s:3076:"%T Detection of tropical forest canopy gaps caused by selective logging using hyperspectral remote sensing 
%A Spaias, L. 
%X The aim of this research is to assess the detectability of small spatial scale selective logging in tropical forests using optical very high spatial and spectral resolution data. Among the forest change processes contributing to the global greenhouse gas emissions, the degradation processes like selective logging represent the most challenging ones to be detected and quantified due to their partial forest canopy removal and their small scale. Furthermore selective logging events are considered as precursor of deforestation and important drivers for reduction of ecosystem services provided by tropical forests. In this research an Unmanned Airborne Vehicle (UAV) with a hyperspectral camera was used to detect small scale canopy gaps originating from selective logging in the tropical regions of Guyana and Indonesia. The UAV-based radiometric detection analysis provides the possibility to calibrate and validate a selective logging radiometric detection method based on airborne or satellite-borne very high spatial resolution data. The data consists of hyperspectral, Digital Surface Model (DSM) images and photos from a commercial camera (RGB). Two flights, a pre-harvest and a post-harvest flight performed with a tree harvested in between on each plot. With the UAV derived DSM and RGB images, the newly created canopy gaps were firstly visually identified. Then their spectral characteristics were explored using the hyperspectral data-cube. Three types of gaps were identified according to their spectral properties, the “Shadowed gap”, the “Non Photosynthetic Vegetation (NPV)-Soil gap”, and the “Understory gap”. Using spectral unmixing (SMACC algorithm), abundance maps were derived. Depending on the type of gap, the corresponding endmember was loaded. A subtraction between the pre- and post-harvest datasets revealed the area that the logging event affected. It was found that the “Understory gap” was the hardest to detect due to its small size and spectral variability. Finally, a detectability analysis gave insight on how much area should the gap cover in a pixel in order to be detected. The Signal to Noise Ratio (SNR) was set at 10 as a threshold to distinguish the change over the background of the image. The analysis showed that the required sub-pixel gap area for detection is 68% for “Shadowed gaps”, 89% for “NPV-Soil gap”, and 100% sub-pixel gap area for “Understory gap”. The gap sizes found in this study were 189 m2 for “Shadowed gaps”, 79 m2 for “NPV-Soil gap”, and 51 m2 for “Understory gap”. Finally, the hyperspectral pixel size required for detection was calculated combining the sub-pixel gap area and the gap size of each gap type found in this study. The pixel sizes were defined as: 9.3x9.3 m2 for “Shadowed gap”, 7x7 m2 for “NPV-Soil gap”, and 1x1 m2 for “Understory gap”. However the pixel size represents only the gaps studied in this research. 
";