Change Detection in Boreal Forests Using Bi-Temporal Aerial Photographs

Increased need for timely forest information is leading to continuous updating of stand databases. In continuous updating, stand attributes are estimated in the field after an operation and stored in databases. To find the changes caused by operations and forest damage, a semi-automatic method based on bi-temporal aerial photographs was developed. The test data were classified into three classes: No-change (952 stands), Moderate-change (163 stands) and Considerable-change (44 stands). The aerial photographs were acquired in years 2001 and 2004 with almost the same image specifications. Altogether 110 features at stand level were extracted and used in change detection analysis. The test data were classified with stepwise discriminant analysis. The overall accuracy of classification varied between 75.3 and 84.7%. The considerable changes were found without error, whereas the Moderate-change and No-change classes were often confused. However, 84.2% of thinned stands were classified correctly. The best accuracy in classification was obtained by using the histogram and textural features extracted from the original, uncorrected images. Radiometric correction did not improve the accuracy of classification. Soil type, characteristics of the growing stock and the location of a stand in an image were found to affect the change detection. Before the method can be applied operationally, issues related to, e.g., confusion between No-change and Moderate-change must be solved.