Terrain Part II: 3D Modeling and Resurvey

Report: Jonathan Luczak
Group 4 Members: J. Luczak; C. McLeish; E. Merkel; B. Miracle; N. Schaffer; S. Smith

Introduction:

   The purpose of this project was to create 3D models of the the terrain that was developed in the first part of this project using ArcScene 10.2 (ArcScene).  Following is a link to the posting for part I of this project:  http://geospat336.blogspot.com/. Using  x-, y-, and z- coordinates obtained from the survey in part I of this project, 3D models were created in ArcScene using the following interpolation methods: inverse distance weighting (IDW), kriging, natural neighbor, spline, and triangular irregular network (TIN). These models were then compared with one another before a decision was made as to which one most accurately represented the earth surface features (ESFs) in the garden box outside. Once the best model was chosen, it was examined for any inaccuracies. When inaccuracies were found, their locations were noted and then they were resurveyed in order to produce a more accurate 3D model in ArcScene, using the interpolation method that seemed to best interpret the data from the field.

Methods:

• Survey Data:

  Survey data from part I of this project were put into an excel spreadsheet as x-, y-, and -coordinates. This data was converted into a point feature class (PFC) using ArcMap 10.2 (ArcMap). The PFC of  the survey data is shown below in figure 1. North, relative to box-direction, is indicated by the red arrow in this and all subsequent figures.

Figure 1



Figure 1 shows the survey data collected in part I of this project as it appeared as a PFC in ArcMap.

• Creation of 3D Models:

   Computer generated models of the ESFs found in the garden box were created using various interpolation methods in ArcScene. Displayed in consistent oblique views in figures 2 through 6, the models produced were, respectively: IDW, natural neighbor, spline, TIN, and kriging. Each of these models were then compared with one another in order to distinguish the one that most accurately portrayed the ESFs in the garden box.

  There are three important details regarding the models shown in figures 2 through 6 that should be noted here:

• All images are being viewed obliquely (i.e. down and towards the east).
• All images are not being viewed at full-extent in order to show detail.
• All images were created using default parameters, except for the kriging image in figure 5 which was created using a linear semivariogram; this setting produced the most spatially accurate model  in this case. 

Figure 2

 Figure 2 shows the peaks and pits produced by the IDW interpolation method generated in ArcScene.

Figure 3

 Shown at roughly the same extent and angle as the IDW model in figure 2, the natural neighbor model in figure 3 displays prominent peaks where there are none.

Figure 4

 Figure 4 shows an oblique view of the 3D model generated via the spline interpolation method in ArcScene.


Figure 5

Figure 5 shows the ArcScene image created using the kriging interpolation method. Input parameters were set to linear semivariogram for this image.

Figure 6


This TIN was produced using the kriging model generated in figure 5 as a template.

• Resurvey:

   3D models of the terrain created in ArcScene revealed errors in the data that was collected. Shown in figure 7a, these inconsistencies are circled on plan-view of the kriging model shown in figure 2.

  x and y coordinates of the erroneous information were found by overlaying the original PFC on the plan-view model of the terrain (figure 7b).

  The resulting box locations were resurveyed with the hope of obtaining more accurate elevation data. In contrast to the survey in part I of this project, resurvey data were collected using a 2.5 cm²  grid in these specific areas; a demonstration of the resurvey of  the garden box is shown in figures 8a and 8b.

Figure 7

a)  

 

 b)

Figure 7a shows a plan-view of a kriging interpolation model using data from the survey conducted in part I of this project. Circled in red are elevation changes that  appear in the model, but not in the garden box. Figure 7b shows how the erroneous points were selected in ArcMap and correlated to the grid of the garden box by using information in the PFC attribute table.


Figure 8

a)
 b)

Figure 8a shows the how the group 4 garden box was resurveyed in order to correct for erroneous points. Figure 8b details how a 2.5 cm² grid was established over the areas to be resurveyed.

• Remodeling:

   Resurveyed data were then entered into Microsoft excel as x, y, and z coordinates from which a new PFC shapefile was created; this new shapefile is shown in figure 9. 

  The new PFC was used to create a 3D model in ArcScene using the kriging method with input parameters set to linear semivariogram  (figure 10), as this was the method that appeared to best fit the ESFs as they appeared in the garden box using the original survey data.

Figure 9

Figure 9 shows the PFC shapefile that was created using data obtained from the resurveying of the terrain. Data from the resurveyed areas appears more dense due to the fact that a 2.5 cm² grid was used to collect the new data while the original used a 5 cm² grid.

 

Figure 10 

 

Figure 10 shows an oblique view of the ArcScene model created using the kriging method. This model was created with the data obtained from the resurvey of the garden box and used the linear semivariogram model.
 

Discussion:

  Using the survey data from the first part of this project, five 3D models were created in ArcScene using the following interpolation methods: IDW, kriging, natural neighbor, spline, and TIN. These models were compared to one another in order to find the one that best represented the ESFs in the actual garden box.

  In order to create models in ArcScene, the original data from the survey had to be imported into ArcMap (or ArcScene) as a table. The table that was imported into ArcMap contained  the x. y, and z data generated from the original survey of the garden box in part I of this project. From this table, a PFC shapefile was created and is featured in figure 1. The shapefile was then used as the foundation on which all the 3D models mentioned above were produced; except for the TIN (figure 6) which was produced using the kriging model displayed in figure 5 model as its template.

  The first model generated in ArcScene was the IDW. Shown in an oblique view in figure 2, the IDW model displayed the overall terrain well, as all the models did in a general sense. For instance, the various hills, plateaus, ridges, valley , and depressions were discernible using the IDW method. However, the problem with using the IDW method to create a model of the ESFs in the garden box was that showed various jagged peaks jutting out of ridges. Also, low lying areas appeared to be pitted in the IDW model, as shown in detail in figure 2.

  In an atempt to correct the problems mentioned above, input parameters, such as "Power", "Output Cell Size", and "Number of Points," were changed numerous times in order to try to smooth out the peaks and pits of the IDW model. However, manipulating these settings did not result in any significant changes in any of the additional IDW models produced.

  The natural neighbor interpolation method was also used to produce a model of the terrain in ArcScene. Like the IDW model, the natural neighbor interpolation produced jagged peaks along hill tops and ridgelines (figure 3). These extraneous features made the natural neighbor model highly inaccurate in terms of its portrayal of the actual terrain. Furthermore, the natural neighbor model was unattractive because it had few creation parameters that could be manipulated compared to other four interpolation methods.

  Figure 4 shows detail of an ArcScene model produced when the regularized spline method of interpolation was used. Although the resulting image was smoother than both the natural neighbor and IDW interpolation method, peaks were still present where the terrain should have been much smoother.

  Numerous parameters were manipulated in an attempt to create a smoother spline model, with none of them producing the desired results. For instance, the weight parameter, which, according to the ArcGIS 9.3 Desktop Help (2012) determines how smooth the model will appear as its value is increased, was varied between 0.01 and its maximum value of 5 with no conclusive results.

   The interpolation method that best represented the terrain as it appeared in the garden box was the kriging method. Displayed in figure 5, the kriging method produced a smoother image when compared to the previous three methods of interpolation. Also, in contrast to the natural neighbor method of interpolation, kriging had numerous parameters that were easily manipulated. For instance, the kriging model shown in figure five was generated using a linear semivariogram. When compared with the other kriging parameters, e.g. spherical; circular; exponential; and Gaussian, the linear semivariogram method produced the model that most accurately represented the terrain in the garden box.

   The final model used to portray the survey data collected in part I of this project is the TIN model shown in figure 6. In order to produce this TIN in ArcMap, a raster template was needed. Since the most visually accurate interpolation method was created via kriging with a linear semivariogram (figure 5), this model was used as the template for the TIN. However, unlike the kriging model, the TIN showed a landscape that contained jagged ridges and was thus inferior to it.

   After the various 3D models were created, inaccuracies in the digital terrain models were noticed. For instance, the plan-view model in figure 7a shows that there is a break in the ridges where the elevation drops to the valley floor. These two elevation drops are represented by the two  eastern-most circles  and is contrary to what actually occurs in the garden box,where the ridges are smooth. Similarly, the depression in figure 7a, which is highlighted by the most westerly circle, should be much rounder than it appears in the model.

  It was decided that in order to correct the inaccuracies exhibited by the 3D models a resurveying of the box at these specific areas would be necessary. The first step in resurveying was to figure out exactly where the problem locations would be in terms of x, y, and z coordinates. Figure 7b shows how the PFC was overlaid onto the terrain model in ArcMap. Once this was done, data points corresponding to the inaccurate ESFs in the model could be selected on the PFC. The selected points could then be viewed in an attribute table of the PFC, also shown in figure 7b. Next, the FIDs could be determined and from them specific x and y coordinates could be related to the corresponding x and y coordinates on the data collection sheet. Finally, the coordinates on the data collection sheet could be related to the physical locations in the garden box; these areas were then be resurveyed.

   It was decided that a greater range of elevation points would need to be taken from the problem areas during the resurvey than in the initial survey. The main reasoning behind this decision was that the areas that displayed inaccuracies were those in which there was a rapid change in topology, i.e. across the ridgeline and along the bowl of the depression. It was hoped that more data would equate to more a more accurate portrayal of elevation variance in the problem areas during the resurvey. Thus a more accurate 3D model could be generated from the new readings compared to those models produced earlier on.

   In order to obtain more data points, 5 cm² grids from the previous survey were divided so that they became  2.5 cm² each. Figures 8a and 8b show some of the details of the resurvey, such as data collection (figure 8a) and the 2.5 cm² grid set-up  (figure 8b). Elevation data  in the resurvey were obtained in a similar manner to the survey data that were collected in part I of the lab. For example, each elevation value in the resurvey was taken from the southwest corner of each new 2.5 cm² grid that was generated. Also, all of the elevation points collected in the resurvey were rounded up to the nearest 0.5 cm as was done in the original survey of the box.
 
   Following the resurvey, the new data points were corrected to sea-level (i.e. 20 cm was added to each negative elevation value, as in part I) and entered as x, y, and z values into an excel spreadsheet. The new spreadsheet was used to create a new PFC (figure 9) and served as the basis from which the final model of the garden box landscape was created in ArcScene.

   Figure 10 shows the final model of the garden box that was generated in ArcScene with data from the resurvey. Since the kriging interpolation method with a linear semivariogram input parameter produced the most accurate representation of the garden box earlier, the same interpolation method was used to generate the final model. The new model in figure 10 is still not an accurate portrayal of the landscape, especially in terms of the ridgelines. However, the ridgelines were improved slightly during the resurvey. For instance, the ridges appear to be less bissected than earlier. The depiction of the depression in the new kriging model appears to be much more accurate than in ones prior to the resurvey.

  The kriging model in figure 10 was further manipulated in ArcScene by changing the symbology from classified to stretched. It seemed  the stretch produced an even smoother model of the terrain than the original linear semivariogram model that used classified symbology. This would be consistent with the symbology information provided by ArcGIS Desktop Help (2009), which stated that the stretched method works best for raster images that represent elevation changes.

Conclusions:

   While the resurvey did not drastically improve the digital representation of the terrain as was hoped, some improvements were achieved, such as the overall shape of the depression and by clsoing up some of the gaps on the ridgeline. It seemed that the linear semivariogram kriging model was the best fit for this particular landscape as it was compared in the computer with at least 15 other models, such as IDW, TIN natural neighbor, spline, and various other kriging models.

   Some improvements that could be made in the future to produce a better representation of the terrain would be to collect more data points on all areas where elevation changes occur. For instance, a 2.5 cm² grid could be set up in areas where there is rapid changes in topology during the first survey.

  Other improvements to the initial survey could be made that may produce more accurate models future. These adjustments include making sure that the thumbtacks are driven into the wooden edges of the box completely and holding the lines tight to ensure that there was no slack in them while measuring elevation. Having grid lines that cross the box at the same elevation might allow future students to take more accurate measurements, enabling them to round the nearest 0.1 cm rather than to the nearest 0.5 cm as our group did.

 The overall project was interesting because it forced one to be very organised. For instance, group 4 collected about 1104 data points. It took a lot of planning and patience to ensure that the data were collected properly in the field and entered correctly into Microsoft excel. With that many data points, it would be no exaggeration to say that the models generated in ArcScene could have looked much worse than they did.

   One challenge with having such large groups work on this assignment, i.e. five people per group, was that scheduling meetings was an issue at times. For instance, sometimes we could not meet all  on a particular day and/or at a particular time. For this reason it was essential that group members communicated all the processes and methods to those members who could not make it to a particular meeting time to ensure that they were caught up. That being said, everyone in the group definitely pulled their own weight by contributing equally to the project in the long run.