Data Preparation
The primary data sources used for this analysis were the 25m DEM for the District of North Vancouver sourced to the GVRD data on the Simon Fraser University local data warehouse. This data set was originally produced by the District of North Vancouver. The other data set used was the LANDSAT image taken in July 2000 and sourced to the Satellite Imagery folder of S.F.U's data warehouse. All other secondary data sources used for this analysis were produced from transformations and reclassifications of these two primary data sources.
The first step in data manipulation was to import the LANDSAT images into IDRISI. The LANSAT data consists of seven separate images. Each image corresponds to one of the seven spectral bands that the LANDSAT can read. The purpose of the LANDSAT data in the overall analysis was to determine the terrain cover in the study area, specifically to distinguish between forested, open scree and rock, water and open grass and smaller vegetation (Gruber & Haefner 1995). Based on LANDSAT imaging literature the most useful spectral bands for distinguishing these terrain coverage features are 2, 3 and 4; corresponding to the visible green, visible red and near-infrared images (Eastman 1999). Each image was imported into IDRISI as separate raster layers. The second step in data manipulation was to synchronize the two data sets. The resolution of the DEM was 25m, while the resolution of the LANDSAT image was 30m. Another difference between the two data sets was the projected reference system; with the DEM being projected in UTM-10n and the LANDSAT image projected in us83tm10. The DEM elevation rater layer and one of the raster layers for were converted into ASCii and imported into ArcMap. Each item was converted to raster in ArcMap and the projection and resolution of the DEM was changed to match the LANDSAT image. The reason the LANDSAT was chosen as the template was because it had the coarsest resolution and could not be reduced further to match the DEM. The number of rows and columns of the DEM was also changed to match the LANDSAT so the two images would have identical specifications enabling future map overlay analysis. The areas on the LANDSAT image in which the DEM had no data would be masked out during further analysis. Both files were converted back into ASCii format and exported back to IDRISI where there were converted into rater layers.
The next step in the data preparation was to create a False colour composite with the three spectral bands in order to differentiate between the different types of terrain cover. The Images were combined using an addition overlay after weights were assigned to each band in order to achieve maximum differentiation. Band 3 (red) and Band 4 (near infrared) were given weights of 0.4 each, while Band 2 (green) was given a weight of 0.2. The reason Bands 3 & 4 were given extra weight is that 3 is the most crucial for vegetation and 4 is the most crucial for identifying water. Band 2 is simply used to compliment the other two bands aiding visible differentiation. The image below is the result of the addition of the three bands.
The image below is the classification of each terrain coverage type derived from the image above. The reclassification of the image below was problematic for two reasons. Some of the values for areas that correspond to water actually scored higher than the values obtained for dense forest cover. The reason for this discrepancy is the shadows created by the topography in the image. The result is some darker sections being classed as water, when they are in fact forest and vice versa. As a result this image was used to only distinguish between, forested, open areas with small vegetation and bare rock and scree.
The solution to this problem was to create a constraint layer that corresponded to all major water areas as well as all urban areas within the study region. This was completed by creating a shapefile in Arcmap in which the areas were digitized from the Landsat 4 image. Once the shapefile was completed, projected and rasterized into a synchronized format, it was imported into IDRISI as a Boolean raster image. This layer was overlayed with the Coverage Composite Classed image to produce a the Full Terrain Coverage Classification Map (see below). The added bonus of the Boolean layer for water and urban areas is that it will serve as a constraint during the Multi-Criteria Analysis.
0- Water & Urban Areas
1-Forrested Areas
2- Open Areas with Small Vegetation
3- Bare Rock & Scree
The rest of the data preparation focused on the DEM and the creation of Aspect, Terrain Shape, Slope and Elevation factors. The Aspect layer was created using the Aspect module of Macro modeler. The input layer was the DEM Project Clip file and the resulting coverage was based on the aspect from 0 to 360 degrees. The slope coverage was creating using a similar approach. DEM Project clip was inputted into the surface module and degrees was selected as the output units. The elevation coverage was the DEM Project Clip itself with no module necessary. The final coverage that needed to be created was the Terrain Shape factor. The DEM Project Clip was first put through a Low pass filter. This smoothed out the data well enough to use the TOPOSHAPE module. The Filtered DEM was inputted into TOPOSHAPE and the following map was produced with 12 Terrain Shape Classifications.
This completed the five input factors; Terrain Shape, Terrain Coverage, Aspect, Elevation and Slope. The data however needed to be reclassified into meaningful information and each factor needed to be standardized leading into the spatial analysis section of the project.