Publication Date:
2011-01-01
Description:
Remote sensing surveys for estimating forest cover may be divided into two approaches: wall-to-wall and sampling. Sampling approaches offer a practical alternative to wall-to-wall mapping, but estimates of forest cover may be affected by the sampling rate of the estimation area. This study aimed to obtain stable estimates of forest cover from satellite data using object-oriented classification at the national level. We investigated a suitable value for the scale parameter in object-oriented classification using eCognition software to identify land cover types, and we evaluated the sampling rate for estimating forest cover at the national level. We used eight different scale parameters when applying object-oriented classification to Landsat data for a set of forty-six 10 km × 10 km sampling tiles centered at each degree of latitude and longitude in Japan. The scale parameter of 10 or less was found suitable for obtaining objects with areas of about 5 ha. Overall accuracy in classification was greater than 75% and greatest when the scale parameter was between 6 and 10. We then analyzed the entire land area of Japan using 10 km × 10 km tiles to evaluate the optimum sampling rate for estimating forest cover. A sampling rate greater than 20% was required to stably estimate forest cover in Japan.
Print ISSN:
0045-5067
Electronic ISSN:
1208-6037
Topics:
Agriculture, Forestry, Horticulture, Fishery, Domestic Science, Nutrition
Permalink