Combining Air Borne LiDAR and Forest Inventory Analysis Data (FIA) to Develop a Forest Carbon Model Using Machine Learning Techniques

Forests play a vital role as a carbon sink, where atmospheric carbon has been stored in the plant bodies via photosynthesis as tree above ground biomass (ABM) and below ground biomasses. Conventional methods of measuring the trees by a network of forest inventory analysis (FIA) data are time, and labor intense particularly in the context of broad-scale carbon modelling studies. Air borne Light detection and Ranging (LiDAR) is a promising way to derive a suite of structural measurements of trees (e.g., mean height, height percentiles), canopy characteristics (e.g., canopy relief ratio and canopy cover), and terrain conditions, such as slope and aspect. The overarching goal of this study is to develop predictive models to estimate ABM by using both airborne LiDAR and FIA field plot data. First, LiDAR derivatives of known FIA sub plot locations will be calibrated using the FIA plot ABM via machine learning (ML) techniques (e.g., random forest algorithm). Secondly, ML models will be validated to extrapolate the forest ABM and to generate a statewide fine-grained carbon map for the state of Connecticut. The study consists of 42 FIA plots, 98 functional FIA subplots and sitewide LiDAR data with mean ABM ranging from 50 Mgha-1 to 1622 Mgha-1. The radius of a FIA subplot is 24ft (7.32m) radius, which is the smallest spatial unit of the study. The resulting state-wide ABM map will serve as a key data layer for multiple stakeholders who involve in forest management decision making process.

Key Words: Forest Inventory Analysis, LiDAR, Forest Carbon, Machine Learning