LandSerf 2.3 Tutorial

You can start LandSerf by using one of the following methods:

• Selecting it from the Start menu (Windows)
• Click on the LandSerf icon on the desktop (Windows)
• Click on landserf.command in the Applications folder (MacOSX)
• running landserf.sh from the command line (UNIX/Linux/MacOSX)
• run or click on landserf.bat (Windows)

Step 1 - Importing an elevation model.

The first stage of any terrain analysis is to import a digital model of the land surface to analyse. Much of LandSerf's functionality concerns processing Digital Elevation Models (DEMs) so we will concentrate on that here. A DEM is simply a matrix of elevation values that describes the height of some surface at regular spatial intervals.

For this tutorial, we shall import a DEM from the United States' National Elevation Dataset representing part of the Columbia River area in Washington state. You may wish to substitute an alternative area of interest (in which case you can find further details on importing data into LandSerf). The DEM we wish to import for this exercise is taken from the United States Geological Survey (USGS) Seamless data server that allows data to be selected from user-defined regions of the American continent. The data are provided in ArcGIS Binary Interleaved Layer (BIL) format and can be found in the data sub-directory of the LandSerf installation.

To import the elevation file:

You should see a square image of the elevation surface as shown in Figure 1. The smaller thumbnail image on the left-hand side of the LandSerf window acts as an 'index' entry. Every time a new object is loaded into LandSerf, a new thumbnail will be added to this index.

Step 2 - Georeferencing Information.

The elevation model provided by the USGS is currently georeferenced using global latitude and longitude values. That is, each of the four corners of the surface are associated with particular locations on the earth's surface. To find out these values we can view the georeferencing information by doing the following:

The problem with using global latitude and longitude values for analysis is that they do not correspond to fixed distances on the ground. One degree of longitude will vary in length depending on the latitude of the measurement. It is therefore useful for us to be able to reproject the surface onto a planar coordinate system where there is a fixed scale between the units used and measurements on the ground. For this tutorial, we will firstly store the fact that the imported file uses global latitude/longitude coordinates, then transform the DEM onto a Universal Transverse Mercator (UTM) projection:

After a few seconds, a new surface should appear as shown in Figure 4. Note that the index area now shows two images - the original as well as the reprojected surfaces. You can select any of the index objects at any time by clicking on the chosen image with the left mouse button. The more rectangular shaped surface now has a fixed resolution where each DEM cell represents a square of 30m x 30m on the ground.

Step 3 - Displaying the Surface.

The images we have seen so far that represent the surface are made by allocating a colour to each cell value in the raster model. So for example, all cells that represent an elevation of around 500m above sea level are coloured green and all those that are 800m above sea level are purple. LandSerf can use this same elevation information to model the amount of light or shade on each raster cell by calculating local slope direction and steepness. This can be combined with the colouring scheme you have already seen to produce a shaded relief map of the surface:

The colour scheme associated with any object can be changed, either by selecting one of LandSerf's pre-defined colour schemes, or by creating a series of colour rules to be associated with the object. At this stage, we will select one of the pre-defined colour schemes:

You should now have a new shaded relief representation of the Lincoln elevation model shown in the main display area. LandSerf has used the key colours you have selected and interpolated new colour combinations between each of them to produce the continuously varying colours you now see.

You will notice that the northern part of our DEM is dominated by part of a large meandering river (the Columbia River) that is coloured green similar to the banks on either side. In order to distinguish the river from its surroundings, we can associate a unique colour with it by finding its elevation and attaching a discrete colour rule to that elevation alone:

Sometimes the level of detail represented by an elevation model is greater than that displayed on the screen. It is therefore useful to be able to 'zoom in' to an image in order to examine it in greater detail. This is one way of exploring how the characteristics of a surface vary with scale.

Step 4 - Attaching metadata to Rasters.

A spatial object in LandSerf consists of the data (elevation values in the DEM we have used so far) and a collection of metadata attached to that object. We have already seen some of those metadata in the form of the georeferencing of the boundaries of the surface, the map projection used, the resolution of each raster cell and the colour rules used to display the raster. We can also create and examine other metadata such as the title, lineage and attribute categories associated with a spatial object.

LandSerf allows several datasets to be stored and displayed simultaneously. We will add a further raster layer representing landcover measured and classified from satellite remote sensing. We will then add some additional metadata to this new layer. Unlike elevation, landcover classes are categorical rather than direct measurements. Each cell is allocated a numeric value that represents a class of landcover (water, shrubland, woodland etc.). The value of that number is not directly important other than providing a unique identifier to a given landcover type.

We will change 4 elements of the new landcover dataset - the attribute labels; the colour scheme; the type of raster; and the lineage notes of the layer. These changes will make it easier to process and interpret any analysis we apply to the data at a later stage.

Step 5 - Simple morphometric analysis.

We have already gained some impression of the landscape represented by our DEM simply by visualising it. We can assume for example that the landscape contains a large meandering river, some apparently mountainous region to the north, several tributaries flowing from the south of the DEM northwards into the river. We might characterise the region as a whole as 'hilly'. We also know from the landcover information that large parts of the landscape are grassland with some patches of woodland and agriculture.

Such description however has obvious limitations. It is subjective (you may have gained a different impression of the same landscape), it is rather vague (what exactly does 'hilly' mean?), and might simply be incorrect (are we sure that is a mountain to the north of the river or is it just a small undulation?). We can use LandSerf to provide us with a more systematic and objective description of the landscape; one that we can use to compare with other landscapes described by other people.

Firstly it is useful to make sure we have a good idea of the scale and extent of the landscape. We can do this by viewing the metadata associated with the DEM.

This vertical range or relief we have identified from the raster information window only gives us a broad summary of how elevation changes over the surface. We can get a more detailed view by examining the frequency distribution of elevation values.

So can we get a better idea of the 'roughness' of the terrain? One way of doing this is to measure the steepest slope at all cells in the DEM and visualise the results:

Slope represents one of the four surface parameters that are often used to characterise surface behaviour. The other three are aspect which identifies the azimuthal direction of steepest slope, profile curvature, which describes the rate of change of slope in profile, and plan curvature, which describes the rate of change of aspect in plan. We shall use LandSerf to view each of these in turn:

Finally, we shall examine one further characterisation of the surface. Rather than measure surface parameters we can classify the terrain into surface features using some of the parameters we have already investigated. One of the commonest (and most useful) classifications is to group all points on a terrain into one of the following:

pits
channels
passes
ridges
peaks
planar regions

This can be done by measuring both the slope and curvature of each cell in the DEM and performing the appropriate classification (for more details on how this may be done see the theory documentation). We will perform such a classification on our DEM:

Feature classifications like this are useful for several reasons. The pattern of channels appears somewhat similar to a drainage network, thus giving us an idea of where water would flow over the surface. Perhaps less obviously the pattern of yellow cells gives us an equivalent ridge network, identifying portions of the landscape where water is likely to flow from. Consequently, the ridge network is related to the drainage divides/watersheds that define drainage basins. In theory at least, the green passes identify points of intersection between the channel and ridge networks. Along with pits and peaks, passes represent what are sometimes called surface specific points - points on the landscape that carry special significance in its structure.

Step 6 - Advanced morphometric analysis.

When we wish to characterise a real terrain using a computer, the closest we can get is to characterise the elevation model that represents it. In doing so we hope that the model is sufficiently accurate for the purposes of analysis. As far as possible we should strive to produce characterisations of landscape that are as dependent as possible on the surface we are interested in, but as independent as possible of the particular model we have used.

One of the most significant limitations of the models and processing we have carried out so far has been the scale of analysis implied by the DEM resolution. You will remember that each cell in the Lincoln DEM is 30m x 30m. Since most of the analytical functions we have used so far work by comparing one cell with its 8 adjacent neighbours, the results of that analysis are likely to be at the resolution of approximately 3 times the grid cell size. So for example, the blue valleys we identified in the previous step will be those that have a width around 30-90m. This scale is really rather arbitrary; we might be more interested in characterisation of features have expression over 1km or 10cm. Unfortunately, the DEM cannot provide us with direct information at a finer scale than its resolution (30m), but we can use it to perform analysis at a broader scale.

The basis for most of the analytical functions in LandSerf is the process of quadratic approximation (see theoretical details on surface characterisation for further information). This involves taking a local window of DEM cells (3 x 3 in the examples we have seen so far), and fitting the most appropriate continuous quadratic function through them. LandSerf then uses this quadratic function to calculate parameters such as slope and curvature. We can set the size of the local window that LandSerf uses before it does any calculation. This allows us to perform analysis at much broader scales than that implied by the DEM resolution.

Firstly, we will view the effect of changing window size by smoothing out features that are smaller than about 330m wide. Since each cell is 30m, we will need to process the surface using window sizes of 11x11 cells rather than 3x3:

The result of broadening the scale of analysis is to smooth out many of the finest scale details while leaving the larger features of the landscape unaltered. You may also notice that the entire raster also appears smaller than the original. This is produced because analysis based on square windows of size nx n cannot process cells within (n-1)/2 cells of the edge. As a consequence, LandSerf fills the border cells of the raster with null values that are not displayed or processed.

We can see the effect of further broadening the scale of analysis by setting the window size to 65 (corresponding to a distance on the ground of 65*30m = 1.95km):

So do the parameters that we can measure from the surface also vary if we calculate them at this new broader scale? We can find out the answer to this question by measuring a surface parameter at many different scales and seeing how it varies:

We will use a similar process to explore the scale dependency of other measures.

One of the advantages of multi-scale processing of terrain is that we can relate our analysis to scales that are of most interest to us. We shall round of this section by calculating the network of surface features at two different scales. First we shall identify the network of ridges and channels at a fine scale:

We can improve the representation of the network by forcing LandSerf to connect some of these ridges and channels together. As this network would describe the topology of the surface, LandSerf will represent it using a vector rather than raster coverage:

Notice how both the blue channel network and the yellow ridge network appear similarly tree-like. The plateau to the south-west is dominated by local passes, pits and peaks (green, black and red dots). This is because at this scale, errors in the DEM tend to dominate flatter regions. We can broaden our scale of analysis, and therefore reduce the impact of DEM error, by repeating the process using a larger window size:

This ends the tutorial. This has been an introduction to LandSerf and the process of terrain analysis. To get the most of the software, you should experiment with some of the other functions available and apply them to your own data. See the user guide for more details.