In: Wagner W., Szdkely, B. (eds.): ISPRS TC VII Symposium - 100 Years ISPRS, Vienna, Austria, July 5-7, 2010, IAPRS, Vol. XXXVIII, Part 7B
In: Wagner
108
and for the Huambo province in Angola, from where there are
high resolution imagery and/or ground data available.
2. STUDY AREA AND DATA PREPARATION
2.1 Study area
16TTE 16-01=
Guinea-Bissau (GB) is a country located in the west coast of
Africa with approximately 36125 km 1 2 . It has a continental
mainland and a group of islands, the Bijagos Archipelago
(Figure 1). This country is characterized mainly by a flat land
surface with altitudes below 100 meters. Two large regions area
found, namely, the coastal low land area, consisting largely of
mangroves and swamps stretching from north to south and the
inland, where open forests, as well as, closed forests dominate
(especially in the southwest). Savanna woodland is present in
the northern and eastern regions of the country (White, 1983).
This distribution is dependent on different soils characteristics
and mainly on differences in rainfall regimes between north and
south, and east and west (Catarino, 2004). Two marked seasons
can be observed in this region, characterized by a tropical
climate: a dry season between November and April and a wet
season between May and October.
The Huambo Province is located on the central plateau of
Angola (Figure 1) and has an area of about 34270 km 2 (USAID,
2008). The region includes the higher mountains and the
highest peak in Angola, the Moco Mountain with an elevation
of 2620 meters, and has a wet season from October and April
and a dry season between May and September. The dominant
vegetation is mainly composed of miombo and savanna
woodlands, with grasslands covering large areas of lower
drainages. The miombo floristic formation is dominated by three
species, Brachystegia spp., Combretum spp. and Julbemardia
spp, whereas in savannas there is a dominant grass layer with
Hyparrenia spp and Androgon spp..
2.2 Landsat data and Field data
For GB, a dataset of five scenes per date of Landsat Thematic
Mapper (TM) and Enhanced Thematic Mapper (ETM+) was
used for 1990, 2002 and 2007, in a total of fifteen images. The
images were from the late dry season, except for 1990. A
geometric correction was performed with ground control points,
resulting in a RMS error less than 1 pixel. The data were
resampled into Universal Transverse Mercator projection Zone
28 north, datum WGS84 with a spatial resolution of 25 meters.
Field data were collected in GB for Mangrove, Closed forest,
Open forest and Savanna-woodland, in 293 geo-referenced
sample plots selected from a stratified random origin regular
net, surveyed in three campaigns conducted in 2007, 2008 and
2009. These data were used for classification accuracy
assessment.
For Huambo, a dataset of four images per date of Landsat TM
was available, free of charge, for the period 1990/1991/1992
and 2000/2001/2002 from the University of Maryland’s Global
Land Cover Facility (http://glcfapp.umiacs.umd.edu/) and
Landsat ETM+ data was available for 2008/2009 from the U.S.
Geological Survey (USGS) Earth Resources Observation
Systems (EROS) Data Center (EDC) (http://glovis.usgs.gov/).
The Landsat images were, already, geometrically corrected to
the Universal Transverse Mercator projection Zone 33 South
and with a spatial resolution of 30 meters.
A few forest inventory plots in areas of forest and savanna were
collected within the Huambo province and were later used in
the classification verification.
2.3 Global land cover maps
Land cover data, produced by three international initiatives
(IGBP, GLC2000 and MODIS) were freely available for this
study. The International Geosphere Biosphere Project (IGBP 1 )
(Loveland et al., 2000) distinguishes 17 land cover classes
according the science requirements of IGBPs core projects and
was produced based on NOAA-AVHRR imagery from 1992-
1993, at a spatial resolution of 1 km.
The Global Land Cover 2000 (GLC2000 2 ) (Fritz et al., 2003)
distinguishes 22 land cover classes, developed based on the
Land Cover Classification System (LCCS) and was derived
from 1 km SPOT4-VEGETATION daily data from November
1999 to December 2000.
The maps produced using the Moderate Resolution Imaging
Spectroradiometer (MODIS 3 ), (Strahler et al., 1999) adopt the
IGBP legend, and have two types of products, one based on
Terra (T) satellite data and one based on Aqua and Terra
combination (AT) satellite data, with a spatial resolution of 1
kilometre and 500 meters, respectively. The last one was
converted into a 1 Km grid database to have the same cell size
of the other global land cover products. These datasets were
produced annually since 2001 until 2007, and in this study two
maps were used for GB in 2002, one from Terra
(MODIST2002) and one from Aqua/Terra combination
(MODISAT2002). Another one was used for 2007
(MODISAT2007). For Huambo, also, two maps were used from
2001 (MODIST2001 and MODISAT2001) and one from 2007
(MODIS AT2007).
3. METHODS
3.1 Landsat land cover maps (LAND maps)
A legend with ten land cover classes was established for the
Landsat image classification in GB: Closed-forest, Open-forest,
Savanna-woodland, Mangrove, Grassland, Croplands/Bare soil,
Wet-vegetation, Burnt, Water, and Sand. This legend was
defined according to expert knowledge and pre-existing maps at
different scales. This information was used to train a supervised
classification tree algorithm. Three Land cover maps were
obtained for 1990 (LAND90), 2002 (LAND2000) and 2007
(LAND2007). Classification accuracy was assessed using data
collected in the field between 2007 and 2009 for four vegetation
classes: Mangrove, Closed Forest, Open Forest and Savanna-
woodland.
1 http ://edc2 .usgs. gov/glcc/tabgeoglobe.php
2 http://bioval.jrc.ec.europa.eu/products/glc2000/products.php
3 https://wist.echo.nasa.gov/
A legend
Huambo, i
vegetation i
miombo,
Croplands/!
supervised
likelihood <
1990 (LAb
The classii
data from
resolution ]
data colled
Table 1 - L
Discover (h
SIMP
Forest
(>30% tree
cover)
Savannas/
Shrublands
(10-30%
tree cover)
Grassland
Croplands/
Barren
Wetland
3.2 Lege
A simplifi
legend ac
cover ma]
Legend cc
Combinin
task due 1
global lan
GLC2000
than 15%.
savannas
between '
For LAN
greater th;