Field Activity 2: Visualizing and Refining Your Terrain Survey

Introduction

This field activity acted as a continuation of field activity 1 in which my group and I made landscapes in planter boxes filled with snow, created a coordinate system and then proceeded to collect elevation data for the entire planter box. In this activity however, we utilized this data to produce 3D models of the surface features using various methods in ArcMap. A more in depth explanation of how the previous procedures were conducted can be found in my previous blog posting, Field Activity 1: Creation of a Digital Elevation Surface.

In this next part of the activity we entered our collected elevation data into Microsoft Excel, imported it into ArcMap and then used it to create a final 3D representation of our planter box landscape. In order to determine which interpolation method was the best way to represent our data we used a total of 5 different methods and compared the results to determine the best one. These methods include IDW, Natural Neighbors, Kriging, Spline and TIN. 

Methods

After all of the elevation data was collected for our planter box landscape we entered it into Microsoft Excel with headers "x", "y" and "z". All of our elevation data is negative because we used the coordinate system we made out of twine as the zero elevation marker. Since we did not have any of our features which went above the twine to keep things simple all of our "z" values are negative.  

(Fig. 1) This Excel spreadsheet shows all the x, y and z data for our group's planter box landscape. In order to utilize this data we imported the sheet into ArcMap as XY data and then later converted it into a shape file to be used in the interpolation methods. 

Once the excel document was in the proper format we added the XY data to ArcMap. After this was done we imported the data to our newly created file geodatabase. To add the XY data we needed to specify which field would represent the z values. Initially we did not realize that we needed to insert the negative into the data table which we imported and it caused our maps to be backwards. After the point shapefile was created out of the x, y, z data (Fig. 2) we could begin to use it to experiment with various interpolation methods.

(Fig. 2) The x, y and z data collected and imported into ArcMap and shows a bird's eye view of our planter box coordinate system points.

Once this data as properly inputted into ArcMap we could apply different interpolation methods in order to both spatially and visually analyze the landscape elevations from our planter box. In this exercise we used 5 different methods including: IDW, Kriging, Natural Neighbor, Spline, and TIN. All of these tools are found in the ArcToolbox under the category titled 3D Analyst Tools and Raster Interpolation. There are two different types of interpolation methods, geostatistical which  use stats to predict surfaces and deterministic methods which base their predictions on formulas and other mathematical data. 

IDW (Inverse Distance Weighted)

Inverse distance weighted interpolation uses linearly weighted combinations of many sample points in an area to determine cell values. This deterministic method gives more weight  to points closer to the processing cell center and points further away from this center are given less importance. 

(Fig. 3) This surface model was produced using the IDW interpolation method. The IDW method is not a good technique to used for this survey because of the number of peaks and valleys it produced in the surface model that are not present in the actual landscape.

Kriging

The kriging interpolation method is a very advanced geostatistical method which creates an estimated surface model based on a scattered set of various points which also have z-values. There is a complex set of statistics used to predict the surface values in this method which offers greater accuracy than that of deterministic interpolation models. This method is usually best for studies of geology or soils. 

(Fig. 4) The kriging was the most method to display the landscape of our planter box because it does not show the distinct elevation differences in features such as the valley, depression and ridge. 

Natural Neighbors

The algorithm which is used in the natural neighbor interpolation determines the closest set of sample points and applies weight to them based on the proportion  areas in order to interpolate an output value. Based on this methodology the output surface that the natural neighbor interpolation creates will not display any of the surface features we incorporated in our landscape that is not explicitly represented in our elevation data. The weights in this method are given based on the amount of area that overlaps rather than distance as the IDW method does.

(Fig. 5) The natural neighbor interpolation method was fairly successful in depicting the proper elevations of the landscape features. One problem with this method though is the peaks which are formed were there are distinct changes in elevation from one sampling point to another.

Spline

The spline method is also a deterministic method like the IDW and Natural Neighbor and uses a mathematical function in order to make the output surface. This interpolation creates a surface model which will pass through each of the elevation data points, causing a more minimized surface curvature in the entire study area. Based on this the spline interpolation produces a smooth  surface. There are two different types of spline, regularized and tension. The main difference between these types is how smooth the output surface is. (The regularized spline produces a much smoother surface compared to the tension spline.)

(Fig. 6) The spline interpolation method provided the most accurate depiction of the landscape surface features of our planter box. It produced a smooth surface between the lower and higher elevation points compared to the other methods which is a much more accurate depiction of the study area than any of the other interpolation methods.

TIN (Triangulated Irregular Network)

A TIN surface is created by placing vertices at various sample points and then connecting the vertices with edges which then form triangles. This results in a contiguous triangle network which also depicts the values between the sample point (vertices) but doesn't change the sample data positioning. 

(Fig. 7) The surface shown in this image was produced using the TIN interpolation method. This was not a good method to use for this particular study because of the network of triangles it creates. Since we used snow to construct these landscapes they were much smoother but TIN interpolation is not able to show any smooth surfaces based on the process of how the surfaces are created.

Discussion

Of the five interpolation methods that we used in this activity the kriging method (as seen in Fig. 4) was the least accurate and produced the worst surface model for our landscape. The kriging method does not properly depict increases or decreases in the elevation of the surface features. The peak was not much greater in elevation than the plane which is an obvious problem with the model. 

In contrast the IDW surface has more peaks in regions where there shouldn't be. This includes the main peak and the ridge region. These surfaces are much more bumpy than they should be. This is a problem because the surface features should be smooth. The natural neighbor surface also has problems with inaccurately portraying high elevations. The ridge in the natural neighbor is shown as sharp points which is not accurate either. 

The TIN surface is fairly accurate in representing the stark changes in the elevation data however it once again does not produce a smooth surface as it should. This is based on the TIN interpolation method which produces a network of triangles. 

The spline method fit the survey the best of all the methods though. The peaks of the mountains were smoothly curved as they should be and reached the proper height while maintaining their shape. As can be seen in Fig. 8, the ridges are properly depicted as continuous, as is the valley feature. The only problem with this surface model is the plane not being uniform. This could be due to the snow and how uneven it was as a medium for constructing the landscape features.

Figures 3- 7 were produced using the original elevation data that we collected in the previous lab however the ridge feature was not well represented. We believed this to be caused because our coordinate system had too large of a resolution to properly capture the shape of this particular feature. We decided to resample the region with a coordinate system of 6 inches by 10 inches. This data was then added to the existing elevation data and helped to produce the final surface model of the survey area (Fig. 8).

Resampling our data was much easier than our initial data collect because we were able to learn from our previous mistakes. We were sure to bring the right amount of twine and just had a better understanding of what we were going to do. It was a slightly warmer day which made the process better as well. We changed the resolution of the coordinate system in only the specific region where we were having issues with the 3D interpretation of the our elevation data, the ridge. Since our initial coordinate system had a larger resolution than our revised one it was not able to capture all aspects of the ridge elevation and therefore made our 3D maps look incorrect compared to our planter box landscape. However after implementing a smaller resolution coordinate system we were able to get a much more detailed and clear view of the ridge feature in our final surface model (Fig. 8).

(Fig. 8) This final surface interpolation using the spline method provided the most accurate depiction of our planter box landscape. The peak and ridge are raised but smooth and the valley and depression are smooth as they should be as well. Overall this model fits this particular survey very well.

Conclusion

After applying all five of these interpolation methods we found that the spline method was the most accurate in representing our study area. This is not too surprising since the spline method creates a surface which will pass through various surface points in order to produce a smooth surface image. We had to conduct a re-survey in order to increase the accuracy of the landscape features. 

It proved to be a very useful to use the top of the planter box as a representative of sea level because we did not have to deal with both positive and negative elevation values. This made the implementation our x, y, z data much easier. There were some inaccuracies due to the fact that we were using snow to create our surface features because it was extremely difficult to create a flat plane and measure it as such. Also, there was some snow fall between when we took our first set of elevation data and when we went to re-survey the planter box landscape. This could have caused inaccuracies in our data as well. 

While our group worked very well together and were able to meet up for the first portion of this field exercise it was difficult to all organize to meet for the re-surveying. Despite this though we were able to stay in contact through email and we learned how useful it is to start a project early in case there are any problems. Overall I really enjoyed this project since we were able to produce a 3D landscape model of what we created in snow in a planter box.