Estimating forest biomass using AIMS lidar and aerial high-resolution imagery

Background/Question/Methods

Increasing atmospheric carbon dioxide (CO₂) levels are a leading cause of climate change (Malhi et al. 2002).  Since at least half of the Earth’s terrestrial carbon is stored in forest biomass (Gower et al. 1996), estimating forest carbon stocks helps us quantify concentrations and potential sources and sinks for CO₂.  One way that ecologists calculate forest biomass is with empirical allometric equations that use species and diameter at breast height (DBH) and divide by two to estimate carbon (Brown and Schroeder 1999, Jenkins et al. 2004).

I hypothesized that I could estimate stand-level biomass using the Airborne Imaging Multispectral Sensor’s (AIMS) high-resolution imagery and lidar height measurements.  To test this notion, I systematically sampled 366 trees, within 20 plots, for species, height, DBH, and canopy data, and compared my height averages with those obtained using the AIMS lidar.  To estimate biomass, I identified species and stem density in georeferenced images of ten 900m² subplots, used ground data to regress DBH from height using the lidar averages, and applied these data into allometric biomass equations.  For the ground validation, I identified species, measured DBH, and recorded place in the canopy in the ten subplots and calculated biomass using the appropriate equations.

Results/Conclusions

The results of a linear regression model using plot-level lidar height averages as a predictor of ground-measured plot-level height averages indicated that lidar was a significant predictor of the average dominant canopy height (p<0.001, R²=0.658).  The linear regression models for estimating DBH from height were all significant, although the R² values ranged from 0.08 (Pinus resinosa) to 81 (Carya spp).  Because of the low R² for P.  resinosa, I was not able to use subplots containing the species.  The linear regression model using remotely gathered data to estimate biomass as a predictor of ground calculated biomass indicated that the remote method was a significant, although not strong, predictor of dominant tree ground biomass (p=0.022, R²=0.499).  Post analyses revealed two major sources of error, the inaccurate identification of Quercus rubra and the miscounting of stems in the aerial imagery. Since, this methodology uses species-specific biomass equations, improving the image identification for dense species, such as Q. rubra, should improve the overall results.  Furthermore, practice and the use of training samples should lead to increased accuracy of image stem count.  Future studies that obtain a larger biomass estimation sample size will help to determine the usefulness of this method.