The kNN method of combining a terrestrial forest inventory with remote sensing image analysis is applied in mountainous terrain. This poses problems of radiometric- topographic normalization, which are addressed in this paper. The SCS (sun-canopy-sensor) method, extended for including diffuse sky radiation, is found to be particularly suited for this. Estimating parameters of radiometric-topographic correction algorithms (such as the ratio of di- rect sun radiation and diffuse sky radiation in case of the extended SCS method) from the image itself by regression of forest pixels arbitrarily selected at different topographic conditions relies on the assumption that the forest pixels selected at different irradiation conditions represent (in the statistical average) identical forest conditions. This assumption can hardly be proved to be valid. The estimation of this ratio by tuning within the kNN cross-validation procedure, on the other hand, represents a sound method, making use of the large amount of forest information available from terrestrial forest inventories. The practicability of this method depends on the number and thematic distribution of field plots on one single scene, for which homogenous at- mospheric conditions can be assumed.
[1]
T. Lin,et al.
The Lambertian assumption and Landsat data.
,
1980
.
[2]
Susan A. Murphy,et al.
Monographs on statistics and applied probability
,
1990
.
[3]
E. Tomppo.
Multi-source national forest inventory of Finland.
,
1994
.
[4]
Peter E. Hart,et al.
Nearest neighbor pattern classification
,
1967,
IEEE Trans. Inf. Theory.
[5]
A. Gillespie,et al.
Topographic Normalization of Landsat TM Images of Forest Based on Subpixel Sun–Canopy–Sensor Geometry
,
1998
.
[6]
M. Kenward,et al.
An Introduction to the Bootstrap
,
2007
.
[7]
P. Teillet,et al.
On the Slope-Aspect Correction of Multispectral Scanner Data
,
1982
.
[8]
J. Colby,et al.
Topographic Normalization in Rugged Terrain
,
1991
.