Forest damage detection using high resolution remotely sensed data

Abandoned forests are increasing in Japan. In abandoned forest, falling and withering of trees may occur easily. Increase of these damaging of forests are troubling the forest administrators who have to keep identifying these damaged areas. However, identification is now implemented by means of direct ground surveying, which is difficult to grasp the damaged areas in the wide forest. Therefore, high spatial resolution remotely sensed imagery and digital surface models (DSM) are anticipated as a cutting-edge solution for supporting the field of forestry. In this study, we develop a forest damage detection method using high resolution remotely sensed imagery and DSM. Multinomial logit model is introduced for classifying the fallen areas and withered tree areas. The logit model is a simple statistical technique that is designed to analyze categorical data. Multinomial logit model can classify multiple categories (more than 3 categories). Explaining variables are (1) Gap areas extracted by DSM and (2) Spectral radiances of remotely sensed imagery. Dependent variables are no damage and damaged area (i.e. fallen area and withered tree area). Forest of Gifu prefecture is chosen as the test site, where a number of forest damages caused by deep snow and Pine beetle are observed every year. IKONOS imagery and LiDAR DSM are used for evaluation. It is confirmed that the multinomial logit model generates higher accuracies for 3 categories. Moreover not only large but also scattered damaged areas are detected.