Silva Fennica

Korpela, I. 2004. Individual tree measurements by means of digital aerial photogrammetry. Silva Fennica Monographs 3. 93 p. This study explores the plausibility of the use of multi-scale, CIR aerial photographs to conduct forest inventory at the individual tree level. Multiple digitised aerial photographs are used for manual and semi-automatic 3D positioning of tree tops, for species classification , and for measurements on tree height and crown width. A new tree top positioning algorithm is presented and tested. It incorporates template matching in a 3D search space. Also, a new method is presented for tree species classification. In it, a partition of the image space according to the continuously varying image-object-sun geometry of aerial views is performed. Discernibility of trees in aerial images is studied. The measurement accuracy and overall measurability of crown width by using manual image measurements is investigated. A simulation study is used to examine the combined effects of discern-ibility and photogrammetric measurement errors on stand variables. The study material contained large-scale colour and CIR image material and 7708 trees from 24 fully mapped plots in Southern Finland. The results of the discernibility analysis suggest that 88–100% of the total stem volume is measurable when using multiple aerial photographs. The structure and density of the forest were found to affect discernibility. The best hit-rates when using the semi-automatic tree top positioning algorithm ranged from 77 to 100% of the visually discernible trees. Systematic underestimation of the crown width was observed and the measurability of crown width was best near the image nadir. Species classification was tested in mixed stands of Scots pine, Norway spruce, and silver birch. The Kappa-coefficients ranged from 0.71 to 0.86. The results of the simulation suggest that very high accuracy at the individual tree level cannot be expected. However, if the photogrammetric measurements are unbiased, the aggregate stand variables can be very accurate. An accurate species recognition method is needed in the mixed stands in order to achieve unbiased estimates for the small strata.