenna turned away from the sky, with 2-D positions and with higher PDOP.

- 58% of positions were 2-D (mean error 29 m), against 15 m for 3-D positions.

- All three models predicted mean error reasonably well, but individual positions varied too much to predict any one of them.

Patagonia

- On open control markers, the three collars differed significantly in mean error, but the differences were small for wildlife work. All three showed directional bias; the ATS collar's positions were the most tightly clustered.

- In the corridors, the SirTrack collar recorded 30 of 45 attempted positions: 21 of 30 on the steppe, 6 of 7 in the canyon and 3 of 8 in the cliff forest.

- Each corridor had one extreme error (395 m, 413 m and 528 m). With these, mean error was 38 m, 80 m and 186 m; without them, 20 m, 14 m and 14 m. The overall mean was 101 m with the outliers and 16 m without.

### 3. What they concluded

- The Televilt claim of 15 m for 90% of 3-D positions is true but misleading, because most positions in wildlife habitat are 2-D and less accurate.

- Dropping 2-D positions loses too much data and could make it look as though an animal does not use an area. The author suggests keeping them and showing each location as an error ellipse sized by its PDOP, rather than as a point.

- PDOP, which collars record anyway, predicted mean error almost as well as measuring canopy and antenna angle, making it a practical screening tool. Positions with a PDOP above 5 should be treated as highly suspect, though not automatically discarded.

- Accuracy depends more on what blocks the sky (trees, rocks, topography, or the animal itself, such as a bear digging with its head in a hole or a cat sleeping on its back) than on antenna angle.

- An animal's behaviour and choice of habitat therefore shape the movement data: species that are reclusive, nocturnal, or use dense cover and steep terrain are more likely to have missed or inaccurate positions, which can distort home range, movement and habitat-use estimates.

- Collars should be tested where they will be used, and before each deployment.

- GPS collars can be used for wildlife research in Torres del Paine, with care over the map scale and the species.

- GPS data are best used for gross movements and locations, followed by ground-truthing with camera traps and tracks and signs surveys to confirm whether a species is present.

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

Tracking was not used. The thesis describes tracking (citing Liebenberg 1990 and Stander 1997) as identifying an animal's tracks and signs and then following its trail, notes that radiotelemetry allows researchers who may be unskilled with tracks and signs to monitor animals, and recommends tracks and signs surveys as one way to ground-truth GPS results.

### 5. Tracker or observer certification

Not needed.

### 6. Other technologies, skills or innovations

- Televilt Simplex, SirTrack, ATS G2100 and BlueSky GPS collars.

- TOPCON Odyssey and Javad dual-frequency survey receivers, post-processed with TOPCON Pinnacle and tied to Continuously Operating Reference Stations (CORS) or International GNSS Service (IGS) stations.

- A stop-and-go base and rover method for real-time survey positions in the field.

- A custom 12-volt battery unit powering six collars at once.

- Hemispherical canopy photos analysed in Gap Light Analyzer (GLA).

- A repeatable method for testing collar accuracy against geodetic control.

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

- Wrong locations can lead to costly or harmful management decisions, for example concluding that a habitat is not needed by a species.

- Errors can be larger than the pixels of the satellite images positions are mapped onto, which can distort habitat and corridor maps, especially for species with small, specialised home ranges.

- Budgets should allow for collar failures, recapture and battery refurbishment.

- GPS may not be the best tool for cryptic species, or for species whose behaviour and habitat block satellite signals.

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

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

- A GPS point shows where to look, not exactly where an animal was. Where canopy, cliffs or the animal's own posture block the sky, the point can be tens of metres or more away, and tracks and signs on the ground can confirm what the collar suggests.

## Cite this summary

Lawrence, K. 2026. Summary of: Global Navigation Satellite System Wildlife Tracking Collar Positioning Accuracy (Lawrence-Apfel, 2009). Original Wisdom, Tracking Literature. https://www.originalwisdom.com/tracking-literature/2009-lawrence-apfel-gnss-wildlife-tracking-collar-accuracy/