This lab was a continuation of our previous lab that was based on the creation of a digital elevation surface. Previously, my partners and I collected elevation points of our landscapes using a grid system we designed. During this lab, we imported our excel data into ArcMap to evaluate its accuracy and to highlight areas that needed to be resampled. After running interpolation tools, our group developed a new sampling approach for our landscape. Methods and results of this resampling technique are included below in this blog.
Methods
Our first step of this week's lab was to import our original landscape data into excel. To do this, we had to verify our data was in the right format for ArcMap. We organized our data in X, Y, and Z points, where X and Y representation the grid square location, and Z represents its generalized elevation. Once the data was displayed as X and Y coordinates in excel, I exported the data. This allows the user to run tools on the newly exported feature class. To change the data from being represented as coordinates to being represented as a continuous surface, interpolation tools were run. After these images were generate, they were imported into Arcscene. This allowed us to identify areas that were not represented accurately in our model. Below in Figure 1 is our sandbox landscape. Following this are the image outputs from tools ran in ArcMap and their 3D outputs from ArcScene.
Figure 1: Our sandbox landscape |
Both IDW, Spline methods, and Natural neighbor all are considered to be deterministic interpolation methods because their methods calculate the smoothness of their output surface based on neighboring values or equations.
1) IDW (Inverse distance weighted) Interpolation:
This method of interpolation estimates cell values by obtaining the average of all data points within a close proximity of each cell. The closer cells to the point being measured have higher influence in the average being calculated, which is bases on an assumed relationships that data points located closer together are more alike than points farter apart. This interpolation model represents our hill, ridge, plain, and depression well however did not accurately model the ridge.
Figure 2: The left image is the IDW output from ArcMap, and the image on the right was created by bringing the left output into Arcscene.
2) Spline Interpolation:
The spline interpolation estimates values using a mathematical function to reduce the curvature of the surface. The end result of the interpolation is a smooth surface, in which passes through all data points. Once again, the interpolation model represents our hill, ridge, plain, and depression well however did not accurately model the ridge. The spline model is smoother than the IDW model, however the valley is not as defined as Natural Neighbor.
Figure 3: The left image is the Spline output from ArcMap, and the image on the right was created by bringing the left output into Arcscene.
3) Natural Neighbor Interpolation:
This interpolation method locates the closest data points to a point of interest, and applies weights to the closest points based on the proportion of the area to calculate an output value. This is known to not interfere with data variation such as peaks, ridges, and valley such as in our landscape, and is known to work well with data that is distributed either normally or irregularly. Our hill, ridge, plain, and depression are represented well however the valley is still not completely accurate once again.
Figure 4: The left image is the Natural Neighbor output from ArcMap, and the image on the right was created by bringing the left output into Arcscene.
Geostatistical Interpolation Methods:
The Kriging method is considered to be geostatistical interpolation methods because it uses statistical methods to find relationships between the data points. Using geostatistical methods, a user can create a prediction surface can also receive a measure of how accurate the predicted surface is.
4) Kriging Interpolation:
The model is used to explain the spatial variation in a surface by analyzing the distance between data points. The kriging model uses a math function to fits all points within a specific location which then develops an output value for each location. This interpolation model represents our hill, ridge, plain, and depression however did not represent the valley at all. Kriging failed to indicate any existance of this feature.
Figure 5: The left image is the Kriging output from ArcMap, and the image on the right was created by bringing the left output into Arcscene.
TIN (Triangular Irregular Networks)
TIN's are a type of vector based digital geographic data that are used to represent surfaces. These triangulate a set of data points to form triangles throughout the surface, which forms a continuous surface of none overlapping triangles. This methods is know to work to model terrains, however, within this output it is apparent our hill, ridge, plain, are represented well but now the valley.
Figure 6: The left image is the TIN output from ArcMap, and the image on the right was created by bringing the left output into Arcscene.
Resampling Methods:
Once we compared these output, we determined all our landscape features were accurately represented in our model except for the winding valley. After discussing, our group decided that resampling using a stratified sampling concentrated on the portion of the sandbox that contains a valley would give us more precise elevation data. We knew that reducing the size of the grid squares to increase the grid’s resolution within our new sampled section would increase the accuracy of the DEM as well.
To begin designing our new measuring grid, we decided the grid squares within Y coordinates 9 through 15 and X coordinates 1 through 14 would be changed from the original 8cm x 8cm dimension to 4cm x 4cm. The box was laid down and leveled in the same location as the previous lab. We used the same origin 2b as the previous lab, and from here we measured out 4cm x 4cm boxes. We created a grid using push pins and string only in the area previously specified. Now, we could begin data collection.
We measured each grid squares elevation using the same methods as the first lab. Due to the fact the box was initially leveled using a leveler at each corner the tight string represented a flat zero level. All elevations sampled were below the sting and therefore received a negative number. Each grid squares value was measured at the top right corner. This data was then entered into an excel spreadsheet. During this we collected a total of 392 new points.
Figure 7: Our resampling grid. New data points were only collectedin the portion of the sandbox where the valley is located. |
Figure 8: Casey gathering elevation samples. |
Figure 9: The left image is the Natural Neighbor output from ArcMap, and the image on the right was created by bringing the left output into Arcscene. The image are generated by using our resampled data, collected for lab 2. This model represents our hill, ridge, plain, depression, and the valley well. I chose this model amongst the other tool outputs because it smoothed areas that need smoothing out without leaving out minor changes in elevation, such as the small ridge running vertically along the left side of the right image.
Metadata:
Who |
Ally Hillstrom, Casey
Aumann, and Morgan Freeburg
|
What
|
Data Collection of
sandbox landscape to create a Digital Elevation Model
|
When
|
Lab 1 data collected
September 18th, 2015
Lab 2 data collected September 22nd, 2015 |
Where
|
UWEC Campus at the
flood plain of Chippewa River in Eau Claire, Wisconsin
|
How
|
Data collected using
a square grid to sample an area of 122cm x 114cm within a wooden box.
Elevation points were taken within each grid square with a measure stick.
|
Discussion
At the beginning of this lab we analyzing our data from the first lab. Five tools (IDW, Spline, Natural Neighbor, Kriging, and TIN) were run in ArcMap, and later imported into ArcScene to determine how accurately they depicted our landscape. All five tools had problems portraying the half of our sandbox that included the winding valley. Each tool can pros and cons of generating our models in comparison to each other.
With our initial data, the least accurate model was the Kriging output. It seemed to generalize the data too much and overly simplified all features. The hills elevation intensity was reduced, and the valley was not present. The TIN model was similar to the Kriging in that our landscape was heavily distorted. At this sample density the TIN inheritantly could not portray smooth surfaces on hills or the valley, and therefore is not a good choice for our DEM. It seems we would have to significantly increase the sample point density to reduce the intensity of the unnatural peaks created within by the TIN.
The IDW model represented most features, however the model created unrealistic peaks and bulges within each grid square. For example, the plain terrain is bumpy, as well as the surface of the hills, which is inaccurate. For this reason, it does not represent our landscape well. Additionally, the elevation of data points in this model are influenced by the elevation of nearby points, which may have lead the model to incorrectly assume elevation points if the neighboring value in reality were not similar.
The Spline method represented the landscape well. It is more detailed than the Kriging model and is smoother than the IDW model. The accuracy of the model is most likely based on the fact the model fits each data point to its output. On the other hand, the elevation changes throughout the output are not as defined as well as they are within the natural neighbor method. The Spline model presents the landscape in a fluid way, however it appears too fluffy.
After comparing all five outputs, our initial landscape's features were most similar to the features in the Natural Neighbor model. Although this model uses neighbor data points to estimate nearby data points, similar to IDW, this model also take the data points proportion of area into consideration. The amount of area the point covers dictate the weight of the point. This interpolation function seems to define our features the more. For example, the areas where the plain meets the valley are more sharply and accurately displayed in comparison to the smooth slopes and edges as the IDW and Spine model.
Attempting to increase the accuracy of our DEM, we resampled our landscape. We chose to reduce the size of the grid squares in the half of our sandbox that contained our valley, to increase the resolution and increase the density of sample elevation points within this area. After collecting our 392 resampled points, we imported the table in to Arcmaps and reran the four interpolation tools. Once these outputs were brought into Arcscene, I determined that Natural Neighbor tool once again best portrayed our landscape's features. This resampling method increased the accuracy of our valley tremendously however it is still not free of error.
Referring to Figure 9, within the valley there is a small stretch of area with a higher elevation than our real life landscape. To attempt to fix this problem, we could have retake sample points of this small area to make sure no data collection errors exist. Otherwise, we could resample the landscapre again by reducing the size of the grid squares dimension further to find even more accurate elevations. Overall, I feel the Natural Neighbor best represents our resampled landscape in comparison to the other tool outputs because it smoothed areas that need smoothing out without leaving out minor, however existing, changes in elevation such as the small ridge running vertically along the left side of the right image (figure 9).
Conclusion
While working on this lab for the past two weeks, I was exposed to new ways to collect geospatial data. My team and I developed a two unique coordinate system grids to collect elevation data of a sandbox landscape. We accomplished our goal of creating a digital elevation method of our landscape by importing our data into both ArcMaps and ArcScene. Throughout the lab I practiced running interpolation tools on our image outputs, while learning how they work and how to interpret them. This week, we were required to think critically to develop a resampling method that would increase the accuracy of our original DEM. This project required a lot of time, patience, team work, flexibility, and troubleshooting. Now that I have practiced gathering geospatial data, importing it into ArcMap, a running Arctools to generate DEM outputs, I feel confident that I would be able to repeat these tasks in future projects, either during another class project or on my own.
Sources
http://help.arcgis.com/en/arcgisdesktop/10.0/help/index.html#/How_Spline_works/009z00000078000000/
http://help.arcgis.com/en/arcgisdesktop/10.0/help/index.html#/How_Natural_Neighbor_works/009z00000077000000/
http://help.arcgis.com/en/arcgisdesktop/10.0/help/index.html#/How_Kriging_works/009z00000076000000/
http://help.arcgis.com/en/arcgisdesktop/10.0/help/index.html#/How_IDW_works/009z00000075000000/
http://help.arcgis.com/en/arcgisdesktop/10.0/help/index.html#//006000000001000000.htm