Mapping vegetation height and identifying the northern forest limit across Canada using ICESat-2, Landsat time series and topographic data

The northern forest-tundra ecotone is one of the fastest warming regions of the globe. Models of vegetation change generally predict a northward advance of boreal forests and corresponding retreat of the tundra. Previous satellite remote sensing analyses in this region have focused on mapping vegetation greenness and tree cover derived from optical multi-spectral sensors. Changes in vegetation structure relating to height and biomass are less frequently investigated due to limited availability of lidar data over space and time in comparison with optical platforms. As such, there is an opportunity to combine lidar and optical remote sensing products for continuous mapping of vegetation structure at high-latitudes, with an emphasis on the forest-tundra transition. In this study, we used lidar data from the Ice, Cloud and land Elevation Satellite (ICESat-2) to classify canopy presence/absence, and predict canopy height across 120 million hectares of the Canadian forest-tundra ecotone at 30 m spatial resolution. Spatially continuous predictors derived from the Landsat satellite archive (2012−2021) and the ASTER (Advanced Spaceborne Thermal Emission and Reflection Radiometer) Digital Elevation Model were used to extrapolate 98th percentile canopy height from the ICESat-2 Land and Vegetation Height (ATL08) product using Random Forests models developed in R (version 4.2.2). Model accuracy was assessed using data from the Land, Vegetation and Ice Sensor (LVIS), a large-footprint airborne lidar system. The overall accuracy of the canopy presence classification was 89%, and canopy presence was detected with 88% accuracy. Models of vegetation height showed an overall R2 of 0.54 and RMSE of 2.09 m. Finally, we used these methods to map the limit of continuous 3 m forest across Canada and compared our model outputs with forest cover from the MODIS and Landsat Vegetation Continuous Fields datasets. This work demonstrates the challenges and potential for mapping horizontal and vertical vegetation structure within sparse, high latitude forests using both lidar and optical remote sensing data.