Large-area mapping of Canadian boreal forest cover, height, biomass and other structural attributes using Landsat composites and lidar plots

Citation data:

Remote Sensing of Environment, ISSN: 0034-4257, Vol: 209, Page: 90-106

Publication Year:
2018
Captures 67
Readers 67
Social Media 224
Tweets 222
Shares, Likes & Comments 2
Citations 4
Citation Indexes 4
DOI:
10.1016/j.rse.2017.12.020
Author(s):
Giona Matasci; Txomin Hermosilla; Michael A. Wulder; Joanne C. White; Nicholas C. Coops; Geordie W. Hobart; Harold S.J. Zald
Publisher(s):
Elsevier BV
Tags:
Agricultural and Biological Sciences; Earth and Planetary Sciences
Most Recent Tweet View All Tweets
article description
Passive optical remotely sensed images such as those from the Landsat satellites enable the development of spatially comprehensive, well-calibrated reflectance measures that support large-area mapping. In recent years, as an alternative to field plot data, the use of Light Detection and Ranging (lidar) acquisitions for calibration and validation purposes in combination with such satellite reflectance data to model a range of forest structural response variables has become well established. In this research, we use a predictive modeling approach to map forest structural attributes over the ~ 552 million ha boreal forest of Canada. For model calibration and independent validation we utilize airborne lidar-derived measurements of forest vertical structure (known as lidar plots) obtained in 2010 via a > 25,000 km transect-based national survey. Models were developed linking the lidar plot structural variables to wall-to-wall 30-m spatial resolution surface reflectance composites derived from Landsat Thematic Mapper and Enhanced Thematic Mapper Plus imagery. Spectral indices extracted from the composites, disturbance information (years since disturbance and type), as well as geographic position and topographic variables (i.e., elevation, slope, radiation, etc.) were considered as predictor variables. A nearest neighbor imputation approach based on the Random Forest framework was used to predict a total of 10 forest structural attributes. The model was developed and validated on > 80,000 lidar plots, with R 2 values ranging from 0.49 to 0.61 for key response variables such as canopy cover, stand height, basal area, stem volume, and aboveground biomass. Additionally, a predictor variable importance analysis confirmed that spectral indices, elevation, and geographic coordinates were key sources of information, ultimately offering an improved understanding of the driving variables for large-area forest structure modeling. This study demonstrates the integration of airborne lidar and Landsat-derived reflectance products to generate detailed and spatially extensive maps of forest structure. The methods are portable to map other attributes of interest (based upon calibration data) through access to Landsat or other appropriate optical remotely-sensed data sources, thereby offering unique opportunities for science, monitoring, and reporting programs.