Effects of stand size on the accuracy of remote sensing-based forest inventory

Abstract

The comparison of results of different forest studies is extremely difficult due to differences in test sites and studied stand characteristics, validation procedures, parameters used as an evaluation criteria, selection of stands, and the number of predictors used to name but a few. All these account for a large variation of the obtained accuracy. Additionally, in most reports inadequate information is given to convert statistically results from one study to the other. Since very few studies, such as Hyyppa/spl uml/ et al. (2000), exist where various remote sensing data sources and methods are verified in the same test site, much of the knowledge of the applicability of various data sources and methods for forest inventory has to be obtained by studies carried out in different tests sites. However, there is a single parameter, stand size, affecting strongly comparisons of forestry inventory results. The effect of stand size on the accuracy of remote sensing-based standwise forest inventory has not been reported extensively. The most dramatic changes occur at the level where stands are small. Not surprisingly, stand size has been successfully utilized as an auxiliary parameter in some studies. This paper describes how the accuracy of estimation is influenced by the stand size. Both spaceborne and airborne data are used in order to show that the effect is not just based on large pixel sizes or the effects of border pixels in spaceborne data. The accuracy of the following remote sensing data, SPOT Pan and XS, Landsat TM, ERS-1/2 SAR PRI and SLC, and airborne data from imaging spectrometer (AISA) is verified as a function of stand size in the range 1 to 20 ha. The paper presents curves that assist in converting results from one stand size to another and compares results of some studies in different test sites. Stand size seems to explain most of the variability of the results; however, for detailed comparison, more carefully described results are needed. Recommendations to design future forest studies are given in order to help the statistical conversion of results from one study to another.