Field Activity #2: Visualizing and Refining Terrain Survey

Introduction:

Emphasis
Field Activity #2 emphasizes assessing the viability of various surface terrain models, determining accuracy of sampling methods, and resurveying. Digital terrain models (DTMs) are used to turn data points into a visual displaying the basic terrain profile per the measurements obtained. Per the instructions below these various models will be displayed, compared, and contrasted in general. In addition, their applicability for the box plot survey from Exercise 1 will be examined.

Instructions

1. Import the XYZ table developed in Field Activity #1 to ArcMAP and make it into a point feature class.
2. Turn the point feature class into a continuous surface.
3. Develop terrain profiles using the following interpolation methods:
• IDW
• Natural Neighbors
• Kriging
• Spline
• TIN
4. Discuss advantages, disadvantages, and observations regarding the 'fit' of each interpolation method.
5. Determine which method best fits the survey

Methods:

Building Initial Terrain Profile
The XYZ table developed in Field Activity #1 was imported as a layer into ArcMAP in order to create a point feature class. Next, this point feature class needed to be displayed to ensure that the points were consistent with what had been collected in the field during our box plot measurement. By using the "Display X,Y data" tool, the points that we had measured could be visualized (as seen in Figure 1). This grid represents all the data points collected. As you can see, the data entered appears to have been displayed consistently with even rows representing the measurement intervals employed in the field.

Because the box plot was merely a model of terrain, no specific projection needed to be applied to this data set.

Figure 1: This figure is the X,Y data points collected from the box plot. These measurements were taken at intervals, thus the points are in even rows throughout. 

Exporting File as Geodatabase
Based on the experience of my fellow students, I initially tried to export my file as a "shapefile". After consulting with several students and my professor, I was still not able to develop the models properly. It was advised that I export my file as a geodatabase (gdb). After doing this, I was finally able to develop the five digital terrain models (DTMs) as specified in this exercise.

Development of Digital Elevation Models in ArcScene

The interpolation process that turns the data points in Figure 1 into the following models is essentially "filling in the gaps" between the points in order to display a continuous surface. Without this process, all that would be displayed would be the elevation measurements at each point. This would in no way resemble the actual surface from which measurements were taken. By smoothing out the areas between measurements, a visualization can be developed to give someone a more accurate representation of an actual surface.

Difficulties
I have initially run into complications trying to to develop the terrain profiles using interpolation methods. Because I am just now taking a GIS class, I am very uncertain of the steps involved. After learning about how to import the XYZ table, turn it into a point feature class, and develop terrain profiles, I tried to carry out the process unsuccessfully. Over the course of two days, I sought  the advice of my peers who walked me through the process step-by-step. Even after seeking help, much to the consternation of those who assisted me, I was still not able to build these terrain profiles.

Assessment of Generated Profiles
After viewing the basic elevation profile based on our XYZ, we were able to locate areas where our measurements may be off. Looking at the profile together, we determined the areas where we would potentially want to resurvey as seen in Figure 1.

(Figure will be inserted with the completion of my profiles)

Re-examination of Physical Terrain
Next, we brought our profile out to our garden planter box to compare with the actual physical terrain.We also brought with us our initially recorded data. By comparing the two, we solidified are choices for resurvey as a group. Then, we plotted our areas to be resurveyed on our initial data set.

Determination of Re-Survey Methods
While still outside, we discussed how we would improve upon our initial survey. The decision was made to divide the identified grids for resurvey into four 2.5 x 2.5 cm grids. In order to avoid confusion, we also decided to use a different color string for the additional grid divisions.

Additional Grid Construction
Following the methods employed in Field Activity #1, we built the additional grid lines with thumbtacks and pink string as seen in Figure 2. In order to ensure accuracy, one member of our group guided the placement of each new grid line based on the plotting we had added to our initial survey.

Figure 2: Original grid squares were divided into four 2.5 x 2.5 cm. grids for re-survey

Re-Survey Process
During our initial survey, the 5 x 5 cm. grids were fairly easy to keep track of and measure. This became much more difficult with the 2.5 x 2.5 cm. grids.

• First, we realized that an original meter stick would be too wide to fit into the smaller grids and had to go get a thinner one.
• Second, it was not as simple as identifying the southwest corner of each grid. In order to provide accurate information to the person recording the data we had to identify a pattern for measuring the four grids that had previously been just one.
• Third, because of the smaller area, it became necessary to place a meter stick along the top of the grid to provide support in order to limit the amount of movement as we measured.

(*these additional measures are shown in figure 3.)



Figure 3: A meter stick was placed along the top of the grid to stabilize for measurement and a thin meter stick was used to take measurements in the smaller grids

After carrying out this res-survey process, the steps to enter data into ArcMAP in order to generate the DEMs was gone through again. Now, I will develop these models to asses there accuracy at displaying our actual box plot surface.

Triangulated Irregular Network (TIN)

For a TIN, a set of points is combined with elevation measurements to create "triangles" with all of the data points collected being located in the corner of the triangle.Each triangle is completely separate from every other triangle, thus there is no overlap. Because of this process a TIN is best for visualizing points in an irregular pattern.

The nature of the features developed in our box plot had far more regularity than regularity, so the TIN is not really the best option for displaying our data points. That being said, to understand the general layout of our box plot, a TIN does a reasonable job (as seen in Figure 4). It is clear to see where the areas of height and depression are located, etc.

Figure 4: This TIN representation of our box plot makes it easy to interpret the general layout of our box plot.

However, when look at from a more "straight-on" view (as seen in Figure 5), many of the peaks are displayed much more sharply than the actual features they represent. This is the result of the triangulation process for interpreting the data. throughout the image, you can see the various triangles for each point making it seem like there is much more variation throughout the surface than there actually is.

Figure 5: This more straight on view of a TIN surface projection makes the triangle-shaped points for each data point much more evident. This is not an accurate representation of our data which is far less sharp and much more even.

Inverse Distance Weighted (IDW)

This interpolation method determines cell values for each data point using a "linearly weighted combination of a set of sample points". With this method, the assumption is that the influence of the variable being mapped decreases with distance. This results in each data point "clumping" in one spot rather than evening out the area between data points (as seen in Figure 6). The surface is basically pock-marked with the data points especially where there is a difference in elevation. In areas where there is no variability in the height measurement the data appears very even. This is probably abnormally even.

Figure 6: This figure shows represents our box plot using the IDW interpolation method. This method assumed influence of a variable decreases with distance. Thus data point height measurements are clumped around the data point rather than smoothed from point to point.

Once again, the more straight-on view of this interpolation method shows abnormally differentiated peaks (as seen in Figure 7). There are not the harsh edges of the TIN method, but the data points still stand out unreasonably compared to the actual surface.

Figure 7: While the peaks are not as sharp as seen using the TIN method, they are still overly pronounced when compared to the actual surface due to the method of decreasing influence away from the data point employed in the IDW interpolation.

Kriging

The Kriging interpolation method generates an estimated surface from a scattered set of points with z-values. It is based on statistical models that employ autocorrelation. In other words, they take into consideration the surrounding measured values and mathematically determine the smoothness of the resulting surface. This smoothing process results in a much more even surface (as seen in Figure 8). Gone are most of the bumps and sharpness seen in both the IDW and TIN interpolations. This makes it a little more dificult to differentiate the surface overall, but this is actually more accurate to the actual surface.

Figure 8: Using the Kriging method results in a much more smooth representation of the actual surface it represents. 

However, when viewed straight-on, the height differentiation throughout the surface is actually very clear (as seen in Figure 9). There are noticeable peaks on the back edge that do not correspond with actual data from our box plot. This is a point of concern. This views is an especially good representation of our actual data.

Figure 9: The straight-on view of the Kriging Method is very smooth, and height differences are very clear as well. This is a very good representation of the data from our box plot.

Natural Neighbor

The Natural Neighbor interpolation method finds the closest subset of input samples to a query point and applies weights to them based on proportionate areas to interpolate a value. Basically it estimates the slope between points and, based on that, determines the height of the points. This results in a very smooth representation of the data (as seen in Figure 10). However, compare to the Kriging method some areas tend to get amorphous and lose their definition altogether such as in the foreground (in green).

Figure 10: This representation of the data using the Natural Neighbor method offers a very smooth display, thus the term "natural" in the name.

The straight-on view of the Natural Neighbors method is also very smooth with the unexplained spike in the background (as seen in Figure 11).  The elevation points are a little bit sharper than seen using the Kriging method. A determination must be made as to which view is a more accurate representation of the data

Figure 11: This straight-on view of the Natural Neigbors interpolation offers a very smooth surface only slightly more sharp than the Kriging method.

Spline

The Spline interpolation method estimates values using a mathematical function meant to minimize overall surface curvature, resulting in a smooth surface that passes exactly through the input points. The resulting surface will minimize any sharp features in the data. It is best used for gently varying surfaces such as elevation, water table heights, and pollution concentrations. As seen in Figure 12, the surface appears almost flattened out. Whereas the IDW and TIN methods overstated peaks, this method has underestimated them. 

Figure 12: Using the Spline Method, sharp features such as peaks are minimized resulting in a flattened out representation of our box plot.

Looking at the Spline interpolation straight on, actually shows a fairly nice representation of our data. Even the unexplained sharp peak in the background is minimized here (as seen in Figure 10). It may flatten out some of the peaked areas, but seen from this view, it gives a really good sense of our box plot.

Figure 12:  The straight-on view of the Spline interpolation method shows minimized peaks, but is an overall fair representation of our actual box plot.

Discussion:

Due to the over-pronounced peaks seen in both the TIN and IDW models, I do not believe they are the best option for displaying the box plot data. The nature of this surface just wasn't irregular enough to make these the most useful methods.

When viewed straight-on, I feel that Kriging, Natural Neighbors, and Spline all do a very nice job representing the box plot. However, when comparing them to each other Kriging still shows peaks being less-sharp than does the Natural Neighbors method.

In addition, the Spline method really does a nice job softening the peaks of our surface. At first, I thought it perhaps rounded out our peaks too much but now actually think this is more accurate than any representation showing sharp features. We just didn't have any very sharp features on our surface. This method is stated to be good for displaying elevation as well. Due to these aspects, I recommend the Spline method for displaying the data from our box plot.

It is quite possible we should have gone over our data more thoroughly and eliminated the unreasonable points such as the peak in the background in the straight-on view using any of the interpolations.

Conclusion:

Assessment of Methods
Throughout Field Activities 1 and 2, our group hoped to build on the experiences of the Field Methods class from 2013. Thus, we immediately employed 5 x 5 cm. intervals, built our elevation using the top of the garden planter box, and discussed our plans thoroughly before implementation.

In order to try improve upon our initial data results, we chose to measure smaller grids to try to better capture the change in elevation in areas we found to be misrepresented in our initial terrain profile.

As we went along, we also became more adept at asking questions regarding our process. Why, how, where, and when may seem trivial, but they were critical questions to address for every step of the process. That way, when the time came to record our experience, it would not just be a guessing game.

Areas for Improvement
Were I to undertake this experiment again, I would do the following:

• designate an official note-taker to fastidiously record not only the details of what we did, , but what we were thinking and why, conditions that were influential, and questions to consider later
• do a more thorough job looking at previous experiments - after the fact I learned that a group last year had used a spray bottle to freeze their terrain surface. As you can see in figure 13, our terrain is checkered with meter stick markings that, no doubt, led to some misrepresentation of the actual elevation of our features.

Figure 13: Our terrain is covered with "incisions" from where the meter stick contacted the terrain surface

• determine whether or not using a range finder would be an option - spraying the surface to freeze it would not be necessary if we could measure elevation without physically touching the terrain
• double and triple check the data set for any possible aberrations leading to obviously wrong data points

Group Assessment
As someone who insists on being thorough, I could not have asked for a better group to work with than John, Emily, and Carolyn and Brendan. Every time we worked together, everyone pitched in unreservedly. Spending the time in planning, preparation, and asking the right questions with one another made it so much easier to record the progression of this experiment. Being someone who has almost no GIS experience, one group member in particular (John) went far out of his way to try and guide me through the process. I would choose to work with this group again if given the opportunity.

Personal Learning Outcomes
I must refer back to my conclusion from Field Activity #1 to explain my learning outcomes which are as follows:

• Assess and plan properly
• Be fluid and creative
• Reflect and learn
• Move Forward

All of the little details of performing a GIS operation, or laying out a measurement grid are merely memorization functions. What separates a person is taking the time to prepare and analyze, think and ask questions, bounce ideas off of one another, reflect on your experiences in order to learn, and make plans for how to adjust in the future.

To be good in the field requires being observant, patient, resourceful, and careful. The integrity of the data you collect will be at the mercy of these elements far more than the technical machinery and methods you employ to obtain them.