In a special case, a small inventory area may fall on one
aerial photo (cluster) as a whole. In this case it is probable
that the classes on higher elevation become overrepresented.
The mean elevations of various classes, however, are not
decisive for the error but the elevations on which the class
borderlines lie. Ifthere are only two classes on a photo the
borderlines for the classes are common and, accordingly, the
estimates for class proportions directly on the photo scale
can be regarded as unbias although not necessarily correct.
Only, if the borderlines are horizontal, the proportions can
be regarded as correct.
Another special case, opposite to the above, would be when the
sampling is made on a photo where the inventory district appears
on the photo border and a part of the inventory district remains
outside of the photo area. In this case the classes on low
elevation tend to become overrepresented.
For the applicability of remote imagery to inventory purposes
it may be of great importance that the sampling directly on the
images can be regarded as unbias. From the pure statistical
grounds an orthoprojectional medium, like a topographic map,
would be preferable because distortions from orthoprojection
tend to increase the variance between clusters. Cost factors,
however, often favor the remote imagery.
Acknowledgements
This paper was made in collaboration with Prof.Dr.G.Hildebrandt
in the Abteilung für Luftbildmessung und -interpretation at the
University of Freiburg and with the financial support of the
Alexander von Humboldt Foundation. I want to express my sincere
thanks for these contributions.
References
Loetsch, F. and Haller, K.E. 1964. Forest Inventory.Vol.1.
München ,
Poso, S. and Kujala, M. 1977. A method for national forest
inventory in northern Finland. Metsántutkimuslaitoksen
Julkaisuja.