Skip to main content
SAF 2024 National Convention Main banner
Final Program


Advanced Brochure
Icon Legend
Back

Poster

Bioenergy, Climate Change, and Carbon

Converging LiDAR, Sentinel-2, and NAIP Data with Machine Learning for Forest Aboveground Biomass Modeling

Thursday, September 19, 2024
9:30 AM – 10:30 AM MST
Location: McKee Hall

    Speaker(s)

  • Shashika Himandi, n/a

    Master's Student/ Graduate Assistant
    Department of Natural Resources and the Environment, College of Agriculture, Health and Natural Resources, University of Connecticut, Storrs, CT.
    Willimantic, Connecticut, United States

The principal objective is to develop predictive machine learning (random forest algorithm) models, integrating LiDAR, Sentinel-2, NAIP, and FIA data to estimate forest aboveground biomass (AGB). The optimum RF model is intended to serve as a robust decision-making tool and increases the efficiency of large scale AGB determination.

Learning Objectives:

  • Upon completion, participants will be able to utilize random forest (RF) algorithm to develop and compare predictive regression models for forest aboveground tree biomass (AGB) using Sentinel-2, NAIP, LiDAR, and FIA data. They will demonstrate proficiency in selecting and integrating variables from remote sensing platforms and FIA inventory data to enhance large-scale AGB estimation efficiency.
  • Participants will be able to identify the techniques of selecting variables when the direct forest measurements are not available in the RS products. Moreover, they will be able to correlate forest structural and qualitative parameters with remote sensing products such as LiDAR heights, slope, and vegetation indices. This study also emphasizes the use of image derived variables and its contribution to increase the accuracy of LiDAR data predictions when comes to AGB.
  • Upon completion, participants will be able to apply RF algorithm techniques, including hyperparameter tuning and variable selection, to develop robust AGB models. They will analyze and interpret model errors, evaluation metrics for model selection, error interpretation and utilizing small sample sizes in machine learning. They will assess how predictions vary with forest ecology and statistical nuances and explore the adaptability of evolving RF models with high-resolution and more accurate training data for future AGB estimation.