Fine-scale canopy height mapping of boreal forest by integrating ICESat-2, PlanetScope & ArcticDEM

Keywords: ICESat-2, Forest canopy height, PlanetScope, ArcticDEM, Machine learning, Random forest, Boreal forest
Abstract Type: Paper Abstract

Abstract

Canopy height is associated with essential biophysical parameters such as primary productivity, biodiversity and above-ground biomass, and is a critical parameter analyzed for forest degradation-restoration, modeling ecosystem and estimating carbon content. In this study, we propose and present the first forest canopy height map at 5-m resolution by integrating ICESat-2 data with very high resolution PlanetScope & ArcticDEM imagery for the year 2020. Our approach utilizes SlideRule to query cloud-optimized ATL03: Global Geolocated Photon Data by accessing NASA’s Common Metadata Repository (CMR) and obtain vegetation height (h_canopy) at a 5-m resolution using PhoREAL algorithm with specified classification schemes. A machine learning-based workflow was developed using the resultant 5-m ICESat-2 derived vegetation height and predictor variables from satellite imagery as input for our selected study site in Alaska. Features constructed for modeling include spectral, vegetation and textural features from PlanetScope and topographic features from ArcticDEM. For validation, the reference canopy height includes NASA Goddard's LiDAR, Hyperspectral & Thermal Imager (G-LiHT), Land, Vegetation, and Ice Sensor (LVIS), & National Ecological Observatory Network (NEON) data. We evaluated the performance between three machine learning models – random forest (RF), artificial neural network (ANN) & gradient boosting machine (GBM). Preliminary results indicate that both RF and GBM models achieved satisfactory accuracy in predicting forest canopy height, with RF proving to be the more accurate regressor. Our research aims to advance fine-scale canopy height mapping for boreal forests to facilitate assessment of forest above-ground biomass & carbon stock.