information provided by the bathymetric LiDAR to perform the ‘‘depth-correction’’
step of hyperspectral imagery preprocessing (Lillycrop and Estep 1995; Bissett
et al. 2005), as described in Sect. 4.2.6. The capability for expanding the integration
of these two data types was accomplished by Lee and Tuell (2003), who developed
a similar methodology for depth-correcting the ‘‘pseudo-reflectance’’ data derived
from bathymetric LiDAR, producing an estimate of seafloor reflectivity at the laser
wavelength of 532 nm. Lee (2003) used maximum likelihood algorithms to derive
habitat maps from both depth-corrected hyperspectral imagery and seafloor pseudoreflectance imagery, and then integrated the output in a high-level data fusion
approach described by Park (2002). This high-level integration yielded a final
classification map of greater accuracy than either of the individual classifications
alone. Tuell and Park (2004) extended the utility of LiDAR data in the preprocessing of hyperspectral imagery by using LiDAR pseudo-reflectance images to
identify the homogeneous seafloor areas required to implement a hyperspectral
depth-correction scheme.
More recently, advanced modeling of environmental and sensor response
functions have provided the capability to extract water column attenuation and
perform radiometric calibration of LiDAR-derived reflectance data (Kopilevich
et al. 2005; Tuell et al. 2005a). These improvements in the processing of bathymetric LiDAR allow for the estimate of absolute seafloor reflectance, as opposed
to the more empirical pseudo-reflectance. Tuell et al. (2005b) demonstrated the use
of LiDAR-derived depth as a fixed constraint, and the LiDAR-derived water
column attenuation and absolute seafloor reflectance as weak constraints, for
implementing the radiative transfer inversion model to solve for absolute bottom
reflectance using hyperspectral imagery (expanding on hyperspectral processing
concepts presented in Sect. 4.3.6). Tuell and Lohrenz (2006) then introduced a data
fusion model to encapsulate the procedures for integrating LiDAR data with
hyperspectral imagery for benthic mapping. Wozencraft et al. (2007) reports that
LiDAR-derived water-leaving reflectance has also been used in a radiometric
balancing step for the hyperspectral imagery.
Current research focuses on using LiDAR-derived parameters to constrain a
combined atmospheric-oceanographic spectral optimization model for a variety of
seafloor and water column properties (Kim et al. 2010). Additionally, Park et al.
(2010) describe ongoing work to extract textural features from the LiDAR data.
All of the LiDAR-derived parameters (e.g., depth, LiDAR seafloor reflectance,
hyperspectral seafloor reflectance, and texture) can then be used as input to a
combined analysis of the receiver operating characteristic (ROC) and linear discriminant functions to identify which LiDAR and hyperspectral features best
inform the benthic classification process. Similar applications on land are using the
fusion of LiDAR data and hyperspectral imagery to improve classification of
landcover types and give new insights into changes in landcover (Reif et al. 2011).
The following sections describe a data fusion model for benthic mapping,
explain how LiDAR-derived parameters can be used to inform hyperspectral
preprocessing, and finally present the integration of hyperspectral imagery and
LiDAR to produce benthic classification maps. Because many of the techniques
176
J. M. Wozencraft and J. Y. Park
step of hyperspectral imagery preprocessing (Lillycrop and Estep 1995; Bissett
et al. 2005), as described in Sect. 4.2.6. The capability for expanding the integration
of these two data types was accomplished by Lee and Tuell (2003), who developed
a similar methodology for depth-correcting the ‘‘pseudo-reflectance’’ data derived
from bathymetric LiDAR, producing an estimate of seafloor reflectivity at the laser
wavelength of 532 nm. Lee (2003) used maximum likelihood algorithms to derive
habitat maps from both depth-corrected hyperspectral imagery and seafloor pseudoreflectance imagery, and then integrated the output in a high-level data fusion
approach described by Park (2002). This high-level integration yielded a final
classification map of greater accuracy than either of the individual classifications
alone. Tuell and Park (2004) extended the utility of LiDAR data in the preprocessing of hyperspectral imagery by using LiDAR pseudo-reflectance images to
identify the homogeneous seafloor areas required to implement a hyperspectral
depth-correction scheme.
More recently, advanced modeling of environmental and sensor response
functions have provided the capability to extract water column attenuation and
perform radiometric calibration of LiDAR-derived reflectance data (Kopilevich
et al. 2005; Tuell et al. 2005a). These improvements in the processing of bathymetric LiDAR allow for the estimate of absolute seafloor reflectance, as opposed
to the more empirical pseudo-reflectance. Tuell et al. (2005b) demonstrated the use
of LiDAR-derived depth as a fixed constraint, and the LiDAR-derived water
column attenuation and absolute seafloor reflectance as weak constraints, for
implementing the radiative transfer inversion model to solve for absolute bottom
reflectance using hyperspectral imagery (expanding on hyperspectral processing
concepts presented in Sect. 4.3.6). Tuell and Lohrenz (2006) then introduced a data
fusion model to encapsulate the procedures for integrating LiDAR data with
hyperspectral imagery for benthic mapping. Wozencraft et al. (2007) reports that
LiDAR-derived water-leaving reflectance has also been used in a radiometric
balancing step for the hyperspectral imagery.
Current research focuses on using LiDAR-derived parameters to constrain a
combined atmospheric-oceanographic spectral optimization model for a variety of
seafloor and water column properties (Kim et al. 2010). Additionally, Park et al.
(2010) describe ongoing work to extract textural features from the LiDAR data.
All of the LiDAR-derived parameters (e.g., depth, LiDAR seafloor reflectance,
hyperspectral seafloor reflectance, and texture) can then be used as input to a
combined analysis of the receiver operating characteristic (ROC) and linear discriminant functions to identify which LiDAR and hyperspectral features best
inform the benthic classification process. Similar applications on land are using the
fusion of LiDAR data and hyperspectral imagery to improve classification of
landcover types and give new insights into changes in landcover (Reif et al. 2011).
The following sections describe a data fusion model for benthic mapping,
explain how LiDAR-derived parameters can be used to inform hyperspectral
preprocessing, and finally present the integration of hyperspectral imagery and
LiDAR to produce benthic classification maps. Because many of the techniques
176
J. M. Wozencraft and J. Y. Park