In the previous lab a landscape was created in a sandbox so that a group of students could take a survey of the sand landscape. The edges of the box were considered sea level, meaning areas above sea level were measured as positives and measurements below sea level were negative. A grid was created so that each measurement could be located in the sandbox by using the grid. A more detailed account of how this survey was conducted can be found in a previous blog post labelled Creating a Digital Survey Model.
The term data normalization refers to the cleaning of data. Making sure all of the excel cells are in the proper format, making sure that there are no spaces in the field titles, and making sure that predetermined symbols are not used, these are but a few of the challenges that are fast when normalizing data. The data normalization is important for this lab because all of the data was collected in the field and the way that the data was collected could have changed slightly during the surveying process. This data needs to be cleaned and put into the same formats across the data.
The data points will be mapped out by using ArcMaps and ArcVisual. The data is in a excel table which will then be mapped in ArcMaps by using the Map by XY data function. These points will be converted to a raster and various types of interpolation will be preformed to show the elevation model in different ways.
Methods
To turn a X,Y,Z data points into an elevation model the first step is to create a geodatabase where all of the data will be stored. Add the excel table with the surveyed data points to the geodatabase. Add the points using the Add XY data function. The data will then be converted to a point feature class. With the new point feature class the points will be converted to continuous data by the means of 5 different interpolation methods: IDW, Natural Neighbors, Kriging, Spline, and TIN. IDW or Inverse Distance Weighted technique of interpolation uses sample points as starting points and further from the sample or starting point you get the more inverse the elevation becomes. The next method of interpolation is Natural Neighbor, which uses central points and take the weighted value of all sample points around the central point. The third method that was used is Kriging, which uses z-values, or the number of standard deviations away from a mean that a point is, to determine the elevation surface. The fourth interpolation method is called Spline, which uses a mathematical function that reduces the surface curvature, making the map look smoother. The final method is called TIN or the Triangular Irregular Networks. TINs are created by triangulating points or technically called vertices, which are then connected by a edges creating a network of triangles. All of these interpolation functions have a 3D version in the 3D Analysis Extension, the 3D version were used to create the continuous data or raster.
After the vertices were interpolated into rasters they were exported into 3D Scene. The rasters were saved at as PNGs with 150 dpi. For each interpolation the orientation needed to be set to that interpolations orientation. This allowed for the data to be displayed properly in 3D Scene. As another step to help make the data look smother, in the setting under display, the display setting was changed to Bilinear Interpolation. This helps to smooth out the pixels in the PNGs that were added to 3D Scene. It does not change effects of the original interpolation that was used. It just smooths out the pixels in the PNG. The color scheme that was used was red to blue dark. This color scheme really helps to differentiate the high elevation points from the low elevation points. Also an aesthetic side note, the color scheme makes the lower elevation points look like water.
Results
Figure 1 looks at the IDW interpolation first. This method turned out the worst. The problem was not with interpolation itself but in the way that the data was collected. When the survey was conducted most of the points were collected in the center of each grid cell. However, when the data was recorded only one part of the XY points were recorded. So for an example, if a point was collected at Y3.1 and X3.2 it was not recorded that way. The mistake that was made was consistent across the grid in this case it was recorded as Y3.1 and X3. So all points collected between Y3 and Y4 all points lined up in a perfect line. When looking at this Figure 1 it is easy to see that a grid pattern has formed. This was caused because of the weighted distance method that the IDW uses. The further from the sample point the lower the elevation. If all points were collected on the grid lines and not in the middle of the grids, like they were suppose to be, this grid pattern appears.
(Figure 1: an IDW Interpolation of the Landscape)
The second method, Natural Neighbor, is Figure 2. Natural Neighbor was one of the more accurate representations of the actual landscape. The method of taking all the points from near by sample points gave it fairly accurate representation. This is probably the best method to counteract the mistake that was made in the data collection. It took all of the "middle" samples that were mistakenly put on the edges of the grid and weighted them giving the interpolation a real look.
(Figure 2: An Natural Neighbor Interpolation of the Landscape)
The third method is Kriging, which used the z-scores of the sample points to create a 3D rendering of the landscape. Figure 3 shows an image of this interpolation. This image was the first created and different color scheme was used, Red to Green Dark. This method ended up giving it a dried dirt look that was not very representative of the landscape. This is due more to the data collection issue that occurred. Using the z-scores should have provided an excellent representation. This method really struggled to form the three hill in the top part of the image. These hills are better represented in the Natural Neighbor and IDW Interpolations.
(Figure 3: The Kriging Interpolatin of the Landscape)
The fourth interpolation was the Spline; Figure 4 shows an image of the Spline Interpolation. The Spline did the best job of capturing the terrain of the top part of the image but misinterpreted parts of it. The ridge running from the middle of the map to the top of the map was suppose to be the west side of a river. None of the other interpolations were able to detect this feature in the landscape. However, this interpolation struggled with the ridges on the bottom of the image that were more clear in the Natural Neighbor Interpolation.
(Figure 4: The Spline Interpolation of the Landscape)
The last method of creating a 3D model was by TIN. The TIN image is located in Figure 5. The TIN looks very similar to the Natural Neighbor Interpolation. It took note of the ridges in the bottom of the image as well as the plateau in the top left section of the image. It did a decent job of detecting the valley between the mountain ridges on the bottom of the image. This method was, overall, as good as the Natural Neighbor.
(Figure 5: TIN Interpolation of the Landscape)
Conclusion
This survey relates to other field based method in that it gave an excellent insight into how to collect data in the field where an employer will expect that the survey gets done properly. It made the group think critically about what is the best way in which to gather the data and where should sea level be. This method differs from actual field work in that sea level will already be determined.
I do not believe that this level of detail will always be possible in the field due to areas being inaccessible due to natural features or the land being owned by another person or corporation. Also if the area that is being surveyed is very large there may not be enough time to go to all desired locations in the area that is being surveyed. To continue off of the last point if all areas do get surveyed it will be at the cost of areas with large elevation changes not getting as surveyed with the detail that it may require.
Interpolation can be used with any types of continuous data, precipitation and temperature are two classic examples. An interesting one that ArcHelp talks about is using the IDW to map out the likelihood of a return customer returning to an retail store. Due to the weighted distance method that is used to interpolate a set vertices it show areas that are further away from the central point. Which correlates to the further away you are from the store a customer is the more unlikely they are to use that store.
No comments:
Post a Comment