loading page

Mapping 30m Boreal Forest Heights Using Landsat and Sentinel Data Calibrated by ICESat-2
  • Tianqi Zhang,
  • Desheng Liu
Tianqi Zhang
The Ohio State University

Corresponding Author:[email protected]

Author Profile
Desheng Liu
The Ohio State University
Author Profile

Abstract

Boreal forest heights are closely associated with the global carbon and energy budget. Existing investigations of boreal forests were mainly carried out at plot scales, which cannot be guaranteed on an annual and regional-scale basis given their sampling schemes. The launch of the Advanced Topographic Laser Altimeter System (ATLAS) onboard the NASA’s Ice, Cloud and Land Elevation Satellite (ICESat-2) enables the measurement of forest vertical structure at a global scale. However, with a photon-counting system, ICESat-2 receives substantially reduced signals over vegetated regions (low albedo), making its applications in forest height mapping challenging. This study made the first attempt to develop a 30-m canopy height model (CHM) for a mountainous forested site (located at the north of Fairbanks, Alaska) by coupling the ICESat-2 observed canopy heights, Hcanopy (response), with Landsat-8 (L8), Sentinel-1 (S1) and Sentinel-2 (S2) data using a random forest regressor. Here, Hcanopy corresponds to the 95th percentile (RH95) of all identified canopy photons within a 100-m segment. Before CHM development, low-quality ICESat-2 tracks were filtered out by comparing with the reference airborne lidar considering factors such as slope, canopy cover, signal-to-noise ratio, and canopy height uncertainty. Results suggest that: 1) ICESat-2 Hcanopy has the highest correlation with airborne lidar RH95 under strong beams; 2) the errors of ICESat-2 tracks become larger under lower signal-to-noise ratios (<5), steeper terrain (slope >20˚), greater canopy height uncertainty (>0.3) and sparser canopy cover condition (<20%); 3) by adopting the aforementioned criteria in filtering the ICESat-2 tracks, the Pearson’s correlation coefficient (R) between ICESat-2 Hcanopy and airborne lidar RH95 has been significantly improved to >0.8 under any beam strength; 4) based on previous results, we find that incorporating features derived from L8, S1 and S2 produces the most desirable CHM (R=0.85), and S2 overall shows a better capability than L8 in predicting regional-scale canopy heights; 5) among all input features, normalized difference vegetation index (NDVI) calculated based on the first red edge band (703.9nm) of S2 is the leading feature on CHM development, whereas land cover appears the least important.