ul 2004
—M
dat set:
1ational
2004.
od with
ind use
(9), pp.
MONITORING THE AMAZON WITH DIFFERENT SPATIAL AND TEMPORAL
RESOLUTION
G. Zimmermann, W. Bijker
ITC, P.O. Box 6, 7500 AA Enschede, The Netherlands - (Zimmermann, bijker)@itc.nl
KEY WORDS: Forestry, Monitoring, Combination, Radar, Resolution, Spatial, Temporal
ABSTRACT:
When monitoring deforestation, frequent images with an optimal spatial resolution are required. But in reality either images with
low spatial resolution and a high revisit frequency or high spatial resolution images with a low revisit time are available. Combining
different spatial and temporal resolutions could solve this problem. The study site in the Colombian Amazon contains small fields
within the forest. The input data consisted of one high spatial resolution AIRSAR image and a time series of nine ERS-1 images of
medium spatial resolution, and their supervised classifications. The AIRSAR image was upscaled with stepwise upscaling based on
interim results and by direct upscaling from the same basis to different levels of spatial resolution. The comparison showed that the
proportion of land cover classes did not change substantially in either of the two upscaling approaches, while the number and size of
the patches showed a clear decrease with continuing upscaling. The direct upscaling approach provided best results. Furthermore,
the conformity of the upscaled AIRSAR land cover map and the ERS-1 land cover maps was determined. For the study area with its
particular land cover pattern, the effect of the spatial resolution on classification was not as important as expected. The fact that
AIRSAR has three fully polarimetric bands, while ERS-1 has only one band and one polarization was a more important cause for
differences between the land cover maps than the differences in spatial resolution.
1. INTRODUCTION
The Amazon forest is the largest tropical forest in the world.
The Greater Amazon region in South America has to cope with
large amounts of damage from deforestation affecting the
region itself, as well as global ecosystems through its influence
on climate and hydrology. Many governmental and non-
governmental organizations are therefore interested in regular
updates of information on the forest. Monitoring based on
remotely sensed imagery is a logical choice, because the area is
vast and inaccessible. Previous research on tropical
deforestation used images with low spatial resolution (e.g.
Cross et al., 1991; Malingreau et al., 1989; Mayaux et al., 1995)
as well as medium spatial resolution imagery (e.g. Skole et al.,
1993).
When monitoring the earth's surface with remote sensing,
problems like high costs for high spatial resolution imagery and
image processing as well as spatial and temporal resolutions
that are sub-optimal for the process to be monitored are
encountered. If monitoring a process over a specific time, one
will need frequent images with an optimal spatial resolution.
But in reality either images with low spatial resolution and a
high revisit frequency or high spatial resolution images with a
low revisit frequency are available. A combination of imagery
with different spatial and temporal resolutions may be
considered to overcome these problems, c.g. to reduce costs and
üme of image acquisition and processing while maintaining
required spatial and temporal detail.
For a number of change processes, both the process itself and
its speed are known or can be predicted across the study area.
Deforestation occurs mainly and most rapidly along the fringes
of the forest and close to roads and rivers.
Much research has already been carried out on integration of
data of different spatial resolution and the generalization of
data. However, the temporal dimension, how this works out in a
957
monitoring system has not received so much attention yet. Thus
it is of particular relevance to focus the research on combining
the spatial and the temporal aspect.
1.1 Research objective
The objective of the research is to assess whether a combination
of low spatial resolution and high spatial resolution imagery
gives better results than only using frequent low spatial
resolution or only infrequent high spatial resolution.
1.2 Research questions
Does a combination of low spatial resolution and high spatial
resolution imagery give better results, in terms of higher
accuracy, more thematic detail, than only using frequent low
spatial resolution or only infrequent high spatial resolution?
* Did the upscaling method affect the results of the
classification of the high spatial resolution data? And
if so, how?
= Did parts of the data or information get lost or could
new information be gained during upscaling?
« What kind of pattern change occurs with a change in
the resolution?
* What is the degree of conformity between low and
high spatial resolution data?
= Can the better spatial detail of the high spatial
resolution images be interpolated over time while
using low spatial resolution images?