Regional Forest Inventory using an Airborne Profiling LiDAR(<Special Issue>Silvilaser)

スポンサーリンク

概要

• 論文の詳細を見る

• A 5,159km profiling airborne LiDAR data set consisting of 56 parallel flight lines (fls) systematically spaced one kilometer apart acquired over the State of Delaware (USA) in y2000 are used to test the accuracy and precision of LiDAR-based forest inventory estimates. Nonparametric techniques is employed to develop simple linear regressions (SLRs) relating ground-measured biomass to laser height and crown closure. The ground-laser models are used to estimate total aboveground dry biomass at the county (3 counties in Delaware - 1,124km^2, 1,542km^2 and 2,539km^2) and State (5,205km^2) levels. The laser estimates are compared to U.S. Forest Service-Forest Inventory and Analysis (FIA) estimates from a 1999 ground-based survey of 215 FIA plots. In addition, the 56-fl data set is treated as a population and subsampled to test three variance estimators. The three variance estimators include weighted versions of the simple random sampling (SRS) estimator, a successive differences (SD) estimator, and a Newton's Method (NM) estimator. Results, constrained to this particular 56 fl data set and post-stratification, indicate the following: (1) Using all 56 fls in conjunction with the nonparametrically derived SLRs, LiDAR-based estimates of biomass are within 4%-24% at the county level and 14%-18% at the state level. (2) Across the 3 counties and State, considering the full range of flight line sampling intensities (from 2 to 28km between fls), the SRS estimator most closely tracks systematic sampling variability. The SD estimator is most conservative, consistently overestimating biomass variability by 〜15%. (3) When a limited, more realistic range of inter-flight line distances from 2-6km between parallel fls is considered, the behavior of the SRS estimator changes markedly. The SD and NM estimators overestimate systematic standard errors (SEs) by 〜18%, whereas the SRS estimator becomes the most conservative, overestimating systematic SEs by 〜30%. This SRS role reversal from least to most conservative as the distance between fls decreases suggests that the fls spaced 2-6km apart are, in Delaware, spatially autocorrelated. We suggest that analysts employ the SD or NM estimators when fls are closely spaced, e.g., 2-6km apart. (4) Inclusion of prediction error, i.e., the residual noise around regression lines used to predict, for instance, biomass as a function of profiling LiDAR height measurements, adds approximately 0.7-1.2 t/ha (an 8-15% increase) to the biomass standard error, averaged across strata and sampling intensities. (5) The positive relationship between the distance between flight lines and the systematic standard error appears to be generally linear (albeit noisy) for a given cover type and study area. Figures are provided illustrating the empirical relationship between flight line distance and systematic SE, by stratum within study area. These may be used to guide the design of airborne LiDAR-based forest surveys on areas from 1,000-5,000km^2.

• 森林計画学会の論文

著者

• Nelson Ross

  Biospheric Sciences Branch Nasa-goddard Space Flight Center

• Nelson Ross

  Biospheric Sciences Branch Nasa's Goddard Space Flight Center

• Gregoire Timothy

  Forest Management School Of Forestry And Environmental Studies Yale University

• Gobakken Terje

  Department of Ecology and Natural Resource Management Norwegian University of Life Sciences

• Stahl Goran

  Department of Forest Resource Management and Geomatics Swedish University of Agricultural Sciences

関連論文

• Estimating Timber Stock of Ehime Prefecture, Japan using Airborne Laser Profiling(Silvilaser)
• Land-cover Classification of Ehime Prefecture, Japan using Airborne Laser Altimetry(Silvilaser)
• Quantifying Variances of Line-Intercept-Sampling Estimators of Percentage Cover(Silvilaser)
• Regional Forest Inventory using an Airborne Profiling LiDAR(Silvilaser)

スポンサーリンク