Biomass Density Based Adjustment of LiDAR-derived Digital Elevation Models: A Machine Learning Approach
Salt marshes are valued for providing protective and non-protective ecosystem services. Accurate digital elevation models (DEMs) in salt marshes are crucial for modeling storm surges and determining the initial DEM elevations for modelling marsh evolution. Due to high biomass density, lidar DEMs in coastal wetlands are seldom reliable. In an aim to reduce lidar-derived DEM error, several multilinear regression and random forest models were developed and tested to estimate biomass density in the salt marshes near Saint Marks Lighthouse in Crawfordville, Florida. Between summer of 2017 and spring of 2018, two field trips were conducted to acquire true elevation and biomass density measures. Lidar point cloud data were combined with vegetation monitoring imagery acquired from Sentinel-2 and Landsat Thematic Mapper (LTM) satellites, and 64 field biomass density samples were used as target variables for developing the models. Biomass density classes were assigned to each biomass sample using a quartile approach. Moreover, 346 in-situ elevation measures were used to calculate the lidar DEM errors. The best model was then used to estimate biomass densities at all 346 locations. Finally, an adjusted DEM was produced by deducting the quartile-based adjustment values from the original lidar DEM. A random forest regression model achieved the highest pseudo R2 value of 0.94 for predicting biomass density in g/m2. The adjusted DEM based on the estimated biomass densities reduced the root mean squared error of the original DEM from 0.38 m to 0.18 m while decreasing the raw mean error from 0.33 m to 0.14 m, improving both measures by 54% and 58%, respectively.
Master of Science (M.S.)
College of Engineering and Computer Science
Civil, Environmental, and Construction Engineering
Civil Engineering; Water Resources Engineering
Length of Campus-only Access
Masters Thesis (Open Access)
Abdelwahab, Khalid, "Biomass Density Based Adjustment of LiDAR-derived Digital Elevation Models: A Machine Learning Approach" (2019). Electronic Theses and Dissertations. 6446.