ITC analysis of aerial images

• Introduction
• Aerial Sensor Technology
• Image needs for computer image analysis
• Synergy with LiDAR
• BRDF Correction and Normalization of Aerial Data Area
• Tests of feature-based BRDF correction curves on two flight lines
• Conclusion

---

Synergy with LiDAR

Although the main focus of ITC analyses has been to extract forest resource information from digital imagery (satellite or aerial), the potential importance of LiDAR data within such a process should not be overlooked. The synergy with LiDAR data can occur at many levels. At one extreme, with experimental high density data (>10 pulses/m2), Individual Tree Crowns (ITCs) can be extracted directly from images derived from the LiDAR data itself (Leckie et al., 2003). Here, we will discuss the benefits of the "soon-to-be" more commom low density (0.5-1 pulse/m2) LiDAR data (already very common in Scandinavian countries).

Airborne LiDAR data are derived from the returns of Laser pulses aimed from the sensor towards the ground and are often pre-separated by suppliers into first and last returns, generally corresponding for forested areas to canopy and ground returns (multiple returns and full-wave LiDAR sensors also exists, but will not be discussed here). One obvious use of such data is to create, by interpolation between the pulses, Digital Surface Models (DSMs) and Digital Terrain Models (DTMs), from which Canopy Height Models (CHMs) can be derived. DTMs have multiple uses in forestry operation planning (e.g., road construction) and provide useful information when linked with other forest inventory attributes (e.g., site quality). CHMs can be used to infer stand heights (occasionaly tree heights) and can constitute an important component of the stand delineation process.

While the ITC Suite strength is its ability to delineate, identify and regroup trees into forest stands (or other environmental strata) while providing a high level of detail on their content, mechanisms are needed to constrain the ITC algorithms to the forested areas (i.e., differentiate tree areas from other image features). Masks of non-forest areas can often be created by specifying simple thresholds on specific image spectral bands and/or on derived images (e.g., NDVI images). Forested areas can often be ascertained by image segmentation, texture analysis or simple unsupervised classification of cruder resolution images. Sometimes, the information from a base map or an older forest inventories (typically already in GIS format) can also be used for such purposes. Unfortrunately, these methods can only provide a crude, first level masking of the non-forested areas. Nevertheless, they are often sufficient to allow the ITC algorithms to function properly.

Using such techniques, vegetated (yet, non-forested) areas are often not successfully masked out (e.g., forest gaps, meadows, clearings, etc.) and some areas of forest canopy are inadvertently masked, often due to the spatial imprecision of some of these techniques (e.g., texture analysis). This can result in either commission or omission errors in the tree isolation or, more generally, poor crown delineation.

Using a simple threshold (< 2m) on a LiDAR-derived canopy model can often lead to a more accurate stratification of forest vs. non-forest areas. The effect of using LiDAR-based masking to constrain ITC analysis to valid forest areas is illustrated in Figure 1. Here, non-forest areas such as clearings and gaps (a) are excluded from analysis using a LiDAR-based mask (b). The ITC tree crown delineation algorithm is constrained to areas of valid forest canopy only (c). Note that such CHM threshold is one of the few techniques that allows the ITC software to deal (often properly) with single trees in open vegetated areas (c). Nevertheless, thresholds on spectral channels are still useful to create specific masks of interest such as roads, landings, right-of-way, or buildings. These features will always add plus-value to the final product.

Figure 1 - Non-forest areas such as clearings and forest gaps (a) are excluded from analysis using a LiDAR-based mask (b); subsequent ITC analysis is constrained to valid forest areas (c). Photo: Silvatech Consulting Ltd., BC

N.B.: A Canopy Height Model (CHM) generated from stereoscopic correlation may work just as well.

---