Horse population estimate supplementary information

Release of raw data underpinning the 2023 scientific survey to estimate the horse population in Kosciuszko National Park.

The National Parks and Wildlife Service has released supplementary information and the raw data collected as part of the 2023 horse population survey in Kosciuszko National Park.

The supplementary statement summarises how the raw data is analysed to generate a population estimate for Kosciuszko National Park.

Copies of the raw data sheets will be made available on request. Please email your request to

Please note that in reviewing the raw data and conducting additional analysis to inform 2024 operations, a small adjustment to the previously published 2023 population estimate has been made because:

  • 18 horses (and 5 clusters) recorded during the survey were not included in the original analysis as a result of a data entry error
  • a small GIS discrepancy in the area of the northern block has been corrected.

The adjustment has resulted in an updated population estimate of 12,797 to 21,760 horses (95% confidence interval [CI]) with a best estimate of 17,393 horses. The adjustment to the best estimate is less than 0.5%.

Please note the supplementary statement below is intended to provide a general overview of how the survey was conducted and the raw data used to generate a population estimate. It does not seek to address detailed technical aspects of the survey, nor the statistical analysis.

How the 2023 survey data were collected and analysed

The 2023 survey generated a population estimate of 12,797 to 21,760 feral horses (95% CI), with a best estimate of 17,393 horses.

  • It is important to note the survey generates an estimate of the horse population for Kosciuszko National Park based on sampling part of the park. It is not a census because counting every horse in such a large, rugged landscape is impossible.
  • The methodology employed – distance sampling – is best practice. The results were peer-reviewed by experts at CSIRO and the Queensland Department of Agriculture and Fisheries.

Horses occur across approximately 365,000 hectares, or 53%, of Kosciuszko National Park.

The 2023 horse population survey was conducted in 4 survey blocks totalling 267,500 hectares, or approximately 39% of the park. That is, the survey produces a population estimate for 39% of the park:

  • because feral horses are known to occur in an additional 14% of the park, the population estimate from the survey is an underestimate of the true population in Kosciuszko National Park
  • there are several reasons the additional 14% of the park is not surveyed – the primary reason is the steep topography makes it unsafe to survey the area using helicopters
  • a map of the horse distribution in Kosciuszko National Park and the 4 survey blocks is provided below (see map 1).

Within the 4 survey blocks, horses were counted across 58,440 hectares, or 21.85%, of the total survey block area. The proportion of each block where horses were counted varies from 19% in the northern block to over 40% in the Snowy Plains block (see table 1).

The survey involved flying a total length of 1,948 km of transects in a helicopter, with trained observers counting horses they observed on either side of the helicopter within 150 metres (300 metres wide in total). Maps 2 and 3 show the transects in the northern and southern survey blocks.

The location of transects follows a systematic random sampling design – that is, the starting point for the transects in each survey block was randomly generated, with layout of the transects then following a systematic pattern. This means that the sampling procedure was unbiased.

In total, 1,926 horses were counted by the observers. This means 1,926 horses were recorded across 58,440 hectares or in 21.85% of the total survey area. Table 1 shows the raw number of horses seen (counted) in each survey block and the proportion of each block that was sampled via transects.

To produce a population estimate for the total survey block area, or 39% of the park, the raw number of horses observed (1,926) needs to be adjusted in 2 ways.

The first adjustment is to reflect the fact that the transects in which horses were recorded covered only a proportion of each survey block area.

  • Because the sample in each block was unbiased, the best science means that we assume a similar density across the remaining area of each of the 4 survey blocks.
  • This first adjustment is a simple multiplication: the raw number of horses seen in each survey block is adjusted for the proportion of that block that was sampled. In the northern block, this means 1,441 horses are multiplied by 5.2 (to convert 19.4% to 100%). This would generate a population estimate in the northern block of around 7,435 horses.
  • This simple adjustment for the proportion of each sampled block provides a total estimate across all 4 survey blocks of around 9,620 horses.

However, this calculation assumes all horses occurring within the 300-metre-wide transect were seen by observers. A second adjustment is required because, in fact, not all horses in the 300-metre-wide transects were seen – that is, there were more than 1,926 horses in the transect area.

  • There are several reasons some horses within the transect area were not seen by observers in the helicopter, including vegetation cover and the distance away from the observer.
  • Distance sampling modelling is used to estimate the proportion of horses on the transect that observers saw. A different 'detection probability' – the proportion of horses seen – was determined for each survey block. However, on average, across all 4 survey blocks, it was estimated that a little more than half of all horses in the 300-metre-wide transects were seen by observers.
  • Once an adjustment is made for the proportion of horses within the transects that were seen by observers, the total population estimate for the 4 survey blocks is 17,393 horses.
  • Recognising there is some variability when estimating the level of detection by observers, the 95% CI for the population is 12,797 to 21,760 horses.
  • We can be 97.5% confident that the horse population in 39% of the park was at least 12,797 horses at the time of the 2023 survey.

The number of horses removed in the initial aerial shooting program in southern Kosciuszko National Park is consistent with the 95% CI generated by the survey results.

Table 1: A summary of the raw counts of horses in the 4 survey blocks

Survey block Raw count % of survey block sampled
Northern 1,441 19.4%
Snowy Plains 82 43.4%
Cabramurra 8 33.8%
Southern 395 20%
Total: Kosciuszko National Park 1,926 21.85%

Table 2: The population estimates (N) for each of the 4 Kosciuszko National Park survey blocks with confidence interval and coefficients of variation (cv%).

Given also with these estimates are the areas surveyed, including the total area of the 4 survey blocks.

Survey block Area (km2) N Confidence interval (95%) CV %
Northern Kosciuszko 1,229 13,212 8,895 to 16,842 16.3
Snowy Plain 161 363 119 to 743
Cabramurra 139 49 0 to 106 58.9
Southern Kosciuszko 1,146 3,769 2,337 to 5,720 23.7
Total: Kosciuszko National Park 2,675 17,393 12,797 to 21,760 13.8


Within each survey block transects were flown by a helicopter with 2 observers on board – one observer on each side of the rear of the helicopter. The helicopter was flown at a ground speed of 93 km h-1 (50 kts) and a height of 61 m (200 ft).

The observers counted the horses within 150 metres on either side of the transect using a calibrated sighting boom.

Key details recorded by the observer include:

  • the number of clusters of horses and the number of horses in each cluster (a cluster can be a single horse if it is on its own)
  • the distance from the centreline (inner edge) of the survey strip – the following categories being used ( 0 to 20 m, 20 to 40 m, 40 to 70 m, 70 to 100 m and 100 to 150 m)
  • vegetation cover at point-of-detection (open or tree covered)
  • observer identification (and seating position) is also recorded
  • weather, including cloud cover and survey aspect (direction of flight in relation to observer seating position).

As indicated above, location of transects followed a systematic random sampling design – that is, the starting point for the transects in each survey block is randomly generated, and the transects then followed a systematic pattern (see maps 2 and 3). This means the sample was unbiased.

In the northern and southern survey blocks, the transects are 1.5 km kilometres apart. If the transects were brought closer together, there would have been an increased risk of horses moving between transects (and hence being double counted) during the survey.

Due to their smaller size, the transects in the Cabramurra block were 900 metres apart and 700 metres for the Snowy Plains block. However, to ensure horses were not double counted, every second transect was flown on one day, with the other transects being flown on the following day.

Not all of the horses within the 300-metre-wide transects were seen by observers – that is, the count (1,926 horses) was less than the true population of horses in the transect area. The proposition that observers do not see every animal on a transect is a proven concept in peer-reviewed scientific papers. It has been verified in published studies performed on animal populations of known size.

The likelihood of detecting an animal decrease with increasing distance from the helicopter – for example, there is a lower probability of seeing a horse at 130 metres distance from the observer compared to 25 metres from the observer.

An adjustment is therefore made to account for the fact that not all of the horses within the 300 metres transect were seen.

This is where distance sampling modelling is used. In general terms, the distance sampling analysis estimates the proportion of horses that were actually seen by observers by developing a detection function.

The overall level of precision of the survey was high (coefficient of variation or CV = 13.8%). As a general guide, a CV of 20% or less is considered very good for large-scale animal surveys such as this one.

The level of sampling was high:

  • 21.7%of the survey blocks were sampled
  • the numbers of observed clusters was high in the 2 major population blocks, with 272 clusters observed in the Northern Kosciuszko National Park block and 120 clusters observed in the Southern Kosciuszko National Park block.

For the southern block in Kosciuszko National Park, the 95% confidence interval for the population estimate was 2,337 to 5,720 horses.

Since the 2023 survey was conducted, National Parks and Wildlife Service has removed 796 horses in control operations (aerial shooting) in only 35.8% of the southern Kosciuszko National Park survey block.

The horse populations in this program area were considered low to moderate density compared to other southern block areas.

Not all horses in the program area have been removed.

While this is a small dataset based on a limited shooting program, extrapolating from this result suggests the actual horse population in the southern block at the time of the survey was within the 95% confidence interval generated by the survey.

As indicated above:

  • the raw number of deer recorded on the transects was 280
  • the raw number of goats recorded on the transects was 18
  • the raw number of pigs recorded on the transects was 9.

The raw number of horses seen on the transect was much higher than the number of deer, goats and pigs.

While the probability of detection for deer is lower than for horses, the survey results indicated that, for the 4 survey blocks combined, the population of horses is substantially higher than the population of deer. Deer also occur in area that were not surveyed.