Vegetation Mapping of Kushiro Wetland in Northeast Hokkaido, Japan: Application of SPOT images, Aerial Balloon Photographs and airborne color near infrared ( CNIR ) Images for classification Methodology The method comprises five main stages, 1) ground truth and taking balloon photos 2) image calibration and mosaicking 3) photo interpretation and classification for base map. 4) application of balloon mosaic photos as training data for classification of aerial NIR color video images. 5) application of SPOT (HRV) images for the classification of whole area of wetland. Processing of satellite imagery and aerial balloon photography were undertaken in the ERDAS Imagine version 8.4 software (REDAS Inc.) digital image processing system and Arc View version 3.2 software (ESRI Inc.).  Figure 2. Ground truth points, vegetation research and GPS survey, July 26-27 in 2001 and Jun 30 - July 2 in 1998 In the first step, The vegetation research (July 26 - 27 in 2001 and Jun 30 - July 2 in 1998) was carried along the wooden path way in the direction of North-South, 59 points of 1m×1m size quadrates and 52 points of 10m×10m size quadrates were set. While in the direction of East-West, 52 points of 20m×20m size quadrates were set every 20m. The ground truth points were located in the " Figure 2 ". The position coordinates were measured using two sets of ground positioning system (GPS, Trimble Corp., 4600LS) by carrier phase DGPS. We tried balloons as platforms for monitoring of wetland vegetation to make vegetation map in detail. Aerial color photographs were taken by two tethered helium filled balloons from nadir view at 100-200 m height on the 2nd of July in 1998 ( Table 1 ). The observation area covers 230m×460m. In the second step, the photos were scanned (600dpi) and corrected for geometric distortions and combined into georeferenced image mosaics (15cm / pixel). The mosaicking of aerial balloon photographs was undertaken in the MOSAIC module ERDAS IMAGIN. These images were projected onto the UTM coordinates system by resampling and digital ortho-photo maps. In the third step, the digital processed vegetation map was constructed by mosaicked photos with manual photo interpretation and field validation was conducted by Arc View GIS version 3.2 software. In the fourth step, NIR color video (SILVACAM, VTT Automation Co., Finland) sequence images (30cm / pixel) were taken by an airplane (Dornier., Do228-200, National Aerospace Laboratory of Japan) from nadir view on 30 June, 1998 at 900 m height( Table 2 ). One 750×480 pixel size image was taken in every 1/30 second and the 500 video sequence images were overlaid using an algorithm [6] of tie points extraction based on image matching and registered to UTM WGS 84 coordination system using ground control points (GCPs). In the fifth step, Two remotely sensed satellite data sets were obtained. SPOT2 (HRV) images were acquired on 10 June 1998 and 17 June 2000. Images were ortho-rectified. Resampling was performed using the nearest neighbor algorithm. The whole of the Kushiro wetland vegetation was classified using the base map., June, July and August. The slight seasonal change was detected to cover the whole of the vegetation community types into the genus and species level. Lastly, the accuracy assessment was done about both of the base map and the satellite images.
Table 2. Performance of the airborne color near infrared (CNIR) video image
|