Figure 1. Top left: standing volume in each cell in m3; top right: Critical Wind Speed (CWS) for Overturning in m s-1, bottom left: Mean Wood Density in kg m-3 and bottom right: Stem Straightness classes (7 best – 1 worst).
By Juan Suárez1, John Fonweban1 and Barry Gardiner2
1Forest Research, Northern Research Station, Roslin EH25 9SY, UK, Corresponding author: juan.suarez@forestry.gsi.gov.uk
2INRA – Unité EPHYSE, 71, Avenue Edouard Bourlaux, 33140 Villenave D’Ornon, France
Abstract
Forest Research UK (FR) has developed an application that will allow foresters to create detailed maps of current woodlands on the public forest estate. This application integrates airborne LiDAR, an active sensor technology also known as light detection and ranging, that can measure the distance to a target by illuminating it with pulses from a laser, with a suite of models developed at FR and the British Forestry Commission’s Forest Enterprise Subcompartment Database (SCDB). The aim is to develop a method to estimate woodland parameters at a very fine scale and using them as input to specific models capable of deriving information about, for example, the timber quality and stability of forest stands. The UK Forestry Commission’s growth models for Sitka spruce (Hamilton and Christie, 1971) have been re-engineered to estimate Site Index, mean diameter at breast height, basal area, standing volume and total number of trees per unit area as a function of top height, measured directly from LiDAR and combined with stand variables extracted from the SCDB such as age and species.
Estimates are standardized by the percentage of canopy cover, which is also derived from LiDAR. The results are presented in units of 10 x 10 square meters, which can be aggregated inside stand boundaries. The example demonstrated here is based on LiDAR data acquired in Aberfoyle (Central Scotland) in 2008. Maps provide a synoptic view of the degree of variability in commercial forests and allow users to run more realistic scenarios using a wide variety of specific models. This new information could allow foresters to implement alternative silviculture methods and, therefore, manage forest resources in a more sustainable and precise manner.
1. Introduction
Forests are dynamic biological systems in continuous transformation, with a succession of episodes of constant growth interrupted by intermediate loss or damage (Gadow and Hui, 1999). Traditionally, forest management is based on a periodic assessment of woodland condition, forecasting and scenario testing. Stand models are constructed to describe trends over time based on field measurements. Future predictions are refined by a constant realignment of models at stand level based on the most up-to-date plot data.
This approach is frequently perceived as being limited in its ability to produce reliable information outside the scope of these models. For instance, in the UK, a large percentage of monoculture even-aged commercial plantations are being converted and managed under Continuous Cover Forestry (CCF). This system is defined by a permanent presence of trees, such as avoiding clearfellings, expanded diameter and height distributions, mixed age classes and selective thinning. Under the new conditions, none of the current stand models can be used reliably to address the new management needs.
Therefore, we have used a new approach that aims at an effective combination of LiDAR, stand data and models. The new method being developed at Forest Research aims to refine forest inventory by creating estimates at sub-stand level, capable of reflecting the degree of variability in stand conditions. Furthermore, the new system seeks to provide better estimates beyond traditional inventory data by running models capable of producing additional and specific information about the quality of the timber being grown, a cost-effective retrieval of wood products and the risk of wind damage for long term retentions. Together, this provides a wealth of information that can substantially increase our knowledge about our forest ecosystems and expand the number of opportunities for a more precise silviculture capable of addressing a wider range of goals.
2. Study area
The study area covers 20,000 hectares in The Loc Lomond and Trossachs National Park in southwest Scotland. (56°10’ N, 4°22’ W). Although the forest district was planted with different species, only Sitka spruce (Picea sitchensis Bong. Carr.) stands were considered for this study (14,450 hectares in 3,645 stands with an average size of 4 hectares). While 15 percent of these stands began transformation to CCF 10 years ago, concerns have been raised about the best way to manage these stands, as well as their timber quality.
The district is frequently affected by endemic windthrow, which is the most important abiotic hazard in the area. Wind affects woodlands by creating new gaps in the canopy, toppling trees and causing a general detriment to the quality of the timber being harvested, such as an increased proportion of compression wood, poor stem form and increased taper. In January 2012, severe wind damage affected 91 hectares and it was calculated that most stands showed some level of damage.
The Forest Enterprise Sub-Compartment Database (SCDB) maps the location and extent of stands larger than 0.5 hectares. Each vector polygon is associated with an attribute database containing data on species, year of planting, initial spacing, thinning regime, and more. These data are queried by an extension to ArcGIS called Forester, developed by the Environmental Systems Research Institute (ESRI) Systems in Redlands, California.
Forest District GIS officers use data from field-measured plots to regularly update the data in the SCDB. Data are stored and managed in an Oracle database (Oracle Corporation, U.S.). A Production Forecast application is run every year to calculate future timber supply using the information contained in the SCDB, combined with growth models developed in the early 1970s (Hamilton and Christie, 1971). This information is all that is currently available to run scenarios and to test the impact of different management alternatives.
3. Materials and methods
LiDAR data were captured in 2008 at a ground density of 1-point per square meter. A version of the SCDB was obtained from Forest Enterprise to match the same year of the LiDAR survey. Point clouds were processed in Terrascan (Terrasolid Ltd., Finland) to discriminate ground hits using last returns only. The selected points were gridded in Surfer (Golden Software Inc., U.S.) using triangulation with Linear Interpolation in cells at 1-meter resolution. First returns also were interpolated in the same way to create a surface model. A normalized Canopy Height Model (CHM) was obtained by subtracting the terrain from the surface model.
A Delphi Pascal routine was written to compute different LiDAR metrics in cells of 10×10 square meters, including percentiles at 5 percent intervals, and descriptive statistics, such as mean, standard deviation, skewness, and densities at 5 percent intervals. The 95th percentile (H95) was selected to represent top height. Validation was undertaken with 12 field plots in the area and seven plots in Kielder Forest in the Northumberland National Park in North England, which showed a systematic level of underprediction by 10 percent (R2 = 0.95) for top height using the H95 height values. Percentage of Canopy Cover (PCC) also was calculated, representing the percentage of those 1×1 square meter cells in the CHM inside each working area of 10×10 square meters, with heights above 2.5 meters.
The Forestry Commission growth models for Sitka spruce were reverse engineered in SAS (SAS Institute Inc., U.S.) to change the Yield Class (YC) Model system used in the UK (as annual increment of timber in m3 ha-1 yr-1) (Hamilton and Christie, 1971) for another one based on top height and Site Index (SI). Top height based growth models are relatively independent of age and management regime, and more reliant on the potential growth in a given site.
Only Sitka spruce stands above YC 2, and with a planting age of above 20-years-old were selected from the SCDB. Top height estimates obtained from LiDAR were used to calculate SI, mean dbh, basal area (G), standing volume (V) and number of individuals. Estimates were normalized by PCC in each working cell and the results mapped in ArcGIS. This procedure created more than a quarter of a million cells.
The resulting inventory data were then entered into the ForestGALES model (Gardiner et al., 2004) and the Timber Quality Model for Conifers, ConTQ (Gardiner et al., 2011). The output of the ForestGALES model mapped the spatial distribution of the critical wind speed (ms-1) for Overturning and Stem Breakage in each cell. The ConTQ model provided information about mean wood density, in kg m-3, and average stem straightness (Gardiner et al., 2011).
4. Results and Discussion
The maps created provide an indication of the degree of the spatial variability of forest inventory estimates inside each forest stand (Figure 1). These cartographic products can also be aggregated to stand level and used to update the SCDB. So, in principle, foresters have, for the first time, the choice of being able to trace the nature of the information normally presented in the SCDB to a higher level of detail in order to better plan interventions or run more comprehensive scenarios. Especially for those areas in transition or already managed under CCF, the possibility of mapping the heterogeneity of stand characteristics should contribute to a better understanding of the long-term effect of selective thinning or variable retentions.
As the new method produces forest inventory estimates that are more dependent on the biological principles contained in current stand models that have been fully tested in the past, the system became more robust and less reliant on empirical fits using field plots. Also, future management scenarios could be run more accurately by replacing a statistical representation of the complexity associated with a stand by a set of small, and in theory, more uniform cells. Moreover, forest managers are no longer constrained to the boundaries of a forest stand as an indivisible management unit. Instead, more opportunities are made available to implement more precise and spatially located alternatives.
The possibility of using basic forest inventory data to estimate other relevant information is an important opportunity available for more robust decision-making. The examples depicted in Figure 1 show how a forester can combine the effect of growing trees up to a certain target volume and balance the risk of wind damage against improving wood quality. This new way of looking at forest stands is likely to expand the choice of management options and provide a better knowledge about the current forest resource.
5. Conclusions
A new method based on the combination of airborne LiDAR, stand information and models has been proposed to provide more precise and comprehensive information about our forest resources. The possibility of obtaining forest inventory estimates at sub-stand level will support well-informed and precise decision-making.
This methodology is simpler and a low-cost solution compared to current methods based on field plots used for training empirical models. Proving that national LiDAR surveys and specific models are available to foresters, they should be able to generate accurate predictions of forest inventory estimates and thematic cartography over large areas, outlining specific information like probability of wind damage, timber quality or wood products.
Future work will concentrate on the retrieval of height and diameter distributions within stands. This development will allow foresters to create new stem assortment tables with log lengths and diameters that can be reclassified into wood products such as sawlog, pallet, pulp and waste for each working cell.
Dr. Juan Suárez is head of the Remote Sensing Applications Programme (RSAP) in Forest Research in Scotland. He has 10 years research experience in airborne LiDAR, hyperspectral analysis in support of the National Forest Inventory (NFI), Terrestrial Laser Scanning, forest modelling and programming. He is currently working on the development of LiDAR-based tools (airborne and satellite) for forest inventory, modelling and monitoring.
References
Gardiner, B.A., Suárez, J.C. Achim, A., Hale, S. and Nicoll, B. (2004). ‘ForestGALES 2.0. A PC-based wind risk model for British forests’. Forestry Commission publications. Edinburgh. ISBN 0855386320. 60 pp.
Gardiner, B. A., Leban, J-M., Hubert, J. and Simpson, H. (2011). ‘Models for predicting the wood density of British grown Sitka spruce’. Forestry 84, pp. 119-132.
Hamilton, G.J. and Christie, J.M., (1971). Forest management tables (metric). Forestry Commission Booklet No. 34. Forestry Commission, Edinburgh.
Von Gadow, K. and Hui, G. (1999). Modelling Forest Development. Kluwer Academic Publishers. Dordrecht, The Netherlands.