Victoria Meyer

Application of Semi-Automated Filter to Improve Waveform Lidar Sub-Canopy Elevation Model

Modeling sub-canopy elevation is an important step in the processing of waveform lidar data to measure three dimensional forest structure. Here, we present a methodology based on high resolution discrete-return lidar (DRL) to correct the ground elevation derived from large-footprint Laser Vegetation Imaging Sensor (LVIS) and to improve measurement of forest structure. We use data acquired over Barro Colorado Island, Panama by LVIS large-footprint lidar (LFL) in 1998 and DRL in 2009. The study found an average vertical difference of 28.7 cm between 98,040 LVIS last-return points and the discrete-return lidar ground surface across the island. The majority (82.3%) of all LVIS points matched discrete return elevations to 2 m or less. Using a multi-step process, the LVIS last-return data is filtered using an iterative approach, expanding window filter to identify outlier points which are not part of the ground surface, as well as applying vertical corrections based on terrain slope within the individual LVIS footprints. The results of the experiment demonstrate that LFL ground surfaces can be effectively filtered using methods adapted from discrete-return lidar point filtering, reducing the average vertical error by 15 cm and reducing the variance in LVIS last-return data by 70 cm. The filters also reduced the largest vertical estimations caused by sensor saturation in the upper reaches of the forest canopy by 14.35 m, which improve forest canopy structure measurement by increasing accuracy in the sub-canopy digital elevation model.

Spatial Distribution of Carbon Stored in Forests of the Democratic Republic of Congo

National forest inventories in tropical regions are sparse and have large uncertainty in capturing the physiographical variations of forest carbon across landscapes. Here, we produce for the first time the spatial patterns of carbon stored in forests of Democratic Republic of Congo (DRC) by using airborne LiDAR inventory of more than 432,000 ha of forests based on a designed probability sampling methodology. The LiDAR mean top canopy height measurements were trained to develop an unbiased carbon estimator by using 92 1-ha ground plots distributed across key forest types in DRC. LiDAR samples provided estimates of mean and uncertainty of aboveground carbon density at provincial scales and were combined with optical and radar satellite imagery in a machine learning algorithm to map forest height and carbon density over the entire country. By using the forest definition of DRC, we found a total of 23.3 ± 1.6 GtC carbon with a mean carbon density of 140 ± 9 MgC ha−1 in the aboveground and belowground live trees. The probability based LiDAR samples capture variations of structure and carbon across edaphic and climate conditions, and provide an alternative approach to national ground inventory for efficient and precise assessment of forest carbon resources for emission reduction (ER) programs.