**2. Satellite lidar profiling**

NASA's Geoscience Laser Altimeter System (GLAS) aboard the Ice, Cloud and land Elevation Satellite (ICESat) is currently the only satellite lidar system to have provided near global sampling coverage over an extended period of time. It therefore offers a unique dataset of vertical profiles of the Earth's surface.

ICESat was launched in January 2003 and the mission continued until the final laser ceased firing in the Autumn of 2009. During this period, the laser was operated for approximately month-long periods annually during Spring and Autumn, and additionally during the Summer earlier in the mission lifetime.

This is a full waveform, lidar profiling system which operated at an altitude of 600km, travelling at 26,000 km h-1 and emitting 1064nm laser pulses at 40Hz. This caused the Earth's surface to be sampled at intervals with footprint centres positioned 172m apart. Footprint diameter and eccentricity have varied considerably between laser campaigns from a major axis of 148.6±9.8m to 51.2±1.7m (Figure 3). Comprehensive information regarding the sensor are available from Abshire *et al.*, 2005; Brenner *et al.*, 2003; NSIDC, 2010a; Schutz *et al.*, 2005; Schutz, 2002. Data plus tools to process them are available free of charge from NSIDC, 2010b.

### **2.1 Characteristics**

The systematic sampling pattern and representation of the vegetation profile within the returned lidar waveform signal (Figure 2, left) enables the spatial distribution of vegetation parameters to be mapped for large areas. The seasonal coverage allows near repeat measurements at a frequency which would not be feasible using conventional survey methods. However the system was designed primarily for cryospheric applications and therefore the configuration is not considered optimal for vegetation analysis. This poses some challenges for forestry applications. Upon sloped, vegetated terrain, vegetation and ground surfaces may occur at the same elevations. This causes the signals from ground and vegetation to be combined within the waveform and, where it is not possible to distinguish a ground peak, this prevents the signal returned from the vegetation from being reasonably identified. Furthermore, dense cloud cover prevents a valid return signal causing gaps in footprint sampling (Figure 3). For regions with high cloud cover such as the tropics, this can serve to worsen the already-sparse sampling density near the equator, produced by the polar orbit.

to wide-ranging variations in sensor design and characteristics as outlined above. In the descriptions within sections 2-4, an example of a satellite sensor is used to demonstrate the principles of large footprint, full waveform profiling data, whilst airborne and terrestrial lidar instruments are used to provide examples of small footprint, discrete return laser scanning.

Lidar remote sensing provides a direct estimation of the elevation of intercepted features. In the context of vegetation, if signals from the ground and vegetation can be distinguished, the relative heights above ground of forest canopies can be calculated. Since an adequate stem diameter and canopy structure are needed to support tree dimensions, vegetation height is closely related to volume and therefore biomass. The sections below provide

NASA's Geoscience Laser Altimeter System (GLAS) aboard the Ice, Cloud and land Elevation Satellite (ICESat) is currently the only satellite lidar system to have provided near global sampling coverage over an extended period of time. It therefore offers a unique

ICESat was launched in January 2003 and the mission continued until the final laser ceased firing in the Autumn of 2009. During this period, the laser was operated for approximately month-long periods annually during Spring and Autumn, and additionally during the

This is a full waveform, lidar profiling system which operated at an altitude of 600km, travelling at 26,000 km h-1 and emitting 1064nm laser pulses at 40Hz. This caused the Earth's surface to be sampled at intervals with footprint centres positioned 172m apart. Footprint diameter and eccentricity have varied considerably between laser campaigns from a major axis of 148.6±9.8m to 51.2±1.7m (Figure 3). Comprehensive information regarding the sensor are available from Abshire *et al.*, 2005; Brenner *et al.*, 2003; NSIDC, 2010a; Schutz *et al.*, 2005; Schutz, 2002. Data plus tools to process them are available free of charge from NSIDC,

The systematic sampling pattern and representation of the vegetation profile within the returned lidar waveform signal (Figure 2, left) enables the spatial distribution of vegetation parameters to be mapped for large areas. The seasonal coverage allows near repeat measurements at a frequency which would not be feasible using conventional survey methods. However the system was designed primarily for cryospheric applications and therefore the configuration is not considered optimal for vegetation analysis. This poses some challenges for forestry applications. Upon sloped, vegetated terrain, vegetation and ground surfaces may occur at the same elevations. This causes the signals from ground and vegetation to be combined within the waveform and, where it is not possible to distinguish a ground peak, this prevents the signal returned from the vegetation from being reasonably identified. Furthermore, dense cloud cover prevents a valid return signal causing gaps in footprint sampling (Figure 3). For regions with high cloud cover such as the tropics, this can serve to worsen the already-sparse sampling density near the equator, produced by the polar orbit.

**1.6 Key concepts for biomass assessment** 

dataset of vertical profiles of the Earth's surface.

Summer earlier in the mission lifetime.

2010b.

**2.1 Characteristics** 

**2. Satellite lidar profiling** 

examples of applications of lidar systems for biomass assessment.

Fig. 3. Multiple GLAS laser campaigns sampling overlaid on a GoogleEarth image of Tambopata, Peru. Missing data are found where dense cloud prevents sufficient energy from penetrating to the Earth's surface and returning to the sensor.

Within each footprint, if the top of the canopy is assumed to be the start of the waveform signal (upper horizontal blue line, figure 2 left), the accuracy with which the signal returned from the vegetation can be identified depends on the ability to identify a representative ground surface within the waveform. Methods to achieve this have included the use of an independent DTM to account for terrain slope within lidar footprints (Lefsky *et al.*, 2005; Rosette *et al.*, 2008) or using Gaussian decomposition of the waveform to locate a peak corresponding to the ground surface (Rosette *et al.*, 2008; Sun *et al.*, 2008a; Sun *et al.*, 2008b). Vegetation height can therefore be estimated as the elevation difference between the start of the waveform signal and the identified ground surface estimated within the waveform. The studies above report RMSE as 3+ metres.

#### **2.2 Applications for biomass estimation**

Most commonly, waveform indices of Height of Median Energy (*HOME*) or relative height percentiles above ground (*RHi*) are calculated using the cumulative energy distribution within this region of the waveform returned from vegetation. More recently, Lefsky *et al.*, 2007, devised an alternative method of estimating a vegetation height parameter which accounts for terrain and vegetation roughness using the waveform leading and trailing edges rather than isolating the signal returned from vegetation.

The sampling measurements produced within the satellite lidar footprints are typically combined with coincident field measurements of biomass. This enables regression equations to be developed using waveform metrics to estimate biomass for the areas

Lidar Remote Sensing for Biomass Assessment 11

from the outer layers of the canopy. However, the degree of penetration is ultimately related to a combination of factors such as the sampling density, beam divergence and

Possible scenarios may be as illustrated in Figure 4 below. With higher density of laser pulses, these difficulties are reduced but at the cost of higher operating expense and flight

Fig. 4. Possible scenarios of lidar point cloud interception of the forest canopy. Source:

d. One of the pulses is intercepted at a lower height due to canopy penetration wrongly

e. Trees on a mound can be assigned a greater height in the absence of a good model of

Despite common perception, lidar does not create tomographic images and, therefore, they cannot be considered as 3-D representations in the strict sense. Only the gaps in canopy cover and transmittance through leaves will allow laser energy to be returned from the ground. As with full waveform data, the critical step for the calculation of vegetation height metrics is to

Since lidar energy penetration through the vegetation canopy will vary with forest structure, density and laser scanning angle in particular, the last return of an emitted lidar pulse may not necessarily be returned from the ground surface. Therefore a means of filtering points is necessary in order to differentiate those returns reaching the ground from those being intercepted at different heights within the canopy (e.g. Kraus & Pfeiffer, 1998; Zhang *et al.*, 2003). A thorough comparison of different approaches and a complete description of filters can be found in Sithole & Vosselman, 2004. This allows the classification of points into ground and vegetation classes (Figure 5). Other algorithms can refine the classification

f. In a situation of sparse density of returns some trees can be ignored completely

distinguish between those points returned from ground and non-ground surfaces.

further to additionally identify features such as buildings, electricity cables, etc.

c. The most likely situation: laser returns do not hit the true top of the tree,

scanning angle (Suárez *et al.*, 2005).

Suárez *et al.*, 2005.

a. Laser hits the true top of the canopy,

suggesting two tree tops,

the ground surface beneath,

b. Small trees close to larger neighbours are ignored,

duration restrictions limiting spatial coverage.

sampled by the lidar footprints. The continuous spatial coverage of optical or radar data permit these estimates to be extrapolated.

Similar data fusion techniques have been used to determine biomass distribution by several authors and encompass a wide breadth of vegetation types and have been applied from regional to continental scales, including a focus on Africa's mangrove forest (Fatoyinbo & Simard, in press 2011), Siberia (Nelson *et al.*, 2009), Quebec (Boudreau *et al.*, 2008), and for mapping throughout Africa (Baccini *et al.*, 2008). Work is currently underway by the latter research group at Woods Hole Research Center, led by Josef Kellndorfer, to estimate tropical forest biomass globally (WHRC, 2011). Additionally, research utilising GLAS is in progress as part of NASA's Carbon Monitoring System initiative (NASA, 2010) to determine biomass distribution within the US (as well as to produce higher resolution biomass maps at a county level using airborne lidar data).

Global vegetation height products derived from GLAS and optical data (Lefsky, 2010; Los *et al.* 2011) or combining GLAS and radar data (Simard, 2011) open possibilities to improve our understanding of global processes (Los *et al.*, 2008) as well as allowing applications for biomass assessment.

However, comparability of data and methods must be taken into consideration. These methods rely on the application of regression equations to extend vegetation parameter estimates across large areas. Nelson, 2010 demonstrates how the calculation of biomass is often sensitive to the equation applied and lidar sensor characteristics. These inconsistencies have implications for repeat analysis and monitoring of change due to the effect of model selection and lidar system evolution on the outcome of biomass assessment.
