Landscapes as continuous entities: forest disturbance and recovery in the Albertine Rift landscape

14 Pages·2011·0.75 MB·English

LandscapeEcol(2011)26:877–890
DOI10.1007/s10980-011-9616-0
RESEARCH ARTICLE
Landscapes as continuous entities: forest disturbance
and recovery in the Albertine Rift landscape
Joel Hartter • Sadie J. Ryan • Jane Southworth •
Colin A. Chapman
Received:3November2010/Accepted:16May2011/Publishedonline:29May2011
(cid:2)SpringerScience+BusinessMediaB.V.2011
Abstract Kibale National Park, within the Alber- surrounding the park. To explore the effect of these
tine Rift, is known for its rich biodiversity. High different levels of forest management and protection
humanpopulationdensityandagriculturalconversion over time, we assessed forest change over the
in the surrounding landscape have created enormous previous three decades, using both discrete and
resource pressure on forest fragments outside the continuous data analyses of satellite imagery. Park
park. Kibale presents a complex protected forest boundaries have remained fairly intact and forest
landscape comprising intact forest inside the park, cover has been maintained or increased inside the
logged areas inside the park, a game corridor with park, while there has been a high level of defores-
degradedforest,andforestfragmentsinthelandscape tation in the landscape surrounding the park. While
absolutechangesinlandcoverareimportantchanges
in vegetation productivity, within land cover classes
are often more telling of longer term changes and
future directions of change. The park has lower
Normalized Difference Vegetation Index (NDVI)
J.Hartter(&) values than the forest fragments outside the park
DepartmentofGeography,UniversityofNewHampshire, andtheformerlyloggedarea—probablyduetoforest
102HuddlestonHall,73MainStreet,Durham,NH03824,
regenerationandearlysuccessionstage.Thecorridor
USA
region has lower productivity, which is surprising
e-mail:[email protected]
given this is also a newer regrowth region and so
S.J.Ryan(&) should be similar to the logged and forest fragments.
DepartmentofEnvironmentalandForestBiology,Illick
Overall, concern can be raised for the future trajec-
Hall,CollegeofEnvironmentalScienceandForestry,
tory of this park. Although forest cover has been
StateUniversityofNewYork,(SUNY-ESF),Syracuse,
NY13210,USA maintained, forest health may be an issue, which for
e-mail:[email protected] futuremanagement,climatechange,biodiversity,and
increased human pressure may signify troubling
J.Southworth
signs.
DepartmentofGeography,LandUseandEnvironmental
ChangeInstitute(LUECI),POBox117315,3141
TurlingtonHall,Gainesville,FL32611,USA Keywords Normalized Difference Vegetation
Index (NDVI) (cid:2) Kibale National Park (cid:2)
C.A.Chapman
Forest fragments (cid:2) Land cover change (cid:2)
McGillSchoolofEnvironment,McGillUniversity,
Montreal,QCH3A2T7,Canada Continuous analyses (cid:2) Reforestation (cid:2) Regrowth
123
878 LandscapeEcol(2011)26:877–890
Introduction because they represent reservoirs of land, resources,
and economic opportunity for people, and simulta-
Tropical deforestation is purported to be a primary neously are often viewed as buffers for PAs by
cause of global environmental change, with substan- managers (Schonewald-Cox and Bayless 1986).
tial projected impacts on biodiversity and world Since very few PAs represent intact ecosystems, it
climate (Geist and Lambin 2002). Protected areas has become increasingly important to consider each
(PAs) are the main means by which tropical forests protected area as a functional component of a larger
are now protected. However, PAs are also often landscape(ParksandHarcourt2002),whichincludes
established after forests have been degraded or non-protected areas.
cleared,eitherthroughlogging, thinning,agricultural Despite the abundant use and relative success of
conversion, or even creation of plantations. These traditional classification methods used in landscape
areas ‘‘in recovery’’ or ‘‘returning’’ to an original or analyses, the use of these methods alone may not
modified intact state present us with interesting alwaysbeoptimal.Discreteclassificationapproaches
challenges in the realm of conservation and preser- areconceptuallysimple,butamajordrawbackisthat
vation of biodiversity. While the purpose of a PA much of the variability within each land-cover type
may be to restore original habitat, understanding the has been removed (Southworth et al. 2004). Another
dynamics of recovery and what the relative value of methodistoincorporatecontinuousdatatorepresent
these areas is, requires serious consideration in order a more constant landscape level change (Southworth
to make informed management decisions. et al. 2004). Continuous data analyses can provide
Particularlyinforests,thenumberanddiversityof more revealing spatial analyses, can focus more on
organisms can be increased by disturbance, as biophysical indicators,and can examinemoresubtle,
pioneer and invasive plant species may support a within-class variability, and not just across-class
panoply of herbivorous insects, smaller mammals, conversion (Foody and Curran 1994; Southworth
andbirds.However,thegoalsofconservationandthe et al. 2004). The Normalized Difference Vegetation
measures of diversity that we currently use are not Index (NDVI) has been widely shown to be a robust
always in concordance. In a study comparing urban and reliable proxy of vegetation productivity and
andestablishedruralplantcommunities,Knappetal. quality in many environments (Jensen et al. 1991;
(2008) discovered greater phylogenetic diversity in Ehrlich et al. 1994; Rey-Benayas and Pope 1995;
established areas, despite greater species richness in Serneels et al. 2001) and an indicator of forage
disturbed areas. This pattern may be important when quality(Petorellietal.2011).Netprimaryproduction
protecting remaining tropical forests. Endemic or is strongly correlated with the NDVI because it
endangered animal species that are tropical old- emphasizes the red and near infrared bands to detect
growth habitat specialists will not fare well in changes in biomass (Roughgarden et al. 1991). In
disturbed habitats and recovered habitat may take dense, closed-canopy mature forests, we would
toolong togrow.Speciesthatcan exploitrecovering expect greenness to be of lower value, due to high
habitat, or which are habitat generalists, and less woody biomass and relatively low leaf area, and
impacted by disturbance, such as crop-raiding pri- remain stable as a sign of health and persistence.
mates or elephants, may slow or prevent sapling Therefore, we can use traditional classifiers to target
recruitment and halt reversion to old-growth entirely specific land cover classes (e.g., forest) and NDVI
(Lawes and Chapman 2006). Thus, it is important to can be used to examine within-class variability.
distinguish old-growth and recovering forest types, In this study, we examine Kibale National Park,
and understand their recovery trajectories and how Uganda. It comprises a mosaic of intact primary
they contribute to the ecosystem. forest and formerly logged areas, and the ‘‘extended
Thepatternsoflandcoverchangeinmosttropical park landscape’’, which contains remnant forest
developing countries relate significantly to anthropo- fragments in an agricultural mosaic surrounding the
genic forcings occurring across multiple spatial and gazetted park. We examine forest productivity of
temporal scales (Woods and Skole 1998; Duncan different components of the PA landscape, using
et al. 1999). Therefore, landscapes around PAs, NDVI. This paper builds on previous land cover
particularly in tropical forested areas, are important change analyses for Kibale using traditional discrete
123
LandscapeEcol(2011)26:877–890 879
classifiers (e.g., Laporte et al. 2008; Hartter and intensive smallholder agriculture, high levels of land
Southworth2009;Southworthetal.2010;Naughton- and resource pressures, and high rates of habitat loss
Treves et al., in press), but represents the first use of and conversion, making it a high priority area for
continuous data in this fashion and the first detailed conservation (Brooks et al. 2001). Kibale National
evaluation of NDVI with respect to areas of known Parkisamedium-altitudetropicalmoistforestwithin
disturbance histories. theAlbertineRiftthatcoversapproximately795 km2
Within the Kibale landscape, we examined four in western Uganda (Fig. 1). This transitional forest
differentforesttypes:in-park,intactforest,whichisold- (between lowland rainforest and montane forest) has
growth; logged-a conglomerate of former logging an average elevation of 900–1590 m and the climate
concessionsinwhichcuttingceasedin1975,andwhich is warm throughout the year, with an average daily
isrevertingtomorecontinuousforest;forestfragments minimum of 15(cid:3)C and maximum of 23(cid:3)C, and a
outside the park within a 5 km buffer (Hartter and mean annual rainfall of 1,662 mm (1970–2010,
Southworth 2009; Hartter et al. 2009), which are in T. Struhsaker and C. Chapman, unpublished data).
varying states of degradation and use; and, a former The bi-modal rainfall pattern produces two major
gamecorridorthatextendsfromthewesternboundary rainyseasons:‘‘short’’(lateFebruary-earlyMay)and
of the park. Using land cover classifications of forest, ‘‘long’’ (late August-late November).
agriculture, wetlands, tea, and crops/bare ground cre- PriortoKibalebecomingaNationalParkin1993,it
ated in previous research (Hartter and Southworth, was designated a Forest Reserve with the main
2009; Southworth et al. 2010), we can compare these objective to provide a sustained production of hard-
fourforesttypes/landscapelocations(park,loggedpark, wood timber (Struhsaker 1997) and softwoods from
forest fragments and corridor) over time, in terms of plantations. Logging was conducted at regular inter-
landscapechangesandchangesinforestproductivityto valsbetween1950and1975atvariableintensities,but
addressthefollowinghypotheses: mostly between 1967 and 1969, resulting in large
forest gaps and variable forest disturbance patterns
1. Intact forest cover will remain stable over time,
(Hartter et al. 2009). Harvested areas were typically
while fragmented, logged, and corridor areas are
allowed toregeneratenaturally, although undesirable
expected to change.
trees were poisoned in some areas. During Uganda’s
2. Differenttypesofmanagement(i.e.,foresttypes)
political upheaval in the 1970s and 1980s, the
will lead to different NDVI signatures.
plantationstypicallyestablishedonanthropogenically
3. Establishment of a corridor area will lead to
createdgrasslandswerenotmanaged(Chapmanetal.
reforestationandchangingNDVIvalues,resulting
2002). Management plans changed when Kibale
in higher NDVI values initially and becoming
became a national park in 1993. Plantations were
moresimilartothoseofintactforestovertime.
harvested from 1993 to 2006 (using manual pit saws
4. Asthepreviouslyloggedareasoftheparkrevertto
and portable sawmills) and the harvested pine areas
moreintact, morematureforest,theproductivity
werelefttoregeneratetonativeforest(Chapmanetal.
signatureofloggedforestwillbecomesimilarto
2002; C. Chapman unpublished data). In 1964 the
theforestintherestoftheparklandscape.
Kibale Forest Corridor Game Reserve was gazetted
5. Forest fragments outside the park will have
and administered as a separate entity by the Uganda
earlier successional stages of plant cover, corre-
GameDepartmentundertheMinistryofTourismuntil
sponding to similar NDVI values as other
it was combined with the Kibale Forest Reserve in
regenerating forest.
1993tocomprisethepresent-dayKibaleNationalPark
(Struhsaker1997).Asearlyas1971,illegaldestruction
andencroachmentoccurredinthecorridor,whichled
Study area toforestconversiontofarms.Theestimatesofpeople
who settled in the corridor varies considerably
TheAlbertineRiftregioninEastAfricaisoneofthe (8,800–170,000),butallwereevictedin1993(Chap-
world’s hotspots for biodiversity (Plumptre 2002; manetal.2010a).
Plumptreetal.2003;2007;Cordeiroetal.2007).Itis Outside Kibale, a mosaic of small farms (most
also one of the most threatened, due to dense \5 ha in size), large tea estates ([200 ha), and
123
880 LandscapeEcol(2011)26:877–890
Fig.1 KibaleNational
Parkandsurrounding
landscapeandareas
classifiedasforest
interspersed forest fragments and wetlands, effec- and resource availability depending on human use
tivelyisolatetheparkfromothertractsofforest.Tea and disturbance (Hartter 2010).
isconfinedtothenorthwestandnortheastsidesofthe Nearly 95% of the population in this area (pre-
park. Most tea lands are held by large multi-national dominately Batoro and Bakiga tribes) sustains their
companies consisting of hundreds of hectares, but livelihoods through farming or other agricultural-
other smaller, individual holdings also dot the based activities (National Environment Management
landscape. A vast network of bottomland forest Authority2001).Thisareaisoneofthemostdensely
fragments and wetlands either extend out from the populated areas in Sub-Saharan Africa (Lepp and
park or are isolated within this mosaic landscape. Holland 2006), with the population around Kibale
Theseareasvaryinsize(\0.5 hato[200 ha),shape, having grown bymorethan 300% between1959and
123
LandscapeEcol(2011)26:877–890 881
1990 (Naughton-Treves 1998) and estimated at an layer.Thefinalfive-classclassification(forest;crops/
averageofapproximately300individuals/km2within bare land—including pastures, cultivated crops, and
5 km of the park boundary by 2006 (Hartter 2010). kitchen gardens; papyrus (Cyperus papyrus L.) and
However,theeasternsidecarrieshigherdensitiesthan elephant grass (Pennisetum purpureum); tea; and
the west. water) obtained an overall accuracy of 89.1% and an
overall kappa statistic of 0.867 (Hartter and South-
worth 2009). The classifications were done indepen-
Methods dently for each image and not based on the same
spectralsignatures due to a lack ofexact anniversary
Image pre-processing dates, although the pattern of land cover signatures
was consistent across all image dates.
SevenLandsatTMandETM?sceneswereobtained: To address our hypotheses, the 1984, 1995, and
26 May 1984, 4 August 1986, 20 August 1989, 17 2003 land-cover composites were used to create
January1995,9January2001,31January2003,and11 change trajectories. Given that we wanted to under-
September 2008. All were acquired at the end of the standthedominantlandcovertrajectories,anddueto
dryseasonwhenfallowagriculturallandscanbeeasily how sensitive this method is to the number of initial
distinguishedfromforests,exceptfor1984,whichwas classes and image dates, we used only three of the
acquiredattheendoftherainyseasonsincethiswas availableimagedateswhichwerespreadouttodetect
theonlyavailablehaze-andcloud-freeimage.Images the patterns of change. (No data were available to
weregeometricallyregisteredto1:50,000scalesurvey include the 2008 image as part of the land cover
topographic maps of the region, followed by radio- analysis and we limited the number of possible
metric calibration and atmospheric correction. All trajectories). Land-cover/use classes of interest here
images were geo-registered to within a root mean are areas forested on all three image dates (forest),
square error of\0.5 pixels (below 15 m accuracy). areas ofwetland andgrassesonallthree imagedates
SinceLandsatTMandETM?satellitesdonotmake (wetland),andareasofagricultureonallthreeimage
geographicallyexactreturnvisits,weoverlaidthegeo- dates (crops or tea) which represents the three stable
registeredimages andclipped outareasfor which, in land cover classes in the region. Next we have two
any year, there were missing data. This mainly conversion classes of interest: reforestation (when
occurredinverysmallareasofthefarnorthandeast, anynon-forestclass(cropsorwetland vegetation)on
giving a slightly truncated edge to the buffer date one or date two is forested by date three; and
(describedbelow)inthoseregions(seeFig. 2). deforestation (any area of forest on date one or date
two, which is non-forest (wetland vegetation, tea or
Image classification and land cover change crops) on date three). Water was excluded from this
detection analysis since the crater lakes predominant in this
region do not vary in area over time. Areas of forest
We created a buffer zone outside the park boundary cover vary in terms of geography and past land use
to a distance of 5 km to describe the surrounding history, and here we discuss four different forest
landscape(769 km2).Thisdistancewashypothesized ‘‘types’’.(1)Intactforest—thisisnaturalforestcover,
asbeingfarenoughtocapturesocio-economiceffects found within the park boundaries, not logged or
ofthepark,asappliedinotherstudies(DeFriesetal. otherwiseimpactedwithinrecenthistory.(2)Logged
2005; Goldman et al. 2008). Land-cover maps were forest—forest areainthe northernmostsectionofthe
derived for each date by independent supervised park which was previously logged. (3) Fragments—
classification for three Landsat scenes: 1984, 1995, degraded remnants of the intact forest found in the
and2003.Duringfieldseasonsin2004and2005,180 landscape around the park. Most of these forest
trainingsampleswerecollectedandusedtoconstruct fragments have seen extensive human use. (4)
a supervised classification using the Gaussian max- Corridor—formergamecorridorconnectingtoneigh-
imum likelihood classifier. The classified image was boringQueenElizabethPark.Ittransectsthewestern
constructed using a layer stack including all bands domesticated landscape, and represents recent regen-
1–5, 7, texture bands of layers 1–5, 7, plus an NDVI erating forest.
123
882 LandscapeEcol(2011)26:877–890
Fig.2 Trajectoriesofland
use/coverchangecategory
occurringwithinthepark
boundariesandsurrounding
landscapefrom1984to
1995to2003.Reforestation
Areasthattransitionedto
forestfromagricultural
land,Forestconversion
areasthattransitionedfrom
foresttoagriculture,
Agriculturecrops,baresoil,
shortgrasses,tea;elephant
grassinallyears,Wetland
wetlandvegetationinall
years,StableForestforest
inallyears
NDVI analysis analysistodetermineandaccountfordifferencesdue
to past management. If we do not separate this out,
We calculated the NDVI value for each image, and the signal from the NDVI change analysis may be a
adjusted NDVI values for the 2001 and 2003 ETM? mixed signal and prove to be confusing. By incor-
images to TM values (Steven et al. 2003; Table 2). poratingthepastuseandmanagement(intact,logged,
To examine the trajectory of productivity over time, regenerating, and fragment) we can then evaluate
wecomparedthemeanNDVIwithineachforesttype their respective impacts.
(Table 1; Fig. 3) for all image dates. Given the Each of the images was captured following
differences in forest type, we then used the NDVI abundant rainfall during the rainy seasons and all
123
LandscapeEcol(2011)26:877–890 883
Table1 Land Cover/Use trajectories for the study region for interest:stableforestcover,stableagriculture,stablewetland/
2003,shownforintactforest(P)andtheloggedarea(L)within grass,reforestationordeforestation,intermsofthepercentof
Kibale,forestfragmentsoutsidepark(within5kmaroundthe thelandscape
park)(F),andthegamecorridor(C),forthefivetrajectoriesof
Trajectory Intactforest(P) Loggedarea(L) 5kmBuffer(F) Corridor(C)
(%area) (%area) (%area) (%area)
Stableforest 79.2 82.5 15.7 20.5
Stableagriculture 4.5 2.6 42.3 42.3
StableWetland 1.2 1.1 5.5 3.4
Reforestationa 11.3 9.0 17.1 16.6
Recentreforestation 6.4 4.4 10.1 9.0
Olderreforestation 5.0 4.6 7.0 7.7
Deforestationb 3.9 4.8 24.7 16.4
Recentdeforestation 1.7 3.6 15.6 10.8
Olderdeforestation 2.2 1.2 9.1 5.6
a OlderreforestationAreasthatbecameforestfrom1984to1995andremainedforest,Recentreforestationareasthatbecameforest
from1995to2003
b OlderdeforestationAreas thatlostforestfrom 1984to1995,Recentdeforestation areasthatlostforestfrom1995to2003and
remained
seven images are not anomalous (not in peak or Results
trough precipitation months). Mean rainfall totals for
the4 monthspriortoimageacquisitionforallimages Spatial patterns of land cover change
were not significantly different than the 4 months
prior to the 2008 image capture (Dunnett’s test, Forest remains the dominant land cover type within
P[0.05). Kibale (not including the corridor—Fig. 2). Over
time, park boundaries have been maintained with no
Change detection evidence of large-scale encroachment. The park
showed a consistent pattern of reforestation since
We used a two-factor (type and year) ANOVA 1984, while only 3.9% of intact forest and 4.8% of
framework to explore differences between mean the logged area deforested (Tables 1, 2). The logged
NDVI values in the four forest types. A post-hoc area had 9% reforestation identified (Table 1).
Tukey’s Honest Significant Difference test (HSD), In contrast, outside the park the situation is quite
a = 0.05 was used to determine which forest types different. This domesticated landscape is in a con-
differed from one another. To assess the increase in stantstateoffluxasaresultofthevariouslivelihoods
similarity of disturbed forest types to the intact type, being supported. Conversion of forest occurred at a
we compared differences in types over time. Differ- high rate, while there is still some effort towards
encesbetweenmeanNDVIvaluesoftheloggedarea, reforestation.Althoughthere isanoveralllossof2%
corridor, fragmented forests, and intact forest were total forest, only about 16% of the forest fragments
calculated for each image year. We regressed these outsideKibalehaveremainedstablesince1984.This
differences upon year (nominal) to determine if the area has experienced higher reforestation rates than
differences were decreasing over the study period. the park, but has also had nearly 25% deforestation,
Sincetheimageswerecollectedatdifferentintervals, suggesting a high rate of landscape turnover. Simi-
wecouldnotperformformaltimeseriesanalyses.As larly,only20%oftheforestinthecorridorremained
thesedataarequiteheterogeneouswithinforesttype, since 1984, while the corridor also sustained 16.4%
wealsocalculatedthemedianvalueforeachtypeand deforestation, with almost 11% of it recent (since a
image, and found that medians differed from the year before being gazetted). At the same time, the
mean by an average of only 1.35%, suggesting that corridor has had 16.6% reforestation, which also
the mean was not biased. indicates a high rate of landscape turnover.
123
884 LandscapeEcol(2011)26:877–890
Fig.3 GraduatedNDVIcompositesforforestonlyfortheparkandthe5kmbuffer
NDVI change The two-factor ANOVA on forest type and year
explained 95% of the variance in the data, with the
Itisevidentthatthereisaspatialpatternofchangein majority explained by year (R2 = 0.95, SS = 0.22,
Y
NDVIfromimagetoimage(Fig. 3).Withinthepark, SS = 0.01), but both factors were significant (forest
T
the highest NDVI values are found in the northern type: P = 0.003, Year: P = 0.0001). The Tukey’s
sectionsofthepark,whichwereloggedbetween1950 HSD showed that the logged area had the highest
and 1975, and are now regenerating forest (Table 2). NDVIvalueswhencontrollingforyear(althoughnot
The rest of the park which was not commercially significantly higher than the fragments), followed by
logged shows NDVI values below the mean. The the corridor, and the intact forest had the lowest
logged area within the park has higher mean NDVI values (although not significantly lower than the
valuescomparedtotheentireparkandthesurrounding corridor; Table 3). The fragments have a higher than
landscapeforallimageyears(Fig. 3).Thereisalsoan average NDVI value in all years except 1989, and in
increasinggradientofNDVIintheparkfromthelower 2001 and 2003, the highest mean NDVI values
elevationsintheeastandsouthtothehighernorthern overall (Fig. 4). In addition, most individual frag-
forests, which may also be related to the logged- ments had above-mean NDVI values for each image
unloggedareasand/orrainfallpatterns(Fig. 3). year (Fig. 2).
123
LandscapeEcol(2011)26:877–890 885
Table2 Forestcover andNDVIvaluesforintactforest(P)andtheloggedarea(L)withinKibale, forestfragmentsoutsidepark
(within5kmaroundthepark)(F),andthegamecorridor(C)
Intactforest(P) Loggedarea(L)
Year Forest NDVI Year Forest NDVI
(%area) Mean StdDev (%area) Mean StdDev
1984 86.1 0.570 0.071 1984 88.5 0.669 0.050
1986 85.8 0.581 0.053 1986 89.6 0.638 0.050
1989 85.8 0.604 0.051 1989 89.6 0.650 0.046
1995 85.8 0.533 0.046 1995 89.6 0.600 0.049
2001a 90.3 0.336 0.067 2001a 91.0 0.386 0.066
2003a 90.3 0.448 0.049 2003a 91.0 0.488 0.050
2008 90.3 0.609 0.049 2008 91.0 0.644 0.046
5kmBuffer(F) Corridor(C)
Year Forest NDVI Year Forest NDVI
(%area) Mean StdDev (%area) Mean StdDev
1984 31.9 0.645 0.063 1984 36.0 0.630 0.068
1986 34.3 0.603 0.071 1986 39.2 0.584 0.070
1989 34.3 0.600 0.085 1989 39.2 0.575 0.091
1995 34.3 0.575 0.045 1995 39.2 0.548 0.045
2001a 29.2 0.413 0.069 2001a 37.2 0.407 0.078
2003a 29.2 0.503 0.051 2003a 37.2 0.376 0.050
2008 29.2 0.629 0.078 2008 37.2 0.637 0.068
a StandardizedNDVIcalculationsfromETM?toTM(Stevenetal.2003),RemoteSensingofEnvironment88,pp412–422
The corridor NDVI values varied about the 3 negative changes and alternating year patterns
landscape average within a similar range to the respectively (Fig. 5).
fragments and logged area, with the striking excep-
tionof2003.Inthecorridor,therewasalowerNDVI
value than any other forest category in 2003, and the Discussion
largest difference from the intact forest of any
category in any year (Fig. 5). Visual inspection of composite satellite images and
Likely due to low sample size (n = 7), the the comparison of NDVI values show that Kibale
regressions of differences between disturbed forest National Park has become an island of forest in a
types and within-park intact forest across years were landscape surrounded by intensive agriculture. This
not significant. However, the formerly logged area, domesticated landscape is characterized by highly
was marginally significant (P = 0.06) and negative fragmentedandrapidlychanging landcovers,imply-
and captured 52% of the variability (logged = inghighlydynamiclanduses.Ourresultsindicatean
3.44 - 0.002*year, R2 = 0.52), while the fragments increase in forest cover inside the park since 1984.
(fragment, P = 0.88, R2 = 0.005) and the corridor Specifically, the majority of Kibale’s in-park forest
(corridor, P = 0.81, R2 = 0.013) showed no trends. has remained intact with only a small proportion lost
In addition, results from the logged area (in which 5 to deforestation over 20 years.
of 6 between-image transitions in difference (83%) Althoughitdoesnotcompletelycapturetheeffec-
were negative, including the last 3 transitions) tiveness of park management, remotely sensed data
suggest convergence to the intact forest signature. does reveal the impacts of various management
In contrast, the fragments and the corridor had 4 and approaches on the vegetation cover and dynamics.
123
886 LandscapeEcol(2011)26:877–890
Table3 Least-squares Mean (LSM) NDVI differences in
forest types, after controlling for year, using a post-hoc Tu-
key’sHSDona2-factorANOVA
Type LSM Group
L 0.582 1
F 0.567 1,2
C 0.537 2,3
P 0.526 3
Summarized by forest type (Type: Logged (L), Fragments in
the surrounding domestic landscape (F), Corridor (C), and
intact forest within the park (P)), least-squares mean NDVI
value (LSM), and grouping (group), where types in the same
group (1–3) are not significantly different from one another.
Fig.5 The differences of mean NDVI value between forest
Overallmodelstandarderror=0.01
typesL(black),F(grey)andC(white)andintactforest(P)at
successiveimagesdatesinthestudy
landscape had significantly higher apparent produc-
tivity (as measured by NDVI) than did the intact
forest within the park. Interestingly, the corridor
appearedtohaveanoverallsimilarityinproductivity,
statistically, to both the forest fragments outside the
park and to the intact portion of the forest within the
park. It is important to note that these images were
obtained in the same season within each year, and
that the precipitation patterns preceding each image
werenotstatisticallydissimilar.Sincetheseasonality
of the area has been changing (Hartter unpublished
Fig.4 Mean forest NDVI values for intact forest within
Kibale,theloggedareawithinthepark,fragmentswithinthe5 data), using the same Julian date in each image year
kmbufferoutsidethepark,andtheformergamecorridor would have been inappropriate. It is thus interesting
to see the variability in the NDVI values between
Forexample,oneofStruhsaker’s(2002)indicatorsof image years, and due to the corrections between
park success isadecreaseinillegalactivities. Large- sensors(TMandETM?),thevaluesarecomparable.
scale encroachment into Kibale has virtually halted It is important to understand the within-class
since formal park establishment in 1993 (Hartter and variability and its potential ecological significance
Goldman 2011) and there are well understood and withinourregions.Itislogicalthattheintactforestin
maintained boundaries. When compared to the sur- the park has lower NDVI values (i.e., NDVI in
rounding landscape, the area within the park has matureforest levels off and then decreases over time
experienced limited land cover change in absolute comparedtonewer,secondarysuccessionalgrowthin
terms,buttherearenotablechangesormodificationsin regenerating areas) than the logged area and buffer,
cover as seen from the NDVI analysis. While effec- butperhapspuzzlingisthatithaslowermeanvalues
tively protected from human exploitation, the park thanthecorridorin5outof7 years.Whilethebuffer
itselfisnotunchanging,withvariationfromoneland- area outside the park is an active transition zone
cover class to another. For example, areas misclassi- where there is ongoing deforestation, there is also
fied as crops in the park are naturally occurring transition of former agricultural lands back to forest.
grasslands, which undergo succession in savanna Sincemanyoftheseforestsareatearliersuccessional
areas,andsomevariationfromwetlandstobareareas. stages, the productivity was expected to be higher
Wefoundthattheformerlyloggedareawithinthe than forests at later successional stages, resulting in
parkandtheremnantforestfragmentsinthedomestic above-mean NDVI values for the forest fragments.
123