Coastal and estuarine habitat mapping, using LIDAR height and intensity and multi-spectral imagery

The airborne laser scanning LIDAR (LIght Detection And Ranging) provides high-resolution Digital Terrain Models (DTM) that have been applied recently to the characterization, quantification and monitoring of coastal environments. This study assesses the contribution of LIDAR altimetry and intensity data, topographically-derived features (slope and aspect), and multi-spectral imagery (three visible and a near-infrared band), to map coastal habitats in the Bidasoa estuary and its adjacent coastal area (Basque Country, northern Spain). The performance of high-resolution data sources was individually and jointly tested, with the maximum likelihood algorithm classifier in a rocky shore and a wetland zone; thus, including some of the most extended Cantabrian Sea littoral habitats, within the Bay of Biscay. The results show that reliability of coastal habitat classification was more enhanced with LIDAR-based DTM, compared with the other data sources: slope, aspect, intensity or near-infrared band. The addition of the DTM, to the three visible bands, produced gains of between 10% and 27% in the agreement measures, between the mapped and validation data (i.e. mean producer's and user's accuracy) for the two test sites. Raw LIDAR intensity images are only of limited value here, since they appeared heterogeneous and speckled. However, the enhanced Lee smoothing filter, applied to the LIDAR intensity, improved the overall accuracy measurements of the habitat classification, especially in the wetland zone; here, there were gains up to 7.9% in mean producer's and 11.6% in mean user's accuracy. This suggests that LIDAR can be useful for habitat mapping, when few data sources are available. The synergy between the LIDAR data, with multi-spectral bands, produced high accurate classifications (mean producer's accuracy: 92% for the 16 rocky habitats and 88% for the 11 wetland habitats). Fusion of the data enabled discrimination of intertidal communities, such as Corallina elongata, barnacles (Chthamalus spp.), and stands of Spartina alterniflora and Phragmites australis, which presented misclassification when conventional visible bands were used alone. All of these results were corroborated by the kappa coefficient of agreement. The high classification accuracy found here, selecting data sources, highlights the value of integrating LIDAR data with multi-spectral imagery for habitat mapping in the intertidal complex fringe.