^{*}

This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution license (http://creativecommons.org/licenses/by/3.0/).

Terrestrial laser scanners (TLS) have the potential to revolutionise measurement of the three-dimensional structure of vegetation canopies for applications in ecology, hydrology and climate change. This potential has been the subject of recent research that has attempted to measure forest biophysical variables from TLS data, and make comparisons with two-dimensional data from hemispherical photography. This research presents a systematic comparison between forest canopy gap fraction estimates derived from TLS measurements and hemispherical photography. The TLS datasets used in the research were obtained between April 2008 and March 2009 at Delamere Forest, Cheshire, UK. The analysis of canopy gap fraction estimates derived from TLS data highlighted the repeatability and consistency of the measurements in comparison with those from coincident hemispherical photographs. The comparison also showed that estimates computed considering only the number of hits and misses registered in the TLS datasets were consistently lower than those estimated from hemispherical photographs. To examine this difference, the potential information available in the intensity values recorded by TLS was investigated and a new method developed to estimate canopy gap fraction proposed. The new approach produced gap fractions closer to those estimated from hemispherical photography, but the research also highlighted the limitations of single return TLS data for this application.

The foliage element of a forest canopy represents the primary surface that controls mass, energy, and gas exchanges between photosynthetically active vegetation and the atmosphere [_{s}_{ns}

Hemispherical photography has been widely and effectively used to characterise biophysical properties of forest canopies for several decades [

Guevara-Escobar

In addition to the difficulties cited above, the need for a three-dimensional (3D) characterisation of forest canopies has led to research focused on the use of Light Detection and Ranging (LiDAR) systems in order to increase the accuracy of biophysical measurements and extend spatial analysis into the third dimension [

A small number of studies have demonstrated that ground-based laser scanners, referred to here as Terrestrial Laser Scanners (TLS), have great potential in improving our ability to remotely measure the biophysical properties of vegetation canopies. TLS can be used for fast acquisition of dense, 3D datasets of entire surfaces, and provide return intensity values based on the reflectivity of objects. Strahler

The objective of this research was to compare methods to derive gap fractions estimates using TLS measurements and hemispherical photography. The research addresses the repeatability and consistency of multi-temporal coincident datasets taken with both sensors at forest stands in the United Kingdom, representing both evergreen and deciduous species. In addition, this paper introduces a TLS method to derive gap fraction estimates based on intensity values recorded by TLS, in order to obtain a better understanding of the interaction between lasers and forest canopies.

Danson

A new method to obtain estimates of canopy gap fraction was also developed using intensity values recorded by laser scanning system. These values are associated with the energy contained in the signal reflected from an object, which may be affected by several factors including the beam divergence property of the lasers, the distance between the sensor and the illuminated object, and the reflectivity of the object [

The TLS approach to calculate gap fractions presented in Danson _{corrected}

The intensity value of the return signal for each measurement is recorded by most airborne and terrestrial laser scanner systems [

Therefore, the intensity (

It can be inferred from ^{2}

An important factor to consider here is the size of the range interval used in the correction of the intensity values. Considering

The study site chosen for this research was Delamere Forest, managed by the UK Forestry Commission and situated about 40 km south-west of Manchester (

The sensor used in this research was a Riegl LMS-Z210i, which is designed for the rapid and accurate acquisition of three-dimensional images by using a two-axis beam-scanning mechanism and a pulsed time-of-flight laser range finder. The data can be recorded by this sensor as either first return or last return of the signal backscattered from the targets, or a combination of both [

In order to capture the temporal dynamics of forest canopies, the TLS datasets for this research were collected on 3 April (leaf off), 13 May (leaf on) and 22 July 2008 (leaf on), and 19 March 2009 (leaf off). The data collection campaign was conducted using the methodology proposed by Danson

In order to obtain estimates of canopy gap fraction, the hemispherical photographs taken coincident with the TLS measurements were processed with Caneye [

For the TLS datasets, the

As examples,

The intensity-range correction process employed involved three stages: (i) from the set of return signals recorded by the TLS at each stand and each date, maximum intensity values were determined for each 1 m interval up to the maximum range detected by the sensor; (ii) the ratio

Matlab™ code was written to implement these three stages. The difference computed in (iii) eliminates the range dependence of intensity values (

This section presents estimates of canopy gap fractions derived from the TLS measurements, calculated using both the point-based and intensity-based methods, and corresponding estimates obtained from hemispherical photography. The convention used in the following figures showing gap fractions is that estimates derived from the photographs (HP) are represented by solid squares, whereas solid and empty circles correspond to fractions calculated using the point-based (PB) and intensity-based (IB) methods respectively.

The measurements taken at the broadleaf deciduous plot indicated a closer agreement (though larger RMSE than for the needle-leaved evergreen plot) between fractions derived from the intensity-based method and the photographs in comparison with estimates obtained from the point-based method for the leaf off data sets. For the leaf on data sets the differences were smaller for the point-based method.

Finally, for the larch plot (needle-leaved deciduous) the intensity-based method again produced smaller gap fraction differences, compared to the photography, than the point-based methods at all dates. The differences were similar in magnitude to those observed for the needle-leaved evergreen stand but they were larger for leaf off dates.

In general, the results presented indicate that gap fractions derived from the TLS intensity-based method were larger than corresponding estimates derived from the point-based method. The reason for this is that the point-based method takes into account the number of hits and misses recorded by TLS to compute gap fractions, but does not consider the portion of the laser beams not intercepted by canopy elements. This factor is taken into account by the intensity-based method and explains the difference in gap fractions derived from both approaches. In order to consider the empty portion of laser beams in the computation of gap fractions, it was necessary to make an assumption about the meaning of the intensity values recorded by the TLS. The assumption consisted of associating the maximum intensity values found at different range intervals to laser footprints that were totally occupied by canopy elements. The reason for making this assumption was the lack of information available from Riegl (the TLS manufacturer) on the echo detection algorithm, noise thresholding or range dependence of the intensity data for the TLS.

In comparison to estimates computed using the point-based method, gap fractions derived from the intensity-based method were closer in magnitude to corresponding results derived from hemispherical photographs. Nevertheless, in some cases TLS estimates were larger than results obtained from the photographs. Specifically, this was observed for leaf on datasets and photographs taken at the deciduous plots, as well as between evergreen datasets and photographs in the region between 35° and 60° zenith. These differences can be explained by the fact that at each range interval the intensity values are corrected with respect to the maximum value found within the interval. This implies that intensity values generated by foliage elements might then be corrected with respect to an intensity value triggered by a non-green (woody) element or

For paired TLS datasets and photographs acquired under leaf off conditions (April 2008 and March 2009) no significant differences between gap fractions would be expected. However,

The comparison discussed in this research was carried out with the intention of obtaining a better understanding of the information that may be derived from TLS measurements and to use that information to develop a refined gap fraction extraction method. The results presented indicate that gap fractions derived from the intensity-based method can be used to characterise temporal dynamics in forest canopies. Moreover, gap fraction estimates can also be used to determine other biophysical properties of forest canopies, such as effective leaf area index and leaf area index, which also require information about the spatial aggregation of foliage elements characterized by the clumping index. However, the fact that TLS fractions obtained from leaf on datasets were consistently larger than results derived from the photographs suggests that more information may be needed to improve the corresponding gap fraction estimates. This indicates a limitation of the sensor used in this research, particularly the separation of green and non-green canopy elements. Separation of canopy elements with different reflectance properties is difficult because intensity is a function of both reflectance and the area of the object in the beam. The results derived from the intensity-based method underline the importance of intensity values recorded by TLS and show the potential for using this information to extract information on forest canopy structure. The use of this information requires a good knowledge of internal mechanisms that measure the intensity of laser returns. As the specification of those mechanisms was not available for the TLS used in this research, an assumption had to be made on the fraction of footprints covered by canopy elements by taking into account the intensity values and ranging information recorded by TLS. The match with gap fraction estimates derived from hemispherical photography may be improved by considering multiple returns that might be triggered for laser footprints with multiple partial hits within a canopy. Therefore, further development of the method presented in this paper is likely to require data from the new generation of full-waveform multiple wavelength laser scanners like the Salford Advanced Laser Canopy Analyser (SALCA) that has recently been developed by the authors [

When dealing with full-waveform systems, the intensity-based method will need to be reformulated in order to account for the information acquired by multiple returns. In this context, a single laser beam may produce multiple intensity values, which will be assessed based on the distance from the sensor at which they are triggered. The information captured by second, third and further returns will provide better characterisation of gap fractions, as a laser beam is partially intercepted by foliage or other elements as it travels through the canopy. If only the first return is taken into account, the portion of the beam that was not intercepted will be considered as a gap. However, as further returns may be triggered, the additional information recorded must be considered for the estimation of gap fractions.

Establishing a direct comparison between the TLS and hemispherical photography methodologies is problematic, as the analysis of hemispherical photographs is based on pixel classification, whereas the processing of TLS datasets involves the examination of patterns and distributions observed in the point clouds as well as the 3D information recorded by the TLS. In addition, results derived from hemispherical photographs depend on other factors such as illumination conditions and a manual intervention to classify the image pixels (sky and canopy elements). In contrast, active sensors like TLS do not depend on external sources for illumination and no manual intervention is required in processing the LiDAR measurements.

Finally, it is worth pointing out that the accuracy of gap fraction estimates derived from both TLS and hemispherical photography could be determined by making comparisons with manual measurements and destructive sampling. However, destructive sampling methods are exceptionally time-consuming and not applicable to this research, as they would have altered the plot conditions and prevented the detection of changes in the canopy. As an alternative to destructive sampling, a detailed structural model of a forest canopy could be developed by means of computer simulation, and the measured gap fraction derived from the virtual forest could be used to validate the TLS results.

The authors would like to thank the Forestry Commission for allowing access to Delamere Forest, and the Manchester Geographical Society for a grant which supported the field data collection campaign. The corresponding author would like to thank the University of Salford and the School of Environment and Life Sciences for supporting this research through the provision of a Graduate Teaching Assistantship. The authors would also like to thank Dr. Vishal Bandugula for his invaluable help in the data collection campaign.

The authors declare no conflict of interest.

Geographic location of Delamere Forest, UK.

Full frame hemispherical photograph (

Intensity-range distributions (maximum values) corresponding to TLS datasets acquired at broadleaved deciduous (

Gap fractions derived from the point-based method (empty circles), intensity-based method (solid circles) and hemispherical photography (solid squares) using datasets collected at the needle-leaved evergreen plot.

Gap fractions derived from the point-based method (empty circles), intensity-based method (solid circles) and hemispherical photography (solid squares) using datasets collected at the broadleaved deciduous plot.

Gap fractions derived from the point-based method (empty circles), intensity-based method (solid circles) and hemispherical photography (solid squares) using datasets collected at the larch plot.

Differences in gap fractions derived from datasets collected at the needle-leaved evergreen plot, (

Differences in gap fractions derived from datasets collected at the broadleaved deciduous plot, (

Differences in gap fractions derived from datasets collected at the larch plot, (

Gap fraction differences between estimates derived from leaf off datasets (solid circles) and photographs (empty circles) obtained at the (

Tree species found in the sampling plots chosen for this research.

Needle-leaved evergreen Plot 4 | Corsican pine ( |
1945 |

Broadleaved deciduous Plot 3 | Sweet chestnut ( |
1899 |

Needle-leaved deciduous Plot 5 | Japanese larch ( |
1981 |

Data acquisition parameters used with the Riegl LMS-Z210i (mounted at an inclination angle of 90°).

| |

Start angle: | 50° |

Stop angle: | 129.7° |

Angular sampling resolution: | 0.108° |

| |

| |

Start angle: | 15° |

Stop angle: | 345.4° |

Angular sampling resolution: | 0.108° |

| |

| |

Number of returns obtained: | 2,261,340 |

Acquisition time: | 4 min |

Root mean square error (RMSE) of gap fractions obtained from both TLS methods

PB v HP | 27.81 | 39.01 | 42.26 | |

IB v HP | 14.15 | 19.75 | 10.00 | |

PB v HP | 33.84 | 17.30 | 23.69 | |

IB v HP | 15.75 | 20.62 | 15.83 | |

PB v HP | 27.49 | 13.89 | 17.93 | |

IB v HP | 16.45 | 19.73 | 17.93 | |

PB v HP | 25.47 | 59.05 | 38.31 | |

IB v HP | 14.79 | 24.42 | 12.37 |