A manuscript entitled “When Big Data are Too Much: Effects of LiDAR Returns and Point Density on Estimation of Forest Biomass”, and co-authored by Dr. Chen was recently accepted by the IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing (J-STARS)!
Abstract:
Assessment of forest aboveground biomass and carbon stocks is increasingly dependent on Light Detection and Ranging (LiDAR) data. However, higher LiDAR point densities and return numbers pose challenges in procuring and processing voluminous LiDAR data for large-area assessments. Data reduction techniques are often used to overcome these challenges but rarely LiDAR returns are analyzed to estimate aboveground forest biomass. We examined the effects of LiDAR returns and data reduction on estimates of aboveground forest biomass in the Charlotte Metropolitan Region of North Carolina, USA. We extracted structural metrics using all returns, first returns, and first return of each laser pulse from reduced point densities (80%, 60%, 40%, 20%, 10%, 5%, and 1%) and the original LiDAR data. We used multiple linear regression to establish statistical relationships between field-measured biomass and LiDAR-derived predictor variables (PVs) for each return and point density combination. We selected non-collinear PVs using the Variance Inflation Factor, and identified the best set of explanatory PVs based on the lowest Akaike Information Criterion value. Using adjusted–R2, the explained variance of biomass models was consistent across all combinations of LiDAR returns and reduced point densities. While the greatest difference between 100% and 1% point densities was 13.3% for the first returns, all three return types performed consistently across all point densities. In addition, the variance of predicted biomass estimates was similar to field measured biomass across the spectrum of models. Our evaluation suggests that for regional-scale assessments of forest aboveground biomass, the use of first returns LiDAR data normalized by digital elevation model is an effective alternative to point density reduction that does not compromise the accuracy of biomass estimates.