Capturing the potential of Lidar

LIDAR has principally been used as a ranging instrument to map elevation. This application requires only the recording of the time taken for the first return. Technological advances mean that Lidar can now be used for more sophisticated measurements such as the description of vegetation canopies. Developments in the constituent technologies (such as digitization and timing) have facilitated the recording of the complete time-varying return. Thus the canopy can be investigated throughout its depth. Furthermore, much data provided by lidar cannot be derived by any other remote sensing method. Its contribution is unique.

What follows is a demonstration of how raw lidar data can be processed to describe vegetation canopies. This demonstration is based on raw data from a NASA experimental instrument (SLICER) as interpreted by CSIRO formulated algorithms. The SLICER instrument was flown over several sites between 1994 and 1997. The data used here are from a single flight line over the BOREAS study area in Canada during 1997. The flight path was over a region of coniferous forest.

The purpose of this exercise is to (1) develop and demonstrate the data processing method; (2) give preparatory insight into future Lidar projects such as VCL; (3) provide strategic direction for an instrument to be built by CSIRO and commercial partners.

Extracting the vegetation signal

The raw data returned by the lidar is the relative intensity of light reflected as a function of time after the outgoing pulse. By translating time into range, we can derive the relative height at which the reflections occurred. Once the ground pulse has been identified, the reflected waveform can be interpreted in terms of height above ground. This is shown in the following figure. The background noise level has been estimated and subtracted from the data shown here. The narrow pulse centred on zero is the reflection from the ground. The asymmetry of this pulse is due to the shape of the outgoing laser pulse. This was designed to have a rapid risetime and asymptotic decay.*

The ground return pulse must be removed in order for the vegetation return to be studied. This can be done by fitting a pulse of the expected shape and subtracting this from the waveform. The following figure shows the same waveform with an asymmetric gaussian subtracted to remove the ground return pulse.

This figure shows the reflected light over a single spot of about 8m in diameter. The first return above noise level tells us the highest point in the canopy within this circular area (about 13m above ground). The shape of this waveform suggests a concentration of foliage (needles and branches) around 10m and an understory of 2-3m in height. This is a plausible result for a coniferous forest.

The spot size of the SLICER instrument is similar to the crown size of the trees, so we expect to see considerable variation in the shape of returns from shot to shot. If trees are clumped i.e. there are groups of trees and gaps between the groups, there will be some areas where the only reflection comes from small plants (such as grasses) and the ground. Also, we expect that the return profile would be quite different for broad leafed canopies (such as eucalpyts) which have both different shaped leaves and canopies.

Interpreting the Lidar profile

For each lidar shot we can derive the fractional cover (the fraction of the vertical view that is occluded by foliage) over the area of that spot. This is calculated as the cumulative sum of returns to each height, divided by the total reflection from foliage and ground. The ground return must be scaled by the relative reflectances of the ground and vegetation. Fractional cover is plotted here against height in the canopy.

This plot shows a fractional cover of 0.82 over the spot sampled by this lidar shot. Looking up from the ground, only 18% of the sky would be visible in vertical view. The shape of this plot tells us something about the shape of the trees. About 60% of the cover lies below 10m, so the top part of the trees must be sparse as cone-shaped coniferous trees are. Also, there is very little contribution to the cover below 2m, so the understory is also sparse.

Fractional cover leads to a quantity that we call gap probability. This is simply 1.0-fractional cover and so represents the fraction of sky visible when looking up through the canopy from different heights. Gap probability is plotted here in red, the blue line is fractional cover as shown above.

Gap probability (Pgap) at different heights through the canopy leads to the apparent foliage profile, or foliage area per unit area at each height through the canopy. The relationship is

This is the vertically projected foliage profile. The actual foliage profile depends on the distribution of the foliage elements (leaves, branches etc) in space. This equation assumes a random distribution, which is an acceptable but not accurate description of the distribution of foliage elements in real trees.

Horizontal extension of the vertical description

The above analysis shows the main steps in processing lidar data from a single shot. The sample SLICER dataset contains data from 4000 laser shots. The small spot size of the SLICER instrument relative to tree size results in significant shot to shot variation. To understand the whole area sampled, it is useful to summarise the results as a series of histograms and scatter plots.

The histogram of canopy heights within the 4000 sample shots reveals two distributions. The dominant one is centred around 12-14m and there is another minor peak at 2-4m. This indicates the proportion of clearings with regrowth or low understory plants, the minor peak, relative to the taller forest canopy.

The distribution of fractional cover sampled is also bi-modal, showing a significant proportion of spots with cover less than 0.1. These are probably related to cleared areas, but could also just be large gaps within the canopy; the sites of treefalls perhaps. This issue could be quantified by using a lidar with a variable spot size. It must be recognised that the area of the lidar pulse/shot determines the accuracy and spatial coherence of all subsequently derived variables. It is important to choose a spot size appropriate for the purpose for which data are being collected.

A common way to display canopy structure data is a structure diagram, which is a plot of canopy height against cover. For this coniferous canopy we find a non-linear relationship with considerable variation. Again we see two populations - canopies and clearings. The variation within the canopy population could indicate local growing conditions or the disturbance history. Whatever the reason, the significant point is that lidar provides a new and unique quantitative description of canopies. The interpretation of these data has just begun.

For more slicer results download this file: SLICER_sample_web.pdf