In: Wagner W., Székely, B. (eds.): ISPRS TC VII Symposium - 100 Years ISPRS, Vienna, Austria, July 5-7, 2010, IAPRS, Vol. XXXVIII, Part 7B 
changes automatically using aerial photography has remained 
unsolved. 
In automatic change detection, the effect of disturbing factors 
should be minimised. These factors include differences in 
atmospheric conditions, sun angle, soil moisture, vegetation 
phenology and, in the case of aerial photographs, the 
differences in viewing angles (Singh, 1989). It is important to 
eliminate factors that might cause differences between similar 
stands in different parts of the images. By using the present 
satellite positioning systems, aerial photographs can be taken 
very close to each other at different times. 
The objective of this study was to investigate whether bi 
temporal aerial photographs taken with similar image 
specifications and adjusted at stand and segment levels are 
useful in change classification, especially in detecting moderate 
changes such as thinnings. 
2. MATERIAL AND METHODS 
2.1 Material 
The study area is located in Western Finland near the town of 
Kauhajoki (22°18’ E, 62°24’ N). The forest holdings are mainly 
privately owned and are in a so called “stripsharing” 
configuration so that the holdings are usually long and narrow. 
The landscape of the area is dominated by flat terrain. The 
elevation varies between 125 and 185 m above sea level. The 
main tree species are pine (Pinus sylvestris L.) and spruce 
(Picea abies (L.) Karst.). The main broadleaved species are 
silver birch {Betula pendula Roth.) and pubescent birch (Betula 
pubescens Ehrh.) There are also several low-density stands in 
the study area. The main site types are fresh (26% of the area), 
dryish (42%) and dry (25%). The site types are according to 
Kuusela and Salminen (1969) and the fertility of peatlands is 
determined and classified using the same system as that for 
mineral soils (Laine & Vasander, 1993). 
The field data consists of 2 362 forest stands. The stand 
attributes were measured during summer 2002 based on a visual 
stand level inventory system that is used in Finland. In this 
system, the stands are initially delineated by visual 
interpretation of aerial photographs. The forest characteristics of 
the stands are then estimated visually with the aid of some 
measurements in the field, and the initial delineation is 
confirmed. 
The changes in stands between 2001 and 2004 were determined 
from different databases of the Regional Forest Centre and all 
stands were checked visually using bi-temporal aerial 
photographs. The uncertain cases were checked in the field. 
Subsequently, all stands were divided into three classes 
according to the type of change. The No-change class consisted 
of stands with no changes other than growth (1 890 stands). The 
Moderate-change class consisted of stands with moderate 
changes (373 stands), i.e. thinning, seed tree felling, tending of 
seedling stand, improvement of young stands, removal of hold 
over trees, soil preparation, slight storm damage, partly operated 
stand and forest road building. Slight storm damage means that 
storms have felled only few trees in the stand. The 
Considerable-change class consisted of stands with major 
changes (99 stands), i.e. clear cutting and severe storm damage. 
Severe storm damage means that storms have felled many trees 
or groups of trees in the stand. 
The image data consisted of two aerial photographs covering 
the study area. The latter photograph was taken as close to the 
former one as possible with respect to time, date and location 
(Table 1). The photographs were acquired using a Leica RC30 
camera and an antivignetting and an infrared radiation filter. 
The nominal scale of the color-infrared photographs was 1:30 
000. The photographs were first scanned with a 
photogrammetric scanner at 14 pm resolution into RGB-images 
with no tone adjustment. Secondly, the images were ortho- 
rectified to a spatial resolution of 0.5 m by using 11 (photoOl) 
and 13 (photo04) control points that were located from digital 
base maps and a digital elevation model. The digital elevation 
model had a resolution of 25 m. The root mean square errors 
(RMSEs) of the rectification were 5.1 m (photoOl) and 3.2 m 
(photo04). These photographs were used as base photographs 
for the extraction of spectral features. 
Aerial photograph 
Latitude 
photoOl 
62.4127 
photo04 
62.4125 
Longitude 
22.3045 
22.3016 
Altitude (m) 
4,751 
4,622 
Date 
23 June 2001 
27 June 2004 
Time (UTC) 
7:55:58 
7:29:06 
Course 
271 
271 
Solar azimuth angle (°) 
52.8 
60.2 
Solar azimuth angle (°) 
42.8 
40.1 
Table 1. The metadata of the aerial photographs 
2.2 Methods 
As change detection is sensitive to location errors, the final 
adjustment of the aerial photographs taken at two time points 
was based on statistical correlations that were computed 
independently for each stand and segment. The geographic 
location of the new photograph was determined by shifting the 
location at the photograph to one of the cardinal points one 
pixel at a time. The search range was ten pixels from the 
original location and covered 21*21 pixels. At each location a 
Pearson's correlation coefficient was computed for the digital 
number values (DNs) of the pixels that referred to a given area 
after the shift. Pixels on the buffer zone in the new photograph 
and pixels that were outside the stand (segment) boundaries in 
the old photograph were not taken into account. The width of 
the buffer zone was ten pixels from the stand (segment) 
boundary. The location that resulted in the largest correlation 
was selected. The process was repeated for each of the channels. 
A stand might not be the most suitable unit for change detection 
because some changes may occur in part of the stand only. 
Change detection was therefore also carried out at the segment 
(=sub-stand) level and results were adjusted to whole stands. In 
the segmentation procedure, the most recent aerial photograph 
was segmented based on the DN values of the three channels 
and the stand boundaries. With the help of the stand boundary, 
the segment was attached to the area of one stand. In the other 
words, the segment boundary did not cross the stand boundary 
(Figure 1).