AIRBORNE MULTISPECTRAL LIDAR DATA FOR LAND-COVER CLASSIFICATION AND LAND/WATER MAPPING USING DIFFERENT SPECTRAL INDEXES

Abstract

Airborne Light Detection And Ranging (LiDAR) data is widely used in remote sensing applications, such as topographic and landwater mapping. Recently, airborne multispectral LiDAR sensors, which acquire data at different wavelengths, are available, thus allows recording a diversity of intensity values from different land features. In this study, three normalized difference feature indexes (NDFI), for vegetation, water, and built-up area mapping, were evaluated. The NDFIs namely, NDFI<sub>G-NIR</sub>, NDFI<sub>G-MIR</sub>, and NDFI<sub>NIR-MIR</sub> were calculated using data collected at three wavelengths; green: 532 nm, near-infrared (NIR): 1064 nm, and mid-infrared (MIR): 1550 nm by the world’s first airborne multispectral LiDAR sensor “Optech Titan”. The Jenks natural breaks optimization method was used to determine the threshold values for each NDFI, in order to cluster the 3D point data into two classes (water and land or vegetation and built-up area). Two sites at Scarborough, Ontario, Canada were tested to evaluate the performance of the NDFIs for land-water, vegetation, and built-up area mapping. The use of the three NDFIs succeeded to discriminate vegetation from built-up areas with an overall accuracy of 92.51%. Based on the classification results, it is suggested to use NDFI<sub>G-MIR</sub> and NDFI<sub>NIR-MIR</sub> for vegetation and built-up areas extraction, respectively. The clustering results show that the direct use of NDFIs for land-water mapping has low performance. Therefore, the clustered classes, based on the NDFIs, are constrained by the recorded number of returns from different wavelengths, thus the overall accuracy is improved to 96.98%.