LiDAR intensity correction and its study on wetland classification

Abstract

Wetlands have received intensive interdisciplinary attention as a unique ecosystem and valuable resources. However, many wetlands in the world are poorly mapped, infrequently mapped or unmapped due to the poor accessibility of wetlands. As a new technology, the airborne LiDAR system has been applied in wetland research. However, most of the studies used only one or two LiDAR observations to extract either terrain or vegetation in wetlands. This research aims at developing new methods to integrate both spatial and radiometric information provided by the airborne LiDAR system to improve mapping and classification of wetlands. To guarantee the accuracy of classification result, the input LiDAR attributes need to be ascertained. For the radiometric information of LiDAR data, proper normalization of the return strength image from the whole survey is needed. In this study, a novel automatic method is proposed to reduce intensity errors in large scale and multiple strips projects. The method considers both intensity discrepancies in strip overlaps and specular reflections in nadir regions. An overlap-driven adjustment is firstly used to remove discrepancies and then, a Phong model weighted filter is used to correct specular reflections in nadir regions. Significant improvement in the radiometric image is demonstrated by a 4 strip project over a wetland area of the Yellow River Delta (YRD), China. After that, the potential of LiDAR's multiple attributes (DSM, DTM, off-ground features, Slope map, multiple pulse returns, and normalized intensity) and other information (aerial photos and tidal data) for wetland classification has been exploited, based on a multi-level object-oriented classification method. By using this method, we are able to classify the YRD wetland into eight classes (wet meadow, forested swamp, Phragmites, Low Land, impervious surface, river, sea, and intertidal zone), which provides much more details than conventional remote sensing methods. The overall classify accuracy is 92.5% which is better or similar to other remote sensing methods.