Evaluation of forest and nonforest classification capability
of ILU image with dirrerent kinds of pixel size and coherence
Generation methodology
Chen Erxue1, Betlem Rosich2, Li Zengyuan3
1, 3Chinese Academy of Forest Beijing, China, 100091
2ESRIN of European Space Agency
Keywords: Forest mapping, Coherence,
Classification, ERS SAR Tandem
Abstract
Forest and non-forest classification
capability using Interferometric Land Use
(ILU) image with three kinds of pixel size
(50m, 75m and 100m) and two kinds of
coherence generation methodology were
evaluated. Ground truth data used included
Land Use Map of the Zengcheng County and
classification result based on Landsat TM
image covering part of the experiment site. It
was shown that there was not so much
difference for ILU image with different pixel
size to classify forest from non-forest. 50m-pixel-
size ILU image was preferred for
forest non-forest mapping. ILU image with
low resolution such as 100m and 75m can
also be used without too much accuracy loss
compared with 50m. But if the National
Stand: Specification for Making Photoplan
of Remote Sensing should be in conformity
to, only the 50m ILU image can be used for
1:250 000 forest mapping; The 75m and
100m ILU image can be used for 1:500 000
and 1:1 000 000 forest mapping. And new
coherence methodology used in this test had
no help to improve the forest classification
accuracy.
Introduction
ERS SAR Tandem data is proved to be
extremely useful for discriminating forest
from non-forest in many areas of the world
[1,
2]. So one project named “Mapping China
Forest with ERS SAR Tandem”
[3]was
proposed with the aim to produce a forest
map of China using ILU images generated
with the Interferometric Quick Look (IQL)
System at ESRIN. The final project output
includes ILU color mosaic image with forest
and non-forest vector layer in different map
scale varying from 1:250 000 to 1:1 000 000.
It’s not necessary and also impossible to
produce forest map of such a big area as one
whole province or the entire China using full
resolution ILU image. So what kinds of pixel
size or resolution ILU image is best for the
generation of one certain scale forest map
should be studied at first. This evaluation
work has been carried out using the
experiment site located in the South China-Zengcheng
County where ground true
database such as Land Use Map, Landsat
TM image, DEM etc. has been established.
Moreover, Two kinds of methodology will
be evaluated here. We want to know if the
classification result will have some
improvement using the coherence image
with topography correction.
ILU Images and Ground Truth Data
ILU Images Used for Evaluation
The experiment site, Zhengcheng County
is located in the South China,
23°06'~23°37'N, 113°29'~113°59' E, the
coverage of this county is about 2800 km
2.
The INSAR data used for the ILU image
generation is showed in the table 1.
Master and slave intensity images with
pixel size of 50m, 75m and 100m and
coherence images processed by an ordinary
methodology without topography correction
and a new methodology with topography
correction were generated using IQL System
at ESRIN.
Table 1 ERS-1 and ERS-2 Tandem INSAR data used
|
|
Items |
Date | Track
| Frame |
|
| ERS-1 |
1996-03-02 | 24211
| 3141
|
| ERS-2 | 1996-03-03
| 4538
| 3141
|
|
For each kinds of pixel size, the master
and slave intensity images were used to
produce Difference intensity image and
Mean intensity image. The Difference image,
Mean intensity image and Coherence image
were combined to produce a color composite
ILU image with Coherence image as Red,
Mean intensity image as Green and
Difference intensity image as Blue. So the
output are six ILU images with the
combination of three kinds of pixel size and
two kinds of coherence methodology.
Ground True Data
Two kinds of ground truth data are
available for the evaluation of forest
mapping accuracy. The first one was the
Land use map of Zengcheng County that
was produced in 1990. As showed in figure
1, there were eight kinds of landuse type:
forest (dark green), orchard (chartreuse), rice
(orange), dry land (tan), grass land (cyan),
farmland (tan), water body (blue) and urban
(pink). As one thematic image layer, landuse
types can be recoded to produce forest (dark
green) and non-forest layer (tan) (Fig. 2),
which would be used for ILU image
classification accuracy evaluation. The
second one was based on one scene of TM
image acquired in March 1996. For one
small part TM image of Zengcheng County,
where forest and non-forest area can be
clearly identified (Fig.3), supervised
classification was applied to produce one
small forest and non-forest map. The
resulted forest map (Fig.4) was used as
another kinds of ground truth data to
evaluate forest and non-forest classification
accuracy of ILU images.
 |
|
Fig.1 Land Use Map of
Zhengcheng County
|
Fig.2 Forest & non-forest
map derived from fig. 1
|
|
|
|
Fig.3 Tm image covering
part of
Zhengcheng County
|
Fig.4 Forest and non-forest
map derived from TM
|
