Assessment of structural attributes of even-aged Eucalyptus grandis forest plantations using small-footprint discrete return lidar data.
Tesfamichael, Solomon Gebremariam.
MetadataShow full item record
Assessment of forest structural attributes has major implications in the management of forestry by providing information of ecological and economic importance. The traditional methods of assessment involve collecting data in the field and are regarded as labour-intensive and expensive. In plantation forestry, field campaigns are generally time consuming and costly, and may compromise profit maximisation. The introduction of lidar (light detection and ranging) remote sensing in forestry has shown promise to add value to the traditional field inventories mainly through large spatial coverages in a timely and cost-effective manner. Lidar remote sensing is an advanced system capable of acquiring information in both the vertical and horizontal dimensions at relatively high resolutions. Numerous studies have established that these qualities of lidar data are suited to estimating forest structural attributes at acceptably high accuracies. The generic approach in most studies is to use lidar data in combination with field data. Such an approach still warrants a high cost of inventory. It is therefore useful to explore alternative methods that rely primarily on lidar data by reducing the necessity for field-derived information. The aim of this study was to derive structural attributes of even-aged Eucalyptus grandis forest plantations using lidar data. The attributes are of significance to timber resource assessments and include plot-level tree height attributes, stems per hectare (SPHA), and volume. The surveyed field data included tree counting and measurement of tree height and diameter at breast height for sample plots. Volume was then calculated using standard allometric models. Small-footprint lidar data of the plantations were also acquired coincident with the field inventories. Mean tree height and dominant height were estimated at a range of simulated lidar point densities between 0.25 points/m2–6 points/m2. Various plot-level distributional metrics were extracted from height values of lidar non-ground points and related with field mean and dominant height values using stepwise regression analysis. The results showed that both attributes could be estimated at high accuracies with no significant differences arising from variations in lidar point density. Estimation of SPHA relied on the exploration of semi-variogram range as a mean window size for applying local maxima filtering to the lidar canopy height surface. A comparative approach of window size determination used pre-determined within-row tree spacing, based on planting information. Two secondary objectives were addressed: comparing spatial resolutions of canopy height surfaces interpolated from non-ground height values and comparison of lidar point densities simulated at three levels. Comparison of spatial resolutions of canopy height surfaces were performed at 0.2 m, 0.5 m, and 1 m using a lidar point density of 5 points/m2. The results indicated that 0.2 m is the most appropriate resolution for locating trees and consequently deriving SPHA. Canopy height surfaces of 0.2 m resolution were created at simulated densities of 1 point/m2, 3 points/m2, and 5 points/m2. While all estimates were negatively biased relative to field-observed SPHA, lidar densities of 3 points/m2 and 5 points/m2 returned similar accuracies, which were both superior to 1 point/m2. It was concluded that 3 points/m2 was sufficient to achieve the accuracy level obtained from higher lidar point densities. Plot-level mean height, dominant height, and volume of trees were estimated for trees located using local maxima filtering approaches at the three lidar point densities. Mean height and dominant height were both estimated at high accuracies for all local maxima filtering techniques and lidar point densities. The results were also comparable to the approach that employed regression analysis that related lidar-derived distributional metrics and field measurements. Estimated dominant height and SPHA, as well as age of trees, were used as independent variables in a function to estimate plot-level basal area. The basal area was then used to compute diameter of the tree with mean basal area, referred to as quadratic mean diameter at breast height (QDBH). Mean tree height and QDBH were used as independent variables in a standard equation to calculate mean tree volume, which was then scaled up to the plot-level. All estimates for the local maxima filtering approaches and lidar point densities returned negatively biased volume, when compared to field observations. This was due to the underestimation of SPHA, which was used as a conversion factor in scaling up from tree-level to plot-level. Volume estimates across lidar point densities exhibited similarities. This suggests that low lidar point densities (e.g., 1 point/m2) have potential for accurate volume estimation. It was concluded that multiple forest structural attributes can be assessed using lidar data only. The accuracy of height derivation meets the standards set by field inventories. The underestimation of SPHA may be comparable to other studies that applied different methods. However, improved estimation accuracy is needed in order to apply the approaches to commercial forestry scenarios. The significance of improving SPHA estimation extends to improved volume estimation. In addition, the potential improvement should also take into consideration the density of lidar points, as this will impact on the cost of acquisition. This research has taken a significant step towards determining if lidar data can be used as a stand-alone remote sensing data source for assessment of structural plantation parameters. Not only does such an approach seem viable, but the lower required point densities will help to reduce acquisition costs significantly.