990
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008
2. MATERIALS AND METHODS
The materials used in this work were:
- Landsat images: MSS, TM and ETM+ from February 1973 to
February 2007, and Aster image (2003); visible, near infrared
and SWIR (short wave infrared) bands.
- Topographic maps (2754-19 and 2754-20-I), from the Militar
Geographic Institute (IGM),
- Digital Elevation Model (DEM (90m), from NASA-SRTM,
- Field data, to collect the structure, spatial arrangement and
plants composition of the forest.
- Softwares: ERDAS Imagine 9.0, Arcview 9.1, Idrisi, ENVI,
and statistics tools.
The work was divided in five steps:
a. Standarization of satellite data: several factors have
influence in quantification and qualification of solar radiation
reflected when different sensors are used along time. The main
sources of error for identification of physonomic-estructural
changes in vegetation among dates are: atmospheric conditions,
errors in image registration, topographic effects, sensor
variability, the abundance, composition and phenological
condition of vegetation. This step included:
a.l. Atmospheric correction: it was evaluated three methods of
Dark object substraction" (DOSl, DOS2, DOS3 and DOS4)
1234) (Chavez,1988; Song et al., 2000) and one of Dense
Dark Vegetation Approach" (DDV) (Kaufman et al.,1988).
a. 2 Geometric correction (georeferentiation): all the images
were co-registered from a Landsat TM 224/79 August, 1989 of
the University of Maryland, (resampling method: cubic
convolution, RMS less than one pixel, cartographic projection
system: UTM (WGS84), zone 20 South).
b. Generation of forest maps: a first categorization of
vegetation from the field data and satellite images processing
was carried out. The categories considered and its descriptions
were:
- Capuera: cane with small trees (height: 5m).
- Grassland: grassland (50-75%).
- Savannah with trees: grassland (50-75%), bare soil (0-20%),
trees (10-25%, Urunday: Anadenanthera colubrina).
- Capueron with one trees layer: cane with trees (height 15m),
dominant species: Helieta apiculata, Trichilla catigua,
Nectandra spp, Ocotea spp., Lonchocarpus leucantus, L.
muehlbergianus.
- Capueron with two trees layer: understorey species, cane,
first trees layer: 20m, second trees layer: 15m, dominant
species: Nectandra saligna, Diadenopterix sorbifolia, L.
muehlbergianus, L. leucanthus, Luhea divaricata, Helieta
apiculata, Trichilla catigua, Allophylus edulis.
- Mixed Forest in low land: dominant species: Helieta
apiculata, Nectadra spp. Patagonula americana,
Bolfourodendron redelianum, Holocalyx balansae,
Parapiptadenia rigida, Rupechtria laxiflora.
- Forest in high land: dominant species: Apuleia leicocarpa,
Enterolobium contorstiliquum, Tabebuia spp, Cedrela
tubiflora.
Grove:dominant species: Helieta apiculata, Nectandra spp,
Acacia spp., Chorisia speciosa, Melia azedarach
- Implanted forest: exotic species: Aleuritis fordii, Melia
azedarach Pinus spp,
In order to apply the mapping methods to wide scale, the forest
units of the first categorization were grouped into five classes: -
- Dense High Forest (MAA)
- Sparse High Forest (MAD)
- Low Forest (MB)
- Capuera
- Implanted forest
c. Generation of models of natural and human variables in
order to interpret the spatial distribution of plant communities
and predict its location in areas with difficult access. In this
point we generated two predictive models: one with natural
variables (height, slope, aspect, NDVI, etc) and other with
human variables (distance to: rivers, villages, roads, etc.).
These models were compared with the forest maps, and
statistical analysis were carried out.
d. Development of a methodology for multi-temporal
comparison of forest: the images were grouped by date in
summer and winter set data. Different classification methods
were evaluated to separate forest vs other covers as the
following:
- MCVT: Variance classification from application of texture
filters.
- CnSI: Unsupervised classification (ISODATA).
- CnSIcp: Unsupervised classification (ISODATA) from PCI
results.
After the selection of the best method, we follow with the
classification of different type of forest (with the same
categories of the forest maps). The methods evaluated were:
SAM, ISODATA, Maximum likelihood, PCI, Tasseled Cap,
NDVI with several combination among the individual results
of each process.
e. Multi-temporal comparison of natural and implanted forest:
this comparison was carried out by Change detection method
among best classified images (previous point). Also, the NDVI
differences'were analyzed, and both results were integrated.
3. RESULTS AND CONCLUSIONS
We can generate a methodology (a protocol) of standarization
of data satellite of different dates and sensors.
The analysis showed that the DOS2 model of atmospheric
correction (Chavez, 1988; Song et al., 2000) was the more
adequate for the study area. The details of the geometric
correction process were as appear in materials and methods.
It was possible from de satellite and field data analysis to
elaborate the forest maps to work in two scale of details.
In the Figure 2 we can see the forest map with nine units of
vegetation in the area of Cuna Piru Reserve with more details
than we can observe in the Fig. 3 where are represented only
four forest classes, while the plantations, the agricultural uses,
grassland, settlements are grouped as unclassified en the
Fig.3.
The methods to generate this last forest map can be used in
other areas near Cuna Piru Reserve to separate forest vs non