Benthic mapping of coastal waters using data fusion of hyperspectral imagery and airborne laser bathymetry

One goal of mapping, the accurate classification of the object space, can be achieved by visual interpretation or analysis of relevant data. Most mapping of earth features relies on the latter method, and is realized using remote sensing. Various airborne sensors are used today for generating topographic and hydrographic mapping products. In this research, we combined data from airborne hyperspectral imagery and airborne laser bathymetry, using data fusion techniques, to map the benthic environment of coastal waters.; Airborne laser bathymetry (ALB) uses laser pulse return waveforms to estimate water depth. These signals are attenuated by the water depth and clarity. A portion of the waveform signal, the peak bottom return, is a function of the bottom reflectance, and therefore, the bottom type. The purpose of this research is to exploit the peak bottom return signal of ALB to obtain benthic information, and then use the information, in combination with spectral imaging information, to aid in benthic classification.; We used AVIRIS hyperspectral data and SHOALS ALB data, obtained over Kaneohe Bay, Hawaii, for this research. After preprocessing the datasets, the water attenuation effects were removed from the AVIRIS data using a radiative transfer model. A variant of this model, developed for this research, was used on the ALB dataset to correct for water attenuation, resulting in a parameter we defined as pseudoreflectance. We classified the resulting datasets using the Maximum Likelihood supervised classification technique. Accuracy assessments of the classifications showed overall accuracies of 80.2% and 66.9% for the AVIRIS classification and the SHOALS classification, respectively. The two classifications were merged using the Dempster-Shafer (D-S) decision-level data fusion method, using a priori weights from the Maximum Likelihood classifications. The resulting D-S classification had an overall accuracy of 87.2%. For comparison, we classified the AVIRIS data (corrected for water attenuation) combined with a depth channel, producing an overall accuracy of 85.3%. Kappa coefficient analysis of all four classifications resulted in 82% confidence that the Kappa coefficients of the D-S classification and the AVIRIS-plus-depth classification are different. Kappa confidence levels greater than 99% were calculated for all the other pairs of classifications.; The results indicate that ALB pseudoreflectance, computed from the peak bottom return waveform signals, contains information that aids in the benthic mapping process, and can be used in a sensor fusion algorithm with hyperspectral data to achieve greater accuracy in bottom classification. Further research into the computation of bottom reflectance from the ALB bottom return waveform may yield additional improvements.