Stratification of natural vegetation for forest and rangeland inventory using Landsat digital imagery and collateral data

Abstract Natural vegetation can be effectively classified and stratified for resource inventory purposes in a Landsat-based automated procedure which utilizes three key elements of photointerpretation—tone, texture and terrain. Inventory analysis is accomplished by employing information derived from registered spectral and digital terrain databases. Tonal information is supplied by one or more Landsat digital images, and textural information is derived from the spectral data as the standard deviation of brightness values within a small window moving across the Landsat image. Terrain data are obtained from the U.S. National Cartographic Information Center digital terrain tapes, containing information derived from 1:250 000 topographic maps. These data are used to construct a digital terrain model from which elevation, slope angle, and slope aspect are obtained. The stratification procedure uses an unsupervised clustering algorithm to identify a large number of narrowly defined classes in the multichannel database which are merged and labelled according to their spectral characteristics and spatial relationships. Terrain channels may be used directly in the clustering procedure, or may be used in a concurrent procedure to model ecological relationships between strata and the landscape elements they prefer. Terrain data may also be used to correct for varying illumination due to topography The stratification procedure has been used for timber volume inventory in the Klamath National Forest, located in northern California, U.S.A., and this paper presents a status report of these efforts. The high relief and diverse ecological conditions throughout this forest have made it an excellent area to test forest classification and stratification techniques. In the stratification of the Klamath National Forest for timber inventory, three attributes characterized each stratum: tree height, tree stocking and regional type. Height and stocking were determined by analyst labelling of timber volume-homogeneous classes obtained by processing of spectral and textural data, Regional type refers to species composition, and is often defined by the dominant species in the stratum; e.g., red fir, douglas fir, ponderosa pine, or mixed conifer. Because forest composition varied systematically with terrain, regional type was modelled using elevation and aspect field data and a registered terrain model of the forest. Although the quantitative evaluation of the accuracy of the Klamath stratification is not yet complete, first results show that the Landsat-based strata have the same high ratio of among-group to within-group variance as do strata obtained by photointerpretation.