Harvard Forest >

Harvard Forest Symposium Abstract 2009

• Title: Remote Satellite-based Prediction of Tree Height relating Waveform Lidar to Multi-Angle Spectral Imagery for Northern Temperate Forests
• Primary Author: Jeanne Anderson (University of New Hampshire - Main Campus)
• Additional Authors: Bobby Braswell (University of New Hampshire - Main Campus); Julian Jenkins (University of New Hampshire - Main Campus); Mary Martin (University of New Hampshire - Main Campus); Scott Ollinger (University of New Hampshire - Main Campus); Lucie Plourde (University of New Hampshire - Main Campus)
• Abstract:

  Recent research from Howland Forest in Maine (Kimes et al. 2006) has demonstrated the use of airborne multi-angle data to predict the vertical structure of forest canopies (as measured by lidar) for a range of forest communities common in northeastern North America. This work opens up the possibility of mapping the spatial variation of forest height structure for large forested expanses by examining the potential of satellite-based multi-angle data to directly model forest height. Developing the means to use airborne waveform lidar, where available within a network of existing research sites, as a link to the broad landscape coverage of multi-angular observations from a satellite-based sensor, could ultimately provide a source of data needed for more accurate modeling of carbon and climate models over continental and global scales.
  
  
  
  Towards this end, we examined the capability of multi-angle spectral radiance data to predict canopy height in two areas of northern temperate mixed conifer and deciduous forest. Waveform lidar imagery was acquired in July 2003 over the 1000-ha. Bartlett Experimental Forest (BEF) in central New Hampshire and 400-ha Harvard Forest (HFR) in central Massachusetts using NASA’s airborne Laser Vegetation Imaging Sensor (LVIS). Local mode multi-view angle imagery was acquired over the Gulf of Maine in May 2005 and over Boston Harbor in September 2006 using NASA’s Multiangle Imaging Spectrometer (MISR). Neural network models were developed to predict the LVIS metric of maximum canopy height (RH100) from 28 bands of MISR multi-angle spectral radiance values. The neural network models, developed separately using spring 2005 and late summer 2006 MISR data, showed a strong relationship to canopy height (r2 = 0.66 and 0.74 respectively). K-fold (10%) validation of each data model indicated that 77% and 75% of the holdback samples, respectively, had residual (absolute) errors of three meters or less. For spring versus late summer models, 82% and 79%, respectively, of all holdback samples had relative errors of 15% or less. Maps illustrating the variation in height structure were developed for twenty forest reserves within New England.
  
  

• Research Category: Conservation and Management
  Ecological Informatics and Modelling
  Regional Studies