Authors: Jarlath O’Neil-Dunne, Sean MacFaden, Anna Royar, Maxwell Reis, Ralph Dubayah, and Anu Swatantran
Despite the plethora of data-acquisition programs collecting remotely sensed high-resolution imagery in the United States, few corresponding high-resolution statewide land-cover datasets have been developed. This is understandable given the challenges inherent in extracting information from massive, highly-variable datasets encompassing heterogeneous landscapes. To overcome these challenges during development of a statewide, high-resolution tree-canopy dataset for Maryland, USA, we designed and deployed a rule-based expert system for mapping land-cover features from multispectral imagery and LiDAR. This object-based approach facilitated integration of imagery and LiDAR into a single classification workflow, exploiting the spectral, height, spatial information contained in the datasets. Rule-based expert systems provided an intelligent approach to feature extraction, ensuring consistency in the output despite variability in collection parameters, data quality, and data completeness, among others. Finally, by distributing the processing load to multiple computing cores, we efficiently extracted land cover from remotely-sensed datasets constituting terrabytes of digital data, covering the entirety of Maryland’s 25,640 km2 (9,900 mi2) land area. We conclude that an object-based approach that incorporates expert systems and enterprise processing is a cost-effective method for statewide land-cover mapping.
You are reviewing the
Download.
Your machine is locate at 18.223.196.171.
