HYPERSPECTRAL ANALYSIS OF JAPANESE OAK WILT TO

Document Sample
HYPERSPECTRAL ANALYSIS OF JAPANESE OAK WILT TO Powered By Docstoc
					 HYPERSPECTRAL ANALYSIS OF JAPANESE OAK WILT TO DETERMINE NORMALIZED
                              WILT INDEX

    Kuniaki Uto, Yuji Takabayashi and Yukio Kosugi                                                                   Toshinari Ogata

                     Tokyo Institute of Technology                                                        Yamagata Prefectural Government
                         uto@pms.titech.ac.jp


                                                                                1. INTRODUCTION

Significant Japanese Oak Wilt has been wide-spreading in the Japanese Islands since the 1980’s [1]. The mass mortality is
caused by the mass attacks of Platypus quercivorus beetles, i.e. vectors of Raffaelea quercivora, into the Japanese oak trees.
The prevention and extermination based on early detection of attacked trees is the only measure to keep the mortality out at
the borders. The principal schemes are 1) early detection of attacked trees based on frequent inspection, 2) exterminate P.
quercivorus before the emergence to keep the mortality out at the borders, 3) prevention by prohibiting taking out the attacked
woods from the attacked area. Although, at present, the wilt distribution are acquired by mapping manually inspected data in
a field survey, the efficiency of the prevention and extermination is restricted by the complication in cost and precision, i.e.
1) difficulty in covering the whole area by manual inspection, 2) difficulty in distinguishing the wilt from autumnal tints by
manual inspection in autumn. We propose an automatic detection method based on remotely sensed hyper- or multi-spectral
data to overcome the problems. In this paper, we firstly introduce a normalized wilt index (NWI) in consideration of spectral
characteristics of withered leaves, and, then, we verify the method by applying the index to hyperspectral remote sensing data
and a multispectral satellite image.

                                                                                2. METHODOLOGY

Normalized difference indices, e.g. NDVI [2], are utilized widely for remote sensing data analyses. Because of the normalized
form which is not affected by the illumination condition much, the normalized difference indices are the most successful ones
to be applicable to a variety of data. In this section, we propose a normalized wilt index (NWI) based on two normalized
difference indices, i.e. NDVI and NDGI [3]. Spectral change of healthy Japanese Oak Tree leaves, according to the evolution of
autumn coloring, is shown in Fig.1(a). The spectral data are observed by an imaging spectrograph of visible and near infrared
(400-1000nm) with 5nm spectral resolution, 121 channels, 484 spatial pixels and 10bit/pixel. The reflectance is estimated based
on the radiance of the standard white reference. The numbers in the figure correspond to ’greenness’ based on manual decision
as shown in Fig.1(b): #1 to #6 are collected before leaf drop, #7 is the fallen dead leaf. The absorption of chlorophyll around
red band, i.e. 670nm, decreases in accordance with the progress of autumnal tints (#1-#6), and the spectral profile results in
smooth monotone increasing curve in 500-1000nm (#7). We pick out two characteristics to differentiate the dead leaves from

                             1
                                   #1
                                   #2
                            0.9
                                   #3
                                   #4
                            0.8    #5
                                   #6
                            0.7    #7
              reflectance




                            0.6

                                                                                                          #1    #2   #3
                            0.5
                                                                                                                          #4   #5
                            0.4                                                                                                      #6   #7
                            0.3


                            0.2


                            0.1


                             0
                             400        500   600    700     800   900   1000
                                               wavelength [nm]

                        (a) Spectral profiles of leaves                                                         (b) Image of leaves


                                                           Fig. 1. Spectral change of healthy Japanese Oak Tree leaves.
                                                                                                                          0.2




                                                                                                                          0.18




                                                                                                                          0.16




                                                                                                                          0.14



  (a) Color image of hyperspectral data                                                                                   0.12




                                                                                                                          0.1




                                                                                                                          0.08




                                                                                                                          0.06
                                     0.05

                                     0.045

                                     0.04

                                     0.035                                                                                0.04
                                     0.03

                                     0.025

                                     0.02
                                                                                                                          0.02
                                     0.015

                                     0.01

                                     0.005

                                     0                                                                                    0




(b) NWI distribution of hyperspectral data      (c) Pseudo color image of ASTER              (d) NWI distribution of ASTER

                                  Fig. 2. NWI distribution of hyperspectral and ASTER data.

the green and yellow leaves: 1)The first derivative between green (550nm) and red (670nm) is positive, 2)The second derivative
at red channel is positive. This tendency was prominent in the case of oak wilt due to P. quercivorus. We propose a NWI, a
product of the first and second derivatives, which is expected to indicate a higher value in dead leaves than in fresh green leaves.
                                  dR(λ)                 d2 R(λ)
                       NWI =                        ·                    = −N DGI · (N DV I + N DGI)                             (1)
                                   dλ        λ=λR         dλ2     λ=λR

Where, λR is wavelength at red (670nm), and R(λ) is reflectance at wavelength λ.

                                                3. EXPERIMENTAL RESULTS

At first, we applied the NWI to hyperspectral data which were observed by the imaging spectrograph in Mogami district, Yam-
agata in Japan on September 3, 2007. The data contains normal trees and dead oak trees by the mass attacks of P. quercivorus
(red regions in Fig.2(a)). The NWI distribution is shown in Fig.2(b) in which the dead trees are extracted appropriately from
a variety of trees. Secondly, the NWI is applied to ASTER VNIR data with 15m spatial resolution which were observed in
Shonai district, Yamagata in Japan on August 15, 2007. The reflectance is calculated based on the ground truth data of airport
apron and the surface of the sea. Fig.2(c) is a part of pseudo-color ASTER data around Mogami district, and Fig.2(d) is the
corresponding NWI distribution. In Fig.2(b),(d), we masked out all NDVI values less than 0.2 to exclude non-vegetation re-
gions. We confirmed that the result is consistent with the wilt map by manual inspection. The more detailed data are under field
investigation.

                                                           4. CONCLUSION

We proposed a normalized wilt index for the purpose of the automatic detection of Japanese Oak Wilt distribution based on
multispectral remote sensing data. We verified that the NWI is an effective index when applyed to the hyperspectral remote
sensing data and multispectral satellite images. Since the NWI is an index which evaluates quite simple characteristics, i.e. the
spectral profiles of dead leaves are smooth monotone increasing curves, more detailed verification based on various filed survey
data is indispensable.

                                                           5. REFERENCES

[1] N. Kamata, “Outbreaks of forest defoliating insects in japan, 1950-2000,” Bull. entomol. res., vol. 92, no. 2, pp. 109–117,
    2002.
[2] J.W. Rouse, Haas J.A., and D.W. Deering, “Monitoring vegetation systems in the great plains with erts,” Proc. Third Earth
    Resources Technology Satellite-1 Symposium, vol. I, pp. 309–317, 1973.
[3] Y. Minekawa, K. Uto, N. Kosaka, Y. Kosugi, H. Ando, Y. Sasaki, K. Oda, S. Mori, and G. Saito, “Salt-damaged paddy fields
    analyses using high-spatial-resolution hyperspectral imaging system,” Proc. IEEE International Conference on Geoscience
    and Remote Sensing Symposium, vol. 3, pp. 2153–2156, 2005.

				
DOCUMENT INFO
Shared By:
Categories:
Tags:
Stats:
views:6
posted:1/5/2012
language:
pages:2