THE MONITOR - Project news: 1.1.3 Remote sensing of forest inventory.

In the last edition of The Monitor, we introduced some of our work investigating the utility of airborne laser scanning (ALS) data for rapid and cost effective stream survey and promised to show some of the results of our endeavours.­

We report on some key findings here, which indicate that laser scanners flown over a piece of land at different heights to aquire various levels of measurement detail (flown in operational configurations), can be used to locate the position of streams very accurately over densely canopied forests grown on reasonably flat terrain.

We see the potential to extend this work to consider the role of other vegetation, terrain and climatic variables which may be available also to predict the locations of stream sources and rates of stream flow. Doing so would add to inventory information and would allow better management of our stream resources in forested areas.

ALS data is being used in milestone 1.1.3 to develop models of canopy height and density that can be employed in forest inventory for improved forest management. Work is also proceeding that examines wider applications for laser scanner and image data in forestry.

We selected an unthinned radiata pine plantation for our first study site. Stream positions were surveyed using air photo interpretation(API) prior to planting. Over bare ground this method identifies visible features to within 12.5 metres of their true position 90% of the time. We compared this bare-ground API survey with a survey undertaken using thinned, last-return laser scanner data once the forest was well established and bare ground was no longer visible from above.

What is last-return laser scanner data, you may ask? The reflected return signal of a laser pulse is waveform. The sensor used in this study recorded the first and last component of this waveform, otherwise known as the first and last return. The last return is data from (hopefully) the most distant object sensed by that particular pulse. This object is (hopefully) the ground, but often it is something else, like vegetation. The last return data is therefore 'thinned' to seperate ground and non-ground pulses.

We used simple interpolation algorithms that are available in open-source software to generate our terrain models, and simple flow direction and accumulation algorithms to locate the stream positions. Here, we report a comparison of two interpolation algorithms and two flight configurations.

The site is a 15 year old radiata pine coupe planted on rolling terrain. Some graphs of coupe condition appear below. Of particular note is the high density of the canopy (Figure 1b). This feature and the availability of a bare ground stream survey make the site ideal for our study. We can assess the usefulness of laser scanner data under extremely dense canopy conditions.

Figure 1: Forest coupe condition

We tested the nearest neighbour interpolator (a simple method which assumes that terrain height equals that of the nearest data point) and the inverse distance weighted interpolator (a slightly more sophisticated method that weights terrain height according to distance from data points). We also tested these interpolators on laser scanner data that was obtained in two flight configurations. These flights differed in flying height and the maximum angle of the laser scanner pulse. Thus, laser footprints differed in size and angle of incidence of reflection. The results of our analysis appear below. The frequency graphs show how the differences between API survey positions and laser scanner survey positions are strongly weighted toward zero.

Figure 2: Positional differences between API and laser scanner survey

This indicates that the laser scanner is as at least as precise at mapping streams in vegetated areas as API is at mapping streams in non-vegetated areas. API is notoriously unreliable in forests but is more than adequate over bare ground, and so we now have a method for stream mapping in forested areas with at least equivalent precision to that of bare ground API stream mapping.

The 90th percentile for each of these appears in the table below. The simpler, nearest neighbour interpolator produces the best results, perhaps because it does not over-smooth the data in the steeper areas around the stream courses, allowing more accurate stream flow simulation across the terrain model. Despite acquiring the data at a wider scanning angle (ie a greater spacing between pulses over the terrain), the lower flying height produced the most accurate result; one which is very similar to the stated accuracy of the API survey.

Table 1: 90% percentile difference between API and laser scanner survey of stream position

Rob Musk
Post-doctoral research fellow.
1.1.3 Remote sensing of forest inventory.