Monitoring Illegal Killing of elephants (MIKE)
Central African Pilot Project
TECHNICAL REPORT No 2
RECOMMENDATIONS FOR GROUND-BASED SURVEY METHODS FOR ELEPHANTS IN THE CENTRAL AFRICAN FOREST REGION
Rene Beyers, Len Thomas, John Hart, Steve Buckland
(Draft, 21 August 2001)
TABLE OF CONTENTS
1 summary and acknowledgements
2.1 Background to Survey Method Questions Addressed by the Pilot Project
3 Introduction to Line Transect Sampling
5.1 Problems Encountered with Survey Data during the Pilot Project
Other problems in conducting inventories
5.2 Recce-transect Survey Results
5.3 Combining Transects and Recces
5.4 Predicting Variation in Encounter Rate at Pilot Sites
5.5 Transect Configuration and Efficiency in Repeated Survey Cycles
6 Conclusions and Recommendations
6.1 Standard Transect Configuration
ANNEX 1: Field data forms.
ANNEX 2: Data reporting forms.
ANNEX 3: Cybertracker as a data collection tool for MIKE.
ANNEX 4: Analysis of data and survey design for the Mike Central Africa Pilot Project: first interim report. (S.T. Buckland and F.M. Underwood).
ANNEX 5: Analysis of data and survey design for the MIKE central African pilot project. Third Report – Part 1: Analysis of Pilot Data (L. Thomas and S.T Buckland)
ANNEX 6: Predicting variation in encounter rates at pilot sites. (L. Thomas).
ANNEX 7: Analysis of data and survey design for the MIKE central African pilot project. Third Report – Part 3: Analysis of Ituri Transects (L. Thomas)
This report provides analyses of pilot project surveys at three sites, Odzala (Congo), Okapi Reserve (DR Congo) and Lope (Gabon). The objectives of these studies were to test and evaluate MIKE protocols for forest inventories of elephants based on ground counts of elephant dung, and to provide estimates of elephant densities for the sampling locations surveyed in the pilot project..
Two inventory methods were tested at the primary Pilot sites: "Recces" (several configurations) and line transects. ("Recce" is short-hand for reconnaissance walk.) In both methods, the observers count dung from a line of compass line of travel. In recces, the observer counts dung along a line of travel that can deviate from strict compass bearing, and can take advantage of paths or other passages of least resistance (including human and animal trails). With transects, the count line follows a strict compass bearing and the perpendicular distance from each observed dung to the line of travel is measured.
A satisfactory analysis of dung-pile density was obtained from the transect data, although there were problems with the distribution of perpendicular distances at all sites. Problems related to data collection were readily resolved following quality control and re-training in the field.
Estimated mean dung densities (per hectare) observed using calibrated recce transect results were 24.5 at Odzala (44 locations), 6.2 at Lope (44 locations) and 12.3 Okapi (14 locations). Using estimates of 13 dung per elephant per day, and a mean disappearance time of 55 days, the dung densities can be converted to elephant densities per km2 of 3.4 (Odzala), 0.9 (Lope) and 1.7 (Okapi). In Ituri, mean dung densities recorded on 5 km transects during pilot surveys were essentially unchanged from those of five years earlier. Year 2000 estimates were 6.7 / ha (coefficient of variation 16.6 %). In 1995, mean density on these same transects was estimated at 5.6 / ha (coefficient of variation 28.7 %). However only a portion of the 1995 transects was resampled. These results should be considered preliminary,
Incorporating the recce data into the density estimates resulted in increased precision at Lope and Ituri, but not at Odzala. This is mainly caused by differences in the spatial correlation between encounter rates on nearby sections of the survey lines, in which recces and transects alternated.
Although the optimal layout of recees and transects will vary by location, we recommend that a standard design be used at all forest sampling locations. A compromise design based on pilot results consists of four, one km transects separated by 3 km of recces. Transects should have well documented starting locations that can be re-located in each survey cycle.
Additional recommendations concerning modifications of field forms, incorporation of data on other species, survey cycle timing, and data quality control and analysis are provided.
We acknowledge and thank the MIKE field teams, in particular team leaders, Paulin Thsikaya (RD Congo), Victor Mbolo (Congo), and Yves Mihindou (Gabon) for their collaboration in this project. Fiona Underwood and Samantha Strindberg contributed to the analysis. The MIKE pilot project was funded by contributions from US Fish and Wildlife Service and Wildlife Conservation Society.
2. Introduction
The overall goal of the MIKE program is to provide information on trends in elephant distribution and abundance, and on the factors affecting these, in particular illegal elephant killing.
The goal of the pilot project was to evaluate several methodological problems associated with assessing trends in elephant populations and elephant poaching in Central African forests, and to test and evaluate reporting forms, data collection protocols and data and information management procedures.
2.1 Background to Survey Method Questions Addressed by the Pilot Project.There have been a number of major elephant surveys conducted in Central Africa forests over the past 15 years. With few exceptions, however, (notably in Garamba National Park in the northern savannas of DR Congo) most studies have been one-off surveys. Very little effort has been given to ensuring repeated standardized counts over time. In addition, most surveys to date made no attempt to assess information on illegal killing. Thus there was in fact little precedent for developing an overall survey strategy for the MIKE program, and in particular in the forest zone.
The need to be able to repeat surveys over time is one of the most important factors to determine overall methodology and approach. The pilot itself could only foresee this, not actually test it, since the pilot was restricted to a single survey cycle. Nevertheless, the experience at one site, where the Pilot constituted the second inventory cycle (Okapi Wildlife Reserve, DRC) indicates that with proper field training, and well marked and documented sampling locations, teams should be able to relocate fixed transects over time in almost all cases.
The only tested and standardized method currently available for surveying elephants in forest is by counts of elephant dung. The pilot project addressed this problem at two different levels: The first level addressed the question of where should surveys be placed, and how many sampling locations are needed for the monitoring program. These questions and recommendations are treated in Technical Report 1. At the second level, the question is how should surveys of dung be conducted at each sampling site. This issue is dealt with in the current report,
Two inventory methods were tested at the MIKE pilot sites: "Recces" (several configurations) and line transects. In both methods, the observers count dung along a compass line of travel. In recces, the observer counts dung along a line of travel that can deviate from strict compass bearing, and can take advantage of paths or other passages of least resistance, including animal and human trails. With transects, the count line follows a strict compass bearing and the perpendicular distance from each observed dung to the line of travel is measured. The major problems addressed regarding these two methods are:
What is the impact on survey results of using biased counts (recces) that do not cover a representative sample of microhabitats and are potentially biased by following human paths, or animal trails, including elephant trails? Is a combination of recces and transects a more efficient configuration than transects only? This question had been answered in the affirmative by Walsh et al (2001) at a limited spatial scale, and the pilot project sought to verify and generalize these results. Can surveys be designed to gather data on other variables of interest to MIKE, including data on habitats and evidence of poaching, in particular illegal killing (of elephants? Can the methods used for surveys of elephant dung be adapted for surveys of other species in forest environments? What are the major field problems associated with conducting ground surveys in the forest, and how can teams be trained and managed to provide the best results?Indirect counts, such as dung counts, require two additional parameters, dung deposition rates and dung disappearance rates, in order to convert estimates of elephant dung densities to estimates of densities of elephants. In the pilot project, we concentrate on producing estimates of trends in elephant dung, and not on converting these to numbers of elephants. We thereby assume that defecation rates and dung decay rates are similar from year to year in the same season at each site, so that trends in dung densities reflect trends in elephants. Although we do not develop the protocols for estimating dung deposition and disappearance rates at this time, we recommend that some effort be made to gather data on dung disappearance rates at least during the next phase of MIKE.
Despite problems, dung counts have some advantages. They are cost effective, can be done in standardized ways at all types of sites, and are correlated with independent measures of elephant abundance. Training field staff for dung counts is relatively straightforward; the survey methods do not rely on sophisticated equipment, and results can be analysed by readily available software such as Distance.
The specific objectives of this report are the following:
-
To present the results of elephant inventories at three pilot sites
-
To report on tests of recce transect design and determine optimal configuration and length of survey units for MIKE elephant surveys in forest ecosystems.
-
To present field protocols, data collection forms and reporting forms developed during the pilot project.
-
To evaluate field team performance, and make recommendations on building capacity of field teams.
During the pilot inventories, in addition to data on elephant dung, field teams recorded observations of ape nests and indicators of human activity (in particular poaching). We limit this report to a presentation and discussion of our elephant dung survey results, although we do comment on the utility of the MIKE protocols for gathering data on these other variables.
3. Introduction to Line Transect SamplingWe used line transect sampling as our primary method to obtain density estimates of elephant dung. This section gives a brief overview of the method; it is explained in detail in the standard work by Buckland et al. (1993, 2001).
Distance sampling on line transects is based on the assumption that the probability of detecting an animal or object decreases with distance from the line transect. It is also assumed that all objects on the line are detected.
Line transects are straight lines cut in the forest using compass and hip chain thread. The observer walks on the transect line and carefully looks for dung on and away from the transect line.
When the observer detects elephant dung he will accurately measure its perpendicular distance to the middle of the transect line (represented by the hip chain thread) (Figure 1). Distances are measured to the nearest centimetre using measuring tape.
Unbiased estimates of densities can be obtained if transects are placed at random or systematically in the survey area and have a known length.
A frequency diagram will show the probability of detecting an object (dung) at particular distances from the transect line. This detection probability can be mathematically modelled as a detection function, g(x). Several models are available to model the detection function and each model can be refined with adjustment parameters to obtain the best fit with the data. During a typical distance analysis different models, and different values of parameters are tested. The model with the best fit to the data is retained. We used the software program Distance 3.5 (Thomas et al. 1998) to model the detection function and select the best model. From the selected model, the effective half-trip width can be calculated which is the area under the detection function curve (Figure 2). It is an estimate of how wide the strip would be if all objects were detected within that strip.
Density (
) of dung can then be calculated
from:
where 
is the number of dung-piles
detected,
is the estimated strip half-width,
and 
is the total length of transect
line.
Data are usually truncated prior to analysis at a certain distance from the line transect to delete outliers that can make modelling of the detection function difficult less robust and less precise (Buckland et al. 1993).
Figure 2. Detection curve of dung piles from the pilot survey in Lope
The biggest advantage of line transects versus strip counts is that objects away from the line can go undetected. Reliable estimates of density can still be obtained when relatively mild assumptions can be met. Strip or plots counts assume that all objects are detected within a strip. For this assumption to be met, the strip must be very narrow, resulting in far fewer observations in total and lower precision.
The assumptions for line transect sampling are:
All objects on the trackline are counted – none are missed. Measurements from the line to the object are exact. Distance from the line to the object is measured either at right angles to the line or the angle between the line and the object is also measured so that a perpendicular distance to the line can be calculated. For elephant dung perpendicular distance from the centre of the dung pile to the transect line was accurately measured to the nearest centimetre using measuring tape. Objects or animals are detected at their initial location. Most animals move away as soon as they detect the observer, but distance has to be recorded to their initial location. Obviously this doesn’t pose a problem for elephant dung! Sighting of individual objects are independent from each other.In addition, for reliable estimation of the detection curve, there should be a "shoulder" extending a small distance on either side of the transect line, where almost all objects are counted (Figure 3). This so-called "shape criterion" is not an assumption of the method, but without it (e.g., Figure 1), reliable estimation is much less certain.
Figure 3. Idealized histogram and detection curve showing a broad shoulder near the transect line.
4. Field MethodsThe evaluation of survey methods was conducted at three pilot sites, Odzala, Lope and Ituri. Survey methods used, and their configurations are summarized in Table 1 and described in more detail below. The surveys were carried out using a systematic placement of sampling locations in Odzala and Lope. In Ituri, sampling locations established in the 1994 – 1995 large mammal inventories were resurveyed for the MIKE pilot. Further details on allocation and placement of sampling locations are available in Report 1.
Table 1. Survey methods utilized during the pilot project.
|
Survey Method |
Odzala |
Lope |
Ituri |
|
Recce transect combination |
5 km units, with subunits consisting of 200 m transects and 1000 m recces |
5 km units, with subunits consisting of 200 m transects and 1000 m recces |
5 km units, with subunits consisting of 200 m transects and 1000 m recces |
|
Line transects only |
Not used |
Not used |
5 km units: 1994-5 transects resurveyed. |
|
Travel recces |
Guard patrols and exploratory mega-transect also placed in survey area |
Variable length recces between sampling locations |
Not Used |
4.1 Recce transects
At each sampling location in Lope, Odzala and Ituri we used a combination of recces and line transects to obtain estimates of dung density. The sampling unit consisted of short line transects interspersed with longer recces. Recces followed a path of least resistance through the forest in a predetermined compass direction not deviating more than an arbitrary angle from this direction.
A number of different spatial configurations of recces and transects can be imagined. For the pilot study we used the following combination (illustrated in Figure 4). Each recce transect unit consisted of 5 x 200 meter transect segments interspersed by 4 x 1 km recces covering a total distance of 5 km. The point of departure was determined in advance and was indicated on a map. The end-point of the total unit was programmed in the GPS. On arrival a first straight line transect of 200 meters was cut in the direction of the end-point using compass, machete and hip chain thread. All signs of mammals and human activity were recorded on the transect line and the perpendicular distance from the observation to the transect line (hip chain thread) was measured. Distance along the thread was also recorded. For elephant dung, distance was measured to the middle of the dung pile. After the first transect a recce was followed along the path of least resistance towards the end-point of the recce-transect. and not deviating more than 45 degrees. For each observation, length along the recce (hip chain thread) was recorded but not perpendicular distance to the observation. At the end of the recce, another short transect was cut towards the end-point followed by another recce etc.
Recces have the advantage of being simpler and less strenuous. A greater area can be covered compared to line transects for the same amount of time. All dung seen from the recce is recorded but perpendicular distance to the transect were not measured.
However two problems arise with recces that can bias the density estimation.
Certain types of vegetation are avoided and a sample based on recces is not representative for the area. The path of least resistance is often an animal or human trail and signs of animals or humans will therefore be overestimated. Observers move faster on recces and so may not observe as carefully Recces do not follow the ideal line (the line where the transect would be in case of random or systematic placement of transects)These biases can be corrected by associating recces with transects. Encounter rates of dung on recces can be calibrated using transect data and a combined encounter rate can be calculated (See below)
Figure 4. Recce-transect combination used in the Pilot Project at each sampling location.
4.2 Transects OnlyIn Ituri both recce-transects and line transects were used for dung surveys at the same sampling sites. Recce-transects followed the same pattern as described above. Line transects, first surveyed in 1994-1995 were re surveyed. The 1994-95 transects were 5 km in length, with one to three transects surveyed at the same location (up to 15 km total). The MIKE surveys covered exactly the same line transects as the earlier surveys, however, the MIKE surveys usually covered the first 2 to 3 km of transect only in many cases. The transects started from the same point as the recce-transects, but went in a different direction. The layout of all of these surveys was essentially like spokes from a hub. This is not a configuration we would recommend in the future, but it did allow us to lay out all of the different test designs with a minimum of logistical costs, and most significantly to compare two data sets separated by 5 or 6 years. This was important since the Ituri site was the only one surveyed in an active area of political instability and insecurity, and the effects of civil war and political instability are potentially among the most important covariates of elephant poaching for the sub-region. The results of this trend analysis are not presented here, as analyses could not be completed in time for this report. In all cases (except one) field teams were able to re-locate the original transects.
4.3 Travel Recces.A simplified form of Recce protocol was used by field teams to collect data on elephant sign, elephant habitats and human activities as the teams moved across the forest between sampling locations. In essence travel recces can be considered "platforms of opportunity" (in the terminology of the cetacean surveys that use this term to describe opportunistic data collection by non survey vessels).
Travel recces are important to the MIKE, teams because they have to cover the ground, and are interested in recording observations and having these utilized as possible. The results of the travel recces are not analysed here, however we consider these results to be potentially useful in spatial modelling of elephant distribution and abundance and as a means of categorizing overall conditions of the sampling locations.
At this point it is clear that travel recce results will only be useful if they are collected with standardized protocols and a defined survey design. We also suggest that other platforms of opportunity, such as exploratory recces ("mega transects") and possibly some types of patrol surveys be treated as potential travel recces.
4.4 Field Forms and InstructionsField forms and instructions utilized during pilot surveys are presented in ANNEX 1. It is important to note that during the Pilot Project data were collected on a range of species and cues other than elephant dung and elephant poaching sign. This was done to allow field testing of methods and team performance and to permit different analyses, including evaluation of satellite imagery. We consider these field and reporting forms to be draft forms only, to be evaluated and modified as necessary in relation to survey objectives.
For the Central African forest zone, MIKE surveys will record observations on additional species than elephants, and must provide data on poaching sign and human impact. Many MIKE survey locations are remote and difficult to access, and likely to contain species other than elephants of high conservation importance. In addition, MIKE teams will usually be collaborating with other monitoring and survey programs, and will be required to provide information of broader conservation use. Nevertheless, we strongly recommend that MIKE surveys be limited to a few of the potentially important species (primarily apes, and possibly a select array of other large fauna in the forest zone). The addition of other species and cues will necessitate additional observers and costs that must be balanced against the need to cover all the required sampling locations within the constraints of available time and budget. These constraints and proposals for data collection during subsequent MIKE protocols are provided in the recommendations section, below.
In particular addition of Distance measures for sign of poaching and illegal elephant killing is strongly advised. Data on these cues were collected during the pilot project, but in a recce format, with perpendicular distances to the cues not measured.
The field forms developed for the Pilot Project were used as a template for data entry. In some cases these were used directly in the field but in other cases field teams modified them to fit their specific requirements, e.g. to fit in smaller-size waterproof field notebooks. In no case did this lead to required data not being collected. We believe that the next generation of field forms can be simplified, and that some protocols, such as habitat classification, might be modified following analyses of the current data set.
In Odzala, field teams used cybertracker to record data electronically (See ANNEX 3). Cybertracker was customized during the Pilot project for collecting recce transect data.
4.5 Field Data Reporting FormsCITES / MIKE Reporting forms developed during the pilot project are presented in ANNEX 2. These are spreadsheets on which field data is transcribed or entered in another way (e.g. transferred electronically in the case of cybertracker). These spreadsheets should eventually feed into a database and may need some further modifications for that purpose.
An alternative approach to data transmission is to develop a relational database with customized entry forms to make data entry easier and more reliable. A prototype of a MS Access database was developed during the Pilot Project and could be updated and adapted to a final reporting form approved by the MIKE Technical Advisory Group.
With Cybertracker it is possible to create links with other databases so that very little manipulation other than data checking may be necessary during the transcription stage (ANNEX 3). Any changes in the field forms will have to be implemented both in Cybertracker and the data reporting spreadsheets.
5. ResultsThe following sections provide an overview of problems encountered in field data collection. This is followed by a presentation of elephant dung densities calculated from MIKE surveys. The program Distance 3.5 was used to develop these results. An updated version, Distance 4.0 will soon be available. This new version has a built-in geographic information system (GIS), and allows survey designs to be generated automatically. In the future, it is hoped to further extend the facilities to allow spatial modelling. Lastly, the results of the evaluation of survey types is presented, and detailed in the Consultant technical reports provided in full in 4, 5, 6 and 7.
5.1 Problems Encountered with Survey Data during the Pilot ProjectThe recce-transect method as described above was generally successfully applied in the field. Trained and motivated field teams in general produced good quality field data. Nevertheless, there were a few problems with the data collection. Most of the problems detected in the pilot data could be traced to violations of the assumptions of the line transect method. The problems encountered are detailed below:
A common measurement error is that observers tend to record all objects within about 50-100 cm from the midline of the transect as on the line (a perpendicular measure of "0"). This may be due to a tendency to count all objects that fall in the area affected by cutting as "on the line" or having a distance of 0 (White et al. 2000). This produces a spike in the frequency histogram at 0 meters and will make modelling difficult. It is solved by stressing to field staff that measurements should be taken from the line to the centre of the dung pile and not to the nearest edge of the pile, or by measuring perpendicular distance to the closest and farthest edge of the dung pile and taking the average. Another common error is rounding of measurements to convenient numbers such as 0, 5, 10 etc. This leads to heaping of data, which is also noticeable on the frequency histogram.
In Odzala and Ituri too many observations were recorded as zero, producing a serious spike in the histogram that makes modelling difficult. For instance in Odzala one particular team recorded 18% of the dung piles as zero (see Figure 5). After this was discovered we changed the measuring procedure by measuring distance to the nearest and furthest edge of the dung pile and divided this number by two. By initiating this measuring protocol, we were able to improve measurement accuracy.
The reverse happened in Lope where people were over-concerned with zero recordings so there was some indication that objects located right on the transect line were given small positive distances to avoid recording them as zero.
Other problems in conducting inventories.
Other problems observed in the field were:
-
Deviating from the compass bearing resulting in a curved instead of a linear path.
Measuring distances in a slight angle from the line instead of perpendicular to the line.
Bending the tape during measurement or holding the tape measure at an angel instead of along a horizontal plane causing inaccuracies in measurement.
Finally, logistical constraints and other problems prevented some teams from surveying all of their planned itinerary and sampling locations.
Figure 5. Example of a spike at 0 m due to recording objects close to but not on the transect line as 0 m (Odzala dataset).
5.2 Recce-transect Survey ResultsThese results apply only to the study area and should not be extrapolated beyond these sites at this time. There were 6 sampling zones in the Lope study area, 3 in Odzala and a single zone in Ituri (See sampling design maps in annex 3 of report 1). The survey zones in Lope were designed to span the range of conditions in the area and should not be considered optimal design strata where the objective is to improve survey efficiency and reduce variation in encounter rates (see Report 1). The three survey zones in Odzala also represented different conditions in the region but with less variation within a particular zone than was the case for Lope.
Table 2 provides a summary of the recce-transect data from the three pilot sites. Besides total dung counts, encounter rates for transects and recces are also shown which makes comparison between sites possible.
Table 2. Summary of recce-transect data from pilot sites.
|
Site |
k |
nt 1 |
nr |
|
|
|
Odzala |
44 |
449 |
1565 |
10.20 |
8.89 |
|
Lope |
44 |
174 |
478 |
3.95 |
2.72 |
|
Ituri |
14 2 |
77 |
266 |
5.5 |
4.75 |
1
k = number of sampling locations, subscript t indicates
transect, r recce, n is total count and 
is
encounter rate
1 These values are before truncation. Table 2 shows counts and encounter rates for transect data after truncation at 4m.
2 22 locations were sampled in Ituri, but 8 did not follow the standard recce-transect sampling plan and were excluded from these analyses
Dung density estimates from the analysis of the transect part of the recce-transects data are given in Tables 3 and 4. Analysis was done on pooled data from all strata combined (Table 3) and on stratified data (Table 4). Encounter rates were estimated by stratum, but the detection function was estimated from pooled data. The stratified results show a slightly lower CV for both Odzala and Lope. Bootstrap estimates of variance are were not generated for the stratified analysis as some strata contain far fewer sampling locations than the 15-20 recommended for reliable bootstrap analysis.
Table 3. Estimates of encounter rate and density from analysis of transect data. Encounter rates for Odzala and Lope have been pooled across sampling zones (strata). Units for density are dung piles/hectare. CI is 95% parametric confidence interval; CIb is the 2.5th and 97.5th percentile from 999 bootstrap resamples, resampling locations.
|
Site |
k |
et |
CV(et) |
D |
CV(D) |
95% CI(D) |
95% CIb(D) |
|
Odzala |
44 |
9.5 |
14.5 |
25.6 |
15.6 |
(18.8, 35.0) |
(18.1, 38.8) |
|
Lope |
44 |
3.5 |
20.8 |
6.22 |
22.0 |
(4.0, 9.6) |
(4.3, 9.6) |
|
Ituri |
14 |
5.0 |
50.0 |
13.9 |
52.0 |
(4.9, 39.4) |
(4.5, 24.7) |
Table 4. Estimates of encounter rate and density from transect data. Encounter rate has been stratified by sampling zone for Odzala (3 zones) and Lope (6 zones). Notation for CI as for Table 3.
|
Site |
ks |
D |
CV(D) |
95% CI(D) |
95% CIb(D) |
|
Odzala |
13, 22, 9 |
23.8 |
13.9 |
(18.0, 31.5) |
(18.0, 32.9) |
|
Lope |
5, 3, 15, 4, 15, 2 |
4.5 |
21.1 |
(2.9, 7.0) |
(3.2, 6.7) |
|
Ituri |
14 |
13.9 |
52.0 |
(4.9, 39.4) |
(4.5, 24.7) |
Tables 5 and 6 show the results for each survey zone in respectively Lope and Odzala. At Lope, many of these estimates are based on very few sampling locations, and so should not be considered reliable (as can be seen from the CVs). At Odzala, there were high dung densities in zone 1 and 2 compared to a low density in zone 3. Zone 1 is inside the protected area far from villages and roads, zone 2 is protected and close to the park headquarters and zone 3 is unprotected and relatively close to roads and villages. (For a more detailed analysis of the spatial patterns in the Odzala data, see the spatial modeling analysis in Technical Report 4).
Table 5. Results from the transect analysis for Lope by survey zone
|
Stratum |
k |
L |
n |
et dung/km |
D dung/ ha |
95% CI (D) dung/ha |
CV (D) |
|
Ivindo |
5 |
5 |
16 |
3.2 |
5.72 |
2.83-11.54 |
27.8 |
|
Leledi |
3 |
3 |
17 |
5.7 |
10.14 |
0.59-174.34 |
74.05 |
|
Lastourville |
15 |
15 |
23 |
1.5 |
2.74 |
1.49-5.03 |
29.24 |
|
Offoue |
4 |
4 |
5 |
1.3 |
2.23 |
0.49-10.28 |
50.83 |
|
Lalara |
15 |
15 |
92 |
6.1 |
10.97 |
6.06-19.88 |
28.68 |
|
Iboundji |
2 |
2 |
0 |
0 |
k: number of sampling locations, l: total transect length, n: dung pile counts for transects, D: mean dung density per hectare, CI: confidence interval, CV: coefficient of variation.
Table 6. Results from the transect analysis for Odzala by survey zone
|
Stratum |
k |
L |
n |
et dung/km |
D dung/ ha |
95% CI (D) dung/ha |
CV (D) |
||||
|
1 (North) |
13 |
13 |
145 |
11.154 |
30.10 |
19.46-46.58 |
20.56 |
||||
|
2 (Middle) |
22 |
22 |
267 |
12.14 |
32.75 |
22.76- 47.14 |
17.86 |
||||
|
3 (South) |
9 |
9 |
6 |
0.67 |
1.79 |
0.80-4.01 |
35.84 |
||||
k: number of sampling locations, l: total transect length, n: dung pile counts for transects, D: mean dung density per hectare, CI: confidence interval, CV: coefficient of variation.
5.3 Combining Transects and Recces
In principal, the recce data that was collected at each location between the sections of transect can be used to improve the precision of the encounter rate estimates. However, encounter rates on recces and transects are unlikely to be the same: recces are more likely to follow elephant trails, or human trails, and so have higher or lower encounter rates; on the other hand observers move more slowly along transects so are less likely to miss dung piles. Therefore, before the recce data can be used, we must correct for any difference in encounter rate between recces and transects.
This is done by estimating a correction factor, 
,
as the ratio of encounter rate on the transects with encounter rate on a subset
of the recces. This ratio is then used to calibrate the remaining recce data.
A combined adjusted encounter rate, 
can then be calculated, using both the transect data and the calibrated recce
data. From this combined density estimates can be calculated. In the analysis
of the Pilot Project data two methods were used to calculate the ratio estimator.
Buckland and Underwood (2000, unpublished) used the data of the entire recce
to estimate while Thomas and Buckland
(2001, unpublished) used the 200 m of recce data closest to each 200 m section
of transect. Details of both methods and calculations are explained in the
respective reports in annex 4 and 5. The first method may give most precision
for estimates of density at each sampling location, and so may be best for
future spatial modeling of the data (although this has yet to be tested).
The latter gives best precision for design-based estimates of density at the
level of the study site, and is the method reported here. Results of the density
estimates for transect only and combined adjusted recce-transect data are
presented in table 7.
Table 7. Comparison of density estimates calculated using transect data only and transect and calibrated recce data.
|
Site |
|
|
|
|
|
Odzala |
25.64 |
15.63 |
24.53 |
15.90 |
|
Lope |
6.22 |
21.99 |
6.24 |
17.15 |
|
Ituri |
13.89 |
51.97 |
12.30 |
37.31 |
Dt : Density on transects; Dc: Density on calibrated recces.
As expected, the adjusted estimates of densities are similar to the estimate
from transect data alone. For Odzala, the CV is also almost identical. For
Lope and Ituri, however, the adjusted estimate is substantially more precise
(
is 22% more precise than 
at Lope and 28% more precise at Ituri).
Thus supplementing transects with recces improved precision in Lope and Ituri but not in Odzala. Dung counts within sampling locations at Odzala were strongly spatially auto-correlated, and the correlation decreased slowly with the separation of the counts. This meant that the extra information from the sections of recce was effectively duplicating information already gained from the transect sections. At Lope and Ituri, however, the auto-correlation decreased rapidly with distance along the line so that the recce sections provided new information. See annex 5 for a detailed explanation of the results.
5.4 Predicting Variation in Encounter Rate at Pilot Sites
The objective of this analysis was to evaluate the recce-transect design in order to make recommendations for an optimal design at the scale of the sampling location.
We looked at spatial autocorrelation in encounter rates between segments on the recce-transect. Spatial autocorrelation means that counts of nearby segments on the recce transect are correlated. Because of this, the gain in precision for a given transect line is lower than would be expected without spatial autocorrelation. Therefore it is often better to have shorter segments of transect spaced further apart than one continuous line. The details of the analysis can be found in annex 6.
The results indicate some advantage of a recce transect as opposed to a transect only design. Principle results of this analysis are summarized in the bullets below:
-
At Odzala counts on nearby segments are highly correlated and this correlation decreases significantly over the space of 1km. A continuous transect of 4 km would be less precise than the 5 X 200 m transect lines separated by 1 km from the Pilot design, even when no recce information was collected. Given that the segments are spaced out, there is no great advantage to collecting and analyzing recce-data in-between the transect segments. Because collecting data on recces slows teams down we would, in this case, recommend short transects 1 km apart without the recces.
At Lope and Ituri, between 2 and 3 kilometers of continuous transect gives about the same precision as the transect-only portions of the pilot design (i.e. 200m of transect, separated by 1km, with no recce information). This is because the correlation between counts on nearby segments of transect is either lower to begin with (in Lope) or doesn’t decrease as much (in Ituri) over the first 1km. (It is correlation of nearby segments that is the most important for determining the overall variance in counts) However, when the recce data from the pilot study are included with the transect data, the recce-transect design is more precise than 2-3 kilometers of continuous transect. Based on preliminary data of time it takes observers to conduct surveys, a field team could expect to cover between 2.6 and 3.35 kilometers of continuous transect in the same time as the 5km recce-transect from the Pilot project.
Recce transects conducted in the pilot project were essentially one-off efforts. There was no attempt to relocate and resurvey these same recce transects at another time. Nevertheless, the purpose of MIKE is to provide survey data over time, through a series of survey cycles. The recce transect configuration used in the pilot project was readily executed by field teams, however, it is not clear that field teams would be able to return to exactly the same recce transect line on subsequent survey cycles. This is because the recce portions of the recce transect unit are not constrained by a fixed compass direction. Thus the beginning point of each transect and recce section, after the first ones, will not necessarily be in the same location.
Given the autocorrelation effect detected on the pilot recce transects, the impact of not following the same travel line between successive surveys may not have a large impact on the precision of the estimate. Further analysis of the pilot data may shed light on this.
Transects with a well-marked starting point and followed correctly on a well controlled compass bearing can be repeated in subsequent survey cycles. This was proven by the Ituri teams who were able to locate and resurvey the exact same transect lines in 2000 that were first laid out five years before.
In principle, with current GPS technology it should also be possible for any well marked transect to be relocated and followed each survey cycle. However in practice it will not be possible for teams to precisely mark and relocate every transect in the forest, and they will incur costs in time for each point they have to precisely relocate. Thus a protocol that has fewer sampling points that must be precisely relocated for each survey cycle is more efficient, and more likely to be correctly executed.
Some compromise must be reached to optimize design at the scale of the sampling location. The pilot results suggest that a design that ensures more, shorter transects, widely separated is a better configuration than a design with fewer, longer lines, or lines closely configured. Although further evaluation of this problem is possible, the pilot results indicate that the best solution will be for teams to start each transect segment from a marked location that can be relocated in each survey cycle.
5.6 Transect Survey Results at IturiOnly a subset of the 2000 data was available at the time of the analysis. Only the 1993-95 transects for which 2000 data was available were included in this analysis. The analysis and results are described in more detail in annex 7.
For the 1995 data, estimated density was 5.59 dung piles/hectare, with CV of 28.7%, 95% parametric confidence limits 2.98, 10.46 and 95% bootstrap confidence limits of 2.49, 8.19. For the 2000 data, estimated density was 6.67 dung piles/hectare, with CV of 16.6%, 95% parametric confidence limits 4.64, 9.57 and 95% bootstrap confidence limits of 4.87 and 8.95. From these results, it is clear that there has been no significant change in dung density at these sampling locations between time periods. A simple z-test of the statistical significance of the difference gives z = 0.567, p=0.57.
It is important not to over-interpret these results, for reasons outlined in annex 7. It is prudent, for now, to treat them as preliminary.
6. Conclusions and RecommendationsFurther analyses of pilot survey data are still possible, and some analyses may help resolve remaining unsettled questions. Nevertheless, the results of the pilot project allow us to make the following conclusions and recommendations for the next phase of MIKE.
6.1 Standard Transect ConfigurationFor the proposed extensive sample program, a consistent configuration should be used across all sampling locations. Based on the pilot data, it would appear that a recce-transect design is better than transects only for estimating density within one time period. However, transect-only designs enable exactly the same lines to be covered during each sampling cycle, and field protocols are easier to implement. In addition, line transect data provides additional information (measures of perpendicular distances) that provide important insights into the quality of data collection and observation. With single line transects it is also feasible to follow the ideal line – a major advantage as failure to perform transects on the ideal line in a recce-transect design can bias the estimates.
Based on results from three sites, we propose a standard design at each sampling location consisting of 4 transects of 1 km each spaced 3 km apart (at which distance spatial autocorrelation between counts is very low). The added precision gained by increasing total transect length above 3-4 km at any given sampling location is not appreciable and should not be done at the expense of increasing sampling location density.
An example of a transect configuration is given in Figure 6.
A recce could be performed on the 3 km stretch between the transects. A recce-transect ratio estimator could be calculated using the data from the 1Km segments of the recces closest to the transects This could then be used as a calibration factor for recce data taken on travels between sampling locations. The latter are useful for spatial modelling of dung density estimates. It would also be possible to calculate combined adjusted encounter rates for transects and recces at the sampling site level, but it is not certain that this would improve precision substantially. After analysis of the first few datasets, the design could be re-evaluated and it can be decided whether to include recces or not. We recommend that detailed records be kept of the time taken to perform transects, recces and non-survey travel so that this decision can be made.
Each transect should have known starting points (documented by GPS), to permit survey teams to return to the same sampling locations in subsequent survey cycles, and to accommodate bias of transects not being located along the ideal transect line.
This same design can be modified to calibrate dung encounter rates on longer travel recces. In this case we would recommend one or more transects to be interspersed in a longer recce as time and resources permitted.
6.2 Field Protocols and Data Collection FormsField forms developed in the pilot project will produce the essential data for an analysis of elephant dung densities. These forms, especially the reporting forms, however, should be simplified. We recommend that simplified forms be developed from the current forms, following review by the MIKE Technical Advisory Group (TAG). Among the important factors affecting the final reporting forms will be the degree to which MIKE field teams will summarize and analyze their field results before submission to CITES. This issue has not been decided to our knowledge.:
For MIKE surveys, the following data should be collected:
-
Dung and direct observations of elephants should be recorded on transects and recces. Presence of elephant trails and tracks should only be noted (in the notes column) to indicate recent or past presence of elephants when no other sign (dung or animals) has been recorded on a particular recce or transect.
All observations of elephant carcasses should be recorded, including those found outside the sampling area. A carcass form should be filled out for each detected carcass and should include a GPS position. All other direct signs of elephant poaching or indirect signs that relate directly to elephant poaching should also be noted, whether on a recce-transect or not. Perpendicular distances to carcasses, and other direct sign of poaching, measured from the travel lines (line transect or recce) should be noted, as it may ultimately be possible to develop detection functions to estimate densities of these cues.
Important direct and indirect observations of elephants (e.g. forest clearings being used by elephants, observations of individuals or herds, etc.) should also be recorded and GPS positions taken where possible.
Habitat data are used to classify satellite imagery. It is important to have this information for the whole study area for spatial modelling (see report 4), however the pilot form may be simplified here. Habitat data can be taken less frequently than was the case during the Pilot Project, e.g. every 200 meters on a transect and every 1 km on recces. Taking GPS positions for habitat data is strongly desired.
-
Information on human sign can be restricted to variables related to illegal elephant killing, or broadened to include other variables related to human impact. The choice of what data to record should be made carefully (see above). Perpendicular distances should be taken for observations of direct poaching sign (snares, poacher camps). These data will be especially important for where unbiased assessments of the rates of illegal activities are desirable, and in particular to compare with more biased surveys of anti-poaching patrol reports (See Technical Report 3).
We recommend that MIKE surveys in the Central African forest region collect data on other selected large mammal sign. During the pilot surveys, field teams recorded observations on apes (nests) primates (direct observations), and selected large ungulates and cats. (indirect and direct observations), as well as observations of selected human activities (snares, poaching sign, etc, discussed above.). We recommend that this broader array of species be surveyed for the following reasons:
-
These observations give important information on the overall state of wildlife, mammal communities, hunting pressure etc. pertinent to illegal elephant killing and potentially useful for modelling exercises.
MIKE surveys are often collaborative with other partners, including government administrations and international NGOs, who may have broader objectives than those of MIKE These collaborations bring essential resources to the surveys, and the surveys must thus meet the needs of all partners.
MIKE surveys will visit some of the most remote areas remaining in the sub-region. Information on wildlife status and human impact is required if these areas are to be considered in a national conservation strategy (and some of these areas undoubtedly should be). It is not cost effective to send multiple surveys into these areas, but rather a single survey will have to provide the information desired for overall recommendations to national administrations.
The following conditions must be met if multi-target surveys are to provide good quality data:
-
For data collected on line transects, all perpendicular distances should be carefully measured. Distances to grouped data (nest ape nest groups) should be treated very carefully. We recommend that each ape nest group be mapped. Details for this can be provided.
If the number of species or categories of cues is to be increased, surveys will require an at least two observers. This increased effort incurs minimal costs in effort travelling to a sampling location and cutting a transect
Additional and species-specific quality controls will be necessary as additional species or other sign are added to the surveys to ensure that line transect assumptions are met.
We recommend that the broadening of survey targets be done with careful consideration of how the data will be used and analysed. It may not be possible to ensure the necessary inputs and logistical costs to add all the information desired, thus a process to prioritise data collection should be developed.
It is also important not to over-burden the field staff with too many responsibilities, as this will slow them down and reduce their ability to observe elephant dung on the transect line. We recommend that at least one observer on each team at any one time be 100% dedicated to looking for elephant dung (this role can rotate periodically to maintain motivation).
A satisfactory analysis of dung-pile density was obtained from data collected by pilot project field teams. Nevertheless there were some problems with data collection at all sites.
Most of the problems detailed in the sections above improved during the course of the pilot project, but our experience highlighted the need for consistent supervision of field work, regular independent evaluation of the data, and follow-up training sessions for field teams as necessary. Specific recommendations include:
-
It is useful to plot and inspect the data as it is collected during the course of the field work in order to discover measurement and detection problems.
For the same reason and to keep teams motivated we strongly recommend that field teams be involved in the analysis as much as possible and that basic analysis be done on-site. (However, given the relative complexity of distance analysis, final analysis of the data should be done by experienced Distance users.)
Field teams involved in the evaluation and analysis of the data, acquired a good understanding of the underlying assumptions of the method, and produced more reliable results. .
The use of electronic data collection (cybertrackers) may help ensure that data collection is maintained at a high level, however, teams should be able to revert to "pencil and paper" when necessary, and no field teams should rely entirely on cybertrackers if they do not know how to protect the data from corruption or loss. See cybertracker report in annex 3).
For the analysis we recommend the use of Distance 3.5 software, or the new Distance 4 when it is released. The Research Unit for Wildlife Population Assessment at the University of St Andrews produces and maintains the software. They provide continuing support for the software and conduct statistical research in further development of Distance sampling and spatial modelling. Another distance sampling analysis package, called Lopes, was created by the Wildlife Conservation Society (Walsh, 1999 ) intended for use by the field teams. This program is probably more suitable for on-site analysis by field teams than Distance. It is easier to use and requires less training than Distance but is much more limited in its analytical capabilities.
We stress that initial analysis of the data can be performed in the field, and should be done while data are still being gathered. Histograms of observed distances (stratified by observer) can be plotted on graph paper and used to diagnose any problems such as heaping at 0 distance. These problems can then be rectified before they compromise the whole survey program.
We recommend that distance sampling methods also be considered for savanna surveys, including aerial savanna counts, and for surveys of selcted human impact and poaching sign (see above) in both forest and savanna
6.5 Survey Cycles.This issue was not addressed directly by the pilot project. However, based on the observed efficiency of the pilot project teams, and taking into account feedback we received from the field teams concerning their needs for time with families, need for physical recovery after some of the surveys, and general health care issues, we provisionally recommend an overall survey cycle that includes 6 to 9 months of intensive inventory effort repeated every two to three years. We consider this recommendation tentative at this point, and contingent upon data requirements of the MIKE Subgroup.
6.6 Dung StudiesThis issue was discussed in Report 1. We reiterate principle points here and add additional recommendations.
We recommend that MIKE continue to use dung counts in forest surveys for the next phase of the MIKE program, but that other independent means of elephant censuses be investigated and developed.
Further studies of dung deposition rates by elephants should be initiated. Some possibilities for this exist in the sub-region, but will necessarily be the focus of a separate study.
Observations of dung disappearance rates should be initiated in each survey cycle, at selected sites at least, in order to calibrate variation in these rates at different sampling locations. Until these results are available, we recommend that for each survey cycle, dung counts be conducted at the same season at a given sampling location. This will help control for the seasonal effect on variance in dung disappearance rates. Nevertheless, factors affecting dung disappearance (decay, leaf fall, rainfall, etc.) should be studied and the process modelled to permit conversion of dung indices to elephant densities.
6.7 Integration of Other Data SourcesThe MIKE program could possibly use information from other sources. This may be especially useful as data are used for spatial modeling of elephant distributions in relation to illegal killing. Some of this data may come from reconnaissance surveys, or well executed anti-poaching patrols etc. These data will be most valuable if they are collected by well defined protocols, and if these can take into account the MIKE objectives.
It is important to note here, that collection of this ancillary data should not divert attention from well-designed field sampling. In addition, other monitoring data should be evaluated in relation to the controlled unbiased surveys we detail in this, and the first technical report. These studies are most likely to be achieved in established MIKE sites or other intensive study areas.
7. BibliographyBuckland, S.T., Anderson, D.R., Burnham, K.P., Laake, J.L., 1993. Distance sampling. Estimating abundance of biological populations. Chapman & Hall
Thomas, L., Laake, J.L., Derry, J.F., Buckland, S.T., Borchers, D.L., Anderson, D.R., Burnham, K.P., Strindberg, S., Hedley, S.L., Burt, M.L., Marques, F.F.C., Pollard, J.H. And Fewster, R.M. 1998. Distance 3.5. Research Unit for Wildlife Population Assessment, University of St. Andrews, UK. Available: http://www.ruwpa.st-and.ac.uk/distance/
Thomas, L., Laake, J.L., Strindberg, S., Marques, F., Borchers, D.L., Buckland, S.T., Anderson, D.R., Burnham, K.P., Hedley, S.L., and Pollard, J.H. 2001. Distance 4.0. Beta 1. Research Unit for Wildlife Population Assessment, University of St. Andrews, UK. http://www.ruwpa.st-and.ac.uk/distance/
Walsh, P.D. 1999, Lopes, a program for analysing transect and recce data. Wildlife Conservation Society, New York, U.S.A.
Walsh, P.D., White, L.J.T., Mbina, C., Idiata, D., Mihindou, Y., Maisels, F., Thibault, M. 2001. Estimates of forest elephant abundance: projecting the relationship between precision and effort. Journal of Applied Ecology, 38 (1), 217-
White, L., Edwards, A. eds.. 2000. Conservation research in the African rain forests: a technical handbook. Wildlife Conservation Society, New York. 444 pp., many illustrations.
8. AnnexesANNEX 1: Field data forms.
ANNEX 2: Data reporting forms.
ANNEX 3: Cybertracker as a data collection tool for MIKE.
ANNEX 4: Analysis of data and survey design for the Mike Central Africa Pilot Project: first interim report. (S.T. Buckland and F.M. Underwood).
ANNEX 5: Analysis of data and survey design for the MIKE central African pilot project. Third Report – Part 1: Analysis of Pilot Data (L. Thomas and S.T Buckland)
ANNEX 6: Predicting variation in encounter rates at pilot sites. (L. Thomas).
ANNEX 7: Analysis of data and survey design for the MIKE central African pilot project. Third Report – Part 3: Analysis of Ituri Transects (L. Thomas)