Airborne discrete-return LIDAR data in the estimation of vertical canopy cover, angular canopy closure and leaf area index

Abstract

Remote sensing of forest canopy cover has been widely studied recently, but little attention has been paid to the quality of field validation data. Ecological literature has two different coverage metrics. Vertical canopy cover (VCC) is the vertical projection of tree crowns ignoring within-crown gaps. Angular canopy closure (ACC) is the proportion of covered sky at some angular range around the zenith, and can be measured with a field-of-view instrument, such as a camera. We compared field-measured VCC and ACC at 15° and 75° from the zenith to different LiDAR (Light Detection and Ranging) metrics, using several LiDAR data sets and comprehensive field data. The VCC was estimated to a high precision using a simple proportion of canopy points in first-return data. Confining to a maximum 15° scan zenith angle, the absolute root mean squared error (RMSE) was 3.7–7.0%, with an overestimation of 3.1–4.6%. We showed that grid-based methods are capable of reducing the inherent overestimation of VCC. The low scan angles and low power settings that are typically applied in topographic LiDARs are not suitable for ACC estimation as they measure in wrong geometry and cannot easily detect small within-crown gaps. However, ACC at 0–15° zenith angles could be estimated from LiDAR data with sufficient precision, using also the last returns (RMSE 8.1–11.3%, bias –6.1–+4.6%). The dependency of LiDAR metrics and ACC at 0–75° zenith angles was nonlinear and was modeled from laser pulse proportions with nonlinear regression with a best-case standard error of 4.1%. We also estimated leaf area index from the LiDAR metrics with linear regression with a standard error of 0.38. The results show that correlations between airborne laser metrics and different canopy field characteristics are very high if the field measurements are done with equivalent accuracy.

Introduction

Forest canopy cover, canopy gap fraction and leaf area index (LAI) are commonly used ecological indicators (Smith et al., 2008). Traditional forest inventories omit these parameters, but airborne LiDAR (Light Detection and Ranging) scanners have been shown to be capable of producing this information with high precision (e.g. Holmgren et al., 2008, Lovell et al., 2003, Solberg et al., 2009).

Both vertical canopy cover and canopy gap fraction are related to the penetration of light through the canopy, but they differ in definition. Jennings et al. (1999) defined canopy cover as the “proportion of the forest floor covered by the vertical projection of the tree crowns”. This definition was adopted by IPCC (2003) and the FAO (2004), and it was further extended by Gschwantner et al. (2009), who presented common tree and forest definitions for the European national forest inventories. They defined first the crown projection area as “the area of the vertical projection of the outermost perimeter of the crown on the horizontal plane” and then crown cover (synonymous to canopy cover) as “the aggregation of the crown projection areas of individual trees (without double-counting of overlapping crown projection areas) divided by the stand area”. This means that when canopy cover is measured in the field, small gaps inside the crown perimeter must be included in the canopy. The crown perimeter, however, can be difficult to distinguish in practice and therefore some subjectivity remains in the field measurement.

The concept of canopy gap fraction (CGF) has several meanings. It can be used to refer to the proportion of canopy gaps in an angular field of view (FOV), which can cover a full hemisphere or be narrower (often 150° or 75° from the zenith). Jennings et al. (1999) used the concept “canopy closure”, defined as “the proportion of sky hemisphere obscured by vegetation when viewed from a single point”, i.e. canopy closure equals 1 – CGF. Canopy closure is a point feature, and it is unique to its XYZ coordinates and the FOV. As a summary: canopy cover is measured in a vertical direction and only includes the between-crown gaps (Gschwantner et al., 2009), whereas canopy closure is obtained from angular measurements and should include all-sizes of gaps in the FOV (Jennings et al., 1999).

Canopy cover and canopy closure are different concepts but they are commonly used as synonyms, especially in older literature (Jennings et al., 1999). As a concept, CGF is easiest to understand correctly, at least when it is mentioned that it is calculated for a conical FOV around the zenith, to separate from the commonly used directional or bidirectional gap fractions. However, we are interested in the proportion of the canopy, and therefore use the concepts canopy cover and canopy closure to keep the figures comparable. We emphasize the difference by using the qualifiers vertical canopy cover (VCC) and angular canopy closure (ACC). Remote sensing (RS) studies often use the undefined concept of ‘fractional cover’, which has commonly been estimated using hemispherical canopy images or optical instruments with a FOV (e.g. Hopkinson and Chasmer, 2009, Morsdorf et al., 2006), i.e. it corresponds to canopy closure as defined by Jennings et al. (1999). However, we prefer to use two separate concepts, VCC and ACC, because they have a better correspondence to field methods and link RS results to ecological and forestry literature.

Estimates of VCC are used as an indicator of suitable wildlife habitats (Cook et al., 1995, Ganey et al., 2008, Gill et al., 2000, Jennings et al., 1999), fire severity mapping (Miller et al., 2009), and in the validation of physically-based reflectance models (Kuusk & Nilson, 2000). The international FAO definition of forest is partly based on VCC (FAO, 2004), making it a central variable in national forest inventories. Traditionally, VCC has been obtained from ground-based measurements (Bonnor, 1967, Bunnell and Vales, 1990, Fiala et al., 2006, Ko et al., 2009) or statistical modeling (Crookston and Stage, 1999, Gill et al., 2000, Korhonen et al., 2007). However, these techniques tend to be too laborious or inaccurate for large areas, where accuracy, time and costs are important (Korhonen et al., 2006).

The most accurate method for the field estimation of VCC is the use of sampling transects and vertical sighting tubes (Korhonen et al., 2006, Paletto and Tosi, 2009, Sarvas, 1953). It is unbiased because the entire area is sampled with vertical observations. However, instruments with a non-zero FOV, such as spherical densiometers (Lemmon, 1956) and digital cameras (Korhonen & Heikkinen, 2009) can also be used if the FOV is less than 20° from the zenith, and a small overestimation in VCC can be accepted (Bunnell and Vales, 1990, Fiala et al., 2006, Korhonen and Heikkinen, 2009, Paletto and Tosi, 2009). However, the within-crown gaps visible in canopy images need to be removed. Otherwise, the inherent overestimation of VCC due to oblique viewing can turn into significant underestimation (Korhonen et al., 2006).

Angular canopy closure (ACC) is better than VCC in describing light conditions under the canopy as it accounts for the transmission of light from all directions (Jennings et al., 1999). Also, LAI, which is a key variable describing vegetation-atmosphere interactions, net primary production and forest health (Running and Coughlan, 1988, Smolander et al., 2000, Turner et al., 2004), can be estimated by hemispherical sensors on the ground (Miller, 1967, Norman and Campbell, 1989). The field estimation of ACC is typically done with hemispherical canopy photographs or instruments such as the LAI-2000 (Jonckheere et al., 2004, LI-COR Inc., 1992) with a FOV of 150° (Frazer et al., 1999, Schleppi et al., 2007). The use of terrestrial LiDARs has also been demonstrated (Danson et al., 2007, Lovell et al., 2003). However, these in-situ methods are prone to sampling errors, they require proper weather, and they become expensive for large areas. Hemispherical images call for correct exposure (Schleppi et al., 2007, Zhang et al., 2005) and are sensitive to the image analysis methodology (Jonckheere et al., 2005).

The estimation of VCC and ACC by remote sensing could offer several advantages. The canopy structure is best observed from the above, and RS also allows wall-to-wall coverage at lower costs. Airborne LiDAR sensors are increasingly used in forest inventories to estimate forestry related parameters (Hyyppä et al., 2008, Korpela et al., 2009, Næsset et al., 2004), and also biophysical canopy variables (Lefsky et al., 2002). Consequently, mapping of canopy gap fraction and LAI with LiDARs has been under investigation since commercial systems became available (e.g. Jensen et al., 2008, Morsdorf et al., 2006, Riaño et al., 2004, Smith et al., 2009, Solberg et al., 2009). The accuracy of estimating VCC using LiDAR data has not been studied as intensively (Holmgren et al., 2008, Holmgren et al., 2003, Wang et al., 2008). One reason may be the relative difficulty of measuring the proportion of vertical between-crown gaps compared to taking canopy images with a fisheye lens. It is also possible that the difference between VCC and ACC (or gap fraction) has not been properly recognized.

As mentioned, two major differences separate VCC and ACC: angularity and observations of within-crown canopy gaps. Typical pulsed discrete-return (DR) LiDARs operate at low (< 20°) scan angles (deviation from the sensor z-axis) in order to guarantee ground observations for topographic mapping. Another important parameter is the diameter of the illuminated footprint, which depends on the beam divergence and ranging distance (Baltsavias, 1999). In small-footprint sensors, typical diameters are 10–30 cm from 1 km altitude. A footprint contains a certain proportion of the energy and typical definitions are 0.63 (1/e) and 0.87 (1/e²). Discrete return LiDARs record on-the-fly 1–4 returns (~ echoes) from the reflected waveform, and the signal-to-noise ratio decreases for higher flying heights and/or lower transmitting power (Hopkinson, 2007). The resulting point patterns (height distribution, number of echoes per pulse) can therefore vary for the same target canopy when important LiDAR parameters are changed between acquisitions. A pulse resulting in one echo produces a single return, which might be seen as a first and last return at the same time. Multi-echo pulses consist of first, last, and intermediate echoes. Because the systems are eye-safe and use moderate power levels, the proportion of gaps along the path of the pulse has to be quite high to produce a measurable ground observation.

In summary, DR LiDARs should be better suited for estimating VCC than ACC. Canopy surface and between-crown gaps can be directly mapped with LiDAR data, using only the first returns. Also, the scan zenith angles (deviation of the pulse from the Z-axis, θ_scan) are relatively small. Thus, a simple VCC estimate can be obtained by calculating the proportion of first returns above a specified height threshold, usually the breast height at 1.3 m. However, even a small scan zenith angle can cause bias that increases towards strip edges: as oblique pulses hit the crowns, unsampled regions remain on the other side (Fig. 1). This effect was simulated by Holmgren et al. (2003), who concluded that the effect of scan zenith angle on the ratio of canopy and ground hits can particularly be observed for species with long crowns. Morsdorf et al. (2008) found no significant bias compared to angular reference data because of a narrow scan angle. The magnitude of the bias and elimination methods, using accurate and vertically sampled field data to ensure the findings, have not been studied so far.

The estimation of ACC and LAI with LiDAR is considerably more difficult as the LiDARs are not designed for these tasks. Existing studies show that the proportion of first returns underestimates the gap fraction, and the proportion of last returns overestimates it (Lovell et al., 2003, Morsdorf et al., 2006). Thus, alternative LAI proxies have been proposed, but a calibration coefficient is needed to match these results to field-measured LAI (Morsdorf et al., 2006, Richardson et al., 2009, Solberg et al., 2009). Good results have also been obtained using different canopy height distribution variables to create multivariate models for LAI (Jensen et al., 2008, Riaño et al., 2004). Alternatively, the LiDAR intensity ratio can be used as an ACC approximation (Hopkinson and Chasmer, 2009, Zhao and Popescu, 2009).

Furthermore, the applied scan angles are usually considerably smaller than the FOVs used in the derivation of gap fractions from hemispherical images. It is not feasible to determine the area of the canopy that is sampled by an image, as each image observation is uniquely related to its XYZ point location and the FOV (Korhonen & Heikkinen, 2009). A comparison of large FOV images and laser data requires the determination of an optimal plot size for the intersection of the laser point cloud (Lovell et al., 2003, Morsdorf et al., 2006, Solberg et al., 2009, Zhao and Popescu, 2009).

Our main aim was to investigate the differences in the estimation of the two different structural canopy parameters, vertical canopy cover and angular canopy closure, from varying types of airborne LiDAR data. Vertical canopy cover and ACC are measured differently, and their estimation from LiDAR data also requires different methods. We hypothesized that VCC obtained with a sighting tube would correlate well with the LiDAR-derived proportion of first return vegetation hits, given the small near-vertical scan zenith angles and the fact that first echoes measure between-crown but not within-crown gaps. The effects of scan zenith angle and a varying pulse density on VCC were examined and quantified. Methods for correcting the angle effect were tested with different LiDAR data sets, including variations in pulse density, sensors and flying altitudes. These LiDAR data were also used to model the ACC at large and narrow FOVs based on the different LiDAR-derived indices. We also present results from LAI estimation using these different data.