Caracterización del Grupo Etario en Estudio.

Figure 50. (a) Comparison of the CC, LX (k = 15°, 360°), and CLX (k = 15°) clumping retrieval methods

with the reference clumping factors computed from HP Pgap at the zenith angle of 55-60°. Each individual marker represents a scene clumping factor – all scenes are presented. (b) PAI errors of the CLX (k=15°) and the LX (k=15° and 360°) clumping methods for all simulated scenes at the zenith angle of 55-60°. The green shade is the ± 20% error.

View zenith angle trade-off

The higher clumping retrieval method errors at the approximate 57.3° viewing angle compared with angles near zenith presents a potential trade-off in this study. The trade-off is between the clumping error at 57.3°, and the unknown or anticipated G estimation error at all other angles, due to G typically requiring estimation at all other zenith angles (Pisek et al., 2013b; Chapter 4). For the Rushworth forest at least, we know that accurate clumping estimates at zenith angles close to nadir was made by the LX and CLX retrieval methods. However, estimating G of leaf and wood at these angles in the field is time consuming and prone to sampling errors (de Wit, 1965; Monteith, 1965; Pisek et al., 2013b; Ryu et al., 2010b). The relative insensitivity of retrieved clumping factors from the CLX (k=15°) method to a range of stem distributions and PAI levels suggests the potential for clumping factor calibration using 3D modelling. In addition, special attention needs to be made to ensure the correct angular values of Pgap,

GT and cos() are utilised for accurate PAI estimation utilising Pgap methods.

Historical Rushworth HP clumping values

An implication of the clumping retrieval method errors found in this study is that these methods are typically implemented by studies worldwide, e.g. (Chianucci & Cutini, 2013; Chianucci et al., 2014; Gonsamo et al., 2010; Leblanc & Fournier, 2014; Macfarlane et al., 2014; Piayda et al., 2015; Pisek et al., 2015; Pisek et al., 2011; Ryu et al., 2010a; Weiss & Baret, 2014). It is expected that the performance of these methods will vary depending on forest type, based on the architecture of the trees and relative positioning of crowns.

128

Pisek et al. (2015) published clumping values from Whroo, which is part of the same patch of Box Ironbark forest from this study, approximately 12 km north of the Rushworth study site. They used the CLX (k = 15°) method at 57.5°; described in further detail in Pisek et al. (2013a). The retrieved clumping value in the field from Pisek et al. (2015) was ≈0.61 ±0.07, which was a close5 match with the clumping value from the simulated HPs using the same retrieval method (0.66 ±0.04, Figure 50a). Although this was the found to be the most accurate method in this study, the clumping error was around 0.19 for the same forest patch as reported in Pisek et al. (2015). This suggests that a degree of caution must be taken when using clumping retrieval methods in forests matching the structural conditions in Rushworth, particularly at higher zenith angles (> 30°) where clumping from the evaluated methods is subject to larger errors (this study; Leblanc & Fournier, 2014).

5

Other factors affecting the comparison between clumping factors include: image resolution, field measured HP image classification accuracy, and sampling design to name a few.

129

5.6

C

This chapter presents an analysis of retrieval methods of two key parameters for LAI estimation through the use of hemispherical photography (HP); namely the proportion of wood-to-total plant material ‘α’, and canopy element clumping based on a 3D modelling and simulation study. The 3D modelling framework was parameterised using a 3D scattering model coupled with highly-detailed 3D explicit reconstructed tree models representative of a sampled forest stand. Virtual scenes comprising a broad range of PAI and stem distributions were produced. The framework enabled angular clumping retrieval methods, based on gap size distribution and logarithmic averaging approaches, and a method to indirectly determine α to be validated against precisely known virtual scene parameters.

The indirect α method utilising classified HP imagery applicable to evergreen forests matched to within 0.01 α of the reference, thus demonstrating its applicability for accurate indirect estimation. Angular dependence on indirect α retrieval was also found; where the entire HP image was needed to produce the most accurate estimate. Conversely, the classified narrow view zenith angle range around 55-60° zenith also provided an α estimate matching the reference. The CLX clumping retrieval method with k equal to 15° was the best performing clumping method; matching closely with the reference values at nadir. A linearly increasing error with zenith angle to >30% PAI at 75° was found for all structural configurations. As such, the performance of the clumping retrieval methods was poorer at larger zenith angles, with PAI errors when derived from the 55-60° zenith angle around 25-30% on average. Therefore, careful consideration of zenith angle range utilised from HP is recommended. The majority of clumping occurred at the within-crown scale, which is often overlooked in studies. Ignoring the impact of α and canopy element clumping for the forest type studied would lead to LAI estimation errors around 40% and 50%, respectively. The findings of this study impact upon indirect LAI retrieval using

Pgap model calculation methods, operating in environments requiring correction for non-randomly

distributed canopy elements and environments with woody canopy elements contributing to the extinction of light.

The ability to indirectly derive clumping is affected by the retrieval algorithm performance in combination with the instrument’s ability to estimate the ‘true’ gap fraction and size distribution, which is a function of instrument resolution, canopy density, gap size distribution, and representativeness of sampling location(s). Therefore, further work is suggested separating the impact of sensor and sampling effects from the canopy structural effects. Specifically, suggested future work includes applying the 3D methodology to different forest types, e.g. tall or multi-layered forests; including species with different woody proportions, leaf angle distributions, and crown characteristics. Finally, canopy element clumping has been described as a complex 3-dimensional problem, traditionally attempted to be resolved in a 2- dimensional manner (Gonsamo et al. 2011). Subsequently, the added ranging information from LiDAR sensors warrants continued investigation.

130