Estimating individual tree leaf area in loblolly pine plantations using LiDAR-derived measurements of height and crown dimensions
Introduction
Forest managers and scientists have long sought efficient ways to estimate leaf area in forested systems. Foliage is the primary site of energy exchange between trees and the environment; thus, leaf area is fundamentally linked to forest productivity (Gower et al., 1992, McCrady and Jokela, 1998). Accurate estimates of leaf area, either at the stand or at the individual tree level, could provide useful information to forest managers, but due to measurement difficulties, it is rarely used in decision-making.
Existing approaches for directly estimating leaf area index (LAI) of a stand have generally proven too costly, untimely or insensitive to support management decisions. Litterfall collections require several months and are generally impractical for management purposes. Allometric equations are often used to estimate leaf area of individual trees, which are then summed across an area to estimate LAI. These equations are expensive to develop and may not apply across wide ranges of sites (Whitehead, 1978, Shelburne et al., 1993). In addition, they require collection of considerable amounts of field data for their application.
Due to difficulties and expense of attaining direct estimates of LAI from traditional approaches, considerable efforts have been made over the past quarter century to develop techniques to indirectly estimate LAI using remote sensing technologies. Vegetation indices derived from satellite images have been significantly correlated to LAI in a wide range of forest types (Spanner et al., 1994, Wulder et al., 1996, Fassnacht et al., 1997). However, several factors introduce error into satellite-based estimates of LAI, based on spectral reflectance.
Continued advances in remote sensing technology are leading to improvements in data generated by satellite-based and other aerial-based platforms, thus enhancing capabilities to estimate forest conditions across large land areas. Newer platforms are capable of finer spatial and spectral resolutions, and new analytical techniques allow increasingly detailed information to be extracted from remotely sensed data (Wu and Strahler, 1994, Wulder et al., 1998). In spite of these improvements, the ability to remotely sense forest structural characteristics, including LAI, with current satellite and aerial-based spectral tools remain limited (Wulder et al., 1998, Holmgren and Thuresson, 1998), due largely to the inability of spectral imagery to adequately characterize the vertical structure of forest canopies (Hall et al., 2005).
Ground-based remote sensing tools are also available for estimation of LAI using approaches based on the penetration of light through the canopy. Estimates of LAI from ground-based sensors, however, while correlated with direct estimates, are often biased (Gower and Norman, 1991, Fassnacht et al., 1994). Ground-based tools currently are incapable of consistently providing accurate estimates of LAI or are limited by the need for species- or site-specific coefficients or correction factors. In addition, the number of sampling points per stand required to estimate LAI with acceptable precision is time consuming and does not easily lend itself to measurements over large land bases.
The past decade has seen increasing interest in the use of light detection and ranging (LiDAR) technologies in forestry applications. LiDAR systems measure the time required for a pulse of laser energy emitted from an aircraft to reflect, or ‘echo’, off surfaces. Time is converted to distance, and through post-processing procedures, these distances provide a sampling of the vertical distribution of the vegetation canopy. The most commonly used LiDAR systems in forest applications are small-footprint, scanning systems. These systems operate by scanning side-to-side while emitting laser pulses resulting in a swath of laser postings through the stand. The width of the swath is determined by the scan angle and the aircraft altitude. The diameter of the footprint of laser energy when it reaches the surface is generally between 0.1 m and 1.0 m. Existing systems are often capable of generating in excess of 4 posts m−2, although posting densities of 1–2 m−2 or lower are more common.
Studies using small-footprint LiDAR to assess forest conditions have typically attempted to estimate average stand conditions (e.g., mean height, average dominant height, stem density, basal area, standing volume, aboveground biomass, foliage biomass) (Hall et al., 2005, Magnussen and Boudewyn, 1998, Means et al., 2000, Næsset, 1997a, Næsset, 1997b, Næsset, 2002, Næsset and Bjerknes, 2001). The most common approach has been to derive various statistical metrics directly from the LiDAR data. These metrics are then included as independent variables in regression analyses that examine correlations with measured stand data. While some of these analyses have resulted in reasonable correlations, the relationships are generally site specific and of limited use in predicting stand values elsewhere.
Improvements in LiDAR technology have led to higher pulse rates and increased LiDAR posting densities, thus making LiDAR analysis of individual tree characteristics possible. One approach to individual tree analysis, as with stand-level analysis, is to correlate tree and crown dimensions with statistical metrics derived from the LiDAR data using regression techniques (Næsset and Økland, 2002). This approach has generally been attempted on relatively large, open-grown trees—not in closed canopy forests. A more direct approach to individual tree analysis involves interpolation of the LiDAR returns emanating from the canopy into a canopy surface model. Peaks in the surface model are identified as trees. This approach has been used to identify individual trees (Andersen et al., 2001, McCombs et al., 2003), and to estimate tree heights (Hyyppä and Inkinen, 1999, McCombs et al., 2003, Popescu and Wynne, 2004) and crown diameters (Popescu et al., 2003). Attempts have also been made to use LiDAR-derived individual tree information to derive stem diameters and basal area (Hyyppä and Inkinen, 1999, Hyyppä et al., 2001), and to estimate stand-level volume and biomass (Popescu et al., 2004).
The success of these efforts has shown that LiDAR is capable of providing structural information at the individual tree level. Roberts et al. (2003) show that individual tree leaf area is reasonably estimated from crown dimensions; which suggests that LiDAR may be capable of providing estimates of individual tree leaf area. If individual tree leaf areas can be estimated with suitable accuracy, then LiDAR may be capable of providing stand-level estimates of leaf area capable of supporting management decisions.
Our goal in this study was to estimate individual tree leaf area using estimates of tree and crown dimensions derived from LiDAR. Our first objective therefore was to evaluate the ability of LiDAR to accurately recover stem and crown dimensions that are used in estimating leaf area. Accuracy was assessed by comparing LiDAR-derived estimates of tree dimensions with ground-based measurements for trees accurately identified with LiDAR. Our second objective was to evaluate the accuracy of LiDAR-based estimates of individual tree leaf area derived from LiDAR-based estimates of stem and crown dimensions. LiDAR-based leaf area estimates were compared to estimates of leaf area derived from ground-measured data.
Section snippets
Study sites
Loblolly pine plantations located in east-central Mississippi and eastern Texas were used in this study. In Mississippi, the Starr site utilized a 16-year-old loblolly pine spacing trial located on the Mississippi State University Starr Memorial Forest (33°16′N, 88°52′W) (Land et al., 1991). The original study included eight replicates of three intertree spacings—1.5 m × 1.5 m (4305 trees ha−1), 2.4 m × 2.4 m (1682 trees ha−1) and 3.0 m × 3.0 m (1076 trees ha−1). Each spacing block within each replicate
Tree identification accuracy
Across all initial tree spacings, an average of 81% of all live loblolly pine trees were accurately identified on plots at the Starr site (Table 3). Initial spacing significantly affected the ability to identify live trees (F2,45 = 90.6, P < 0.0001). The accuracy of live tree identification was 68.1% on the 1.5 m plots, 88.8% on the 2.4 m plots and 93.1% on the 3.0 m plots. Commission error rates ranged from 4.9% on the 2.4 m plots to 9.9% on the 3.0 m plots, although these differences were not
Discussion
LiDAR-based approaches based on direct estimation of individual tree structural parameters have advantages over statistical approaches that use multiple regression techniques to draw correlations between mean stand parameters and statistical metrics derived from the LiDAR data. While correlations developed under regression approaches, in some instances, have been relatively strong, they require the development of new correlations for each set of unique stand conditions, thus limiting their
Conclusions
Initial tree spacing significantly affected the ability of LiDAR to estimate several tree and stand parameters. Thus, knowledge of approximate tree spacing prior to LiDAR analysis is important for setting appropriate focal filter sizes for tree identification and height determination. However, spacing is a design parameter in most plantations and is often verified after planting, which would provide needed density information. Other studies have developed approaches that automatically set
Acknowledgements
The authors would like to express appreciation to Dr. Robert Parker and Mr. Curtis Collins for helpful comments on an earlier draft of this paper. This research was supported by NASA grant NAG13-99018. Approved for publication as Journal Article No. FO-271 of the Forest and Wildlife Research Center, Mississippi State University.
References (41)
- et al.
Short-term changes in soil C, N, and biota following harvesting and regeneration of loblolly pine (Pinus taeda L.)
Forest Ecol. Manage.
(2002) - et al.
A comparison of optical and direct methods for estimating foliage surface area index in forests
Agric. Forest Meterol.
(1994) - et al.
Estimating the leaf area index of north central Wisconsin forests using the Landsat Thematic Mapper
Remote Sens. Environ.
(1997) - et al.
Estimating stand structure using discrete-return LiDAR: an example from low density, fire prone ponderosa pine forests
Forest Ecol. Manage.
(2005) Determination of mean tree height of forest stands using airborne laser scanner data
ISPRS J. Photogramm. Remote Sens.
(1997)Estimating timber volume of forest stands using airborne laser scanner data
Remote Sens. Environ.
(1997)Predicting forest stand characteristics with airborne scanning laser using a practical two-stage procedure and field data
Remote Sens. Environ.
(2002)- et al.
Estimating tree heights and number of stems in young forest stands using airborne laser scanner data
Remote Sens. Environ.
(2001) - et al.
Estimating tree height and tree crown properties using airborne scanning laser in a boreal nature reserve
Remote Sens. Environ.
(2002) Estimation of tree heights and stand volume using an airborne LiDAR system
Remote Sens. Environ.
(1996)
Family influences on leaf area estimates derived from crown and tree dimensions in Pinus taeda
Forest Ecol. Manage.
The effects of site, stand density, and sapwood permeability on the relationship between leaf area and sapwood area in loblolly pine (Pinus taeda L.)
Forest Ecol. Manage.
Aerial image texture information in the estimation of northern deciduous and mixed wood forest leaf area index (LAI)
Remote Sens. Environ.
Automated individual tree measurement through morphological analysis of a LIDAR-based canopy surface model
Equations for estimating loblolly pine branch and foliage weight and surface area distributions
Can. J. Forest Res.
An evaluation of the uniform stress hypothesis based on stem taper in conifer stems
TREES
Rapid estimation of leaf area index in conifer and broad-leaf plantations
Ecology
Carbon dynamics of rocky mountain Douglas-fir—influence of water and nutrient availability
Ecol. Monogr.
Satellite remote sensing for forestry planning–a review
Scand. J. Forest Res.
Detecting and estimating attributes for single trees using laser scanner
Photogramm. J. Finland
Cited by (102)
Extracting sinkhole features from time-series of TerraSAR-X/TanDEM-X data
2019, ISPRS Journal of Photogrammetry and Remote SensingCitation Excerpt :The availability of high resolution (1 m) three-dimensional (3D) information derived from LIDAR provided the researchers with an opportunity to detect and delineate surface depression in much finer resolution, in comparison to other DEM-based studies (Liu et al., 2005). However, there are some limitations in using airborne LIDAR data including the requirement of low flight altitude, narrow swath (almost up to 20° off-nadir) and high cost, which make using them very challenging for regular surveys at regional scales (Geofisica and Giacaman, 2009; Roberts et al., 2005). The launch of TanDEM-X mission by German Aerospace Center (DLR) opened a new era for creating global high-resolution DEMs using space-borne SAR data.
Review of indirect optical measurements of leaf area index: Recent advances, challenges, and perspectives
2019, Agricultural and Forest MeteorologyCitation Excerpt :Multi-scale platforms, including ground-based, airborne, and space-borne LiDAR, were employed in relative studies. The first studies exploiting the LIDAR signal were based on allometric methods empirically established between LAI and some structural parameters directly derived from the LIDAR signal, such as the canopy volume and height (Lefsky et al., 1999; Lim et al., 2003; Riano et al., 2004; Roberts et al., 2005). However, these relationships are applicable only to a single species in a given geographical area (Richardson et al., 2009).
Estimating the leaf area of an individual tree in urban areas using terrestrial laser scanner and path length distribution model
2018, ISPRS Journal of Photogrammetry and Remote SensingCitation Excerpt :This method is less laborious and more efficient than the traditional allometric method because this method uses LiDAR-derived tree dimensions rather than manually measuring the tree dimensions for regression. However, the regression-based method still relies on manually measuring the leaf area of several trees to establish a regression relationship, and the accuracy of this method is largely limited by the accuracy of the LiDAR-based estimates of crown dimensions (Roberts et al., 2005). The voxel-based 3D modeling method is used to estimate the LAI and leaf area density by directly counting the contact frequency in each layer of the studied tree (Beland et al., 2014; Grau et al., 2017; Hosoi and Omasa, 2006).
Comparison of high-density LiDAR and satellite photogrammetry for forest inventory
2018, ISPRS Journal of Photogrammetry and Remote SensingCitation Excerpt :Remote sensing has greatly enhanced terrestrial sampling methods, allowing improved assessment of forest resources at a range of scales (Dash et al., 2015a; Breidenbach and Astrup, 2012). Airborne laser scanning (ALS or lidar) has emerged as a particularly useful technology for accurate characterisation of forest properties at resolutions ranging from stand level (Hall et al., 2005; Naesset, 2002) and plot level (Holmgren et al., 2003; Hyyppä et al., 2001; Lim and Treitz, 2004; Popescu et al., 2004) to individual tree level (Chen et al., 2006; Coops et al., 2006; Holmgren and Persson, 2004; Persson et al., 2002; Popescu and Zhao, 2008; Roberts et al., 2005; Yu et al., 2004). As ALS provides very detailed three-dimensional information around forest structure and the underlying terrain, this type of data is considered to be the most useful form of remote sensing for forest inventory (Vauhkonen et al., 2014) and ALS can predict inventory metrics far more precisely than satellite imagery (Dash et al., 2015b).
Predicting temperate forest stand types using only structural profiles from discrete return airborne lidar
2018, ISPRS Journal of Photogrammetry and Remote SensingCitation Excerpt :Lidar data for forest management has largely been acquired using airborne platforms (i.e. Hopkinson et al., 2013; Detto et al., 2015; Wilkes et al., 2015) with a growing body of research for future incorporation of terrestrial (i.e. Yang et al., 2013; Calders et al., 2015; Newnham et al., 2015) and spaceborne (i.e. Krainak et al., 2012; Hayashi et al., 2013; Tang et al., 2014) sensors. Lidar data has been used to measure a number of forest structural characteristics including tree density, stem volume, aboveground carbon, leaf area index, and basal area (Nӕsset, 1997, 2002; Andersen et al., 2005; Roberts et al., 2005; Hudak et al., 2006; Coops et al., 2007; Silva et al., 2014, 2016). Discrete airborne lidar returns are commonly aggregated over small areas, similar in size to field plots, and height and canopy density metrics are calculated.