Effect of slope on treetop detection using a LiDAR Canopy Height Model

https://doi.org/10.1016/j.isprsjprs.2015.02.013Get rights and content

Abstract

Canopy Height Models (CHMs) or normalized Digital Surface Models (nDSM) derived from LiDAR data have been applied to extract relevant forest inventory information. However, generating a CHM by height normalizing the raw LiDAR points is challenging if trees are located on complex terrain. On steep slopes, the raw elevation values located on either the downhill or the uphill part of a tree crown are height-normalized with parts of the digital terrain model that may be much lower or higher than the tree stem base, respectively. In treetop detection, a highest crown return located in the downhill part may prove to be a “false” local maximum that is distant from the true treetop. Based on this observation, we theoretically and experimentally quantify the effect of slope on the accuracy of treetop detection. The theoretical model presented a systematic horizontal displacement of treetops that causes tree height to be systematically displaced as a function of terrain slope and tree crown radius. Interestingly, our experimental results showed that the effect of CHM distortion on treetop displacement depends not only on the steepness of the slope but more importantly on the crown shape, which is species-dependent. The influence of the systematic error was significant for Scots pine, which has an irregular crown pattern and weak apical dominance, but not for mountain pine, which has a narrow conical crown with a distinct apex. Based on our findings, we suggest that in order to minimize the negative effect of steep slopes on the CHM, especially in heterogeneous forest with multiple species or species which change their morphological characteristics as they mature, it is best to use raw elevation values (i.e., use the un-normalized DSM) and compute the height after treetop detection.

Introduction

Information on individual trees is critical for a variety of forest activities and for environmental modeling at the local and regional scales (Lichstein et al., 2010). In the last decade, airborne Light Detection and Ranging (LiDAR) has become a reliable remote sensing technique for estimating individual tree parameters, due to its capability to generate detailed and very precise three-dimensional tree information (Hyyppä et al., 2008, Lim et al., 2003).

As an initial and important step in any analysis of LiDAR data on individual trees, treetop detection has attracted much attention and research (Hosoi et al., 2012, Hyyppä et al., 2012, Jing et al., 2012, Kaartinen et al., 2012, Popescu and Wynne, 2004, Vastaranta et al., 2011). Identifying the correct treetop can provide accurate information on crown characteristics and the tree height information, which in turn are useful inputs for growth and volume estimation models (Gebreslasie et al., 2011, Vastaranta et al., 2011, Wulder et al., 2000). A widespread approach is to identify local maxima, which generally correspond to the location and height of individual trees, and then to construct crown segments (Falkowski et al., 2006, Næsset and Økland, 2002, Solberg et al., 2006, Véga and Durrieu, 2011).

The local maxima are typically obtained from the height variation of a LiDAR-derived Canopy Height Model (CHM), also known as a normalized Digital Surface Model (nDSM) (Forzieri et al., 2009, Li et al., 2012, Persson et al., 2002, Yu et al., 2011). There are two ways to create a CHM: with rasters or with point clouds (Li et al., 2012, Persson et al., 2002). When working with rasters, the LiDAR ground returns are used to create a raster DTM (Digital Terrain Model), and the highest or first LiDAR returns are used to create a raster DSM (Digital Surface Model). Then the raster DTM is subtracted from the raster DSM to create the final raster CHM (Lim et al., 2003). When working with point clouds, the classified LiDAR is height-normalized by replacing the raw elevation of each return (i.e. its z coordinate) with its height above the DTM (Khosravipour et al., 2014, Van Leeuwen et al., 2010). Either way, the end result is the absolute canopy height above the bare-earth terrain surface.

Although the procedure of computing local maxima from a CHM is conceptually simple, the accuracy of its result largely depends on the quality of the acquired LiDAR data, its processing and/or post-processing, and the forest conditions (Kaartinen et al., 2012). For example, the use of a higher density of laser pulse footprints improves the chance of the laser hitting the treetops (Hyyppä et al., 2008, Lefsky et al., 2002), and the use of an efficient local maxima technique enhances treetop identification by reducing commission and omission errors (Chen et al., 2006, Kaartinen et al., 2012, Vauhkonen et al., 2012). A new study suggests that the accuracy of treetop detection can be improved further by removing height irregularities in the CHM (Khosravipour et al., 2014). Moreover, a number of studies indicate that the various forest conditions (e.g., crown sizes, ages, site types, tree species and forest density) can significantly influence intermediate LiDAR derivatives and thereby the performance of tree detection algorithms (Falkowski et al., 2008, Pitkänen et al., 2004, Popescu and Wynne, 2004, Vauhkonen et al., 2012, Yu et al., 2011).

Complex forest terrain presents a challenging problem, as it affects the performance of the height normalization step by distorting the CHM, which can reduce the accuracy of extracted tree biophysical parameters (Vega et al., 2014). On steep slopes, the raw elevation values located, for example, on either the downhill or the uphill part of a tree crown are height-normalized with parts of the DTM that may be much lower or higher than the tree stem base, respectively (Breidenbach et al., 2008). Therefore, in the CHM, the downhill part of the crown will “rise” while the uphill part will “sink”, causing the entire tree crown to be systematically distorted. In treetop detection, the “rising” branch overhanging lower terrain in the downhill part can turn into a “false” local maximum that is distant from the true treetop. This problem was posed in Isenburg’s keynote speech at Silvilaser 2012 (Isenburg, 2012). He found a CHM that overestimated true tree height by more than double: eucalyptus trees on steep and eroded slopes in the Canary Island of Tenerife were estimated as being 51 m tall whereas their true height was 25 m. Takahashi et al. (2005) and Véga and Durrieu (2011) also reported that one of the sources of tree height overestimation from LiDAR-derived CHM is a horizontal offset error between field and LiDAR treetop detection, particularly on steeper slopes. They concluded the difference may be due to the LiDAR-derived treetop simply being identified as the maximum value of CHM within the crown area on steeper slopes. Heurich et al. (2003) pointed out that this error increases for leaning trees and/or steeper terrain slopes. Breidenbach et al. (2008) reported an increasing underestimation of the CHM-derived height with steeper upward slopes and vice versa for downward slope, which can cause tree height – one of the most important stand characteristics determined in forest inventory – to be misinterpreted, thereby affecting estimates of subsequent biophysical parameters such as biomass, volume and carbon sequestration. The recent study of Vega et al. (2014) suggested using un-normalized elevation values (i.e. using the DSM), and computing the height after a tree crown segmentation step, to avoid the undesirable effect of steep slopes on the CHM. However, until now, the influence of the normalization process on treetop detection and height estimation has neither been studied nor quantified.

The aim of this study was to quantify the effects of slope gradient on the accuracy of treetop detection when using a LiDAR-derived CHM. We first present a simplified theoretical model to illustrate how normalization causes a systematic error in CHM-based treetop detection when an individual tree is located on a slope. We then assess the accuracy of treetop detection by using both the CHM (i.e. the normalized elevations) and the DSM (i.e. un-normalized elevations). Next, we compute the positional difference between the same tree detected in both the CHM and the DSM, in order to investigate the influence of the slope on the horizontal displacement of CHM-detected trees and its effect on subsequent height estimation.

Section snippets

Theoretical model

The systematic error in CHM-based treetop identification can be quantified by using a conceptual model that is based on field measurement of tree heights. In the field, the original tree height is determined as a vertical distance from tree apex to the upslope root crown (Husch et al., 1982). According to the model (illustrated graphically in Fig. 1), the height of a tree is calculated as the magnitude (length) of a vector h originating at the base of the tree and ending at the treetop. Without

Study area

Bois Noir (44°23′N, 6°45′E) is a forest in the Barcelonnette basin in the southern French Alps (Fig. 3). The northern part of Bois Noir is characterized by irregular rugged topography with slope gradients ranging from 10° to 35° (Thiery et al., 2007) and the southern part is characterized by extensive steep scree slopes of up to 70° (Razak et al., 2011b). The test area was 1.30 km2, with a cover predominantly of mountain Pine (Pinus uncinata) and Scots pine (Pinus sylvestris), at an elevation

Experimental results

In order to create a CHM for detecting individual treetop position, the normalization process was applied to the elevation of the LiDAR points. Fig. 4 shows the distorting influence of slope on this LiDAR height normalization on the original morphological structure of the crown of a Scots pine and the crown of a mountain pine. As can be seen, the Scots pine tree, which has a wider, more irregular crown pattern and weaker apical dominance, is affected more than the mountain pine tree, which has

Discussion

One of the challenges when generating a CHM through height normalization is that a systematic error appears whenever trees are located on complex, sloping terrain. Based on this observation, we theoretically and experimentally quantified the effect of slope on the accuracy of treetop detection and its effect on the tree height estimation before and after height normalization.

The theoretical model presented the systematic error in LiDAR-derived treetop detection and height estimation as a

Conclusions

We have theoretically and experimentally shown in this paper that the LiDAR height normalization process systematically reduces the accuracy of treetop detection when trees are located on sloping terrain. Our experiments also showed that the effect of slope-distorted CHMs on treetop detection strongly depends on the tree’s crown shape, which is largely determined by its species. It would be interesting to investigate the effect of height normalization on various other non-conical crown shapes

Acknowledgments

The authors would like to express their gratitude to Menno Straatsma for delivering high quality LiDAR data. We also would like to thank Joy Burrough for editing the language of a near-final draft of the paper.

References (57)

  • C. Véga et al.

    Multi-level filtering segmentation to measure individual tree parameters based on Lidar data: application to a mountainous forest with heterogeneous stands

    Int. J. Appl. Earth Obs. Geoinf.

    (2011)
  • C. Vega et al.

    PTrees: a point-based approach to forest tree extraction from lidar data

    Int. J. Appl. Earth Obs. Geoinf.

    (2014)
  • M. Wulder et al.

    Local maximum filtering for the extraction of tree locations and basal area from high spatial resolution imagery

    Remote Sens. Environ.

    (2000)
  • X. Yu et al.

    Predicting individual tree attributes from airborne laser point clouds based on the random forests technique

    ISPRS J. Photogr. Remote Sens.

    (2011)
  • P. Axelsson

    DEM generation from laser scanner data using adaptive TIN models

    Int. Arch. Photogr. Remote Sens.

    (2000)
  • G.S. Biging et al.

    Stochastic models for conifer tree crown profiles

    Forest Sci.

    (1997)
  • J. Breidenbach et al.

    Quantifying the influence of slope, aspect, crown shape and stem density on the estimation of tree height at plot level using lidar and InSAR data

    Int. J. Remote Sens.

    (2008)
  • Q. Chen et al.

    Isolating individual trees in a savanna woodland using small footprint lidar data

    Photogr. Eng. Remote Sens.

    (2006)
  • Y.X. Chen et al.

    Feature modelling of high resolution remote sensing images considering spatial autocorrelation

    Int. Arch. Photogr. Remote Sens. Spatial Inform. Sci.

    (2012)
  • P.F. Doruska et al.

    Modeling the diameter and locational distribution of branches within the crowns of loblolly pine trees in unthinned plantations

    Can. J. For. Res.

    (1994)
  • M.J. Falkowski et al.

    Automated estimation of individual conifer tree height and crown diameter via two-dimensional spatial wavelet analysis of lidar data

    Can. J. Remote Sens.

    (2006)
  • M.J. Falkowski et al.

    The influence of conifer forest canopy cover on the accuracy of two individual tree measurement algorithms using lidar data

    Can. J. Remote Sens.

    (2008)
  • M.T. Gebreslasie et al.

    Individual tree detection based on variable and fixed window size local maxima filtering applied to IKONOS imagery for even-aged Eucalyptus plantation forests

    Int. J. Remote Sens.

    (2011)
  • Heurich, M., Schneider, T., Kennel, E., 2003. Laser scanning for identification of forest structure in the Bavarian...
  • B.K.P. Horn

    Hill shading and the reflectance map

    Proc. IEEE

    (1981)
  • F. Hosoi et al.

    Accurate detection of tree apexes in coniferous canopies from airborne scanning light detection and ranging images based on crown-extraction filtering

    J. Appl. Remote Sens.

    (2012)
  • B. Husch et al.

    Forest Mensuration

    (1982)
  • J. Hyyppä et al.

    Review of methods of small-footprint airborne laser scanning for extracting forest inventory data in boreal forests

    Int. J. Remote Sens.

    (2008)
  • Cited by (0)

    View full text