An approach to monitor pine
caterpillar using TM imagery
Dai Changda, Lei Liping, Hu Deyong, Wang Jiesheng
Remote
Sensing Satellite Ground Station
Chinese Academy of Sciences
P.O.
Box 8701 Beijing, China
Abstract
This paper
deals with an approach to monitor pine caterpillar damage by using TM
image. Through numerical analysis of samples, a set of TM data image
processing methods were developed accordingly. Calculating and classifying
the normalized perpendicular vegetation index and greenness change index
provided an effective measure to detect insect damage of forest. A damage
map with three levels varying from severe, light and unaffected was mode
out. The area of each level and its attribute percentage were counted
based on pixels and had satisfied accuracy comparing to the field
investigating results. This map showed comprehensively the caterpillar
damage status and could meet the requirement of practical utilization.
Introduction
In recent years the pine caterpillar
spreads widely and rapidly in China and causes damage to the forest not
less than the forest fire. We have carried out an experimental study on
monitoring caterpillar by the TM data with an aim to get timely precise
pest information which is important in taking action against the pests.
Study site and TM data
The study site is a state-owned
Gushan forest farm in Chuzhou county, Anhui province, located in the lower
middle reaches of Yangzi river. In that area valleys were cultivated for
growing rice, wheat, rape and other crops. The hill slopes were afforested
dominantly with Masson pine trees after 1950. Now the man-make pine forest
has grown up, but caterpillar becomes a problem. In the spring of 1988
more than half of the forest suffered different disaster of pest damage.
After treatment, the over sintering caterpillar were under controlling the
spring of 1989. Two WRS 123/28 scenes of TM data of April 23, 1988 and
April 26, 1989 were chosen. Comparison can, therefore, be made not only
among different field locations for 1988, but also between the same field
locations of 1988 and 1989.
Image processing and information
extraction
Caterpillars Consume pine needles, causing reduction of
leaf are and changing thermal status in the plant. This will decrease
brightness of TM4 and increase that of TM 3,5 and 7, . But these changes
are influenced by many interfering factors caused by complicated ground
objects and atmospheric conditions. The following processing stapes are
taken to enhance our required information.
- Statistical Analysis of Image
The statistical analysis of
the study site windows for 1988 and 1989 TM data shows: the mean
brightness of all bands except TM4 and the minimum brightness for all
bands are higher in 1988 while ratio of TM4 and mean brightness of TM4
are higher in 1989. This indicates the reduction of biomass in 1988 by
caterpillars.
- Rationing and Smoothing of Data
Spectral analysis of
forest areas in different pest-affected levels and non-forest ground
objects such as residential area, cultivated field, water body, etc.,
shows that they are very complicated in spectral characteristics and
dispersed widely in brightness. TO eliminate the interference from these
non-forest ground object, ratioing of TM4 to TM3 was carried out and
followed by 5X5 template smoothing.. These processing provides a means
to differentiate the pine forest by the complicated background.
- Perpendicular Vegetation Index (PVI) Calculation.
Sunlight can still seep through masson pine forests even though
they are very closed. This means that the spectral data of pine forest
on TM image include certain information of Soil. It also jam the damage
information. A couple of treatments including ratio-based indices and
PVI were tested and the PVI was proved to be the most sensitive symptoms
of the damage. Sample pixels in unaffected forests were of highest PVI
values, while the value decreased from those lightly affected to those
severely affected. Therefore using PVI to detect leaf area and biomass
may be favorable. PVI is the distance of a point in TM3/TM4 two
dimensional space to the soil (non-vegetation) line, while the soilline
is fitted from the points of non-vegetation ground object in the two
dimensional space [1]. For the chosen TM scenes of 1988 and 1989 the
soil lines are
| and |
Y = 19.11 + 0.83X
Y = 11.1 +
1.03X |
respectively.
The PVI value
for each forest pixel is calculated. The lower is the PVI value, The
more severe is the pest problem.
- Normalization of PVI and Use in Classification
As PVI
values were derived from non calibrated CCT data, which were still to
influence of noise by atmospherical condition sensor behavior and
elevation, etc., Normalization of PVI is necessary in order to make
comparing analysis more reliable and precise. According to formula of
Caloz (2).
NPVI = PVI/SE
Where SE is the deviation of
the soil line. The NPVI values for images of 1988 and 1989 are
calculated. Pixels with NPVI values between 96 to 120 belong to healthy
forest with the caterpillar density less than 1 piece per tree and
needle damage less then 10%. Light affected areas have NPVI in region of
75 to 95, with caterpillar density less than 3 pieces per tree and
needle damage less than 10-30% . The severely damaged areas have NPVI
vlues between 51 to 74, with caterpillar density lager than 3 pieces per
tree and needle damage more than 30%. The damage map for the
overwintering caterpillar in 1988 was made.
- Calculating and Classifying Greenness Change Index
In
order to study the feasibility of extracted change information of
damage, we conducted a precise geometrical correction according to
ground control points. The NPVI value of 1989 minus that of 1988 was
called Greenness Change Index (GCI), which indicated the increase of
biomass resulted from normal growth of pine trees from 1988 to 1989 and
the regenerated needles due to elimination of caterpillar by synthetical
measures after the plague. It was obvious that GCI values were usually
larger in severely affected area. Those in light or not affected areas
were comparatively small. According to sample data study, GCI values in
severely affected area were all larger than 20, light area 7 to 19. GCI
values less than 7 were associated with healthy growing plants. In this
way another damage map in the spring of 1988 was derived from data of
both years.
- Output of Damage Map, Counting Areas of Each Damage
Level
Two damage maps mentioned above showed similar distribution
of effected forests as expected. GCI method was not favorable to be
popularized to be popularized for its doubled costs of two temporal data
and their processing work. So we took only the classified NPVI damage
map as output by designating severely affected, lightly affected and
unaffected areas with red, yellow and green, while non forest areas with
black. Results and conclusion
The final damage map and
statistics of damage levels of 1988 indicates that total affected area
reaches 52% of the total pine forest, among which severely affected area
is larger than lightly affected area. They are 29% and 23% of total forest
area respectively.
In order to assess the accuracy of this result,
a hazard map based on field investigation conducted by the technician of
that farm was reduced to the same scale as the TM image. In spite of the
fact that the hazard map was drawn on the basis of forest subcompartment,
while the TM damage map was derived pixel by pixel, these two maps give
fairly good coincidence in results both in percentage of each damage level
(cf. table), and in distribution status. This proves reliable and can meet
the requirement of practical utilization in caterpillar prevention and
control.
Table. Results of detecting pine caterpillar hazard by two
methods
| Hazard level |
Percent of total forest area |
| From TM image |
by ground survey |
| Server |
29.10 |
30 |
| Light |
22.84 |
28 |
| Not affected |
48.06 |
42 | Literature
- Wang Jiesheng et al, Remote Sensing of Environment, China, Vol. 4,
No. 4, 1989. pp. 243-248.
- R. Caloz et al, Proceedings of IGARSS' 86 Symposium, Vol.III,p.p.
1471-1475.
|