Traditional wetland mapping methods are in need of modernization. They typically depend solely on few dates of optical imagery, cloud-free data acquisition, and therefore surface features of interest are often obstructed, inaccurately mapped, or not present during data acquisition. Despite the limitations of data acquisition during cloud-free daylight only, multi-temporal multi-spectral optical data are still highly valuable for mapping wetlands and classifying wetland type. However, radar sensors are unique in that they are insensitive to atmospheric and low light conditions, and thus can offer more consistent multi-temporal image acquisition. Unique characteristics about surface scattering mechanisms, such as saturated extent of wetlands, can be found by utilizing both the intensity and phase information from multiple polarizations and multiple wavelengths of radar data. In addition, information from lidar can reveal important details about the variability and structure of surface features, and the potentiality of water to collect in certain areas.
The research presented in this dissertation will show important developments in wetland mapping by integrating several platforms of remotely sensed data, including: two sources of radar data including fully polarimetric RADARSAT-2 data (C-band) and dual-pol PALSAR data (L-band); two sources of optical data including Landsat TM imagery and aerial orthophotos, lidar point cloud data with intensity and derived topographic indices. Decision tree classification using the random forest model will be utilized to take advantage of the unique differences in these data. Assessments of outputs from random forest will be used to identify the most significant data sources for two levels of land cover classification: discriminating between water, wetland and upland areas, and sub-classifying wetland type. It is expected that results from this research will deliver a valuable, affordable, and practical wetland probability tool to aid manual photo interpretation, not to replace it.
University of Minnesota Ph.D. dissertation. May 2013. Major: Natural Resources Science and Management. Advisor: Dr. Joseph F. Knight. 1 computer file (PDF); xi, 179 pages.
Corcoran, Jennifer Marie.
Integrating data from several remotely sensed platforms to accurately map wetlands.
Retrieved from the University of Minnesota Digital Conservancy,
Content distributed via the University of Minnesota's Digital Conservancy may be subject to additional license and use restrictions applied by the depositor.