Fine-scale vegetation classification with remotely sensed imagery: How much does spatial resolution matter?

As the use of remote sensing imagery for ecosystem management on large-scale (> 500 ha) tracts of land continues to increase, little work has been conducted to assess its utility for smaller parcels.  Questions concerning how much detail is appropriate when choosing imagery to use for classification are of particular interest to land managers.  In this study, a dense grid-based field sampling method (resulting in  ca. 9.0 % property coverage) followed by agglomerative cluster analysis and ordination was used to identify all vegetation alliances and associations on a 148-ha demonstration tree farm in Maple Creek, CA.  A supervised classification using automated feature extraction was then performed on 3 image types of varying spatial resolutions (0.15 m 4-band aerial photo, 0.60 m 4-band satellite image, and 1 m 3-band satellite image). Resultant classifications were compared with the detailed vegetation map, derived from plot and image data, to assess accuracy.  Preliminary results show differences in classification accuracy between the 3 images with the 0.15m aerial photo producing the highest accuracy (77%), followed by the 0.6m image (68%) and the 1m image (62%) when taken to an alliance level.  These results highlight the potential benefits of using high spatial resolution (≤ 1m/pixel) imagery when attempting to classify vegetation at a fine scale.