Validation of ICESat-2 terrain and canopy heights in boreal forests

NASA's Ice, Cloud and Land Elevation Satellite-2 (ICESat-2) launched in the fall of 2018 and is collecting ve-getation canopy height and terrain measurements globally. In this paper we validate the terrain and canopy heights estimated from 11 months of ICESat-2 data using airborne lidar data collected in southern Finland. Overall, the terrain heights from the ATL08 data product agreed with the airborne lidar with vertical errors less than 75 cm (mean = -0.07 m; MAE = 0.53 m, RMSE = 0.73 m; n observations = 909,467). ATL08 terrain heights had positive bias (33 cm) when permanent snow cover was present compared to the airborne lidar. Canopy heights derived from ICESat-2 varied significantly by the beam selection (i.e. strong versus weak). Use of the weak beams increased the underestimation of canopy height by 1.06 m and increased RMSE% by 8.08%. The lowest canopy height errors were associated with strong beam/night/summer ATL08 acquisitions, which underestimated canopy height by 0.56 m (Bias% = 3.18; RMSE% of 13.75%), followed by strong beam/day/summer acquisitions (Bias% = 2.85%; RMSE% = 14.77%). For this study area with managed, coniferous dominated boreal forests, we found that ATL08 canopy height errors were lowest for canopy cover ranging from 40 to 85% and within this range, we found that on average, ICESat-2 missed the top 11-13% of canopy heights. At low canopy cover (< 40%), photons are more likely to be reflected from the terrain surface rather than the canopy due to the lower amount of vegetation present. Although not designed for canopy height retrievals, herein we have demonstrated that ICESat-2 can provide a useful source of canopy height information in boreal forests. By quantifying the accuracy with which terrain and canopy heights are characterized by ATL08, we provide insights on the capabilities and limitations of these data for these applications. Our results indicate that end users interested in canopy height retrievals in particular should avoid the use of weak beam data in boreal forest conditions.