Cluster analysis of structural stage classes to map wildland fuels in a Madrean ecosystem

• Published in: Journal of environmental management 2003-07, Vol.68 (3), p.239-252
• Main Authors: Miller, Jay D, Danzer, Shelley R, Watts, Joseph M, Stone, Sheridan, Yool, Stephen R
• Format: Article
• Language: English
• Subjects: Animal, plant and microbial ecology; Aves; Biological and medical sciences; Cluster Analysis; Decision Making; Desert Climate; Ecosystem; Fires - prevention & control; Forestry - statistics & numerical data; Fuels; Fundamental and applied biological sciences. Psychology; General aspects. Techniques; Geographic Information Systems - utilization; Hierarchal cluster analysis; Madrean vegetation; Maps as Topic; Models, Theoretical; Principal Component Analysis; Remote sensing; Risk Assessment; Southwestern United States; Southwestern US; Structural stage; Teledetection and vegetation maps; Trees - classification; Vegetables; Vertebrates: general zoology, morphology, phylogeny, systematics, cytogenetics, geographical distribution; Wildland fuels
• ISSN: 0301-4797; 1095-8630
• DOI: 10.1016/S0301-4797(03)00062-8

Geospatial information technology is changing the nature of fire mapping science and management. Geographic information systems (GIS) and global positioning system technology coupled with remotely sensed data provide powerful tools for mapping, assessing, and understanding the complex spatial phenomena of wildland fuels and fire hazard. The effectiveness of these technologies for fire management still depends on good baseline fuels data since techniques have yet to be developed to directly interrogate understory fuels with remotely sensed data. We couple field data collections with GIS, remote sensing, and hierarchical clustering to characterize and map the variability of wildland fuels within and across vegetation types. One hundred fifty six fuel plots were sampled in eight vegetation types ranging in elevation from 1150 to 2600 m surrounding a Madrean ‘sky island’ mountain range in the southwestern US. Fuel plots within individual vegetation types were divided into classes representing various stages of structural development with unique fuel load characteristics using a hierarchical clustering method. Two Landsat satellite images were then classified into vegetation/fuel classes using a hybrid unsupervised/supervised approach. A back-classification accuracy assessment, which uses the same pixels to test as used to train the classifier, produced an overall Kappa of 50% for the vegetation/fuels map. The map with fuel classes within vegetation type collapsed into single classes was verified with an independent dataset, yielding an overall Kappa of 80%.