Selecting Representative High Resolution Sample Images for Land Cover Studies. Part 2: Application to Estimating Land Cover Composition

We tested the effectiveness of the Purposive Selection selected using the AVHRR tiles. A random sample was also selected for comparison. The domain AVHRR cover Algorithm (PSA, described in the companion first article) to accurately estimate land cover composition over a type fractions were then corrected using TM maps for the selected tiles, following the method of Walsh and Burk large area. The knowledge of land cover distribution over (1993). The land cover composition obtained through the large areas is increasingly more important for numerous combined “domain AVHRR/sample TM” data was then scientific and policy purposes. Unless a complete detailed compared with the domain TM coverage. We found that map is necessary, a sampling approach is the best stratPSA provided a representative sample to correct the egy for determining the relative proportions of individual AVHRR map, particularly for small sample sizes. Comcover types because of its cost-effectiveness and speed of pared to the random selection, PSA yielded more accuapplication. With coarse resolution land cover maps at rate results at all tested sampling fractions (up to 30% of continental or global scales increasingly becoming availall tiles). With a PSA sample of 7% (18%), the average able, the possibility arises of using such maps synergistiabsolute difference per class between the correct and the cally with a sample of high resolution satellite coverage. estimated fraction was 0.058% (0.043%). For the same The goal of such synergy would be to obtain accurate sample fractions, the average relative error per class was estimates of land cover composition over large areas as 16.1% (9.8%) for PSA and 24.5% (18.7%) for random well as the knowledge of local spatial distribution. We selection. The difference between PSA and random selecevaluated PSA performance for sample selection over a tions was significant at the 0.001 probability level. It is 136,432 km2 area (domain) in the BOREAS Region of concluded that the PSA strategy is an effective way to Saskatchewan and Manitoba, Canada. Two maps were combine coarse and fine resolution satellite data to obtain prepared for the domain, one based on NOAA Advanced expedient and cost-effective land cover information over Very High Resolution Radiometer (AVHRR, 1 km pixels) large areas. An important benefit of the synergistic comand one on LANDSAT Thematic Mapper (TM, 30 m). bination of the two maps is knowledge of land cover disAfter dividing the area into 134 tiles, a PSA sample was tribution at the landscape level. This is because the coarse resolution map provides the overall distribution * Canada Centre for Remote Sensing, Ottawa, Ontario patterns across the domain, while the fine resolution map † Intermap Technologies, Inc., Ottawa, Canada supplies the average composition of the coarse resolution ‡ Canadian Forest Service, Quebec City, Quebec pixels in each cover type. Thus, each coarse pixel can be § Canadian Forest Service, Victoria, British Columbia statistically divided into the component high resolution Address correspondence to Josef Cihlar, Canada Centre for Remote Sensing, 588 Booth St., Ottawa, Ontario, Canada K1A 0Y7. classes. Crown Copyright  2000 Published by Elsevier E-mail: [email protected] Science Inc. Received 14 December 1998; revised 5 May 1999.