Acuerdos de convivencia en el aula

In order to answer the primary research question, and determine how LiDAR can improve the assessment of forest structure, the potential impacts of scale must be examined. From a review of the literature in Chapters 1 and 2, it was identified that the use of basic geographic entities enables the potential impacts of scale to be investigated. As outlined in the research design, this thesis uses basic geographic entities of tree and tree components as the basis for enhancing forest structural assessments. The following section describes the development of the tree and component scale geographic entities.

As outlined previously (Lee et al., 2001 – Appendix C), there were a number of issues with using canopy height models (CHM) for tree and stand level assessment, especially when there was a requirement for sub-canopy stem assessment. Therefore, to provide an alternative method for assessing the stand, a Height Scaled Crown Openness Index (HSCOI) method was developed. The analysis of field and LiDAR data was undertaken in several successive stages (Stages I – IX; Table 11). Specifically, the tree scale modelling combined with field data were used to:

1. Evaluate the potential of the LiDAR data for identifying stem locations, including those associated with sub-canopy trees,

2. Support the generation of empirical relationships between LiDAR-derived data (i.e., the canopy height models and HSCOI) and field data tree height and diameter for all identified stems, and

3. Evaluate the potential of the LiDAR for retrieving stand level estimates of stem density, predominant stem height, basal area, and cover and (crown cover and foliage-branch cover).

Each of the stages (except stage IX) was comprehensively described in Lee and Lucas (2007) (Appendix C).

Table 11: Overview of processing stages for the HSCOI

Stage Purpose I Calibration and validation strategy To ensure a) a sequence of reliable inputs to models

and b) data to test model outputs across a range of forest types and environments.

II Calculation of field plot-based stem density, cover, and sub-canopy tree

assessment

To provide calibration and validation sets at the plot level and to establish, based on field data, the

accuracy of retrieval. III Conceptual development of the Height

scaled Crown Openness Index (HSCOI) IV Calculation of the HSCOI

Matrix generation and intersection of LiDAR points.

Calculation of LiDAR penetration.

To conceptualise and demonstrate the steps required for calculation of the HSCOI.

V Smoothing of the HSCOI and generation

of minima. To allow detection of stems regardless of crown dimensions and position in the vertical profile. VI Stem location using the HSCOI:

Stem identification and extraction Filtering multiple scales utilising tree

crown area and HSCOI thresholds.

To allow mapping of stems by locating HSCOI minima and refinement of these maps by utilising empirical functions and field-measurements (e.g.,

height, crown area) for different species. VII Crown/cluster area and cover estimation To identify the forest/non-forest boundary and

crowns/crown clusters contained within. VIII Estimation of stem height and plot-scale

attributes (including density and predominant height).

To indicate stem size at the tree and stand level, thereby facilitating tree and stand level assessment of

growth (successional) stage and estimation of biomass.

IX Estimation of stem diameter (D130) and

biomass To provide required inputs for allometric equations, and for SAR simulation calibration.

HSCOI Stage IX: Calculation of stem diameter

Most LiDAR studies have used various metrics of height to determine stem diameter, either at an individual tree level, or at a plot or stand level via basal area. This research used field data to generate a function that predicted stem diameter (at 130 cm: D130) from tree height.

A random sample (80 %, n = 3,016) of field-measured live trees with D130 ≥ 5cm, (from 33

field plots) were used (Figure 20a) such that: ) * 1189 . 0 ( 130 3.9806*exp H D = Equation 2

All species were included in the investigation, and the empirical non-linear (power) function was generally applicable across the main genus types. When evaluated against the remaining 20 % of live trees (D130≥ 5cm, n = 755), predictions of D130 were reasonable (r2 =

0.60, RSE = 5.97 cm, with the slope and intercept of the best-fit line being 1.06 and 0.43 respectively; Figure 20b). Whilst some scatter was evident, the slope of the best-fit line indicated that stem D130 derived from height measurements approximated actual field-measured

D130. The function could be further refined by splitting into multiple species-specific classes.

However, the application would also require species classification and mapping of the remotely sensed data (Lucas et al., 2008).

Figure 20: (a) Height-to-D130 translation function using 80% of field measured stems and (b)

validation using remaining 20% of field stems.

A hollows factor was also applied to the stems, based on a small sample of field data (Figure 21). The hollows factor only applied to trees with the D130 of greater that 30cm D130, as

no trees smaller than this were observed to have substantial hollows that would impact on overall tree biomass using existing allometric equations. The range of observations for the hollows factor could easily utilise a linear function also, which reflects the relatively low number of observations.

Figure 21: Hollows function derived from field data measurements (outside field plots), and applied to LiDAR derived stems.

Additionally, the sample was relative small, so further development of this function was recommended. As the hollows factor was only applied to a relatively small set of the trees due to the D130 threshold, the amount of error contributed to the plot or stand total from uncertainty in this function was likely to be small, but should be investigated as a future research option.

3.5.2 Individual crown segmentation and delineation

Conceptual overview

As shown in the thesis conceptual flowchart (Figure 7, section 3.1), crown delineation was a key component of developing basic geographic entities. These entities form the building blocks of the multi-scale strategy to improve the assessment of forest structure using LiDAR. An individual tree was defined as the spatial unit of hierarchal Level 0, and the previous section (Lee and Lucas, 2007 – Appendix C) has described how individual tree stems can be mapped. This section outlines the LiDAR methods developed to delineate individual tree crowns, thus completing the second component of Level 0.

Tree crowns were used to provide bounding areas for the tree component modelling, so that field data derived empirical functions can be applied at a tree level. This also allows voxels, branches, and stems to be assigned and constrained to a single crown. Crown cover estimates were generated for the field plots and larger assessment areas (e.g., primary sampling unit or ICESat footprints), to compare to field data. Tree crown cover was also up-scaled and compared with other cover metrics such as foliage-branch cover and foliage projective cover, estimated from a range of sources. As with the initial development of the Height Scaled Crown Openness Index (HSCOI), the crown delineation methodology has primarily been developed for the Queensland Injune data. Due to time constraints, the delineation methods have not yet been calibrated for the NE Victorian field data.

There were four main stages in the LiDAR crown delineation method. These are briefly described below, and illustrated in (Figure 21).

• Empirical functions were deriving from field data for two main tree structural types, thus allowing a simple template approach such that an expected crown area for a given tree height can be utilised.

• HSCOI crown edge segments were generated, as described in Section 3.5.1.

• Crown segments were classified into different structural types utilising LiDAR apparent vertical profiles and multi-scale spatial assessments.

• Individual crown objects were generated from segments using a range of attribute and spatial context criteria. The use of multi-scale adaptive templates allows the merging of small segments or splitting of large crown objects into more accurate crown representations as appropriate. Final crown dimensions were then calculated, and validated against field data.