Super High Resolution Airborne Remote Sensing for Evaluating the Decomposition Function of Ecosystem of Temperate Forest in Japan

Forest ecosystem is sustained by nutrients cycle among trees, floor vegetation, litter, and soil etc. One of important driving mechanisms for such nutrients cycle is the decomposition of the fallen trees by fungi, and this process would play an important function in the biogeochemical cycle of the environment. This study challenged to identify the position and size of fallen trees in a temperate forest in Japan based on super high resolution (less than 1cm) visual images taken from a camera aboard a helicopter. Field campaign was carried out on November 29, 2011 at the experimental forest (6 ha, 300m x 200m, 36° 56' 10.5'N, 140° 35' 16.5'E) in Kitaibaraki, Ibaraki, Japan. According to the census survey of the forest, deciduous broad leave trees are dominant. There was almost no leaf in the forest crown on the day of the field campaign, and that brought a high visibility of the floor from the sky. The topography of the forest site is characterized by a small valley with a river flowing north to south at its bottom. An unmanned helicopter (Yamaha RMAX G1) flew over the forest in north-south lines with a speed of 3m/s at height of 30-70m from the ground surface. The interval between adjacent two lines was 20m. A consumer grade camera (Canon EOS Kiss X5 with 55mm lens; 5184 x 3456 pixels) was fixed with the vertically looking down direction on the helicopter. The camera took forest images with 5 seconds interval. The helicopter was also equipped by a laser range finder (LRF) (SkEyesBOX MP-1). Based on the point cloud created by the LRF measurement, 1 x 1m digital elevation model (DEM) of the ground surface was established by finding the lowest point value of the point cloud in each 1 x 1m grid of the forest. The forest was covered by 211 images taken by the camera. Each image was orthorectified by using the DEM and the data of the position and orientations of the helicopter, and then they were mosaicked into one image. Fallen trees with the diameter more than 10cm were targeted for the analysis, and we confirmed that some of fallen tree was easily to identify visually, but some of them was hard to identify because the branch and trunk of trees hided fallen trees. We are going develop the automatic detection algorithm by decision tree and object-based classifications, and then validate the algorithm with in situ information.