The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
Publications Copernicus
Download
Citation
Articles | Volume XLII-3/W3
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLII-3/W3, 181–184, 2017
https://doi.org/10.5194/isprs-archives-XLII-3-W3-181-2017
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLII-3/W3, 181–184, 2017
https://doi.org/10.5194/isprs-archives-XLII-3-W3-181-2017

  20 Oct 2017

20 Oct 2017

DETECTING FORESTS DAMAGED BY PINE WILT DISEASE AT THE INDIVIDUAL TREE LEVEL USING AIRBORNE LASER DATA AND WORLDVIEW-2/3 IMAGES OVER TWO SEASONS

Y. Takenaka, M. Katoh, S. Deng, and K. Cheung Y. Takenaka et al.
  • Institute of Mountain Science, Shinshu University, 8304, Minamiminowa-Village, Kamiina-County, Nagano 399-4598, Japan

Keywords: Pine wilt disease, Tree mortality, Airborne laser scanning, WorldView-2, WorldView-3

Abstract. Pine wilt disease is caused by the pine wood nematode (Bursaphelenchus xylophilus) and Japanese pine sawyer (Monochamus alternatus). This study attempted to detect damaged pine trees at different levels using a combination of airborne laser scanning (ALS) data and high-resolution space-borne images. A canopy height model with a resolution of 50 cm derived from the ALS data was used for the delineation of tree crowns using the Individual Tree Detection method. Two pan-sharpened images were established using the ortho-rectified images. Next, we analyzed two kinds of intensity-hue-saturation (IHS) images and 18 remote sensing indices (RSI) derived from the pan-sharpened images. The mean and standard deviation of the 2 IHS images, 18 RSI, and 8 bands of the WV-2 and WV-3 images were extracted for each tree crown and were used to classify tree crowns using a support vector machine classifier. Individual tree crowns were assigned to one of nine classes: bare ground, Larix kaempferi, Cryptomeria japonica, Chamaecyparis obtusa, broadleaved trees, healthy pines, and damaged pines at slight, moderate, and heavy levels. The accuracy of the classifications using the WV-2 images ranged from 76.5 to 99.6 %, with an overall accuracy of 98.5 %. However, the accuracy of the classifications using the WV-3 images ranged from 40.4 to 95.4 %, with an overall accuracy of 72 %, which suggests poorer accuracy compared to those classes derived from the WV-2 images. This is because the WV-3 images were acquired in October 2016 from an area with low sun, at a low altitude.