FIGURE 37: HEIGHT PROFILES AND REFERENCE ELEMENT AREA FREQUENCY

(a) _(b)

Figure 37. (a) Height profile comparison of measured ‘meas’ ALS flown at Rushworth using the mean of

all nine 100 m x 100 m plots centered on the plot locations (green line) and ± 1 standard deviation (SD; green error bars), with height profiles from the simulated ‘sim’ plots using librat. Simulated scenes were grouped into PAI levels 0.6-1.8 (grey line) and PAI = 2.4 (black) with ± 1 SD error bars. The bin size is 1m. All returns are non-ground 1st return. (b) Element area frequency for the simulated scenes of leaves (green lines), wood (brown lines), and plant (leaf and wood together; black lines). Element area was calculated from the summation of the 3D tree model facet area comprising a scene. The first three PAI levels (PAI 0.6, 1.2, 1.8) have the same element area frequency (solid lines) and the fourth PAI level (PAI 2.4) has a different frequency (dashed lines). The bin size is 0.5m.

The ALS profile was smooth and normally distributed with a comparatively large degree of variance to the simulated scene PAI levels profiles. One explanatory factor was due to the finite sample size of tree

2

librat was used to simulate the height profiles, derived from the simulated scene element cover maps (90 m x 90 m, 1cm resolution), which also provide the height of the first intercept from above the canopy. Recall that only first returns were used from both methods, in addition to a narrow FOV ALS acquisition leading to similar occlusion effects.

107

models used in the virtual scenes (n = 51), which were cloned to produce a higher stem density and LAI level within the scene, compared to a larger degree of natural variation of trees that takes place in the field. In addition, the same population of trees was used for each simulated PAI level, and the same tree proportions were used for PAI = 0.6, 1.2, and 1.8. PAI = 2.4 was displayed separately from the first three PAI levels (PAI = 0.6, 1.2, 1.8) due to selecting larger trees to increase the scene PAI, while keeping the stem numbers and distributions equal to PAI = 1.8. This factor caused the standard deviations of the simulated scenes height profiles to be comparatively small to the ALS profile. The small variance in the simulated height profiles was mainly a function of the slight variation in occlusion of elements from the different stem distributions for each level. This was also a reason why the two ‘peaks’ start to appear in the simulated data, because the same tree model proportion was used for each scene comprising a specific PAI level, rather than varying the tree models selected comprising each scene. Constant tree model proportions for each PAI level was deliberate to aid with interpreting stem clumping results. This key methodological step prevented biasing results from implementing different tree models that may have variable levels of within-crown clumping. It is also noteworthy that 1st returns from librat simulated height profiles were derived from an infinitely small beam, whereas ALS had a larger beam diameter (~20-30 cm diameter at the canopy level). This may impact on the vertical distribution of the retrieved canopy element profiles.

5.4.4 W

-

The indirect α estimation method following Sea et al. (2013) from simulated reference classified HPs was investigated in this chapter (Section 5.3.1). α following this method was estimated as the proportion of woody cover to total plant cover i.e. α = ∑woody pixels / (∑woody pixels + ∑leaf pixels). The method accuracy is determined by direct comparison of the method’s estimates with the scene α values, calculated from precisely known woody and leaf element area of all tree models. The impact of restricting the HP field-of-view (FOV) will then be analysed to determine whether the entire FOV is required to obtain an accurate estimate. The sensitivity of the method to the simulated PAI levels and stem distributions will be established through grouping α estimates and significance testing, further explained in Section 5.4.6.

5.4.5 C

The LX, CC, CCW and CLX clumping retrieval methods described in the methods section were selected to be evaluated (Section 5.3.2). Only clumping retrieval methods applicable to 2D gap size measurements from instruments or collection methods such as HP, TRAC, and LAI-2000/2200 were investigated. The clumping retrieval method values were computed using DHP.exe version 4.8 and TRACWin.exe version 5.1.0 (Leblanc, 2008). The 13 HPs simulated per virtual scene were grouped in the processing software to produce one clumping factor for each retrieval method per zenith annuli per scene. The LX and CLX methods were parameterised with different segment sizes k. Azimuthal segment sizes of k = 15, 45, and 90° applicable to HP images were chosen in this study to encompass a range of k values, and to be

108

consistent with segment sizes explored in previous studies (Leblanc & Fournier, 2014; Pisek et al., 2011). Optimal segment size is an area of ongoing research.

Overall, angular clumping values for 24 scenes were produced for each clumping retrieval method. CC, LX, and CLX clumping values were presented for the zenith angle range 7 - 75°. This avoided segments less than 10 times the mean foliage element width at low zenith angles, and also avoided edge effects of the simulated scenes higher zenith angles, respectively. The reference clumping factors for the same angular range were also computed using Eqn. 30 from the simulated HP Pgap and scene element cover map data, in addition to known GT() (Figure 38) and PAI scene values. This enabled a direct comparison

of clumping retrieval methods against the estimated scene reference clumping over coincident zenith angles. The clumping factor difference between the retrieval methods and reference clumping is equivalent to the PAI error, due to the linear relationship of clumping with PAI and LAI (Eqn. 15). For example, a 0.1 clumping factor difference between a retrieval method and the reference clumping value equates to a 10% error in PAI. Sensitivity of the clumping retrieval methods to virtual scene PAI level and stem clumping will be established through controlling for factors during significance testing (Section

5.4.6).