Assignment 5- Visualizing and Refining Terrain Data

Introduction

This assignment required the data from the previous sandbox assignment.  In the previous assignment, each group was to create a landscape in a sandbox.  The sandbox was 114 cm on each side and it was divided into 6 cm boxes to create a grid across the sandbox.  Below in Figure 1 is a picture showing the grid.

Figure 1: The grid overlaying the sandbox for this assignment.

Each square was measured, recorded, and then input into an excel file.  It was realized later that a measurement for each square was not needed.  This is because sampling wasn't taken into consideration.  Systematic sampling was used in a way.  This type of sampling is organized distribution of measurements, which means not having to measure each square.  This was sort of done in the previous assignment.  There were some squares where it was obvious they would have the same measurements as the previous squares.  Otherwise, each square was measured, which ended up creating very precise results.  There is nothing wrong with very precise results especially considering that it's a small area of study.  However, that degree of preciseness just wasn't always needed.

When the measurements for the previous lab were being input into an excel file, data normalization had to be taken into consideration.  Data normalization is important when inputting data.  It is the process of organizing columns and tables of database in order to improve data integrity.  If the measurements were input wrong into the excel file, that could make the results turn out completely different.  Luckily, the results turned out correct the first time and the measurements taken were accurate.

In this lab, the measurements were input into ArcMap and a point feature class was created.  From there, interpolation methods were used to create the digital elevation model (DEM) of the sandbox.  There are multiple different interpolation methods that produce different looking results.  Each were tried out until it was decided which method was most accurate to display the sandbox data.

Methods

1. IDW

• IDW (Inverse Distance Weighting) was the first interpolation method that was tried out.  This method uses weighted distance, which means the average needs to be between the highest and lowest numbers.  This method estimates values by averaging the values in the same area of each processing cell.  When a point is closer to the center of an estimated cell, it will have a greater influence on it's appearance.  This means IDW cannot create ridges or valleys if they have not been measured thoroughly.  This method relies on sampling that is distributed evenly in order to produce accurate results.  Below in Figure 2 the sandbox data is being displayed using IDW interpolation. This method looks a bit too precise.  Perhaps if the measurements followed the systematic sampling better rather than taking a measurement for each square, then the resulting image might look more smooth.

Figure 2: IDW interpolation from the side.

2. Kriging

• This interpolation method is more advanced and intense.  It generates an image from the set of z-values.  This method produces very precise results.  It produces even more precise results when more measurements are taken, like this scenario.  This method creates an output that flows together better than some other choppy looking methods.  Below in Figure 3 the kriging interpolation method can be seen.  This looks a like a very accurate and realistic representation of the sandbox landscape.

Figure 3: Kriging interpolation from the side.

3. Natural Neighbor

• In this interpolation method, it finds the closest subset of input samples to a point.  It then applies weights to them, which is based on areas to interpolate a value.  This technique will not produce any unrepresented data.  If data is missing it is averaged with the data it has to produce a model that is close enough to the real thing.  This can be seen below in Figure 4.  This method is similar to IDW, but without some many divots.  This method smooths out the data more, producing a better looking result.

Figure 4: Natural Neighbor interpolation from the side.

4. Spline

• This method estimates values using a mathematical function.  This function minimizes surface curvature, which produces a very smooth looking surface which still is accurately representing the data.  The great number of measurements means the smother the surface of the output.  This can be seen below in Figure 5.  It is of course smoothed out more than the sandbox, but the features seem to be displaying the most accurate representation compared to other interpolation methods.

Figure 5: Spline interpolation from the side.

5. TIN

• The final interpolation method is a triangular irregular network (TIN).  A TIN is a vector-based digital geographic data.  In this method, it is create by triangulating a set of vertices with vector data.  The vertices are connected through edges to form a network of triangles.  This is seen below in Figure 6.  This method is practically the same as the above methods, but it is just a different form of data (vector vs. raster).  This still produced a very accurate representation of the sandbox landscape.

Figure 6: TIN from the side.

Discussion

After working with all the interpolation methods, it was decided that the spline method (Figure 4) produced the most accurate representation of the landscape.  This method is smooth and the proportions to the actual landscape were accurate rather than too low or too high.  It didn't have weird bumps or unexplainable results.  A runner up was the TIN (Figure 5).  This also displayed a very accurate representation of the sandbox.  however, it just seemed too bumpy.  It wasn't smooth enough to show an accurate output.  If it were smoother it would be just as good, if not better, than spline.  Next was the kriging method (Figure 2).  This type of interpolation produced an interesting result.  It looked like a real life landscape rather than the sandbox.  This did produce an intriguing result, but it just wasn't accurate with the heights and depths.  They are in the correct areas, but the mountains aren't high enough and the valleys aren't low enough.  Natural neighbor (Figure 3)ranked next because it looks similar to the spline method, but the output had odd shapes all over the place.  This result wasn't accurate compared to the actual landscape.  Finally, IDW (Figure 1) was ranked last.  This is because the output turned out to be rather unsatisfying.  It does look like the landscape in some ways, but the measurements for this interpolation method were too accurate for it to display smoothly.

The data that was used for this assignment, which was collected two weeks ago, was very accurate.  This was nice because it produced fantastic results.  The interpolation methods were interesting to work with because they all look so different representing the same thing.  If different sampling methods were used, some interpolation methods may have worked better than the others.  But, overall the results were satisfactory.

Conclusion

This assignment produced great results and was very good experience.  There were some mistakes made, like not precisely following a sampling method.  This didn't necessarily mess up the results, but it would have been easier to take less measurements.  In assignment 1, the landscape was created, a grid was overlayed, each square on the grid was measured and then recorded.  These measurements were input into an excel file.  In assignment 2, the excel file was input into ArcMap which created a point feature class.  From there, interpolation methods were used to produce different looking DEMs.  Ultimately the spline interpolation method was chose as the best method because it displayed the most accurate representation of the sandbox landscape.  This was really good experience when it come to collecting geospatial data.  It helped to work as a team and talk out questions and concerns about the assignment.  A lot of knowledge was gained through these assignments.