Full text of DOI:10.1080/01431160117141

This article was downloaded by: [Florida International University]
On: 26 December 2014, At: 03:53
Publisher: Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number:
1072954 Registered office: Mortimer House, 37-41 Mortimer Street,
London W1T 3JH, UK
International Journal of
Remote Sensing
Publication details, including instructions for
authors and subscription information:
http://www.tandfonline.com/loi/tres20
Forest change monitoring of
a remote biosphere reserve
S. A. Sader a , D. J. Hayes b , J. A. Hepinstall b ,
M. Coan c & C. Soza d
a Department of Forest Management , University
of Maine , 201 Nutting Hall, Orono, ME,
04469-5755, USA E-mail:
b Maine Image Analysis Laboratory, Department
of Forest Management , University of Maine , 201
Nutting Hall, Orono, ME, 04469-5755, USA
c Seven Islands Land Company , Ashland, ME,
04732, USA
d ProPeten Program , Conservation International ,
Flores, Guatemala
Published online: 25 Nov 2010.
To cite this article: S. A. Sader , D. J. Hayes , J. A. Hepinstall , M. Coan & C. Soza
(2001) Forest change monitoring of a remote biosphere reserve, International
Journal of Remote Sensing, 22:10, 1937-1950
To link to this article: http://dx.doi.org/10.1080/01431160117141
PLEASE SCROLL DOWN FOR ARTICLE
Taylor & Francis makes every effort to ensure the accuracy of all
the information (the “Content”) contained in the publications on our
platform. However, Taylor & Francis, our agents, and our licensors
make no representations or warranties whatsoever as to the accuracy,
completeness, or suitability for any purpose of the Content. Any opinions
and views expressed in this publication are the opinions and views of
the authors, and are not the views of or endorsed by Taylor & Francis.

The accuracy of the Content should not be relied upon and should be
independently verified with primary sources of information. Taylor and
Francis shall not be liable for any losses, actions, claims, proceedings,
demands, costs, expenses, damages, and other liabilities whatsoever
or howsoever caused arising directly or indirectly in connection with, in
relation to or arising out of the use of the Content.
This article may be used for research, teaching, and private study
purposes. Any substantial or systematic reproduction, redistribution,
reselling, loan, sub-licensing, systematic supply, or distribution in any
form to anyone is expressly forbidden. Terms & Conditions of access
and use can be found at http://www.tandfonline.com/page/terms-and-
conditions

int. j. remote sensing, 2001, vol. 22, no. 10, 1937–1950
Forest change monitoring of a remote biosphere reserve
S. A. SADER
Department of Forest Management, 201 Nutting Hall, University of Maine,
Orono, ME 04469-5755, USA; e-mail: sader@umenfa.maine.edu
D. J. HAYES, J. A. HEPINSTALL
Maine Image Analysis Laboratory, Department of Forest Management,
201 Nutting Hall, University of Maine, Orono, ME 04469-5755, USA
M. COAN
Seven Islands Land Company, Ashland, ME 04732, USA
C. SOZA
Conservation International, ProPeten Program, Flores, Guatemala
(Received 29 July 1999; in Ž nal form 14 March 2000)
Abstract. A forest monitoring framework using Universal Transverse Mercator
(UTM) grid cells to report forest change estimates derived from time-series
satellite imagery was established for the Maya Biosphere Reserve (MBR) in
northern Guatemala. Five dates of Landsat Thematic Mapper imagery were
acquired and digitally processed to quantify forest change for four time periods:
1986 to 1990, 1990 to 1993, 1993 to 1995, and 1995 to 1997. Time-series change
2
estimates are reported for 215 UTM grid cells approximately 100km each. For
the period 1990 to 1997, after the designation of the MBR, the percentage of grid
cells with detectable annual forest clearing increased from 38% (1990–93) to 41%
(1993–95) and 45% (1995–97). Prior to the establishment of the MBR (1986–90),
none of the grid cells exhibited greater than 4.0% annual forest clearing. However
in the next three time periods, 7.0% , 8.8% and 9.3% of the grid cells had clearing
rates exceeding 4.0% per year. The accuracy of detecting forest clearing was
86.5% over all time periods (Kappa 0.82). Estimates of forest change and user’s
and producer’s accuracy are reported for each time period between 1990 and
1997. The time-series forest change and spatial arrangement of grid locations
indicate hot spots where rates and trends of agricultural expansion can be mon-
itored. The baseline survey and the establishment of the UTM grid network to
localize forest change estimates provides a framework for future satellite estimates
of forest and land cover conversion to be monitored through time. The UTM
grid is proposed as the Ž rst level in a multi-level ecological monitoring system
for the Maya Biosphere Reserve where there are few permanent landmarks in the
remote forest region.
1. Introduction
The utility of satellite imagery for monitoring tropical forest change has been
reported by many authors, of which this is not an exhaustive list (Singh 1986,
International Journal of Remote Sensing
ISSN 0143-1161 print ISSN 1366-5901 online © 2001 Taylor & Francis Ltd
/
http: www.tandf.co.ukjournals
//
/

1938
S. A. Sader et al.
Fearnside 1986, Sader et al. 1990, Skole and Tucker 1993, Lambin 1996, Foody et al.
1996, Brondizio et al. 1996). In remote tropical regions, satellite remote sensing (via
radar or optical sensors) may be the only feasible method to monitor forest clearing
and land cover/land use conversion over large areas. Access to remote forest regions
by surface roads or rivers is often limited, and aerial photography is either non-
existent, outdated or infeasible to acquire for large regions.
The Maya Biosphere Reserve (MBR) in the Peten District of northern Guatemala
(Ž gure 1) was established in 1990 by the Congress of Guatemala (Stuart 1992). The
Peten District (36 000 km ) is the largest and most remote region of Guatemala. The
2
northern part of the district, along with adjacent Mexican states (Chiapas and
Campeche) and Belize (Rio Bravo Conservation District), constitutes the largest
contiguous tropical moist forest left in Central America. As in other parts of Central
America, the once remote tropical forests of Guatemala are being cut and burned
to create new farmland (Sader et al. 1996).
Given the need for landscape-level forest change information in the MBR, the
US Agency for International Development and non-government organizations
(NGOs), sought to develop a forest change monitoring system. Kristensen et al.
(1997) proposed a three level monitoring system for the MBR, which included
satellite forest change detection at the regional level, and ground-based methods for
monitoring biological parameters and non-timber forest product extraction at the
second and third level.
The International Union of Forest Research Organizations (IUFRO) deŽ nes
forest monitoring as ‘the periodic measurement or observation of selected physical,
chemical and biological parameters (of forests) for establishing baselines for detecting
and quantifying changes over time’ (Paivinen 1994). The establishment of permanent
forest plots has been an important component of continuous forest inventory pro-
grammes for decades (Stott 1968). A permanent plot is established, monumented
and documented in such a manner so that the exact area or objects in the plot can
be remeasured at a later time and for which there is an intent and plan for remeasure-
ment (Avery 1967, Paivinen 1994).
In the 1940s, when aircraft remote sensing (aerial photography) became an
important tool in forest inventory and monitoring, the concept of photo plots was
established, whereby photogrammetric techniques and double sampling statistical
procedures allowed the extrapolation of ground plot measurements to the forest
stand types interpreted from aerial photography (Bickford 1952, Bonner 1975). A
Zoom Transfer Scope or similar instrument enabled the transfer of photo plot
locations to a georeferenced base map, therefore the photo plot could be relocated
on future aerial photos to update changes that may have occurred.
The IUFRO developed International Guidelines for Forest Monitoring (Paivinen
1994); however they did not specify or endorse a speciŽ c sampling frame or perman-
ent plot design applicable for global or regional monitoring using satellite imagery.
The United Nations Food and Agricultural Organization (FAO), in their 1990 Forest
Resource Assessment of the Tropical Regions (FAO 1993), adopted the Landsat
Worldwide Reference System as the sample frame to inventory and monitor pan-
tropical forest status and change. The US Environmental Protection Agency (EPA)
uses a system of 640 km hexagons as the spatial unit for ecological sampling and
2
biodiversity analysis (White et al. 1992). The FAO and EPA frameworks incorporate
large sample units suitable for global level monitoring. Fearnside (1986) applied a
grid system to report forest change estimates for 1
1 latitude and longitude
°× °

Forest change monitoring of a remote biosphere reserve
1939
Figure 1. Location of the Maya Biosphere Reserve and buVer zone in the Peten District of
Guatemala.
quadrants in the Brazilian Amazon, however, to our knowledge this was a one-time
application for reporting purposes and was not designed as a continuous monitoring
framework.
In this paper we employ a simple yet accurate time-series satellite change detection

1940
S. A. Sader et al.
method to monitor forest change in the Maya Biosphere Reserve. A Universal
Transverse Mercator grid network provides a framework to localize change estimates
in a remote forest region with few roads or physical landmarks. We propose the
network of UTM grid cells as the reference frame for level 1 forest monitoring (e.g.
Kristensen et al. 1997) in the MBR. The grid cells are relatively large (100 km ) yet
2
manageable to localize the forest change data in 221 discrete units that cover the
entire MBR and buVer zone. The UTM grid is a familiar mapping unit on the
Guatemala 1:250 000 quadrangle map series. Given the importance of forest change
data for conservation planning in the MBR (Kristensen et al. 1997), the grid cells
can serve as the level 1 reporting unit or as the sample frame to select level 2 or 3
samples in a multi-stage design to link satellite and ground data for more detailed
ecological studies.
The study area encompasses the MBR and buVer zone to the south of the Reserve
(Ž gure 1). The MBR contains three types of management units: national parks,
biological reserves and a multiple use zone. Dominant natural vegetation types
include tropical broadleaf evergreen forest, semi-deciduous broadleaf forest and
scrub-shrub broadleaf vegetation, the last two occurring in seasonally  ooded
swamps (Lundell 1937).
Socio-economic and cultural driving forces in uencing the recent migration and
forest conversion have been reported by Schwartz (1990), Stuart (1992), Southgate
and Basterrechea (1992), Sader et al. (1994) and Whitacre et al. (1995).
2. Methods
2.1. Change detection
The time-series change detection technique applied in this study relies on the
computation of the normalized diVerence vegetation index (NDVI) at each date
using the following formula:
NDVI (near IR visible red)/(near IR visible red)
(1)
=
-
+
The NDVI has been shown to be highly correlated with green biomass, crown
closure, and leaf area index, among other vegetation parameters (Tucker 1979, Sellers
1985). The NDVI suppresses diVerential solar illumination eVects of slope and aspect
orientation (Lillesand and Kiefer 1994) and helps to normalize diVerences in bright-
ness values when processing multiple dates of imagery (Singh 1986, Lyon et al. 1998).
Lyon et al. (1998) compared seven vegetation indices to detect vegetation change in
a Chiapas, Mexico study site. They reported that the NDVI was least aVected by
topographic factors and was the only index that showed histograms with normal
distributions. These authors concluded that the NDVI diVerence technique produced
the best change detection results based on lab and Ž eld comparisons.
A requirement for time-series change detection is the availability of relatively
cloud-free satellite imagery at each date. To reduce variability in spectral re ectance
patterns related to vegetation phenology over the annual cycle, the imagery needed
to be collected within anniversary windows. Portions of three Landsat scenes—
identiŽ ed by the Landsat Worldwide Reference System as path 19, row 48; path 20,
row 48; and path 21, row 48—cover the width of the Maya Biosphere Reserve and
the adjacent buVer zone. The middle scene (path 20, row 48) encompasses approxi-
mately 90% of the entire area. Our study utilized Landsat Thematic Mapper (TM)
images collected from late February to mid-May in 1986, 1990, 1993, 1995 and
1997. All images were collected in the dry season representing a relatively stable

Forest change monitoring of a remote biosphere reserve
1941
phenological status of the vegetation at these acquisition dates. Well-distributed
ground control points were selected to georeference and resample (nearest neighbour)
all imagery to a previously rectiŽ ed 1986 TM base image at a 30-m pixel resolution.
Unsupervised classiŽ cation of NDVI time-series images incorporated post-change
detection GIS merging and screen digitizing to edit known ‘false change’ errors
(Sader et al. 1996). The method draws from both automated classiŽ cation and visual
interpretation techniques. The RGB-NDVI classiŽ cation attempts to stratify change
versus no change in a time-series. This is similar to ratio image diVerencing (Jensen
1996) except that it is performed in one classiŽ cation run with three dates rather
than computing a series of two-date diVerenced images. An earlier version of the
RGB-NDVI method and the concept of additive colour theory to interpret the
results were reported by Sader and Winne (1992). Satellite change detection studies
in Guatemala test sites indicated that three-date RGB-NDVI classiŽ cation methods
(Sader et al. 1996) produced higher accuracy (kappa statistic) than PCA and NDVI
image diVerencing (Hayes and Sader, unpublished). The RGB-NDVI classiŽ cation
method was also considered to be the fastest and easiest to perform.
The NDVI was computed for each of the Ž ve dates. The near-IR and visible red
wavebands (equation (1)) correspond to TM band 4 and 3, respectively. An unsuper-
vised (clustering) classiŽ cation technique with a maximum likelihood decision rule
(ERDAS 1996; Jensen 1996) was performed separately on three-date NDVI datasets
(1986–90–93 and 1993–95–97). The classiŽ ed images were one-channel Ž les, each
representing 45 multitemporal NDVI cluster classes. The change, no-change class
labels were assigned to each cluster class through observation of the NDVI cluster
means and standard deviation at each date in reference to known ground locations
in the study area and guided by interpretation of TM false colour composites of
each date. Major shifts in NDVI due to forest clearing and vegetation regrowth can
be detected for Landsat pixels that changed between two or more dates. Many
of the clusters showed no signiŽ cant change, as indicated by nearly equal NDVI
values at each date. Clouds and water were masked out prior to classiŽ cation;
however screen digitizing of cloud shadows and water edges was required as a
post-classiŽ cation procedure to eliminate false readings.
Lillesand and Kiefer (1994) suggested haze correction prior to any image ratio
calculation to reduce additive eVects of atmospheric attentuation in the separate
wavebands. Bootstrap atmospheric corrections using a minimum value subtraction
from deep, clear water pixels (Jensen 1996) were applied to the visible red and
re ected infrared bands prior to computation of a normalized diVerence vegetation
index (NDVI). Radiometric normalization is often recommended for change detec-
tion studies especially when the objective is to identify types of land cover change
(Hall et al. 1991, Coppin and Bauer 1996). Radiometric corrections were not applied
because the large NDVI value change between a pixel representing a forest canopy
at one date and the complete removal of the canopy at the next date was above the
noise levels that would be compensated for using a calibration procedure (Sader and
Winne 1992, Cohen et al. 1998). Related studies on the accuracy of radiometrically
corrected and non-corrected NDVI datasets for the middle scene (path 20, row 48)
indicated no signiŽ cant diVerence. The objective of this study was to develop a
relatively simple procedure to detect and quantify forest clearing in a time-series and
not to monitor land cover change, which may justify a calibration component.
A 5 5 majority Ž lter was applied to eliminate isolated pixels and consolidate
×
the boundaries of the cleared patches. A 5 5 pixel Ž lter was preferred over a 3
3
×
×

1942
S. A. Sader et al.
Ž lter because the latter retained smaller patches. The 5 5 Ž lter eliminated clearings
×
of less than approximately 1 ha. As the average size of new forest clearings in the
MBR exceeded 2.5 ha (Schwartz 1990, Sader et al. 1994), Ž ltering should have
minimal impact on the detection and estimation of forest change at the biosphere
level. The Landsat scenes were analysed separately and merged after classiŽ cation,
editing, Ž ltering and recoding.
Zone 16 UTM vector intersections were converted to raster grid cells to report
forest change estimates for each grid location. Many of the UTM grid cells approxi-
mate 100 km except along the reserve boundaries and where two UTM zones meet
2
(Zone 15 and 16) near the middle of the study area. Each cell was assigned a unique
number beginning at one for the most north-westerly cell and ending with UTM
cell 221 in the south-west portion of the MBR buVer zone. Unfortunately, six cells
in the south-east corner of the buVer zone were not covered by the Landsat imagery
acquired for this study.
2.2. Accuracy assessment
To have some conŽ dence in the accuracy of the change detection map and to
promote the methods employed as a basis for current and future forest monitoring
in the MBR, an accuracy assessment was needed. The accuracy of the multi-temporal
change detection was tested using an approach reported by Cohen et al. (1998). This
method relies on the interpretation of Landsat-TM colour composite images at each
date for a set of stratiŽ ed randomly selected sample plots. There were no historical
aerial photos, GIS coverage, or ground data that could be used as reference data
for an accuracy assessment of the change detection map. Therefore, the interpretation
of multi-date colour composites provided the reference data to generate an error
matrix (Congalton 1991). Cohen et al. (1998) found this method to be as accurate
as using historical vegetation coverage for identifying forest clearcuts in the PaciŽ c
Northwest region of the USA.
The MBR and buVer zone are covered by portions of three Landsat scenes;
however the centre scene represents over 90% of the land area. All accuracy assess-
ment samples were drawn from the centre scene to simplify the multi-date interpreta-
tion procedure. The remaining part of the MBR and buVer zone not represented
contained a higher proportion of roadless and inaccessible forest where forest change
was minor ( 5% of the total change) compared with the centre scene.
<
We drew two stratiŽ ed random samples from the change detection map and had
each set of sample plots interpreted by a diVerent forestry graduate student. The two
students received a brief training session and were provided with examples of plots
that exhibited forest clearing at diVerent dates. None of the training plots were used
as accuracy test samples. The samples were 3 pixel 3 pixel blocks that had a
×
majority of nine out of nine pixels in either the forest change or no change categories.
In total, 325 test plots were drawn from four strata: 1990–93 change; 1993–95 change;
1995–97 change; and no change (all dates). The 1986–90 period was before the
establishment of the Maya Biosphere Reserve (Stuart 1992) and was not included
in the accuracy assessment sampling. In the ERDAS-Imagine software package, the
sample plot locations were displayed in four colour composite images (RGB 453),
geographically linked and enlarged on a 24 inch display monitor. In an eVort to
reduce interpretation bias, the interpreters did not know which change detection
stratum was represented by each test plot. Also the interpreters were not involved

Forest change monitoring of a remote biosphere reserve
1943
in the image classiŽ cation. The interpreters recorded their observation for each plot
on a prepared form for easy compilation of the results.
3. Results
The rates and trends of forest change can be observed by comparing the four
change-detection time periods. The annual forest clearing rates for the MBR (exclud-
ing the buVer zone) increased from 0.04 to 0.23% from 1986–90 and 1990–93 then
remained nearly constant at 0.33 and 0.36% per year in 1993–95 and 1995–97,
respectively. Forest clearing in the last two time periods for the multiple use zone,
were equal at 0.25% per year. The clearing rate in the buVer zone increased from
0.74 to 2.71, 3.28 and 3.78% per year in the four sequential time periods. The time-
series clearing rates for each management unit in the MBR are provided in table 1.
Although the forest clearing rates were roughly equal when reported at the
Reserve, multiple use zone and buVer zone level (especially for the two most recent
time periods, 1993–95 and 1995–97), the reporting at the UTM grid cell level
provides spatial distributions of forest clearing on a more local scale (Ž gure 2). For
example, in the 1986–1990 time period, 28% of the UTM grid cells had detectable
forest clearing ( 0.1% year ). In the next three consecutive time periods (1990–93;
Õ1
>
1993–95; 1995–97), the percentage of grid cells with detectable change increased to
38% , 41% and 47% . From 1986 to 1990, none of the cells exhibited greater than
4.0% forest clearing per year. In the next three consecutive time periods, 15 cells
(7.0% ), 19 cells (8.8% ) and 20 cells (9.3% ) experienced greater than 4.0% forest
clearing per year (Ž gure 3). Of the cells where we detected greater than 4.0% per
year forest clearing in 1993–95 and 1995–97, less than one half of these were the
same cells in the two time periods. Thirteen cells had more than 6.0% per year
cleared, between 1993 and 1995, but only six were in this category between 1995
and 1997. All thirteen cells with
6.0% clearing from 1993 to 1995 had lower
>
Table 1. The Maya Biosphere Reserve management units: area and annual forest clearing
rates (% per year) from 1986 to 1997.
1986 Forest area Percentage land
(ha) of
management
unit
area in
management
unit
1986– 1990– 1993– 1995–
1990 1993 1995 1997
Management units
Tikal National Park
Laguna del Tigre
National Park
El Mirador National
Park
Sierra del Lacandon
National Park
55 005
289 912
2.62
13.94
0.00
0.01
0.00
0.06
0.00
0.28
0.00
0.57
55 148
2.63
9.57
0.00
0.13
0.00
1.18
0.00
1.26
0.00
0.78
191 867
Rio Azul National Park
Cerro Cahui Reserve
El Zotz Reserve
Dos Lagunas Reserve
Laguna del Tigre
Reserve
61 763
555
34 934
30 719
45 168
2.95
0.03
1.69
1.47
2.17
0.00
0.20
0.04
0.00
0.00
0.00
0.06
0.05
0.00
0.00
0.00
0.00
0.03
0.00
0.01
0.00
0.26
0.18
0.00
0.08
Multiple use zone
BuVer zone
826 351
408 973
40.08
22.85
0.05
0.74
0.16
2.71
0.25
3.76
0.25
3.28
Total area
2 000 396

1944
S. A. Sader et al.

Forest change monitoring of a remote biosphere reserve
1945
clearing rates in 1995–97 (table 2). The reduced clearing rate in 1995–97 does not
appear to be a function of the absence of large forest tracts remaining to sustain the
continued high clearing rates. Most of the cells with the highest clearing rates were
in the buVer zone. Grid cells with higher rates of forest clearing were associated with
closer proximity to roads and river access corridors. A previous study (Sader 1995)
indicated that over 90% of new forest clearings between 1986 and 1990 were within
3 km of a road or river.
Laguna del Tigre National Park (Ž gure 1) provides a prime example of the road
and forest clearing relationship. Laguna del Tigre National Park experienced very
low forest clearing rates in the earlier time periods (1986–90, 1990–93) but the rate
increased to approximately 0.6% per year from 1995 to 1997 (table 1). A closer look
at these grid cells (Ž gure 4) indicates that the recent activity is clustered around an
oil road that enters the park from the south. Several of the cells exhibit forest clearing
in the 1.0–4.0% category with one cell exceeding 4.0% per year (Ž gure 2).
A Z-test on the error matrices of the two image interpreters resulted in no
signiŽ cant diVerence, therefore the test plots of both interpreters were pooled for
reporting. The overall accuracy of the change detection map (1990–1997) was 86.5%
and the estimated Kappa coeYcient was 0.82 (table 3). The user’s accuracy for three
Figure 3. Number of UTM grid cells in four forest clearing categories for four time periods.
Table 2. Number of UTM grid cells with diVerent levels of forest change in four change
detection time periods.
Forest change (per year)
1986–90
1990–93
1993–95
1995–97
>
>
>
0.1%
4.0%
6.0%
61
0
0
81
15
9
89
19
13
100
20
6

1946
S. A. Sader et al.
Figure 4. Spatial distribution of time-series forest clearing for nine UTM grid cells along a
road transecting the Laguna del Tigre National Park, Peten, Guatemala.
out of four categories was over 90% , however the 1993–95 user’s accuracy was lower
at 71.1% . Producer’s accuracy ranged from 80.2% (1990–93 change) to 92.9%
(1995–97 change). The user’s and producer’s accuracy are related to commission
and omission errors in the mapping results (Congalton 1991).

Forest change monitoring of a remote biosphere reserve
1947
Table 3. Accuracy assessment error matrix with overall percentage correct, user’s and
producer’s accuracy and Kappa statistic.
Reference data
Forest change
1993–95
No change
Forest
ClassiŽ ed data
Forest change
User’s
accuracy
1995–97
1990–93
Total
1995–97
1993–95
1990–93
No change forest
Total
79
2
3
1
85
5
64
2
0
71
0
22
93
1
116
3
2
3
87
90
101
47
325
90.8%
71.1%
92.1%
95.7%
45
53
84.9%
Producer’s accuracy 92.9%
90.1%
80.2%
Overall classiŽ cation accuracy: 86.5%
Kappa statistic: 0.82
4. Conclusions
A limitation of image diVerencing or ratio image diVerencing as a change detection
method is that it does not specify the ‘from-to’ land cover type change information
(Jensen 1996), but only detects that a change did or did not occur. Another disadvant-
age of image diVerencing is the subjectivity and diYculty in deŽ ning a threshold
value that unambiguously separates change and no change events. The RGB-NDVI
classiŽ cation produces distinct change/no change cluster classes and circumvents the
subjective and often tedious decision criteria to determine a threshold digital value
between change and no change events at each time sequence. In labelling and
grouping the change versus no change cluster classes, we intentionally included
questionable ‘change’ clusters (potential commission error) to reduce the possibility
that we omitted a cluster class that included potential forest change pixels. We
allowed commission in our initial cluster grouping because we included a procedure
to edit out known false change readings at the second stage of analysis (GIS screen
digitizing). These areas were usually located along the edges of water bodies or in
and around seasonally inundated wetlands.
As the overall accuracy of the change detection results exceeded 85% correct,
the level that the US Geological Survey (Anderson et al. 1976) uses as a threshold
to deŽ ne acceptability, we conclude that our results are of suYcient accuracy to
recommend the satellite change detection method to monitor the rates and spatial
location of forest change in the MBR.
A simple, fast and eVective change detection technique was desired to facilitate
transfer of methods to the NGOs and government agencies in the Peten, Guatemala.
One of the major NGOs, Conservation International’s ProPeten programme, has
adopted the satellite forest change detection mapping in their ecological monitoring
programme. Kristensen et al. (1997) stated, ‘for ProPeten this map has been our
most powerful monitoring tool’. GIS specialists and forestry technicians will perform
future change detection updates relying on minimal training and computer resources
with modest memory and hard disk storage capacity.
The NDVI technique does not utilize the full spectral information content avail-
able on the Landsat-TM sensor (particularly the middle infrared re ected wave-
bands). However, the forest clearing events occurring in northern Guatemala are

1948
S. A. Sader et al.
almost exclusively the proximate source of slash-and-burn farming or clearing for
pasture development. The ‘clearcut’ removal of all trees between time sequences is
an abrupt canopy alteration that is highly detectable on visual RGB-NDVI colour
composite images and can be categorized through the unsupervised RGB-NDVI
classiŽ cation method. This Ž rst level forest change detection method does not cat-
egorize land cover change. Monitoring of time-series land cover/land use change
and second growth forests are the topic of on-going research in the Peten.
We have reported the time-series forest change estimates at the management unit
level, which is important to policy makers and NGOs responsible for conservation
programmes. Forest clearing estimates in some of the larger units can appear to be
deceptively low when averaged over the entire unit because the majority of the forest
is still inaccessible and relatively undisturbed (e.g. core areas and multiple use zone).
Many of the forest clearing activities are conŽ ned to local areas around towns and
road or river access corridors in these larger land units. Reporting by UTM grid
cell better deŽ nes the active fronts of clearing or ‘hot spots’ which may be targeted
for immediate conservation management decision making through community-based
participation to mitigate the continual loss of forest cover.
The UTM grid network is proposed as the Ž rst level of a multi-level forest
monitoring system for the MBR to provide a localized unit for reporting forest
change estimates and as a sampling frame for selection of second and third stage
samples for more detailed ecological studies at the forest community and species
level (Kristensen et al. 1997). Development of the multi-level monitoring system is
still in the conceptual stage, however independent monitoring has been initiated at
each level.
References
Anderson, J. R., Hardy, E. E., Roach, J. T., and Whitmer, R. E., 1976, A land use and land
cover classiŽ cation system for use with remote sensor data. Geological Survey
Professional Paper 964. (Washington, DC: US Geological Survey).
Avery, T. E., 1967, Forest Measurements. (New York: McGraw-Hill).
Bickford, C. A., 1952, The sampling design used in the forest survey of the Northeast. Journal
of Forestry, 50, 290–293.
Bonner, G. M., 1975, Cluster sampling with large-scale aerial photography in forest inventor-
ies. Info Rep. FMR-X-80, Environment Canada Forest Service, Forest Management
Institute, Ottawa, Ont.
Brondizio, E., Moran, E., Mausel, P., and Wu, Y., 1996, Land cover in the Amazon estuary:
linking of thematic mapper with botanical and historical data. Photogrammetric
Engineering and Remote Sensing, 62(9) 921–929.
Cohen, W. B., Fiorella, M., Gray, G., Helmer, E., and Anderson, K., 1998, An eYcient
and accurate method for mapping forest clearcuts in the PaciŽ c Northwest using
Landsat imagery. Photogrammetric Engineering and Remote Sensing, 64, 293–300.
Congalton, R. G., 1991, A review of assessing the accuracy of classiŽ cations of remotely
sensed data. Remote Sensing of Environment, 37, 35–46.
Coppin, P. R., and Bauer, M. E., 1996, Digital change detection in forest ecosystems with
remote sensing imagery. Remote Sensing Reviews, 13, 207–234.
ERDAS, 1996, ERDAS Imagine, version 8.2 (Atlanta: ERDAS, Inc).
Fearnside, P. M., 1986, Spatial concentration of deforestation in the Brazilian Amazon.
Ambio, 15, 74–81.
Food amd Agriculture Organization of the United Nations (FAO), 1993, Forest
Resources Assessment 1990. Tropical Countries. FAO Forestry Paper 112 (Rome: FAO).

Forest change monitoring of a remote biosphere reserve
1949
Foody, G. M., Palubinska, S. G., Lucas, R. M., Curran, P. J., and Honzak, M., 1996,
Identifying terrestrial carbon sinks: classiŽ cation of successional stages in regenerating
tropical forest from Landsat TM data. Remote Sensing of Environment, 55, 205–216.
Hall, F. G., Strebel, D. E., Nickeson, J. E., and Goetz, S. J., 1991, Radiometric rectiŽ cation:
toward a common radiometric response among multidate, multisensor images. Remote
Sensing of Environment, 35, 11–27.
Jensen, J. R., 1996, Introductory Digital Image Processing, a Remote Sensing Perspective, 2nd
edn (Upper Saddle River, New Jersey: Prentice Hall).
Kristensen, P. J., Gould, K., and Thomsen, J. B., 1997, Approaches to Ž eld-based monitoring
and evaluation implemented by Conservation International. Proceedings and Papers
of International Workshop on Biodiversity Monitoring, Brazilian Institute for
Environment and Renewable Natural Resources, Pirenopolis, Brazil, 22–25 June 1997,
pp. 129–144.
Lambin, E. F., 1996, Change detection at multi-temporal scales: seasonal and annual
variations in landscape variables. Photogrammetric Engineering and Remote Sensing,
62, 931–938.
Lillesand, T. M., and Keifer, R. W., 1994, Remote Sensing and Image Interpretation, 3rd edn
(New York: John Wiley and Sons).
Lundell, C. L., 1937, The Vegetation of the Peten. Publication No. 478 (Washington, DC:
Carnegie Institute).
Lyon, J. G., Yuan, D., Lunetta , R. S., and Elvidge, C. D., 1998, A change detection
experiment using vegetation indices. Photogrammetric Engineering and Remote Sensing,
64(2), 143–150.
Paivinen, R., 1994, IUFRO International Guidelines for Monitoring. IUFRO World Series,
Vol. 5 (Vienna, Austria: International Union of Forest Research Organizations
(IUFRO) Secretariat).
Sader, S. A., 1995, Spatial characteristics offorest clearing and vegetation regrowth as detected
by Landsat Thematic Mapper imagery. Photogrammetric Engineering and Remote
Sensing, 61, 145–151.
Sader, S. A., and Winne, J. C., 1992, RGB-NDVI colour composites for visualizing forest
change dynamics. International Journal of Remote Sensing, 13, 3055–3067.
Sader, S. A., Sever, T., and Smoot, J. C., 1996, Time series tropical forest change detection:
A visual and quantitative approach. Proceedings of the International Symposium on
Optical Science, Engineering and Instrumentation, Denver, Colorado. In Multispectral
Imaging for Terrestrial Applications, vol. 2818, edited by B. Huberty, J. B. Lurie, J. A.
Caylor, P. Coppin, and P. C. Roberts (Bellingham, Washington: International Society
of Optical Engineering), pp. 2–12.
Sader, S. A., Sever, T., Smoot, J. C., and Richards, M., 1994, Forest change estimates for
the northern Peten region of Guatemala—1986 to 1990. Human Ecology, 22, 317–322.
Sader, S. A., Stone, T. A., and Joyce, A. T., 1990, Remote sensing of tropical forests:
An overview of research and applications using non-photographic sensors.
Photogrammetric Engineering and Remote Sensing, 55, 1343–1351.
Schwartz, N. B., 1990, Forest Society (Philadelphia: University of Pennsylvania Press).
Sellers, P. J.,1985,Canopy re ectance,photosynthesis and transpiration. International Journal
of Remote Sensing, 6, 1335–1372.
Singh, A., 1986, Change detection in the tropical forest environment of northeastern India
using Landsat remote sensing and tropical land management. In Remote Sensing and
Tropical Land Management, edited by M. J. Eden and J. T. Perry (New York: John
Wiley & Sons), pp. 237–254.
Skole, D., and Tucker, C., 1993, Tropical deforestation and habitat fragmentation in the
Amazon: Satellite data from 1978 to 1988. Science, 260, 1905–1910.
Southgate, D., and Basterrechea, M., 1992, Population growth, public policy and resource
degradation: The case of Guatemala. Ambio, 21, 460–464.
Stott, C. B., 1968, A short history of continuous forest inventory east of the Mississippi.
Journal of Forestry, 66, 834–837.
Stuart, G. E., 1992, Maya heartland under siege. National Geographic, 182(5), 94–107.
Tucker, C. J., 1979, Red and photographic infrared linear combinations for monitoring
vegetation. Remote Sensing of Environment, 8, 127–150.

1950
Forest change monitoring of a remote biosphere reserve
Whitacre, D. F., Madrid, J., Marroquin, C., Dubon, T., Jurado, N. O., Tobar, W. R.,
Gonzalez, B., Arevalo, A., Garcia, G., Schulze, M., Jones, S. L., Sutter, J., and
Baker, A. J., 1995, Slash-and-burn farming and bird conservation in northern Peten,
Guatemala. In Conservation of Neotropical Migratory Birds in Mexico, Miscellaneous
Publication 727, edited by M. H. Wilson and S. A. Sader (Orono: Maine Agricultural
and Forest Experiment Station), pp. 215–255.
White, D., Kimmerling, J., and Overton, W. S., 1992, Cartographic and geometric compon-
ents of a global design for environmental monitoring. Cartographic Geographic
Information Systems, 19, 5–22.