Mondi recently acquired an airborne lidar dataset covering 27 000ha in the Melmoth area close to Richards Bay. Being a new investment, the first of its kind for Mondi, it was important that the maximum value be extracted from this lidar acquisition.
by Mark Norris-Rogers – Mondi Ltd and Ron Behrendt – Behron LLC
mark.norris-rogers@mondigroup.co.za; behron@centurytel.net

Starting in 1994, Mondi began utilising geographic information system (GIS) software from Esri (esri.com), and it is now one of its core management information systems. It was therefore opportune to use ArcGIS to process the lidar data to perform quality assurance on the data received and derive practical products that would aid forest managers in their work. These products included the following:

• Digital elevation models (DEM): Continuous surfaces that accurately represent the bare earth terrain features;

• Digital surface models (DSM): Continuous surfaces that accurately represent the above ground features such as tree canopies, buildings, towers, and power lines;

• Canopy height models (CHM): Continuous surfaces that accurately represent tree canopy heights derived by subtracting a DEM from its equivalent DSM; and

• Aspect and slope data.

All these products provide useful input information for forest planning, site classification, and environmental monitoring processes, among other applications.

The results from this project demonstrate how quickly and easily forest management information can be extracted from lidar data using out-of-the-box ArcGIS tools and functionality. The focus of this article addresses the business outcomes from using lidar data, giving forest and land managers an understanding of what information products can be produced. A companion white paper provides a more detailed how-to guide on creating products such as DEMs, DSMs, CHMs, and aspect and slope data and is aimed at GIS technical personnel.

Why lidar?
Forest management is fundamentally spatial in nature, not only in horizontal 2D space (x,y axes) but also in vertical space (z axis) (i.e., elevation and height). Area measurements in hectares or acres, volume measurements in cubic metres per hectare/acre, and distance and height/elevation measurements in metres/feet are critical for forest management. Therefore, the ability to accurately measure these is an important function for forest management staff.

A major challenge historically for operational forestry applications has been the ability to derive elevation data that accurately describes terrain conditions (elevation, slope, aspect, ground roughness, and more) over large areas where the ground surface is often not visible due to dense canopy cover. These terrain conditions have major cost and safety implications for forestry operations, ranging from site classification and sustainability issues to harvest planning and machine operations. In addition, other critical forest descriptors requiring a good understanding of the vertical (z) element include the tree stand structural characteristics such as tree heights, diameter at breast height (DBH), canopy diameter, and stocking (trees per hectare/acre). Data and tools that can provide and analyse this information are therefore vital to forest managers.

The advent of airborne lidar technology has provided forest management staff with the ability to create the previously mentioned input layers that are more detailed and accurate than similar layers produced with earlier technology, such as contour interpolation and auto-correlation techniques.

Lidar data acquisition and processing
For Mondi's first lidar collection, data was acquired at a minimum point density of six points/m2, together with false colour infrared imagery at a ground resolution of 10cm. The lidar data was delivered as LAS-formatted1 files with the points classified as ground or non-ground. A classified point cloud is required to be able to separate the point data into relevant products (e.g., ground returns are used to create a bare-earth DEM surface). The primary reason false colour infrared imagery was specified was due to the unique responses that vegetation returns in the near-infrared band, enabling the imagery to be used to classify and extract growth vigour and other information about the vegetation.

On receipt of the lidar data from the acquisition vendor, the files were ingested into Esri's ArcGIS 10.1 LAS datasets. Beginning with ArcGIS 10.1, a new data format was developed specifically for lidar data called a LAS dataset, having a .lasd file extension, which enables ArcGIS to organise and manage lidar files in their native LAS format. Depending on data volume, one might need to work with several LAS datasets to cover the full lidar area.

Quality control
Quality assurance/quality control (QA/QC) processes were required to ensure that the acquisition specifications were met. These checks included assessing whether 100% ground coverage was achieved during the lidar acquisition; whether any outlier data, such as points falling below ground level or points far above maximum canopy height (usually from returns reflected off flying birds), were present; and if the minimum point density was achieved. Outlier points were reclassified as required.

Area coverage, as well as point spacing, was also calculated to ensure the lidar data conformed to the acquisition specification.

Derivative products created from the lidar and associated benefits
Lidar data in the form delivered by the acquisition vendor is simply a 3D point cloud that has limited use in ArcGIS beyond simple visualisation. However, by using selected classes from the point cloud, continuous surfaces can be created to provide a variety of useful products. Using the functionality in the ArcGIS 3D Analyst extension and selecting only ground returns, a DEM was produced. Similarly, by selecting only the first returns, a DSM was created. From these derived datasets, canopy height models, slope, slope class, and aspect datasets were created. For better terrain visualisation, hillshade representations of the DEMs and DSMs were also produced. These visualisation datasets provided very detailed information, which gave forest managers an easy and intuitive understanding of the terrain in which they had to operate.

Terrain visualisation and ground roughness
An important part of forest harvest planning is to have a good understanding of the terrain where harvesting operations are to be conducted. Terrain characteristics influence machine accessibility, log extraction and storage, machine/terrain matching, and other factors. Historically, this information has been quite challenging to acquire without extensive ground surveys, which are often too expensive and impractical to implement since terrain features are often hidden by vegetation or topography. Even when infield, it is difficult to obtain a holistic view of the area to be worked on.

The ability of lidar pulses to find their way through very small gaps in the canopy enable a lidar survey to produce an extremely high-definition map of the bare ground, not only in terms of micro terrain features, such as dips and hollows, terraces, and similar uneven features, but it can also provide indications of where obstacles exist, such as large boulders or stumps that can affect machine access within a stand. Apart from usually being hidden under canopy, these features are generally below the scale at which most contour or topographic maps are produced, and are therefore not mapped when using traditional cartographic methods.

Figure 2 illustrates the variability of roughness within these stands. With lidar data captured at six points per square metre, sufficient ground strikes were obtained to reveal the actual terrain features that lie hidden under the forest canopy. Comparing the forest stand in the lower right-hand corner of the image to the stand on the lower left-hand side of the image reveals much rougher terrain conditions in the former stand. Locations where features need to be verified or additional details obtained can be identified from the lidar data, and focused ground visits can be planned, resulting in considerable time and financial savings for these field-based activities. Other features that can be highlighted include drainage channels, small wetlands, or poorly drained areas, and similar features that would impact on road design and maintenance.

Slopes and slope classes
Slope plays a major role in many forest planning and operational activities, influencing such factors as safety, productivity, machine selection, site sustainability, and accessibility. Therefore, the ability to derive detailed slope and slope class information is important and can have a significant impact on the forest plan.

Figure 3 illustrates how slope affects the operation of mechanical harvesters. Slopes between 45 and 60% require self-levelling harvesters that have a cost premium over conventional harvesters. There are therefore major cost implications where self-levelling harvesters are required, hence the need to map these areas accurately. In addition to the cost factor, there are also major safety implications, where incorrect machine/terrain matching could cause a serious accident or injury.

Figure 4 illustrates the increase in detail and accuracy of slope class definition. The left-hand image shows the slope classes derived from conventional 10m contours captured using a stereoplotter. Twenty-metre cell-size raster DEMs were derived from this contour data and used to classify the slope classes. The right-hand image shows the slope classes derived from a 1m lidar-derived DEM. One can see an increase in the 45-60% slope class, compared to the 20m DEM slope classes, as well as in the >60% slope class, which was absent in the 20m DEM slope class data. These improved slope classifications will have a significant impact on future harvest planning.

Figure 3. How slope affects the operation of harvesters. Slopes of 45-60% require self-leveling harvesters (right-hand picture).

Figure 4. Twenty-metre slope classes compared to lidar-derived one-metre slope classes.

Machine access
Another significant aspect of successful harvest planning is the ability to assess the accessibility of mechanised equipment to enter forest stands. This is particularly the case in steep terrain where high road banks or berms can hinder access. The level of detail provided by lidar data enables these obstacles to be identified during the planning process, thereby improving planning efficiencies. Figure 5 illustrates a slope map derived from a 1m resolution lidar DEM, where the slopes are graded from level (green) to very steep (red). High road banks or berms can be seen as narrow strips of red running parallel to the road lines.

Figure 5. Identification of road banks affecting stand access.

The heights of these road banks can be determined from querying the DEM heights, making it possible to acquire a reasonable estimate of the height of these banks. The adjacent photograph shows the road bank in relation to how it is illustrated on the slope map. Potential access points can be identified from the map, thereby reducing the amount of time required for in-field verification. Access points with acceptable slopes can be highlighted and operational stand harvest plans developed appropriately.

Log storage
Steep road banks also hinder log storage and landing space off the road area, resulting in extracted timber from a forest stand having to be stacked along the road surface. Knowing acceptable access points to forest stands reduces distances to suitable landings or depots and lowers extraction costs, thereby improving planning and operational efficiency.

Aspect
Aspect is another terrain feature with influence on silvicultural practices and affects site conditions, such as soil humidity, temperature stress, and similar tree growth conditions, as well as risks, such as fire intensity. For example, in the southern hemisphere, hot and dry northerly aspects should be planted earlier in the summer planting season than cooler, more moist southerly aspects. Accurate definition of aspect can impact the quality of decision-making when such planning and operational activities are required. The percentage of each aspect class within every forest stand is provided as a basic planning input for Mondi's foresters and planners. Figure 6 illustrates the improvement in detail of aspect classification of the 1m lidar-derived DEM compared to that derived from the original 20m DEM. This greater detail shows that the northern aspect is actually the dominant aspect compared to a perceived dominant easterly aspect that resulted from the 20m DEM, in part because small terrain features were smoothed out, giving a different result than was actually the case.

Figure 6. Twenty-metre aspect classes compared to lidar-derived one-metre aspect classes.

Contours
Contour data provides a simple, intuitive picture of the terrain and is particularly useful in communicating elevation heights. Knowledge of elevation above sea level is important in forest management decisions. For example, most species have specific elevation range requirements for optimal growth, which need to be taken into account when selecting species to be planted in forest stands. By reducing the risk of off-site planting and the resultant stress conditions, one can minimise the impact of pests and disease infestation due to this stress.

When compared to the 10m derived contours that Mondi has used historically, the increase in terrain detail available from a lidar-derived DEM enables much more accurate contour data to be produced. Figure 7 illustrates this very well. The blue lines in the image show the original 10m contours, while the red lines are 2m contours derived from the 1m lidar DEM. One can see that while the general form is similar, the blue contours miss much of the fine detail that is captured by the red lidar-derived contour data. Smaller drainage lines and shallow valleys are lost by the coarser 10m contours but are accurately reflected by the 2m contours.

Figure 7. Comparison between 10m contours (blue) and lidar 2m contours (red).

Improved forest road and stand delineation
For many years, stand boundaries, forest roads, and other topographic features have been mapped using on-screen digitisation/tracing of these features from digital orthophotographs. However, it is often difficult to accurately capture the necessary features because they are hidden under the tree canopy. This is particularly the case when mapping roads, as their true alignment is often masked by tree shadow or canopy, and the operator has to estimate where the road actually runs. Using GPS data can assist in some cases but is not always available, as canopy cover can also prevent adequate GPS reception from being achieved.

The identification of road misalignment has been highlighted as a result of the lidar data acquisition. The DEM hillshade product is able to accurately visualise road networks and can be used as the base layer to enable GIS editors to correctly capture road alignments. It also provides a very intuitive dataset for operational foresters to verify their roads. Figure 8 illustrates the road alignment issues. The brown lines represent the roads in the GIS database, and when overlaid on the DEM hillshade, correcting any misalignment is easily done. Missing roads can also be quickly identified. Once a correctly aligned dataset is achieved, not only are distance measurements accurate but other road metrics, such as minimum turn radii of sharp bends, can be calculated. This is an important measurement when long-length logging trucks are used to haul out timber, as these vehicles have specific minimum turn radii requirements to be able to travel along routes (see Figure 9). Knowing this factor can improve route planning, road maintenance and upgrade plans and extraction cost calculations.

Figure 8. Road misalignment highlighted on lidar DEM hillshade. Figure 9. (right) Logging truck Illustrating the length of these vehicles.

In Mondi's case, it was found that these road alignment issues were particularly noticeable in steep areas compared to flatter terrain, where there was better alignment of the roads. This fact reinforces the importance of lidar in improving the roads data because of its ability to accurately capture detailed terrain features, especially in steep areas.

Stand tree heights
A critical measure for foresters is stand tree height, and a core function of lidar is the ability to record height data accurately. Subtracting a DEM from a DSM results in a CHM, and in addition to providing tree height data, this product also provides a very good indication of variability within a stand. Summary height statistics, such as the mean height and range at a stand level, can be derived using the statistical functions within ArcGIS Spatial Analyst. Where additional statistical relationships, such as height/DBH curves are available, further analyses can be done to derive stand tree size and volume estimates. Figure 10 illustrates a CHM, where different-aged stands can be identified, as can height variation within a single stand. In mature stands, the brown shades indicate higher trees than the yellow and green shades, while the purple colours represent young stands 1 to 3m high.

Figure 10. Canopy height model, or CHM, showing different height classes. The false-colour infrared image is provided for comparison purposes.

More detailed tree height information within individual forest stands was obtained by deriving a height class frequency distribution for each stand. This table was then processed in a statistical package to extract the required statistical data, removing outlier data and deriving very accurate dominant tree height data for each stand. Having 100% height coverage of the individual stand in place of sample plot heights provides a much more robust measurement of the tree heights, as well as providing a spatial representation of where height variability occurs within a stand. Stands with yellow/green shades, as well as purple shades, indicate wide variability within the stand, usually indicating gaps in the stocking of a stand (i.e., a poorly stocked stand).

Stand boundaries
Stand boundary delineation can also be improved using either the canopy height model data or the DSM hillshade data. When the GIS forest stand boundary vector data is overlaid on either of these datasets, any misalignment is quickly highlighted. Figure 11 illustrates this. The blue line is the GIS forest stand boundary data, and deviations from the actual standing trees are clearly seen. Correcting the stand boundary delineation can be done by tracing the edge of the trees highlighted in the CHM or DSM data.

Figure 11. Stand boundary misalignment highlighted in the CHM and DSM hillshade.

The impact of incorrect stand boundaries can be significant, particularly where payment for work done is based on area worked. These area values are usually derived from the GIS database and, if incorrect, can lead to under- or overpayments being made. There can also be a multiplier effect, where several different operations are conducted on a stand. Stand boundaries often change across rotations (and even within rotations, where damage occurs from such events as fire, insect attack, or frost), and it is vital to monitor these changes.

Another use of the CHM is to highlight stands where harvesting has not been completed. The area of these unfelled patches can be extracted and calculated. Figure 12 illustrates this aspect, where the brown, yellow, or green colours indicated standing trees within a felled area.

Figure 12. Identification of unfelled patches within a forest stand.

Environmental applications
Lidar data can also provide useful information for environmental management. Applications include the identification of erosion channels; land-use planning at a landscape level; high conservation value (HCV) forest identification and mapping; and invasive alien plant mapping, particularly where this occurs in grassland areas. More advanced areas of application would be in natural forest stand structure research and classification. Other uses include very accurate hydrological stream identification and mapping.

Mondi plans to undertake an indigenous tree planting programme along selected riparian zones that have had commercial tree stands removed to improve available stream flow. If left open, these areas tend to become weed infested, but this risk is reduced by planting indigenous trees in these areas. The lidar data will assist in identifying these riparian zones (called A Channels).

Figure 13 illustrates how well erosion channels can be highlighted from a lidar DEM, even though hidden by vegetation cover when seen on the orthophoto. Indications as to severity, depth, and extent of the erosion can be assessed from the imagery and corrective measures planned.

Figure 13. Identification of erosion ditches from lidar DEM hillshade.

Figure 14 shows a riverine forest area that could be classified as an HCV forest. Other characteristics, such as canopy density and height classes, could be extracted from the lidar data.
Conclusion

Figure 14. HCV riverine forest patch identified from DEM hillshade.

Often, an initial reaction to acquiring a lidar collection is concern regarding the cost of obtaining such data, frequently followed by concerns with working with such large volumes of data. However, the detail and accuracy of information produced from lidar are such that the cost-benefit is positive and very real. In the case study described in this report, the improvement in slope class definition and its impact on Mondi's ability to improve harvest planning justified the cost of obtaining the data. These harvest plans were no longer based on estimated areas but on actual calculated areas. In addition, there was a very significant reduction in accident risk due to incorrect machine/terrain matching. Areas requiring the use of self-levelling harvesters were now clearly defined, and the risk of a non-self-levelling harvester being used in terrain too steep, was greatly reduced.

Together with the ability to minimise the impact of road banks affecting access to stands and problems stacking timber on roadside, this has enabled forest planners to produce more cost-effective harvesting plans. The added benefits of improved road and forest stand boundary data, the ability of the DEM hillshade products to enable users to clearly visualise the terrain they needed to work in, the provision of stand tree height data, as well as a better understanding of variations within stands all resulted in improved management focus and better management decisions.

All the products described in this report were derived using standard ArcGIS 10.1 functionality available in the core ArcMap or in the 3D Analyst extension, including such tools as LAS Dataset To Raster, Hillshade, Slope, and Raster Math functions. The software had the ability to handle the data volume of just under 400GB.

The success of this initial project has led Mondi to expand its lidar mapping programme. In addition, planning is under way to develop more sophisticated products that will assist in deriving quantitative forest stand structure data that can be used to produce timber volume estimates from the lidar data.

Acknowledgements
The authors would like to acknowledge the input and assistance of the following people and organisations:

• Cody Benkelman, Imagery Product Manager, Esri Inc. for his technical guidance and input;

• Peter Eredics, Forestry Manager, Esri Inc. for his approval of and support for this project;

• Mondi Ltd, for making the lidar data available.

Published in June 2013