Elephant poaching in Niassa Reserve, Mozambique:population impact revealed by combined surveytrends for live elephants and carcasses

V E R N O N R . B O O T H and K E V I N M . D U N H A M

Abstract Trends in the populations of large herbivores inNiassa Reserve, Mozambique, have beenmonitored throughbiennial aerial surveys since . The elephant Loxodontaafricana population has been subjected to intensive illegalhunting since . We used a simple population modelto mimic the observed trends in the numbers of live anddead elephants to demonstrate the impact of poaching.The number of fresh or recent carcasses recorded wasused in the model as an index of the annual mortalityrate. Amaximum likelihood analysis to compare populationmodels revealed that the best fit to the survey estimatesof both live elephants and old or very old carcasses wasa model that started with , elephants in . Thisnumber increased through births by .% annually anddecreased through deaths from natural and anthropogeniccauses. In the best-fit model, the mean mortality rate inany year was . times the observed + carcass ratio(ratio for carcasses in age categories and ), and carcassesremained visible for a mean of years. The model suggestedthat c. elephants were poached during – andanother c. , during . Population estimates for liveelephants and carcasses are now routine outcomes of aerialsurveys conducted as part of the CITES programme forMonitoring the Illegal Killing of Elephants, and our methodcan be applied to any population with a time series of esti-mates for live and dead elephants.

Keywords Aerial survey, carcass, elephant, Mozambique,Niassa, poaching, population model, population trend

This paper contains supplementary material that can befound online at http://journals.cambridge.org

Introduction

There have been notable increases in both the poachingof African elephants Loxodonta africana and the illegal

trade in ivory since , mainly because of the risingdemand for ivory in the fast-growing economies of Asia,

particularly China and Thailand (UNEP et al., ).Elephant populations in Eastern and Southern Africa thatwere regarded as secure are now threatened, as a wave ofpoaching seems to be spreading east and southwards acrossAfrica. The Monitoring the Illegal Killing of Elephants(MIKE) programme (CITES, ) collects data on ele-phant poaching and law enforcement from. sites acrossAfrica (Burn et al., ). However, this programme dependson the law enforcement authority at each site collecting andreporting the carcass data used to judge the effectiveness oftheir law enforcement. Aerial surveys can provide both anindependent check and an assessment of elephant popu-lation trends over a large area. The MIKE standards for aer-ial surveys of elephant populations (Craig, ) require thecollection of data on both live elephants and elephantcarcasses.

Douglas-Hamilton & Hillman () were among thefirst to draw attention to the importance of carcass numberas an indication of the trend in a population of live ele-phants. Later, Douglas-Hamilton & Burrill () showedthat the proportion of dead elephants was correlated withthe percentage change in the number of live elephants dur-ing the previous years. Wells () modelled a theoreticalelephant population under different mortality regimes. Hesuggested that temporal patterns in the number of dead ele-phants (as a proportion of live plus dead elephants) permit-ted natural mortality alone to be distinguished frommortality from both hunting and natural causes, and forselective hunting to be distinguished from unselectivehunting.

Here we describe a method that uses a simple populationmodel to reveal the trend in an elephant population by com-bining survey estimates for live elephants with survey esti-mates for elephant carcasses. A previous model of anelephant population (Dunham, ) is extended, firstlyby varying the annual mortality rate in the model to reflectthe observed variation in the number of fresh or recent car-casses of elephants, and secondly by mimicking the ob-served variations in both the number of live elephants andthe number of old or very old carcasses. We illustrate themethod using long-term aerial survey data for the savannahelephant population in the Niassa National Reserve innorthern Mozambique.

The Niassa population is the largest elephant populationin Mozambique and the only one that has been surveyedregularly in recent years. The trend in the estimated number

VERNON R. BOOTH (Corresponding author) 9 Glenara Avenue North, Highlands,Harare, Zimbabwe. E-mail vernonrbooth@gmail.com

KEVIN M. DUNHAM PO Box CH385, Chisipite, Harare, Zimbabwe

Received May . Revision requested June .Accepted July .

© 2014 Fauna & Flora International, Oryx, Page 1 of 10 doi:10.1017/S0030605314000568

of live elephants in the Reserve has not been clear (Craig,), however, with the sample survey estimates suggestingthat the number of live elephants increased from c. ,during to c. , during , before declining toc. , during (Table ). Real changes of this magni-tude over just a few years are unlikely. However, it is not un-usual for the trend in population number to be unclear,especially if the population estimates are derived from arelatively small number of sample surveys. We determinethe likely trend in the Niassa elephant population by usinga model that mimics both the number of live elephants andthe number of carcasses. In the process, our analysis revealsthe recent impact of illegal hunting on the elephantpopulation.

Study area

Niassa National Reserve (c. , km) is in northernmostMozambique, covering part of Cabo Delgado Province andnearly one-third of Niassa Province (Fig. ). It is the coun-try’s largest conservation area and the third largest protectedarea in Africa, and has the highest concentration of wildlifein Mozambique. Tanzania’s Selous Game Reserve (c. ,km) lies km north of the Rovuma River, which formsthe international border as well as the northern border ofNiassa National Reserve. Approximately , km ofcommunal land in southern Tanzania, lying between theNiassa and Selous reserves, is designated the Selous–Niassa Corridor (Baldus & Hahn, ). Niassa has a trop-ical climate, with mean monthly temperatures of c. °Cduring October and November, falling to c. °C duringthe cold season (June–August). Rainfall is restricted to thehot season, beginning in late October and ending duringApril or May. Altitude is lowest in the east ( m) at theconfluence of the Rovuma and Lugenda Rivers, wheremean annual rainfall is c. mm. The ground rises steadilyto ,m in the west, where annual rainfall is c. ,mm.

The Reserve is noted for spectacular granite inselbergsthat stand above the surrounding landscape. Drainage isdominated by the Rovuma and Lugenda Rivers, which arelarge, braided, sand rivers with strong perennial flows. Acentral watershed between these rivers feeds an extensivenetwork of seasonally inundated wetlands or dambos aswell as numerous seasonal rivers lined with dense riparianwoodland. Most of the Reserve is covered by miombo(Brachystegia) woodland interspersed with drier areas ofbushed savannah. Fires started deliberately by people, andfuelled by m high grass, burn annually during the latedry season (August–October; Ribeiro et al., ).

Niassa National Reserve was created in but aban-doned after hostilities started in . After the nationalpeace accord the Government of Mozambique enteredinto an arrangement with a private company to manage T

ABLE1Summaryoftheresults

ofaerialsamplesurveysofelephantsL

oxodontaafricana

andelephant

carcassesinNiassaNationalR

eserve

inno

rthern

Mozam

biqu

e(Fig.).The

carcassesare

categorizedaccordingto

age,as

describedin

theMetho

ds.

Year

Live

elephant

Carcass

1Carcass

2Carcass

3Carcass

4Carcass

3+4

Source

Estim

ate

Variance

Estim

ate

Variance

Estim

ate

Variance

Estim

ate

Variance

Estim

ate

Variance

Estim

ate

Variance

1998

8,707

937,556

14172

00

336

4,204

Gibson(1998)

2000

11,828

1,144,751

663,710

27225

644

7,643

Gibson(2000)

2002

13,061

1,540,443

1085

30484

81722

524

4,792

605

5,514

Craig&Gibson(2002)

2004

12,477

1,160,375

00

00

156

1,715

305

3,293

461

5,008

Craig&Gibson(2004)

2006

11,833

1,326,122

00

00

266

2,890

322

3,333

588

6,223

Craig(2006)

2009

20,364

1,929,691

43623

971,015

400

4,562

356

4,560

756

9,122

Craig(2009)

2011

11,971

1,662,047

1092

252

2,549

281

2,645

2,084

27,419

2,365

30,064

Craig(2011)

2 V. R. Booth and K. M. Dunham

© 2014 Fauna & Flora International, Oryx, 1–10

the Reserve. Established during , the Sociedade paraa Gestão e Desenvolvimento da Reserva do Niassa(SGDRN) was an innovative partnership between the publicand private sectors, with the Government of Mozambiqueretaining ownership of % while providing SGDRN withexclusive management and development rights. Duringthe last years SGDRN undertook basic research and de-veloped the Reserve by securing private investment intourism.

The – management plan divided the Reserveinto management units: nine hunting blocks, six photo-tourism blocks and two zones of high biodiversity value(SGDRN, ). The presence within the Reserve of nineadministrative districts, three towns and . villages,which support ., people, introduced an unusual di-mension to management of the Reserve. The livelihoodsof these communities are centred on crop production,with shifting agriculture being the norm. These communi-ties face considerable challenges in terms of their remote-ness, limited access to physical and social infrastructure,and widespread poverty; meanwhile continued clearing ofnatural vegetation for agriculture is a significant threat tothe conservation goals of Niassa (Cunliffe et al., ;Mandondo & Ngarivhume, ).

Niassa is a study site for the MIKE programme (CITES,). Each year – elephant bulls are shot legally by sporthunters in the Reserve. In addition, – elephants are shotannually during problem-animal control (Marufo et al.,). To the south and east of the Reserve lie communallyoccupied areas that still support wildlife, including ele-phants, but for which there are no formal managementarrangements or plans.

Methods

Aerial surveys of wildlife

Stratified systematic transect sampling of the wildlife inNiassa National Reserve was undertaken during ,, , , , and (Gibson, ,; Craig & Gibson, , ; Craig, , , )using the methods described by Norton-Griffiths ().One light aircraft (Cessna or ) equipped with aradar altimeter and a Global Positioning System (GPS)receiver was used for each survey. The surveys were conduc-ted during October–November and took c. days to coverthe , km study area. The location of the first transect

Quirimbas National Park

IndianOcean

FIG. 1 The location of Niassa NationalReserve in northern Mozambique.

Elephant poaching in Mozambique 3

© 2014 Fauna & Flora International, Oryx, 1–10

in each stratumwas determined randomly. Maps of the stra-ta boundaries and transects were uploaded to a portableGPS and used for accurate navigation along transects. Anominal height of ft ( m) above ground level wasmaintained using the radar altimeter.

The boundaries of search strips on each side of the air-craft were defined by pairs of rods attached to the aircraftlift struts. The width of the strips was measured empiricallyby flying at various heights at right angles across an airstripon which large-sized numbers had been painted at m in-tervals. Each observer called out the outermost and inner-most numbers seen within his strip. The rods wereadjusted so that the difference between the numbers setthe calibrated strip width at c. m on each side when ata height of ft. The nominal sampling intensity wasc. % but the mean sampling intensity was .% (n = ,range .–.).

To achieve a search rate of c. minute per km the aircraftwas flown along transects at< km per hour. Two experi-enced observers seated behind the pilot and recorder calledout sightings of animals within the search strips. The re-corder, seated next to the pilot, recorded these sightings ofanimals and carcasses, noting the species, number and lo-cation in degrees and decimal minutes from a secondGPS. The height as indicated by the radar altimeter wasnoted at -s intervals, to allow the calculation of themean height for each transect. The time at which the flightsalong each transect started and ended was also recorded toprovide mean speed.

Elephant carcasses

The observers noted the approximate time since death foreach elephant carcass and allocated each carcass to an agecategory. During – three categories were used: ,fresh; , recent; , old (Douglas-Hamilton & Hillman,). Since the previous category has been replacedwith a new category representing old carcasses and a newcategory for very old carcasses. The new categorization isthat recommended for elephant surveys in MIKE sites(Craig, ). Generally, carcasses in categories and

were of elephants that had died during the year of the survey(mostly since the end of the rainy season) and carcasses incategories and were of elephants that had died duringearlier years or possibly during the previous rainy season.

The survey reports gave the carcass ratio sensuDouglas-Hamilton & Burrill (; although it is actually aproportion or percentage, not a ratio), which was the esti-mated number of all elephant carcasses expressed as a per-centage of the estimated number of all elephants (i.e. aliveplus dead). This all-carcass ratio is an index of the elephantmortality rate during the several years prior to the survey.We calculated the carcass ratio for categories + , whichis an index of the elephant mortality rate during the year

of the survey (including all causes, both natural and anthro-pogenic). The carcass ratio for categories + is defined asthe estimated number of elephant carcasses in age category or , expressed as a percentage of the sum of this numberand the estimated number of live elephants.

Data analysis

The population estimates and variances that we extractedfrom the survey reports had been calculated using method

of Jolly (). We calculated the % confidence limits forall mean estimates of population number as: populationestimate ± (tv.

√total variance), where t = student’s t, and v =

degrees of freedom estimated by Satterthwaite’s rule(Snedecor & Cochran, ; Gasaway et al., ). If, for anysurvey, the calculated lower confidence limit was less than thenumber of individuals seen in the search strips, this latternumber was substituted for the lower confidence limit.

Trends in elephant number were determined assumingthat an exponential model was appropriate for estimatingthe rate of population change. The exponential rate of popu-lation change per annum (r) was calculated using themethodof Gasaway et al. () based on weighted regression of nat-ural logarithms of the population estimates against time, withthe variance of r based on the sampling variances of the popu-lation estimates. The percentage rate of population change perannum was × (er− ). The population was considered tohave increased significantly over a time period if both lowerand upper % confidence limits of r were positive.

Population model

The numbers of live elephants and elephant carcasses inNiassa were mimicked with a model that started with an as-sumed population number. The number of elephants dur-ing the following year was then determined by increasingthe starting population by an assumed birth rate anddecreasing it by an assumed mortality rate. Thus:

Nt = (1+ b).(1−mt−1).Nt−1

where Nt = number of elephants in year t; b =mean birthrate of population per year (i.e. the mean proportion of ele-phants in the population that produce a calf during theyear), assumed to be constant; and mt− =mean mortalityrate of elephants of both sexes and all ages, as a proportionof the population, in year t− . Consequently, (mt−.Nt−) isthe number of elephants that died during year t− and thisnumber is included in the equation to determine the num-ber of elephant carcasses (see below).

The + carcass ratio provides an index of the mortalityrate during the survey year. Carcasses in age categories andhavemost of the skin still present.Hence for surveys conduc-ted at the end of the dry seasonmost if not all elephants in car-cass categories and had probably died since the end of the

4 V. R. Booth and K. M. Dunham

© 2014 Fauna & Flora International, Oryx, 1–10

last rains, because the skinwouldhave decomposedduring thewet season. The carcasses of some elephants that died after therains may have lost their skins and thus been included incategory or, if the bones were scattered, category . Thus,whereasmost (if not all) category and carcasses were of ele-phants that died after the rains, perhaps not all elephants thatdied during the survey year were in categories or .

The number of live elephants plus the number of cate-gory + carcasses is the number of elephants that werealive at the end of the rainy season. The number of thesethat had died by the time of the late dry-season surveywas the number of category + carcasses. No category

or carcasses were seen during the or surveys(Table ) but, given the large size of the population, someelephants must have died during these years and so thezero estimates probably represent sampling error. During, and the lower confidence limit of the esti-mated number of category or carcasses was close to zeroand therefore similar to the zero estimates of and .Hence, we assume that the mortality rate was low duringthese years and that the best estimate of mortality rate wasprovided by the mean of the values for the + carcass ratiofrom the five surveys during –. Both the mortalityrate and the number of elephants dying increased dramat-ically during the period after the survey.

In the model population, the mortality rate was calcu-lated as

mt = f .[1+ 2 carcass ratio]t/100where f = constant, in effect a correction factor to allow forthe fact that the carcasses of elephants that died during therainy season early in the year might have lost their skin andthus be considered category carcasses.

Category and carcasses remain for several years untilthe bones decompose or are scattered such that the carcassesare no longer recognizable to aerial observers. It is notknown for how many years elephant carcasses remain rec-ognizable and so this period is the fourth unknown to beincluded in the model: c is defined as the mean durationof the time period (in years) between the death of an ele-phant and the disappearance of its carcass. Hence, the numberof carcasses in any year is the sum of the numbers of ele-phants that died during the last c years.

In the model, the number of category + carcasses wascalculated as the total number of elephants that died duringthe last c years (i.e. the current year and c × previous years)minus the number of elephants that died during the currentyear and the carcasses of which are still at the category or stage. Thus,

Dt = [(mt.Nt) + (mt−1.Nt−1) + (mt−2.Nt−2)+ (mt−3.Nt−3) . . .+ (mt−(c−1).Nt−(c−1))]− [Nt .(1+ 2 carcass ratio)t/100]

where Dt = number of category + carcasses present inyear t.

The numbers of both live elephants and carcasses pre-dicted by the model were compared to the survey estimatesof elephant and carcass numbers, using Martin’s ()maximum likelihood estimator (Dunham, ). For eachyear during which a population was surveyed the analysiscompares the population number predicted by the modelto the survey estimate. It does this by relating the differencebetween the predicted and estimated numbers to the vari-ance of the mean estimate of population. The outcome ofthis analysis is an estimator that equals one if the predictedand estimated numbers are identical during all survey years,and declines towards zero the less perfect the fit between thepredicted and estimated numbers. When several modelsare compared, the one that gives the greatest index valueis the model that provides the best fit to the survey estimates.

The model was run for –. The start year was to calculate the number of elephants that died during the years preceding the survey. The Microsoft Excel add-in Solver was used to determine the values of the start-ing population during (N) and the constants b(mean birth rate), f (factor relating + carcass ratio tomortality rate) and c (mean number of years that carcassesremain recognizable) that maximized the index value.Executing Solver was equivalent to running a series of mod-els with different combinations of N, b, f and c. In themodel the + carcass ratio was taken as . (the mean ob-served ratio during the – surveys) until , afterwhich it increased at an exponential rate of . annually(as calculated from a value of . during and theobserved values during and ).

Preliminary analyses with the evolutionary solving meth-od in Solver revealed that Solver would run only with con-straints attached to the four unknowns in the model. Foreach unknown we deliberately selected minimum and max-imum values that were outside the range of values that weconsidered likely for the Niassa population. We reasonedthat this would allow Solver to run but would have little in-fluence on its final choice of the best-fit values for the un-knowns. The unknowns in the model were constrained asfollows: N as an integer, –,; b, –.; f, .–; and c as an integer, –.

The initial starting values of the unknowns were varied,with three possible starting values for each unknown: thevalue of theminimum constraint, the value of the maximumconstraint, and the mean of these two values. Hence, thepossible starting values for N were , , and,; for b were , . and .; for f were ., .and ; and for c were , and . Thus, there were com-binations of starting values and we ran Solver for eachcombination (scenario). For non-smooth optimizationproblems, Solver’s evolutionary solving method uses algo-rithms to seek a ‘good’ solution, not a solution that can be

Elephant poaching in Mozambique 5

© 2014 Fauna & Flora International, Oryx, 1–10

proved to be optimal (Frontline Systems, ). The evo-lutionary solving method differs from classical optimizationmethods in several ways, including its use of random sam-pling. Consequently, it may yield different solutions ondifferent runs.

Approximately % (/) of the scenarios produced thesame best-fit values for the unknowns, regardless of the start-ing values (Supplementary Table S). Another % ofscenarios produced similar best-fit values (e.g. N pre-dicted to be ,, , or , instead of ,). Solverruns that predicted very different valueswere always a poorerfit to the survey estimates (as shown by very small index va-lues) and these scenarios were rerun, sometimes more thanonce. The analysis revealed that the variation in the best-fitvalues between the first run for each scenario was not attrib-utable to the starting values for the unknowns but was prob-ably a result of the random element involved in Solver’sevolutionary solving method.

Themodelling exercise was repeated to determine the va-lues for N, b and f that produced a trend that was the bestfit to the observed numbers of live elephants only. The fitbetween the modelled and observed numbers of carcasseswas noted but no attempt was made to fit the modelledtrend to the observed numbers.

Rainfall

Temporal trends in annual rainfall over the Reserve were de-termined using the African rainfall estimation algorithm(Herman et al., ), which provides satellite-derived andrain-gauge-coupled rainfall estimates for southern Africa.Decadal data for July –June inclusive were down-loaded from the Famine Early Warning System website(FEWS NET, ) and summed to provide rainfall esti-mates for each July–June climate year. For each climateyear, mean rainfall was calculated for the pixels of × km that covered Niassa National Reserve.

Fire

Interannual variation in the proportion of the Reserve thatwas burnt during survey years was determined usingMODIS (Moderate Resolution Imaging Spectroradiometer)burnt area data. For each survey year from we calculatedthe total area of all × mpixels lyingwithin theReservein which fires were recorded (i.e. pixels with a QualityAssessment of or ) during May–October inclusive, andhence the percentage of the Reserve that was burnt. MODISdata were not available for . During , , and the percentage of the Reserve burnt was also re-corded during aerial surveys of the wildlife (Craig & Gibson,, ; Craig, , ). The percentage burnt duringother survey years was not reported.

Results

The number of elephants estimated by the surveys variedfrom , during to , during (Fig. a,Table ). The observed annual exponential rate of increase(r) was .–.% (% CI .–.) during –. However, the estimate was particularly high;when only the – data were considered, observedr was .–.% per annum (% CI .–.).That the lower confidence limit is just above zero suggeststhat r may not be significantly different from zero.The number of fresh and recent elephant carcasses (agecategories and ) estimated by the surveys increased some-time after (Fig. b) and the number of old and very oldcarcasses (age categories and ) more than tripled between and (Fig. c, Table ).

(a)

(b)

(c)

FIG. 2 (a) Number of live elephants Loxodonta africana, (b)number of elephant carcasses in age categories and , and (c)number of elephant carcasses in age categories and , based onaerial sample surveys of Niassa National Reserve, Mozambique(Fig. ) during –. Vertical lines indicate % confidenceintervals of estimates. The solid line in (b) is the assumed trendline for the + carcass ratio that was used in the populationmodel. Data sources are provided in the text.

6 V. R. Booth and K. M. Dunham

© 2014 Fauna & Flora International, Oryx, 1–10

The maximum likelihood analysis to compare popu-lation models for – revealed that the best fit to theobserved elephant and carcass numbers was provided by amodel population of , elephants during , which in-creased by births at .% annually and decreased at an an-nual mortality rate that was . times the + carcass ratio,with the carcasses of elephants remaining visible for a meanof years after death (Fig. ). Generally, there was little vari-ation in these numbers, regardless of the starting values usedwhen running Solver (Supplementary Table S).

When the modelling exercise was repeated to find themodel that provided a trend line that was the best fit to the ob-served number of live elephants only (ignoring the fit to carcassnumber), the model population started with , elephantsduring , increased by births at .%annually and decreasedat an annual mortality rate that was . times the + carcassratio (Supplementary Table S). This trend line (Fig. a) was abetter fit to live elephant numbers than the trend producedwhen fitting a model to both live elephants and carcasses(Fig. a) but it did not fit the observed number of carcasses,over-predicting them during and , even if they re-mained visible for a mean of just years after death (Fig. b).

The trend of the modelled population that gave the bestfit to the survey estimates of both elephant and carcass num-bers is likely to be a reasonable representation of the trend

exhibited by the elephant population in Niassa since .This trend line suggested that the population increased dur-ing – but then levelled off or declined during–, as the mortality rate increased dramaticallyafter . The population model suggested that the survey overestimated elephant number but not carcassnumber.

The surveys revealed a large increase in the mortality rateof elephants (as indexed by the + carcass ratio) during– and this was reflected in an increase of c. ,old or very old carcasses observed during . In the modelthat provided the best fit to numbers of both live and deadelephants, , elephants died during –. If themortality rate had not increased sometime after themodel suggests that just would have died during theseyears, and therefore an additional , elephants died dur-ing – as a consequence of the increased mortalityrate. This increase was exponential and the samemodel sug-gests that , of these additional deaths occurred during.

Mean annual rainfall over the Reserve was mm dur-ing – (Table ). Rainfall was well below the meanduring only climate years (–, mm; –, mm). The MODIS burnt area product suggestedthe percentage of the Reserve burnt annually increased from

(a)

(b)

FIG. 4 Outputs of a model providing best fit only to the numbersof live elephants: (a) the number of live elephants; (b) thenumber of elephant carcasses in age categories and . Themodelled population initially comprised , elephants during, which increased by births at .% annually and decreased atan annual mortality rate . times the + carcass ratio(expressed as a proportion). The trend line for carcasses assumesthat the carcasses remained visible for a mean of only yearsafter death. Data points indicate numbers estimated during thesurveys, with vertical lines indicating upper and lower %confidence intervals of estimates. Sources for survey data areprovided in the text.

(a)

(b)

FIG. 3 Outputs of a model providing best fits to both thenumbers of live elephants (a) and the numbers of elephantcarcasses in age categories and (b). The modelled populationinitially comprised , elephants during , which increasedby births at .% annually, and decreased at an annual mortalityrate . times the + carcass ratio (expressed as a proportion),with carcasses remaining visible for a mean of years afterdeath. Data points indicate numbers estimated during thesurveys, with vertical lines indicating upper and lower %confidence intervals of estimates. Sources for survey data areprovided in the text.

Elephant poaching in Mozambique 7

© 2014 Fauna & Flora International, Oryx, 1–10

–% during – to . % during –(Table ). However, fires and burnt areas recorded duringthe aerial surveys of wildlife suggested that –% of theReserve burnt annually. It is beyond the scope of thiswork to explain the discrepancy between these two datasources but VRB’s personal observations of the Reserveduring the dry season are closer to the records from theaerial surveys.

Discussion

Cause of increased mortality of elephants

Increased mortality of African elephants and a subsequentpopulation die-off often follow severe drought, as observed,

for example, in Tsavo (Kenya) during – (Corfield,). However, annual rainfall in Niassa was average orabove-average after (Table ), which suggests thatthe observed increase in mortality during that period wasnot drought-related. We considered the possibility that in-terannual variation in the percentage of the Reserve that wasburnt influenced the visibility of elephant carcasses to aerialobservers, but the increase in the number of carcasses re-corded during could not be explained by an increasein the proportion of the Reserve that was burnt (Table ).We believe that there is no natural explanation for the in-creased mortality observed after the survey but thatit resulted from illegal hunting. Hence, it is likely that theadditional , elephants that the models suggest diedduring – (including the additional , that diedduring ) were poached. The figure is equivalentto a mean of . elephants poached daily. Our estimate of, elephants poached during is much greater thanthe number of poached elephants () found by fieldstaff and reported to the MIKE programme that year(CITES, ).

Population model

Our study is the first to model simultaneously the numbersof live and dead elephants and to match the modelled trendsto survey estimates of elephant and carcass numbers.Application of this approach was facilitated by the consist-ency between years in survey design, execution and person-nel during the Niassa survey programme. The importance ofthis approach is revealed by the model that mimicked thetrend in the number of live elephants only. Although thefit between the modelled and observed elephant numbers(Fig. a) was better than the fit produced by the model thatmimicked trends in both elephant and carcass numbers(Fig. a), the best-fit estimates for the unknown parameterswere not realistic in themodel that fit only live elephants, be-cause they implied that only % of elephants that died wererecorded as category or carcasses during a survey in theyear that they died, and that elephant carcasses disappeared(at least to an aerial observer) within years, on average. Thepopulation model assumes that the Niassa elephant popu-lation is essentially a closed population, with no large-scalemovements of elephants between the Reserve and surround-ing areas. Radio-tracking studies of elephants in the Selous–Niassa corridor have revealed that individuals living close tothe Rovuma River (the international border) cross the river,moving between Mozambique and Tanzania, because theirhome range spans the border (Hofer & Mpanduji, ).There is no evidence of large numbers of elephants movingbetween Niassa and Selous Reserves. Similarly, tracking ofcollared elephants in the communal lands south-east ofNiassa, and in Quirimbas National Park to the east, show

TABLE 3 Percentage of the area of Niassa National Reserve (totalarea , km; Fig. ) burnt during May–October inclusive foryears when aerial surveys of wildlife were carried out. Data for– were provided by the MODIS burnt area product (noMODIS data are available for ). The percentage of areaburnt was also recorded during aerial surveys of wildlife during–. Blank cells indicate the absence of data.

Year

% area burnt

MODIS burnt area product Aerial survey

19982000 17.22002 15.9 522004 27.8 612006 36.1 592009 38.3 632011 33.4

TABLE 2 Mean rainfall in Niassa National Reserve (Fig. ) for eachclimate year (July–June) from – to –, determinedusing the African Rainfall Estimation algorithm (Herman et al.,).

Climate year Mean estimated rainfall (mm)

1997–1998 4291998–1999 8881999–2000 4702000–2001 9352001–2002 7792002–2003 1,0402003–2004 8072004–2005 8492005–2006 6432006–2007 1,0722007–2008 7912008–2009 7672009–2010 8382010–2011 862Mean 798

8 V. R. Booth and K. M. Dunham

© 2014 Fauna & Flora International, Oryx, 1–10

that the elephants remain there and do notmove to and fromNiassa (van Aarde et al., ).

Our model suggests that the number of elephants dyingin any year is . times greater than the estimated number offresh and recent carcasses. The presence (age category ) orabsence (category ) of the skin is the most easily observeddifference between these age categories. Hence, it seemslikely that, at least in Niassa, the skin of dead elephants de-composes in, year, particularly for elephants that die dur-ing the rainy season.

Conservation and management implications

Population estimates for live elephants and carcasses arenow routine outcomes of aerial surveys conducted toMIKE standards (Craig, ), and ourmethod of determin-ing the population trends by combining survey estimates forelephants and carcasses can be applied to any populationwith a time series of estimates for live and dead elephants.

The elephant population in Niassa is still the largest inMozambique. Controlling poaching is complicated by thesize of the Reserve and its difficult terrain, limited man-power and financial resources, human settlements withinthe Reserve, the long and porous international borderwith Tanzania, and weak national policy and legislationthat do not facilitate substantial benefits for local communi-ties. These factors combine to fuel the illegal ivory trade (vander Westhuizen, ). Poachers focus on big trophy bulls(Kambako Safaris, pers. comm.), the loss of which affectsthe viability of the safari industry and thus the managementbudget for the Reserve.

Management authorities in Niassa must deal with adistrict-driven rural development agenda on one hand and aconservation agenda on the other. The people living withinthe Reserve see little value in elephants and thus passivelyencourage poaching, which brings benefits through income,meat and the removal of potential human–elephant conflict.Reversing this attitude requires the conservation authorities toadopt strategies that give greater benefits to local communitiesfrom wildlife, and introduce livelihood strategies that stemvegetation clearance for the expansion of agriculture.

Acknowledgements

We thank Anabela Rodrigues, former Director of SGDRN,for providing access to the Niassa National Reserve aerialsurvey and law enforcement databases. This analysis isbased on the findings of surveys conducted by Colin Craigand Debbie Gibson under contract to SGDRN. Those sur-veys were funded by USAID, the U.S. Fish and WildlifeService, and Fauna & Flora International. We also thankthe safari operators and SGDRN field staff who provided ob-servations on carcass decomposition. Fay Robertson and

Peter Frost commented on drafts of this article, and Peteralso provided valuable discussion about MODIS fire data.

References

BALDUS, R.D. & HAHN, R. () The Selous–Niassa Wildlife Corridorin Tanzania: Biodiversity Conservation from the Grassroots.Practical Experiences and Lessons from Integrating LocalCommunities into Trans-boundary Natural Resources Management.FAO, Rome, Italy, and CIC, Budapest, Hungary.

BURN, R.W., UNDERWOOD, F.M. & BLANC, J. () Global trends andfactors associated with the illegal killing of elephants: a hierarchicalBayesian analysis of carcass encounter data. PLoS ONE, (), e.

CITES ()Monitoring of Illegal Hunting in Elephant Range States.A report to the th meeting of the Conference of the Parties toCITES. CoP Doc. .. CITES Secretariat, Geneva, Switzerland.

CITES () Elephant Conservation, Illegal Killing and Ivory Trade.Supplementary Information on Document SCDoc .. A report tothe nd meeting of the Standing Committee of CITES. SCInf. . CITES Secretariat, Geneva, Switzerland.

CORFIELD, T.F. () Elephant mortality in Tsavo National Park,Kenya. African Journal of Ecology, , –.

CRAIG, G.C. () Aerial Survey of Wildlife in the Niassa Reserve andSurrounds, Mozambique, October . Sociedade para a Gestão eDesenvolvimento da Reserva do Niassa, Maputo, Mozambique.

CRAIG, G.C. () Aerial Survey of Wildlife in the Niassa Reserve andAdjacent Areas, Mozambique, October . Sociedade para aGestão e Desenvolvimento da Reserva do Niassa, Maputo,Mozambique.

CRAIG, G.C. () Aerial Survey of Wildlife in the Niassa GameReserve, Mozambique, October . Sociedade para a Gestão eDesenvolvimento da Reserva do Niassa, Maputo, Mozambique.

CRAIG, G.C. () Monitoring the Illegal Killing of Elephants: AerialSurvey Standards for the MIKE Programme. Version .. CITESMIKE Programme, Nairobi, Kenya.

CRAIG, G.C. & GIBSON, D.St.C. () Aerial Survey of Wildlife in theNiassa Reserve and Hunting Concessions, Moçambique, October. Sociedade para a Gestão e Desenvolvimento da Reserva doNiassa, Maputo, Mozambique.

CRAIG, G.C. & GIBSON, D.St.C. () Aerial Survey of Wildlife in theNiassa Reserve and Surrounds, Mozambique, October .Sociedade para a Gestão e Desenvolvimento da Reserva do Niassa,Maputo, Mozambique.

CUNLIFFE, R., MANDONDO, A., GAMES, I., NGARIVHUME, J. & DORÉ,D. () Reconciling Conservation Goals with Agriculturally BasedLivelihoods: A Proposal for Future Development of the NiassaNational Reserve and Surrounding Areas. Main Report (Volumes I& II). Shanduko, Centre for Agrarian and Environmental Research,Harare, Zimbabwe, on behalf of: República De Moçambique,Província De Niassa; Fauna & Flora International, Cambridge,England; and Sociedade Para a Gestão e Desenvolvimento DaReserva Do Niassa, Maputo, Mozambique.

DOUGLAS-HAMILTON, I. & BURRILL, A. () Using elephant carcassratios to determine population trends. In African Wildlife: Researchand Management (eds F.I.B. Kayanja & E.L. Edroma), pp. –.International Council of Scientific Unions, Paris, France.

DOUGLAS-HAMILTON, I. & HILLMAN, A.K.K. () Elephantcarcasses and skeletons as indicators of population trends. In LowLevel Aerial Survey Techniques. Report of an InternationalWorkshop, – November , Nairobi, pp. –. InternationalLivestock Centre for Africa, Addis Ababa, Ethiopia.

Elephant poaching in Mozambique 9

© 2014 Fauna & Flora International, Oryx, 1–10

DUNHAM, K.M. () Detection of anthropogenic mortality inelephant Loxodonta africana populations: a long-term case studyfrom the Sebungwe region of Zimbabwe. Oryx, , –.

FEWS NET () Http://earlywarning.usgs.gov/fews/downloads/index.php?regionID=af&productID=&periodID= [accessed

August ].FRONTLINE SYSTEMS () Basic Solver—Non-Smooth Optimization.

Http://www.solver.com/content/basic-solver-non-smooth-optimization [accessed February ].

GASAWAY, W.C., DUBOIS, S.D., REED, D.J. & HARBO, S.J. ()Estimating moose population parameters from aerial surveys.Biological Papers of the University of Alaska, , –.Http://winfonet.alaska.gov/sandi/moose/surveys/documents/ua_biology_papers_nr_.pdf [accessed August ].

GIBSON, D.St.C. () Aerial Survey of Wildlife In and AroundNiassa Game Reserve, Moçambique, October . Sociedade para aGestão e Desenvolvimento da Reserva do Niassa, Maputo,Mozambique.

GIBSON, D.St.C. () Aerial Survey of Wildlife in the Niassa Reserveand Hunting Concessions, Moçambique, October/November .Sociedade para a Gestão e Desenvolvimento da Reserva do Niassa,Maputo, Mozambique.

HERMAN, A., KUMAR, V.B., ARKIN, P.A. & KOUSKY, J.V. ()Objectively determined -day African rainfall estimates created forfamine early warning systems. International Journal of RemoteSensing, , –.

HOFER, H. & MPANDUJI , D.G. () Distribution and Movements ofElephants and OtherWildlife in the Selous–NiassaWildlife Corridor,Tanzania. Tropical Ecology Support Programme. DeutscheGesellschaft fur Technische Zusammenarbeit (GTZ) GmbH,Eschborn, Germany.

JOLLY, G.M. () Sampling methods for aerial censuses of wildlifepopulations. East African Agriculture and Forestry Journal, ,–.

MANDONDO, A. & NGARIVHUME, J. () Reconciling ConservationObjectives with Agriculturally Based Livelihoods in the NiassaNational Reserve: Mobility, Settlement, Infrastructure, Markets,Livelihoods and Institutions. Annex . Shanduko, Centre forAgrarian and Environmental Research, Harare, Zimbabwe, onbehalf of: República DeMoçambique, Província De Niassa; Fauna &Flora International, Cambridge, England; and Sociedade Para aGestão e Desenvolvimento Da Reserva Do Niassa, Maputo,Mozambique.

MARTIN, R.B. () Maximum likelihood analysis of the elephantpopulation estimates for Matabeleland North, Zimbabwe. InElephant Management in Zimbabwe, nd edition (eds R.B. Martin,

G.C. Craig & V.R. Booth), pp. –. Department of National Parksand Wildlife Management, Harare, Zimbabwe.

MARUFO, M., JORGE, A. & RODRIGUES, A. () Characteristics andExperiences of Human–Wildlife Conflict. Niassa National Reserve.Internal report prepared by Sociedade para a Gestão eDesenvolvimento da Reserva do Niassa, Maputo, Mozambique.

MODIS. Active Fire & Burned Area Products. Http://modis-fire.umd.edu/index.html [accessed August ].

NORTON-GRIFFITHS, M. () Counting Animals, nd edition.African Wildlife Leadership Foundation, Nairobi, Kenya.

RIBEIRO, N.S., SHUGART, H.H. & WASHINGTON-ALLEN, R. ()The effects of fire and elephants on species composition andstructure of the Niassa Reserve, northern Mozambique. ForestEcology and Management, , –.

SGDRN () Zoning management. In Niassa National ReserveManagement Plan, Part II: Management Framework and ActionPlan. Ministry of Tourism and Sociedade para a Gestão eDesenvolvimento da Reserva do Niassa, Maputo, Mozambique.

SNEDECOR, G.W. & COCHRAN, W.G. () Statistical Methods, thedition. Iowa State University Press, Ames, USA.

UNEP, CITES, IUCN & TRAFFIC () Elephants in the Dust—The African Elephant Crisis. A Rapid Response Assessment. UnitedNations Environment Programme, GRID–Arendal, Arendal,Norway.

VAN AARDE, R., OTT, T., NTUMI, C. & TRIMBLE, M. () ConflictBetween People and Elephant—Zonation as a Potential Solution.Presentation at International Seminar on theMitigation of Human–Wildlife Conflict, – May , Lichinga. Direcção Nacional deTerra e Florestas, Ministry of Agriculture, Maputo, Mozambique.

VAN DER WESTHUIZEN, H.F. () Review: Law EnforcementProgramme Niassa National Reserve. Internal report prepared forSociedade para a Gestão e Desenvolvimento da Reserva do Niassa,Maputo, Mozambique.

WELLS, M.P. () The use of carcass data in the study andmanagement of African elephants: a modelling approach. AfricanJournal of Ecology, , –.

Biographical sketches

VERNON BOOTH is a freelance wildlife management consultant withparticular experience in administration and economics of the tourismindustry, protected area management planning, natural resource proj-ect evaluation and appraisal, and community-based natural resourcemanagement. KEV IN DUNHAM is a wildlife biologist interested inthe management and conservation of large mammals.

10 V. R. Booth and K. M. Dunham

© 2014 Fauna & Flora International, Oryx, 1–10