The estimation of forest inventory parameters from small-footprint waveform and discrete return airborne LiDAR data
Authors: Sumnall, M.J., Hill, R.A. and Hinsley, S.A.
Conference: Silvilaser 2012
Dates: 17-19 September 2012
Abstract:The quantification of forest structure is important for a variety of disciplines, including ecology and management, as vegetation structure is related to a wide variety of ecosystem processes. This research investigates the estimation of forest structural variables from small-footprint airborne LiDAR capturing both discrete return (DR) and full waveform (FW) data. The study site is in southern England and contains a variety of forest structural types. The DR and FW LiDAR data were acquired simultaneously using a Leica ALS50-II system, under both leaf-on and leaf-off conditions. The DR data had up to four discrete returns per laser pulse, with an average of ~4 pulses (i.e. first/only returns) per sq. metre. The FW data were acquired at a rate of ~2.5 pulses (i.e. full waveform) per sq. metre. Point data were generated from each waveform through Gaussian decomposition, yielding 1-11 returns per pulse (mode = 5). Metrics derived from each return point included: x-, y- and z-coordinates, plus intensity (DR only), or amplitude and width (FW only). Field data collection was conducted in 21 field plots (30 x 30m in size), located across a stratification of forest structural properties within the study site. A number of forest inventory metrics were recorded per plot, including: (i) tree height; (ii) tree stem count; (iii) tree diameter at breast height (DBH); (iv) tree crown base height; (v) canopy cover; (vi) standing deadwood (snag) volume; and (vii) seedling number. This dataset formed the basis for ‘training’ subsequent statistical models. For each field plot area several hundred metrics were extracted from both the DR and FW LiDAR data, using both individual tree crown and area-based approaches. These LiDAR metrics were used for direct comparison with field data and in least-squares stepwise linear regression analysis to model forest structure. Compared with field data, dominant height and canopy cover were derived from the airborne LiDAR data with an RMSE of 1.31m and 10% (FW) and 1.32m and 13% (DR) respectively. Of the ten field variables modelled with regression analysis, five were modelled with a higher R2 value using FW LiDAR data, two were modelled with a higher R2 value using DR LiDAR data, and three were modelled with very similar R2 values (i.e. within 0.05) using both LiDAR datasets. Although further analysis and validation is still required, these results suggest that full waveform LiDAR data describe the structural components of woodland to an equal (or better) level of accuracy than discrete return data, in spite of a considerably lower spatial sampling rate due to the waveform digitisation process.
Source: Manual
Preferred by: Ross Hill