Developing a Drone-Based Machine Learning for Spatial Modeling and Analysis of Biomass and Carbon Sequestration in Forest Ecosystems
Abstract
This study aims to utilize drones and machine learning for spatial modeling and analysis of biomass and carbon storage in forest ecosystems. The research employs Red Green Blue (RGB) imagery captured by drones as a tool for data analysis, using Agisoft PhotoScan software to process data in conjunction with field measurements to create models for estimating above ground biomass (AGB) and carbon storage. The machine learning techniques applied include Canopy Height Model (CHM) and Segment Mean Shift (SMS). The findings reveal that field data surveys identified a total of 1,241 tree species, with an estimated carbon storage of 213.53 tonnes of CO₂ equivalent (tCO2e). Results from machine learning using the CHM technique showed a carbon storage estimation of 212.51 tCO2e, with an error margin of 0.48% and a carbon storage difference of 1.01 tCO2e. Meanwhile, the SMS technique estimated carbon storage at 207.01 tCO2e, with an error margin of 3.15% and a carbon storage difference of 6.52 tCO2e. It can be concluded that the CHM technique demonstrates higher accuracy compared to SMS in estimating carbon density. Additionally, further analysis of CHM results showed an accuracy of 0.597, precision of 0.902, recall of 0.638, an F1-score of 0.747, and an overall accuracy of 74.737%.