Cm. Trotter et al., ESTIMATION OF TIMBER VOLUME IN A CONIFEROUS PLANTATION FOREST USING LANDSAT TM, International journal of remote sensing, 18(10), 1997, pp. 2209-2223
Optimisation of economic return from forests requires that comprehensi
ve forest inventory data be available to support the design of harvest
ing strategies. Such inventory data can potentially be obtained by rem
ote sensing. This study investigates the accuracy with which wood volu
me (m(3) ha(-1)) in a plantation forest can be calculated from Landsat
TM data at the pixel and forest-stand spatial scales. Wood volumes we
re estimated from regression analysis, nonparametric line-fitting, and
an N-dimensional K-nearest-neighbour classification scheme. At the pi
xel scale, relations between Landsat data and measured wood volume wer
e found to be significant but weak, with r(2) values of less than or e
qual to 0.3, and with correspondingly poor estimates of wood volume (r
oot-mean-square errors-rmses-of > 100 m(3) ha(-1)). By averaging the p
ixel-scale estimates, wood volume estimates of acceptable accuracy wer
e obtained for forest-stand areas of about 40 ha (rmses of less than o
r equal to 46 m(3) ha(-1)). Parametric regression performed slightly b
etter overall than non-parametric line fitting techniques for estimati
ng wood volume. Estimates of similar accuracy to those obtained by reg
ression were also given by NK-classification at the pixel-scale, provi
ded K was large (greater than or equal to 15), although the classifier
produced biased results at the forest-stand scale. It is concluded th
at Landsat TM only provides an acceptable data source for estimating w
ood volumes in plantation forests for areas of about 40 ha and larger.
The very low dynamic range in the Landsat data is probably a signific
ant factor limiting its use for inventory at more detailed scales.