Establishing a Long Term System
for Monitoring the Illegal Killing of Elephants (MIKE)
Annex 3: Methods of counting elephants
Numbers and distributions of animal species, along with information on habitat and land use parameters, are essential for drawing up management and conservation plans and making informed decisions. This section gives a brief description of the four main methods of obtaining data on elephant numbers (taken from Studying Elephants, African Wildlife Foundation 1996).
Elephant census techniques fall into two classes. The first comprises those surveys where the elephants themselves are counted. These are 'direct counts'. The second class includes surveys where signs of elephants (dungpiles, tracks, feeding signs) are counted. These are 'indirect counts'.
Direct counts of elephants can either be carried out from the air or from the ground. In savannah habitats, aerial counts remain the most effective means of elephant census (DouglasHamilton et al. 1992). There are two kinds of aerial counts: sample counts and total counts. In a sample count only part, or a sample, of the area is searched and counted, and the number of animals in the whole area is then estimated from the number in the sampled area (NortonGriffiths 1978). In a total count, on the other hand, the whole of the designated area is searched, and it is assumed that all groups are located and counted accurately (NortonGriffiths 1978).
Aerial sample methods are today widely used for censusing elephants and monitoring their movement and habitat use. It is also only by aerial methods that areas that are not accessible on the ground can be censused. The choice of whether to use total or sample aerial counts will depend on the area to be covered, the size of the populations and the resources available in terms of trained manpower, aircraft, funding and time available. Sample counts tend to be cheaper than total counts, simply because only part of the area is searched. Total counts are, however, suitable in relatively small study areas (of the order of 1,000 km^{2}), and the results are easy to understand because they are not confounded by the statistical assumptions of sample counts.
Where it is impossible to count elephants directly, as in the extensive forests of Asia and west and central Africa, signs of elephants such as dung piles are used to provide an estimate of elephant numbers.
Elephants themselves can be counted from the ground either on foot or from a vehicle. Ground counts from vehicles are practicable and give excellent results in small to medium sized areas where the country can be traversed by vehicles, and where the vegetation is reasonably open and the animals tame to vehicles (NortonGriffiths 1978). Carrying out counts on foot is not common nowadays, but where resources are limiting they can provide good information on a population.
The appropriate technique to use in counting elephants, thus depends on the type of habitat (i.e. vegetation density and topography), the size of the area to be surveyed, the elephant density, and also the type of estimate required. Does one need an accurate estimate, one that approaches the true population size, but may have wide confidence limits, or does one need a precise estimate, one that may be biased, but has narrow confidence limits?
Managers require an accurate estimate, preferably at regular intervals, for a population subject to legal offtake, in the form of safari hunting and culling operations. In most cases, however, a precise estimate will be sufficient, and will enable one to monitor population trends.
Counting elephants from the air  sample counts
In a sample count, only part of the study area is searched and a count made. A series of samples, which are representative of the study area are taken (Cochran 1963; Campbell 1967; NortonGriffiths 1978). The study area, or the census zone, is the whole area for which the elephant population count is to be carried out, e.g. national park, district, etc, while the sample zone is that part of the census zone in which the elephants are actually searched for and counted. The total number of elephants in the census zone is then extrapolated from the number counted in the sample zone.
In a sample count, we take a few observations, but the conclusions we draw have a wider application. In other words, we observe a sample, but apply the conclusions to a population. For example, the assumption might be that if 10% of the area is sampled, then it will contain 10% of the elephants in the census zone.
The foregoing would hold if elephant distribution and vegetation conditions were uniform, in which case any kind of sample would give similar results. However, elephant numbers and distributions are far from uniform in any one census area. Similarly, elephants will be more easily seen, and thus counted, in open areas as opposed to thickly vegetated country. The sample zone, i.e., that portion of the census zone in which the elephants are counted, must, therefore, reflect any variations as much as possible.
The census zone is divided into sample units which are chosen at random, meaning that every one unit, n, has an equal chance of being selected for sampling from the possible N such units in the census zone (Cochran 1963; NortonGriffiths 1978). The sample zone is, therefore, randomly distributed in the census zone, thus, theoretically, representing the variations in elephant numbers and distribution.
The population estimate of the elephants is then calculated, based on the average counted number of animals in the sample units. Since the units are randomly selected, the average number of elephants per unit in the sample will correspond to the average number in the whole population. The total population estimate is then obtained by multiplying the sample mean by the total number of units in the census zone.
Sample counting assumes that the area actually sampled (sample zone) contains a corresponding percentage of the 'true' population in the census zone. Due to various factors, however, this may not be the case. To start with, elephants (as indeed other animals), are not evenly distributed. Thus, different sample units in the census zone will contain varying numbers of elephants. It follows then that different population estimates will be obtained depending on the units selected for sampling, i.e. there will be large numbers of alternative estimates. This result is due to what is referred to as sampling error, and the larger the variation in numbers of elephants between the units, the larger the range of alternative estimates or confidence limits. Sampling error results from the uneven distribution of animals and the sampling technique used (NortonGriffiths 1978).
In addition to sampling error, biases also affect the population estimates. Biases are errors in one direction, e.g. underestimating. They result from various factors  such as spotting and counting, photo counting, aircraft operations, etc.
At this point let us examine the words accuracy and precision. Consider a hypothetical population of 94 elephants. Suppose that during three different surveys, we get 50, 72 and 160 elephants, giving an average of 94; alternatively we could also get 92, 97 and 93, also giving an average of 94. The latter is more precise, as the 'true' population lies within a narrow range, i.e. the confidence limits are low. On the other hand, an accurate estimate is very near the 'true' population, but the confidence limits may be wide.
Whether we aim for an accurate or precise estimate is determined by the objectives of the survey. Accurate estimates are more important if a culling operation is to take place, while precise ones are important for detecting changes in population trends. The ideal estimate would be one that is both accurate and precise.
Stratification
Elephants tend to be clumped in distribution, such that even when sample units are randomly selected, the estimate will have high variances. Stratification or division into areas or strata of more or less homogenous elephant density reduces the variance. Stratification can also be carried out according to vegetation type or density or other major sources of variation. Through stratification, sampling effort can be more efficiently allocated to areas of greater interest or ecological importance. The strata so identified are then sampled separately and the estimates combined for the entire census zone (Cochran 1963).
Counting elephants from the air  total counts
Total counting of elephants has been adopted in many national parks, reserves and other parts of the elephants' range in Africa. One of the reasons that total aerial counting of elephants wins favour is that elephants, being large animals, are relatively easy to spot and count compared to other animals.
The aim of an elephant total aerial count is to scan the entire surface of a selected census area, and to record the position and number of each elephant or group of elephants. A total count is similar to sample block counts but in this case the blocks, when joined, cover the whole census zone.
The flight lines should be designed with the intention of being able to spot all the elephant groups and individuals; there are a number of variations as to how this may be done.
Errors can arise in failing to spot elephant groups, counting them inaccurately, or in double counting the same groups. These errors can be greatly reduced by training and careful attention to technique.
The census zone should be divided into discrete counting blocks. By common practice these are usually defined by features such as roads, cutlines, mountains, protected area boundaries or rivers. Rivers, however, are unsuitable as block boundaries because they tend to attract concentrations of elephants. A movement of elephants across the river while the count is going on could cause that group either to be doublecounted, or to be missed altogether. It is better to use watersheds as boundaries, as is done in the Kruger National Park in South Africa, because elephants tend to be relatively sparsely distributed near them.
A block should usually be of a size that can easily be covered by one aircraft in one flying day. In the case of Kenya's Tsavo National Park, blocks vary in size from 500km^{2} to 1,500km^{2}, but the average size is 1,100km^{2}. Each flight crew should be allocated one or more blocks to be counted per day and should be provided with flight maps of the blocks. In the Tsavo elephant count of 1994 flight crews on average spent 5.5 hours a day counting with another 13 hours flying to and from the block. Scanning rates on average were 210km/hr (DouglasHamilton et al. 1994)
These days it is highly desirable to use a Geographical Positioning System (GPS) in the aircraft, both to assist in navigation and for recording waypoints (a waypoint is the location of an observation point along one's line of flight).
Estimating forest elephant abundance by dung counts
Dung counts are the most common type of indirect census method for counting elephants. Since the early 1980s, as interest quickened in the status of elephants in the forests of west and central Africa, more and more dung counts have been conducted. In the late 1980s researchers in India and then in Southeast Asia turned to dung counts for estimating the numbers of Asian elephants. The proliferation of forest elephant surveys on both continents has stimulated the rapid evolution of dung survey techniques. These methods have been described previously by Barnes and Jensen (1987), Dawson and Dekker (1992), and Barnes (1993).
Many of the concepts involved in dung counts are similar to those already described in aerial surveys, i.e. one goes through the same stages of stratification, arranging the layout of transects within each stratum, collecting the data on the transects, and then analysing the data. However, with dung counts one then has the further problem of converting estimates of dungpile numbers into estimates of elephant numbers.
A major difference between direct counts of elephants and dung counts is that the methods for direct counts have been worked out and standardised, and the improvements now consist of finetuning. On the other hand, the general methods of dung counts are still evolving.
A dung survey can be used in two ways. First, one may use dung as an index of elephant abundance or relative distribution. This can provide a considerable amount of valuable information about the biology of elephants in your study area (e.g. Barnes et al. 1991). For many purposes you do not need an estimate of the actual number of elephants. An estimate of the number of dungpiles, the relative distribution of dungpiles, or changes in the number of dung piles over a period of years will give you all the information you need to manage the survey area.
The second option is to translate the dung data into numbers of elephants. To do this will require considerably more time and effort.
To obtain an estimate of elephant numbers you will have to go through four stages:
1

estimate the numbers of dungpiles, or the density of dungpiles per km^{2}

2

estimate the defecation rate of elephants

3

estimate the mean rate of dung decay

4

combine the above three estimates to give an estimate of elephant numbers or the density of elephants.

Transect counts
The simplest form of estimation of numbers from observation data uses linear extrapolation. That is, having surveyed a defined area within a region, such as a transect with a fixed width, and assuming that all the animals within that area have been seen, applying the calculated density to the whole region. This method produces the best results in open country where there is no visibility problem. In all other cases the method will be inadequate in at least two ways that result in error in the estimate of animal abundance:
i)

it is difficult to define accurately the area that has been surveyed; and

ii)

one assumes that all individuals have been seen in the surveyed area. This assumption, however, is not realistic when using a transect of fixed width in woodland habitats, for instance. In this case the population estimates will be negatively biased, that is, one will estimate fewer elephants than there actually are in an area.



These problems can be overcome by using variable fixedwidth transects, whereby the width of the transect is adjusted according to the vegetation density. In open country, the width of the transect may be as much as 500m, while in areas of dense vegetation the fixed width may be reduced to 100m. This technique, however, has the same sources of error as the fixedwidth method described above. King's method was the first technique to use this variable visibility profile, taking the average sighting distance as half the effective strip width or half the width of the strip censused. Although the method is weak and usually produces overestimates of density (NortonGriffiths 1978), it does not require much training to carry out the field procedure and the data analysis.
Linetransect counts
In linetransect sampling the observer progresses through the area following a straight line of known length (transect). He or she records each animal, notes the distance of the animal from the observer when spotted and using a compass, its bearing, which is then converted to a sighting angle relative to the transect. As a result, the observer is able to calculate the perpendicular distance of each animal from the transect. The width of the transect is not fixed and changes constantly according to the visibility or density of the vegetation along that particular segment of the transect. The width of the transect also differs for each species of animal when multispecies counts are conducted.
The data from a linetransect survey are a set of distances and angles and the resultant sample size itself (i.e., number of groups seen and number of transects walked). The set of distances and angles are transformed to a set of perpendicular distances of the elephants from the transect line. These perpendicular distances are then used in a statistical model to calculate the elephant density for the area. The basic idea underlying such a model is that the probability of detecting an elephant decreases as its distance from the transect line increases.
The major advantage of the linetransect sampling technique is the relative ease of its implementation in the field. The placement of transect lines may be either temporary or permanent. Permanent transect lines, delineated by markers, should be considered if the transects are to be surveyed periodically. Use of permanent transects enables pairing of the data for the analysis of differences in density over time and thereby increases the power of such analyses. When the survey areas have been selected, the layout of transects must be determined. That layout will depend on statistical design requirements, but considerations of logistics, supplies and access will in practice often determine the final survey design.
Direct counts of elephants from the ground
The most direct way to estimate the abundance of an elephant population is to count all individuals in a defined area. An estimate of population density is obtained simply by dividing the number counted by the size of the area censused, and the density figure obtained in this way can then be applied to surrounding areas with similar characteristics, such as soil types and vegetation. Census methods based on this approach are usually called quadrant, plot or strip sampling methods.
Defining an area or establishing a plot and then counting all the elephants within it on foot or from a vehicle can be very time consuming and impractical, and certainly impossible if the target elephant population is mobile or if individuals are widely scattered. As an alternative, transect and linetransect methods have been devised to estimate animal abundance. Both can be carried out on foot or from a vehicle, and the principles that apply are very similar to those used in estimating elephant abundance using dung counts.
Notes on estimating elephant mortality in forest habitats (Richard Barnes, pers.comm.)
There are two major problems that must be confronted in designing an elephant monitoring system for central Africa. The first problem is that, unlike savannah regions of east and southern Africa, there is no welldeveloped method for detecting elephant carcasses. Although some methods (e.g. foot patrols or aerial reconnaissance of forest clearings) show promise, there is currently little or no data available on the rate at which elephant carcasses can be detected with a given method. Without this data on carcass encounter rate, it is impossible to estimate how much field effort will be required at each site in order to amass a sample large enough to estimate trends in the rate of illegal killing with any confidence.
Thus, until more research on carcass detection rates is collected, the best method available for detecting changes in the rate of illegal killing will be to monitor the size of populations at the selected sites.
Welldeveloped methods for forest elephant population monitoring currently exist, and are increasingly sensitive to changes in elephant abundance. In fact, these methods are now sensitive enough so that, with a reasonable amount of field effort, changes in population size of the magnitude commonly observed in heavily poached areas can be detected. More importantly, data are already available for estimating the amount of field effort necessary to detect changes of abundance of a stipulated magnitude. Furthermore, one newly developed population monitoring method ('forest reconnaissance' or 'recce') uses a field protocol quite similar to the foot patrols commonly used for carcass detection.
This means not only that once carcass detection methods are developed it may be possible to implement population monitoring and carcass detection simultaneously, but that the training and infrastructure investment necessary to implement population monitoring should all apply equally well to carcass detection. Note also that because the visibility of elephants is extremely low in forest areas, the live animal encounter rate figure used in carcass ratio estimator of killing rate may need to be replaced with the dung counts used by population monitoring methods. We expect that with more research it will become more efficient to use carcasses.
REFERENCES
Barnes, R.F.W. & Jensen, K.L. 1987. How to count elephants in forests IUCN African Elephant & Rhino Specialist Group Technical Bulletin 1, 16. Version in français "Comment compter les éléphants de forêt?". traduction Dr. H. Mertens.
Barnes, R.F.W. et al. 1991. Man determines the distribution of elephants in the rain forests of northeastern Gabon. African Journal of Ecology 29, 5463.
Barnes, R.F.W. 1993. Indirect methods for counting elephants in forest. Pachyderm 16, 2430.
Campbell, R.C. 1967. Statistics for Biologists. Cambridge University Press.
Cochran, W.G. 1963. Sampling Techniques. Wiley, London.
Cochran, W.G. 1977. Sampling Techniques (Third Edition), Wiley, New York.
Dawson, S. & Dekker, A.J.F.M. 1992. Counting Asian elephants in forests. FAO, Bangkok
DouglasHamilton et al. 1994. Tsavo Elephant Count. Kenya Wildlife Service. Typescript report.
Kangwana, K. (Ed.). 1996. Studying Elephants, Technical Handbook Series No. 7, African Wildlife Foundation.
NortonGriffiths, M. 1978. Counting Animals, Handbooks on techniques currently used in African wildlife ecology. No.1, African Wildlife Leadership Foundation.
nnex 4 Selection of a representative sample of sites
Annex 2 Data requirements