Because population estimates drive nearly all management decisions pertaining to wild horses and burros, accuracy is important. Several widely used techniques exist for conducting aerial population estimates of wildlife, but individually, each has important limitations. Researchers at the U.S. Geological Survey, Fort Collins Science Center (FORT), are evaluating combinations of these techniques that compensate for those limitations and balance cost-effectiveness and precision in estimating population numbers of wild horses and burros.
The techniques under consideration for surveying wild horses and burros are mark-resight using natural markings (horses only), simultaneous double-count, sightability bias correction modeling, and distance sampling. All of these techniques are amenable to standard statistical survey design methods, including stratification and subsampling of the study area. Additionally, for each approach the sample unit is a horse or burro group rather than an individual animal, and capture or handling of the animals is not required. Beyond these common benefits, each has distinct advantages and disadvantages. Combining two or more techniques that integrate the individual aspects of each is a recent development that promises to alleviate the shortcomings inherent in a given technique by complementing it with information from one or more of the others.
Mark-Resight
Mark-resight techniques use the number of marked animals that are observed or missed during a survey as the basis for estimating population size. Mark-resight techniques require that the population be closed (animals do not move in or out of the area between marking and the survey), that the marks are not missed by the observer or lost from the animals, that the sample of marked animals is representative of the population, and that sightability of the animals is not enhanced or reduced by the marks. Mark-resight that includes recognition of uniquely marked animals during the survey has greater power and ability to estimate the population, and fewer resighting survey flights are required.
Artificial marking devices, such as radio collars, have not worked well on wild horses in the past. Furthermore, conventional mark-resight was found to be inaccurate in trials during the 1980s. Temporary markings could be applied during gathers, but this would provide only for post-gather population estimates. Alternatively, some groups of horses have sufficient distinctive natural markings (body color, face markings, and leg markings) to be uniquely identifiable, making artificial marking unnecessary. The makeup of individuals in a harem can be used to uniquely identify that harem, even from a helicopter, using photographs. Identifiable wild horses constitute the “marked” sample during a subsequent aerial or ground survey. Therefore, estimate precision can be improved by “marking” or photographing a large portion of the population during the first survey. This also means that the cost of marking (sighting once) is no higher than for resight: allocating roughly equal effort to the sighting and resighting surveys should produce the most cost-effective estimate.
Advantages of mark-resight are that the technique is well-established, suitable software is available, and there is high statistical power when the technique is applied properly. Using natural markings could be too costly in large, widely dispersed herds; however, in herds with a high proportion of distinctively colored animals, natural markings offer an excellent alternative for estimating small to medium-size populations, which can be very difficult to estimate precisely using any of the other techniques under consideration.
See Lubow and Ransom (2009) for results from tests where aerial population estimates were compared to populations of known size.
Simultaneous Double-Count
This method is a form of mark-resight. Two observers in an aircraft independently observe and record groups of wild horses or wild burros. Sighting rates are estimated by comparing sighting records of the two observers. Groups seen by both observers must be identified, either by communication with a third observer during the flight, or by matching observations based on time, location, and other characteristics of the sighting. Those animals seen by one observer are the “marked” groups; those that are also seen by the other observer are “resighted.” Sighting probabilities for both observers can be computed from this information. Standard mark-resight or Lincoln-Petersen calculations are used to generate a population estimate.
The simultaneous double-count possesses several distinct advantages. No capture or handling of animals or pre-survey work is required. Observer differences are estimated and accounted for. To ensure that observations are independent (uncorrelated), aerial observers can be isolated through the use of barriers and adherence to strict guidelines and discipline. The variable visibility of different groups of animals is a bigger challenge; however, a recent advance in the statistical methodology removes much of this concern. Finally, this technique has been tested on a limited basis where estimates could be compared with a known number of animals, and the results are promising.
Sightability Bias Correction Model
This technique, also known as the Idaho Sightability Model, requires building a model of the sighting probability for groups of horses or burros. Surveys, or sighting trials, are conducted to determine which covariates (related factors) influence sightability. A model is developed based on the sightability rate of marked or ground-truthed animals. Typically, a model is developed that predicts sighting probability of individual groups of animals based on a set of covariates such as group size, percent tree and shrub cover that will hide animals, percent snow cover, observer experience, survey intensity, etc. The “model” or logistic regression of the sighting covariates is then applied to the sighting conditions recorded for each count unit and each entire survey. For example, sighting rate typically increases with group size for elk. If most of the elk groups are large during a particular survey, the correction factor is lower. On the other hand, when elk groups are small or are located under tree cover, as is typically the case when snows are deep, sightability decreases. In these cases, the correction factor will be higher, and the variance of the estimate will increase.
Sightability models assume that groups are sighted independently and that all meaningful covariates affecting sighting rates are included in the model. Model development requires an adequate sample size (100–200 groups). This technique may not be the best choice for ungulates in situations where detection rates are low, especially when detection rates average less than 50 percent. Sightability models are most useful if sighting rates for all groups are more than 60 percent. Models are specific to covariates affecting sighting rates; hence, different model variants are required for different aircraft, different seasons, and different landscapes. This lies at the heart of the model’s shortcomings in that considerable time, cost, and effort goes into model development and calibration. Further, the model must hold over space, time, and new observers for estimates to be valid, and populations within each count unit must remain constant during the survey. However, this accounting for multiple sighting factors is also the sightability model’s chief advantage.
Distance Sampling
Distance sampling is a method for estimating population density directly, from which total population is easily calculated. Unlike other survey techniques, the primary data are distances to animals rather than counts of the animals themselves. The distances are measured perpendicularly from a transect line to observed animals. Informally, it is intuitive that distances from observers to animals in low-density populations will be higher, on average, than in high-density populations. This fact is the basis for estimating density. Another important distinction is that distance sampling is based on sampling with replacement. Consequently, natural movement of individuals and even double-counting of groups do not degrade the accuracy of estimates.
Traditional distance sampling requires that three assumptions be met: all animals on or near the transect line must be seen; distances from the transect to the animals are measured accurately; and animals are spotted and accurately located before they move in response to the approaching observer. Typically, two observers located on either side of an aircraft record the size of all groups seen and the perpendicular distances from the transect line to each group. Transect lines must be placed randomly (or systematically with a random start) but do not necessarily have to be straight; contour lines can be flown instead. Location of and distances from the transect line can be measured using GPS and other technologies.
Advantages of this technique are that each population estimate requires only one overflight; natural movements and double counting are not a problem; the technique is robust to variable visibility of animals within and between surveys; and the technique, sampling protocols, and statistics have been extensively developed. The biggest challenge is likely to be movement of wild horses in response to the approaching aircraft prior to detection by observers.
Combining and Testing Aerial Survey Techniques for Wild Horses
At the request of the U.S. Bureau of Land Management (BLM), scientists at the U.S. Geological Survey Fort Collins Science Center (FORT) are testing combinations of four techniques for estimating wild horse populations: mark-resight, simultaneous double-count, sightability bias correction modeling, and distance sampling. Field methodology to implement an integration of techniques has been developed and tested in Antarctic marine fauna populations; however, the equipment and aircraft used are not suitable for wild horse survey needs. Developing a similar approach using more readily available aircraft and less expensive technology could produce a very user-friendly and more accurate survey method than is currently available to wild horse managers.
Advantages of Combining Techniques
The combination of multiple sources of information overcomes most of the deficiencies of the separate techniques alone and provides greater power and efficiency. For example, the major difficulty in the double-count technique—ensuring similar sighting probabilities for all animals—can be resolved by modeling sightability using covariates in a manner similar to the sightability bias correction model. However, unlike the traditional model, multiple observers provide sufficient information to estimate sighting models for each covariate from a single survey. Therefore, no pre-calibration of the model is required, and the assumption that the initial calibration applies uniformly over space, time, and observers is eliminated. Similarly, the requirement in distance sampling that all animals near the transect line are spotted can be eliminated by applying data from a double-count to estimate the detection probability on the line. Finally, both the mark-resight and distance sampling methods can be made more precise by incorporating sightability covariates.
Simultaneously using two or more methods increases the complexity of data collection for aerial crews. For example, combining sightability and double-count would mean that observers record all of the same data that are recorded for sightability models, with the addition of a GPS location used to later match the sightings of the two observers. Still, by including one additional measurement—perpendicular distance from the transect to observed animal groups (distance sampling)—all three techniques can be combined, tested, and compared during the same survey.
Testing the Combinations
With input from various Federal and State wildlife survey experts, FORT’s aerial research team completed a work plan in the summer of 2003 and was in the air conducting surveys by October of that year. By testing the techniques at the fertility control field-trial sites, the research team had the advantage of knowing the true population numbers, against which the aerial survey results could be compared. Wild horses at these sites are known individually and observed from the ground 6–7 months per year.
Mark-resight sampling combined with sightability bias correction modeling has now been tested against known populations at the McCullough Peaks Herd Management Area (HMA), Wyoming; the Little Book Cliffs Wild Horse Range, Colorado; and the Pryor Mountain Wild Horse Range, Montana and Wyoming. Aerial photography was used to create "marked" individuals, followed by additional flights to collect "resight" data. Because horses exhibit unique coloration and markings, a photographic image served to "mark" an individual; then, photos from each flight were matched up to determine which animals were resighted. Results to date show this to be a promising technique for smaller populations and for habitats consisting of considerable tree cover and rugged terrain, where only a limited percentage of the animals may be visible. See Lubow and Ransom (2009) for details on the results of these tests.
In Nevada and Wyoming, several flights have been conducted pairing simultaneous double-count with sightability bias covariates. The HMAs surveyed were very large and more representative of large western herd areas than the fertility control field-trial HMAs. Since the true populations of these areas were not known, surveys were conducted before and after a removal of horses, and the statistical estimates compared to the known removal number. The results of these flights guided the development of a more statistically sound population estimate for each HMA, but also provided considerable data on individual observers and the effect position in the aircraft has on sightability of horses. The distance to and size of the horse groups also played a key role in the modeling. See Ransom (2012) for additional information on factors affecting sightability of horses. A manuscript on the final combined technique—using simultaneous double-count, sightability bias correction modeling, and distance sampling—is in preparation.
One other technique, forward-looking infrared (FLIR), was integrated into the study in 2004. FLIR has been used in surveying many other wildlife species and can potentially improve the sightability of wildlife in more vegetated or rugged habitats. Our test, conducted in Nevada, resulted in a less satisfactory population estimate than the simultaneous double-count survey performed at the same herd area. This technique might be more suitable for surveying wild burros, which are challenging to count because of their more solitary nature and preference for rocky, rugged habitat.
Because population estimates drive nearly all management decisions pertaining to wild horses and burros, accuracy is important. Several widely used techniques exist for conducting aerial population estimates of wildlife, but individually, each has important limitations. Researchers at the U.S. Geological Survey, Fort Collins Science Center (FORT), are evaluating combinations of these techniques that compensate for those limitations and balance cost-effectiveness and precision in estimating population numbers of wild horses and burros.
The techniques under consideration for surveying wild horses and burros are mark-resight using natural markings (horses only), simultaneous double-count, sightability bias correction modeling, and distance sampling. All of these techniques are amenable to standard statistical survey design methods, including stratification and subsampling of the study area. Additionally, for each approach the sample unit is a horse or burro group rather than an individual animal, and capture or handling of the animals is not required. Beyond these common benefits, each has distinct advantages and disadvantages. Combining two or more techniques that integrate the individual aspects of each is a recent development that promises to alleviate the shortcomings inherent in a given technique by complementing it with information from one or more of the others.
Mark-Resight
Mark-resight techniques use the number of marked animals that are observed or missed during a survey as the basis for estimating population size. Mark-resight techniques require that the population be closed (animals do not move in or out of the area between marking and the survey), that the marks are not missed by the observer or lost from the animals, that the sample of marked animals is representative of the population, and that sightability of the animals is not enhanced or reduced by the marks. Mark-resight that includes recognition of uniquely marked animals during the survey has greater power and ability to estimate the population, and fewer resighting survey flights are required.
Artificial marking devices, such as radio collars, have not worked well on wild horses in the past. Furthermore, conventional mark-resight was found to be inaccurate in trials during the 1980s. Temporary markings could be applied during gathers, but this would provide only for post-gather population estimates. Alternatively, some groups of horses have sufficient distinctive natural markings (body color, face markings, and leg markings) to be uniquely identifiable, making artificial marking unnecessary. The makeup of individuals in a harem can be used to uniquely identify that harem, even from a helicopter, using photographs. Identifiable wild horses constitute the “marked” sample during a subsequent aerial or ground survey. Therefore, estimate precision can be improved by “marking” or photographing a large portion of the population during the first survey. This also means that the cost of marking (sighting once) is no higher than for resight: allocating roughly equal effort to the sighting and resighting surveys should produce the most cost-effective estimate.
Advantages of mark-resight are that the technique is well-established, suitable software is available, and there is high statistical power when the technique is applied properly. Using natural markings could be too costly in large, widely dispersed herds; however, in herds with a high proportion of distinctively colored animals, natural markings offer an excellent alternative for estimating small to medium-size populations, which can be very difficult to estimate precisely using any of the other techniques under consideration.
See Lubow and Ransom (2009) for results from tests where aerial population estimates were compared to populations of known size.
Simultaneous Double-Count
This method is a form of mark-resight. Two observers in an aircraft independently observe and record groups of wild horses or wild burros. Sighting rates are estimated by comparing sighting records of the two observers. Groups seen by both observers must be identified, either by communication with a third observer during the flight, or by matching observations based on time, location, and other characteristics of the sighting. Those animals seen by one observer are the “marked” groups; those that are also seen by the other observer are “resighted.” Sighting probabilities for both observers can be computed from this information. Standard mark-resight or Lincoln-Petersen calculations are used to generate a population estimate.
The simultaneous double-count possesses several distinct advantages. No capture or handling of animals or pre-survey work is required. Observer differences are estimated and accounted for. To ensure that observations are independent (uncorrelated), aerial observers can be isolated through the use of barriers and adherence to strict guidelines and discipline. The variable visibility of different groups of animals is a bigger challenge; however, a recent advance in the statistical methodology removes much of this concern. Finally, this technique has been tested on a limited basis where estimates could be compared with a known number of animals, and the results are promising.
Sightability Bias Correction Model
This technique, also known as the Idaho Sightability Model, requires building a model of the sighting probability for groups of horses or burros. Surveys, or sighting trials, are conducted to determine which covariates (related factors) influence sightability. A model is developed based on the sightability rate of marked or ground-truthed animals. Typically, a model is developed that predicts sighting probability of individual groups of animals based on a set of covariates such as group size, percent tree and shrub cover that will hide animals, percent snow cover, observer experience, survey intensity, etc. The “model” or logistic regression of the sighting covariates is then applied to the sighting conditions recorded for each count unit and each entire survey. For example, sighting rate typically increases with group size for elk. If most of the elk groups are large during a particular survey, the correction factor is lower. On the other hand, when elk groups are small or are located under tree cover, as is typically the case when snows are deep, sightability decreases. In these cases, the correction factor will be higher, and the variance of the estimate will increase.
Sightability models assume that groups are sighted independently and that all meaningful covariates affecting sighting rates are included in the model. Model development requires an adequate sample size (100–200 groups). This technique may not be the best choice for ungulates in situations where detection rates are low, especially when detection rates average less than 50 percent. Sightability models are most useful if sighting rates for all groups are more than 60 percent. Models are specific to covariates affecting sighting rates; hence, different model variants are required for different aircraft, different seasons, and different landscapes. This lies at the heart of the model’s shortcomings in that considerable time, cost, and effort goes into model development and calibration. Further, the model must hold over space, time, and new observers for estimates to be valid, and populations within each count unit must remain constant during the survey. However, this accounting for multiple sighting factors is also the sightability model’s chief advantage.
Distance Sampling
Distance sampling is a method for estimating population density directly, from which total population is easily calculated. Unlike other survey techniques, the primary data are distances to animals rather than counts of the animals themselves. The distances are measured perpendicularly from a transect line to observed animals. Informally, it is intuitive that distances from observers to animals in low-density populations will be higher, on average, than in high-density populations. This fact is the basis for estimating density. Another important distinction is that distance sampling is based on sampling with replacement. Consequently, natural movement of individuals and even double-counting of groups do not degrade the accuracy of estimates.
Traditional distance sampling requires that three assumptions be met: all animals on or near the transect line must be seen; distances from the transect to the animals are measured accurately; and animals are spotted and accurately located before they move in response to the approaching observer. Typically, two observers located on either side of an aircraft record the size of all groups seen and the perpendicular distances from the transect line to each group. Transect lines must be placed randomly (or systematically with a random start) but do not necessarily have to be straight; contour lines can be flown instead. Location of and distances from the transect line can be measured using GPS and other technologies.
Advantages of this technique are that each population estimate requires only one overflight; natural movements and double counting are not a problem; the technique is robust to variable visibility of animals within and between surveys; and the technique, sampling protocols, and statistics have been extensively developed. The biggest challenge is likely to be movement of wild horses in response to the approaching aircraft prior to detection by observers.
Combining and Testing Aerial Survey Techniques for Wild Horses
At the request of the U.S. Bureau of Land Management (BLM), scientists at the U.S. Geological Survey Fort Collins Science Center (FORT) are testing combinations of four techniques for estimating wild horse populations: mark-resight, simultaneous double-count, sightability bias correction modeling, and distance sampling. Field methodology to implement an integration of techniques has been developed and tested in Antarctic marine fauna populations; however, the equipment and aircraft used are not suitable for wild horse survey needs. Developing a similar approach using more readily available aircraft and less expensive technology could produce a very user-friendly and more accurate survey method than is currently available to wild horse managers.
Advantages of Combining Techniques
The combination of multiple sources of information overcomes most of the deficiencies of the separate techniques alone and provides greater power and efficiency. For example, the major difficulty in the double-count technique—ensuring similar sighting probabilities for all animals—can be resolved by modeling sightability using covariates in a manner similar to the sightability bias correction model. However, unlike the traditional model, multiple observers provide sufficient information to estimate sighting models for each covariate from a single survey. Therefore, no pre-calibration of the model is required, and the assumption that the initial calibration applies uniformly over space, time, and observers is eliminated. Similarly, the requirement in distance sampling that all animals near the transect line are spotted can be eliminated by applying data from a double-count to estimate the detection probability on the line. Finally, both the mark-resight and distance sampling methods can be made more precise by incorporating sightability covariates.
Simultaneously using two or more methods increases the complexity of data collection for aerial crews. For example, combining sightability and double-count would mean that observers record all of the same data that are recorded for sightability models, with the addition of a GPS location used to later match the sightings of the two observers. Still, by including one additional measurement—perpendicular distance from the transect to observed animal groups (distance sampling)—all three techniques can be combined, tested, and compared during the same survey.
Testing the Combinations
With input from various Federal and State wildlife survey experts, FORT’s aerial research team completed a work plan in the summer of 2003 and was in the air conducting surveys by October of that year. By testing the techniques at the fertility control field-trial sites, the research team had the advantage of knowing the true population numbers, against which the aerial survey results could be compared. Wild horses at these sites are known individually and observed from the ground 6–7 months per year.
Mark-resight sampling combined with sightability bias correction modeling has now been tested against known populations at the McCullough Peaks Herd Management Area (HMA), Wyoming; the Little Book Cliffs Wild Horse Range, Colorado; and the Pryor Mountain Wild Horse Range, Montana and Wyoming. Aerial photography was used to create "marked" individuals, followed by additional flights to collect "resight" data. Because horses exhibit unique coloration and markings, a photographic image served to "mark" an individual; then, photos from each flight were matched up to determine which animals were resighted. Results to date show this to be a promising technique for smaller populations and for habitats consisting of considerable tree cover and rugged terrain, where only a limited percentage of the animals may be visible. See Lubow and Ransom (2009) for details on the results of these tests.
In Nevada and Wyoming, several flights have been conducted pairing simultaneous double-count with sightability bias covariates. The HMAs surveyed were very large and more representative of large western herd areas than the fertility control field-trial HMAs. Since the true populations of these areas were not known, surveys were conducted before and after a removal of horses, and the statistical estimates compared to the known removal number. The results of these flights guided the development of a more statistically sound population estimate for each HMA, but also provided considerable data on individual observers and the effect position in the aircraft has on sightability of horses. The distance to and size of the horse groups also played a key role in the modeling. See Ransom (2012) for additional information on factors affecting sightability of horses. A manuscript on the final combined technique—using simultaneous double-count, sightability bias correction modeling, and distance sampling—is in preparation.
One other technique, forward-looking infrared (FLIR), was integrated into the study in 2004. FLIR has been used in surveying many other wildlife species and can potentially improve the sightability of wildlife in more vegetated or rugged habitats. Our test, conducted in Nevada, resulted in a less satisfactory population estimate than the simultaneous double-count survey performed at the same herd area. This technique might be more suitable for surveying wild burros, which are challenging to count because of their more solitary nature and preference for rocky, rugged habitat.