Application of geospatial machine learning model for vegetation species image classification and above ground biomass estimation in upper and eastern gulf of Thailand mangrove forests

Abstract

The estimation of above-ground biomass in evergreen shrubs has traditionally been acquired from the ground measurements done by a labor-intensive method. Measuring such data requires diameter breast height (DBH), canopy cover, tree height, tree species, and density to be added to the above-ground biomass allometric equation. The method is both costly and time-consuming. To adapt to this rapidly changing climate caused by Global Warming, it is important to know how much carbon (greenhouse gas) has been stored. In 2015, Thailand entered the Paris Agreement, which consequently put pressure on the country to assess the total above-ground biomass of Thailand by the year 2025. However, the traditional measuring method is inadequate for the task, preventing the country from accurately assessing the total above-ground biomass. This research uses high-spatial-resolution data acquired from unmanned aerial vehicle (UAV) images to help measure the amount of above ground-biomass in forests. The assumption is that UAVs data could help reduce the cost of ground measurement and save time. Through orthophoto by point cloud reconstruction, DSM (Digital Surface Model), DEM (Digital Elevation Model) , it is assumed that the equation for above ground-biomass could be developed. Moreover, UAV images can be put into machine learning to distinguish the species and calculate above-ground biomass (ABG) emission at a significantly lower cost, higher accuracy, and better efficiency. This research uses UAV acquired data to train AI to distinguish vegetation species and calculate the above-ground biomass. IMPLICATION OF THE THESIS: This research assesses the application of UAV images and AI to estimate above-ground biomass/carbon stock and carbon sequestration in the Mangroves area in the provinces of Samut Sakhon, Phetchaburi, Samut Songkhram, and Chon Buri. The derived data (predicted DBH and trees height derived from CHM) and artificial intelligence are used to identify species from UAV images, which can then be evaluated using ground live data for accuracy. At the same time, UAV data provides very high spatial resolution, cost-effectiveness, time efficiency, and multi-temporal acquisition.