Accuracy of digitized aerial photographs for assessing forest habitats at plot level

Abstract

The objective of the present study was to determine the accuracy of digitized colour‐infrared aerial photographs for predicting forest habitats defined by dominant tree species, stand age and ground vegetation in order to assess the structural diversity of forests. The solar direction and viewing angle at exposure time exert strong effects on illumination by causing a bidirectional reflectance effect. As a result, the same forest habitat will have totally different reflectance values, depending on its position in the photograph. Digital tone values were thus calibrated prior to the stratification, using empirical regression‐calibration or rationing methods. Linear regression calibration to the principal‐point level of the photographs, in which the mean pixel value was modelled as a function of sun and sensor position at the time of exposure, was shown to be the most effective method. The results showed that regression calibration significantly improved stratification accuracies. Separation of 12 habitat types was accomplished with 85.3% accuracy and 48 habitat types with 57.7% accuracy. The results also indicated that final stratification accuracy is very dependent on the prestratification needed, since the bidirectional reflectance effect in aerial photographs differs in various habitat types.