Raster Modeling

Goal and Objectives:

The goal of this lab was to determine possible future locations for frac sand mines in Trempealeau County, Wisconsin. This was done using a number of raster and vector data sets to create models of sand mining suitability and risks for the county. The results of the two models were then overlaid to find the best locations for mining with minimal environmental risks.

Data sets and sources:

The data sets used in this lab were acquired from the USGS and Trempealeau County's GIS office. In order to limit the size of the output data, I only performed analysis on southern 2/3 of the county (Study Area).

Study Area

Methods:

The first model I created was the sand mine suitability model. The data I used to determine sand mining suitability were: geologic units, land use/land cover, distance to rail terminals, slope, and water table depth. First, I converted every vector dataset to raster, as it would allow for easier suitability calculations. Next, I determined which criteria were most important for sand mine suitability, and I reclassified each raster to the values between 1-3, with 3 having high suitability and 1 having low suitability (Table 1).

Table 1: Suitability Criteria

The first suitability criteria I assessed was geologic unit. In Trempealeau County, frac sand is only found only in the Wonewoc Formation and formations of the Trempealeau Group. As a frac sand mine can only feasibly exist over these formations, I set these two values to high suitability, and all other formations were reclassified to "NoData" (Figure 1a). The reclassification to "NoData" will prevent any future suitability calculations from even being performed over the non-frac sand bearing formations (Figure 1b).

Figure 1a: Reclassifying by geologic unit.

Figure 1b: Reclassified geology,
showing only frac sand bearing formations.

The second suitability criteria I assessed was land cover type. I used the "reclassify" tool to make a new raster with values adjusted for the suitability of land cover types for the creation of a frac sand mine (Figure 2a,b). The most suitable land cover types were more open, and the least suitable were mainly water or developed lands (Table 1).

Figure 2a: Reclassifying by land cover type.

Figure 2b: Reclassified land cover.

In order to limit the cost of shipping and impact of the mine on the county's roadways, it is important for a future mine to be positioned close to a rail terminal. In order to assess the distances from rail terminals, I first created a buffer around Trempealeau County, and clipped a feature class of rail terminals, creating a new feature class containing only rail terminals within 15 miles of the processing area (Figure 3a). Next, I used Network Analysis tools to make a service area around the rail terminals, with driving distance rings of 5, 10, 15, 20, 25, and 30 miles. I copied the polygon output to my geodatabase, where I projected it from WGS84 into the Trempealeau County coordinate system. After projecting, I clipped the polygons to the processing extent before converting the polygons to a raster data set. I then reclassified the raster based on <10, 10-20, and >20 mile drive time areas (Figure 3b).

Figure 3a: Creating rail terminal drive times using Network Analysis.

Figure 3b: The resulting <10, 10-20,
 and >20 mile distance rings

A steeply sloped mine would provide additional dangers to the workers, so areas with gentle slopes will be prioritized. Before creating a slope raster, I resampled the DEM from 10m to 30m cell size, as all of the other analysis is being performed at 30m (Figure 4a). Next, used the slope tool generate a slope raster of the processing area. I then used the block statistics tool to create an averaged surface with 90m cell size. I then reclassified the output raster based on slopes: <8 degrees, between 8-16 degrees, and >16 degrees (Table 1, Figure 4b).

Figure 4a: Generating a normalized slope

Figure 4b: The resulting areas of <8 degree slope,
8-16 degree slope, and >16 degree slope.

Fran sand mines require large amounts of water in order to keep dust contained, so it is important that the water table is relatively high, to keep well-drilling costs down. In order to calculate water table depth, I first acquired water table contours in a .e00 file. I converted the .e00 file to an ArcInfo coverage file using the import from e00 script tool. After running the tool, I added imported the file into my geodatabase before using it as the input for the topo to raster tool. The topo to raster tool output a raster showing the water table elevation in feet, so I used raster calculator to convert the values from feet to meters (Figure 5a). Next, I used the minus tool to subtract the water table elevation from the DEM values to create a water table depth raster. I reclassified the water table depths based on an equal interval classification method (Table 1, Figure 5b).

Figure 5a: Calculating water table depth

Figure 5b: The reclassified water table depths

Next, I used raster calculator to multiply the values of all of the suitability rasters, generating a suitability index model (Figures 6a,b).

Figure 6a: The complete suitability index model.

Figure 6b: The output surface generated
by the suitability index model

The second model I created was the sand mine risk index model. The data I used to determine sand mining risks were: distance to streams, impact on prime farmland, distance to populated areas, distance to schools, risk of groundwater contamination, and aesthetic implications on parks and trails. Next, I converted every vector dataset to raster, and reclassified each raster to the values between 1-3, with 3 posing high environmental risks and 1 posing low environmental risks (Table 2).

Table 2: Risk criteria

Sand and other particulates from frat sand mines can quickly become washed away into nearby streams, negatively effecting their wildlife populations. If I calculated distance to streams without first limiting the streams to certain characteristics, the end result would be meaningless as the streams feature class has features ranging from perennial streams to seasonally inundated ditches. I first used a select tool to make a feature class of only perennial streams. Next, I used the polyline to raster tool to convert the stream features into a raster (Figure 7a). I then used the euclidean distance tool to generate a raster with values showing the distance from a perennial stream. I used the reclass tool to create three different risk categories: areas less than 0.5 miles from a stream are high risk, areas between 0.5 - 1 miles are medium risk, and areas further than 1 mile from a stream are low risk (Table 2, Figure 7b).

Figure 7a: Determining risk to streams.

Figure 7b: Reclassified stream impact

The next environmental risk factor I assessed was the impact on prime farmland. Mines shouldn't be created over high producing farmland, as the environmental costs would be steep. In order to assess the impact on prime farmland, I first converted the prime farmland polygon feature class into a raster (Figure 8a). Next, I reclassified the raster based on criteria described at length in table 2 (Figure 8b).

Figure 8a: Assessing the impact on prime farmland.

Figure 8b: Reclassified impact on prime farmland 

Fran sand mines should be situated away from populated areas, to limit any possible health issues from affecting the citizens. I used NLCD2011 to specify the locations of developed areas, using the extract by attributes tool, with "Developed, Medium Intensity" and "Developed, High Intensity" as my determination of developed areas. Next, I used euclidean distance to generate a raster with values showing the distance from a developed area (Figure 9a). I used the reclass tool to set the values for all areas within 640 meters of a populated area to NoData, as it is illegal for a mine to be within that zone (Table 2). I also created subsequent classes at 2 and 4 times that value (Table 2, Figure 9b).

Figure 9a: Assessing the risk to populated areas

Figure 9b: Reclassified distance to populated areas.
Areas within 640m appear the color of the background

Not all of the schools are located within high or moderately developed areas, so I used parcel data to locate schools within Trempealeau County. I used the select tool to extract only parcels with "Owner's last name" equaling a wildcard value of "SCHOOL". Next, I used polygon to raster to prep the data for the euclidean distance calculation (Figure 10a). After calculating euclidean distance, I reclassified the distances substantially more conservatively than for populated areas (Table 2, Figure 10b).

Figure 10a: Determining the impact to schools.

Figure 10b: Risk to schools.

Every mining operation poses certain risks to ground water, and I decided to factor this risk into my risk index model. First, I clipped the Wisconsin DNR served groundwater contamination susceptibility model (GCSM) polygon feature class to my processing area before projecting it into the county coordinate system (Figure 11a). Next, I converted the projected feature class to a raster dataset. I reclassified the GCSM values based on the Jenks classification method (Table 2, Figure 11b).

Figure 11a: Assessing groundwater contamination risk using the GCSM

Figure 11b: Risk of groundwater contamination.

Any future mine site would be best placed in a location where it wouldn't be visible from any of the existing parks or trails. In order to assess the aesthetic impact to recreational areas, I first used the feature to point tool to create centroid points for each park in Trempealeau County. Next, I clipped the points feature class to the processing boundary to prevent the model from processing necessary information. I used the add field tool to add fields named "observer A" and "observer B" (Figure 12a). I used the calculate field tool to assign "observer A" and "observer B" height values of 6ft and 75ft respectively. Next, I ran the viewshed tool, which generated an output raster showing areas visible from the parks (Figure 12b). I reclassified the viewshed raster, with areas visible from the parks reclassified to "NoData", and areas not visible from the parks classified to 1. To assess visibility from trails, I first used the select layer by location to select only trails whose centroid was within the processing boundary (Figure 12a). Next, I used the select tool to select only horse or bike trails. Next, I used the feature to point tool to convert the lines to points. I then added and calculated fields for observer a and b, just as for the parks viewshed. After the fields were calculated, I ran the observer points tool. The output of the observer points tool is a raster with an attribute table specifying which observers can see each point. I used the add field and calculate field to tabulate the number of observers that can see each area of the raster (Figure 12c). I used the reclassify tool to assign values based on the total number of observers, with areas viewable by 4+ observers classified as "NoData", areas not visible to the observers classified as 1, areas visible to 1 or two observers  classified as 2, and areas viewable by 3 observers classified to 3.

Figure 12a: Assessing visibility from parks and trails using viewshed and observation points.

Figure 12b: Total area viewable from the centers of the parks.

Figure 12c: Visibility of the trail points,
by total number of observers.

Next, I used raster calculator to multiply the values of all of the risk rasters, generating a risk index model (Figures 6a,b).

Figure 13a: The complete risk index model.

Figure 13b: The output surface generated
from the risk index model

After both index models were calculated, I used raster calculator to overlay the results of the two models (Figure 14a). After creating the combined index raster, I used the extract by attributes tool to select only the cells with values greater than or equal to 239. Next, I converted the resulting cells to polygons using the raster to polygon tool. I used the dissolve tool to merge all connected polygons into single areas. I used the add field and calculate field tools to create a field with the ideal location's areas converted from square feet into acres. The select tool was then run to make feature classes showing all of the potential mine locations greater than 1 acre in size, and all of the potential mine locations greater than 20 acres in size (Figure 14a).

Figure 14a: Creating the combined index raster, and locating areas of different sizes.

Results:

The resulting polygon feature class shows ten ideal locations for future frac sand mines. The locations vary in size from just barely larger than 20 acres, to 130 acres in size (Figure 15). None of these locations are visible from any of Trempealeau County's parks, nor are they visible from any of the trail observation points.

Figure 15: The 10 ideal locations for future frac sand mines.

Conclusions:

This project provided a good opportunity for me to apply the knowledge I've learned about raster tools in very practical ways. Although the end results are purely hypothetical, this project was fantastically applicable, especially as frac sand continues to be extracted from the hills and valleys of western Wisconsin. The designing these models was difficult, forcing me to synthesize all of the information learned throughout the semester in order to create a viable end result.

Network Analysis

Goal and Objectives:

The goal of this assignment was to learn how to perform network analysis, and use model builder to automate the process. In order to apply network analysis to the over-arching theme of frac sand mining, we assessed the cost incurred by several counties in Wisconsin as a result of increased truck traffic from overland shipping of frac sand between mine and rail terminals.

Although the calculations were performed using hypothetical numbers, the workflow is still the same as if the calculations were performed with proper values.

Methods:

In order to estimate the damage incurred per county, it was first necessary to determine the routes between the mines and rail terminals. This was done using the  closest facility analysis, a technique that uses a street network dataset to determine which facility is closest to a specific incident, and is also used for everything from emergency service dispatching to pizza delivery. to input the streets network dataset, and the Add Locations tool to load the locations of the mines and rail terminals. The mines were added as "incidents", and the rail terminals were added as "facilities". Next, the solve tool was run to calculate the route. The "Select Data" and "Copy Data" tools were used to add the calculated routes to a geodatabase. Before any further calculation could be performed, it was necessary to project the routes into a projected coordinate system, so their lengths wouldn't be calculated in decimal degrees.

Figure 1: The completed data flow model.

 In order to calculate the total road length per county, I used the "Intersect" tool to separate the routes into individual sections per county. The Summary Statistics tool was used to obtain a table featuring the total length of the routes per county. These lengths were output in meters, however, so some calculations were needed before cost-per-county could be determined. First, the Add Field tool was used to add a float column named "Distance_Miles". Next, I used the Calculate Field tool to write an equation converting the distances from meters to miles and adding the calculated values to "Distance_Miles". Once the distances were calculated, I added another float field "Annual_Cost" to hold the estimated cost value. I used the Calculate Field tool to write an equation so "Annual_Cost" = "Distance_Miles" * 50 (hypothetical number of trips per year) * 2 (to and from the terminal) * $0.022 (using a hypothetical cost per truck mile of 2.2 cents).

Results:

Figure 2: Haul distance per County.

The table showed that five of the counties had routes longer than 100 miles, two counties had routes totaling between 50 to 100 miles, with 10 counties having under 50 miles (Figure 2). Three of the counties have mines or rail terminals almost exactly on their borders, limiting their haul distances to 2.57, 1.28, and 0.96 miles, respectively (Figures 2,5).

Figure 3: Hypothetical Annual Cost per County

The annual cost per county is directly related to the haul distance per county (Figures 2,3.

Figure 4: Hypothetical Annual Cost per County.

Figure 5: Frac sand mine locations, rail terminal locations, and the routes between the two.

Discussion:

It was extremely important to use a float field for the calculations, as the default long integer cannot hold decimal values. Other students were having computational difficulties as a result of choosing the wrong field type.

The data showed very different results from what I was expecting, as counties' roads weren't only suffering damage from traffic from mines within the county, but also from mines in surrounding counties. The extra traffic on Chippewa County's roads from mines in Barron County potentially double the cost incurred by Chippewa County (Figure 5).

Conclusions:

This lab provided an opportunity for me to learn how to perform network analysis. Although the results are completely hypothetical, the workflow is accurate.

Sources:

Mine location information, County Boundaries, and was acquired from the Wisconsin Department of Natural Resources.
http://midamericafreight.org/wp-content/uploads/FracSandWhitePaperDRAFT.pdf

Geocoding Frac Sand Mines

Goal and Objectives:

The goal of this lab was to geocode a subset of frac sand mine locations in Wisconsin. This was to gain an understanding of data normalization and the accuracy of data collected by others.

Methods:

Before geocoding could occur, the addresses for the mines needed to be normalized (Figure 1). This involved adding additional columns to the excel spreadsheet, and copying different portions of the address field to them. This was done so the geocoder tool would be able to properly locate the mine's positions.

After all of the addresses were normalized, I added the excel sheet to the geocoding tool, and matched the fields to match the geocoding style. The geocoder then matched the addresses to their corresponding locations according to Esri's "World Geocoding Service". All of the addresses matched locations in the "World Geocoding Service", but weren't exactly accurate, so manual re-positioning of all of the points was necessary. In order to do this, the "address inspector" was used to "Unmatch" the point from its geocoded location, and the correct point was selected using the "Pick Address from Map" tool. Public Land Survey System (PLSS) information was used with the address information to identify the true locations of the mines.

Next, I compared my results to my colleagues. First, I used the "merge" tool to combine all of their shapefiles into one shapefile. I then used the "near" tool to determine the distances between each of my points and the closest point collected by my colleagues.

Results:

Figure  1: The table before normalization

Figure 2: The normalized table

The resulting table from the normalizing procedure was easier for the computer and persons to interpret (Figures 1,2) .

The initial geocoding results were extremely erroneous, with one of the initial results being located in Scotland, rather than its true location in Wisconsin (Figure 3).

Figure 3: One of the points was erroneously located in Scotland.

When comparing my results with those of my classmates, the differences between our resulting locations varied between 3 meters to 9 kilometers (Figure 4). 

Figure 4: A table showing the distances between my point locations
 and my classmates' point locations.

Discussion:

different types of errors
how can we know what points are correct?

Conclusion:

Post 2: Python

Introduction:

Python is a scripting language that allows easier automation of tasks in ArcGIS.

Scripts:

Script 1: 

For lab 5 I created a python script that projected, clipped, and loaded a list of rasters into a geodatabase. I used a for loop to iterate through the list of rasters. After a raster was loaded, the script would determine the Datum of the rasters's projection, in order to decide whether or not a transformation was needed for the projection and apply the proper transformation.

Script 1 part 1

Script 1 part 2

Script 2:

For lab 7, some data preparation was needed before network analysis could be performed, necessitating a script. The script was to select mines based on the following criteria:

• The mine must be active
• The mine must not also be a rail loading station
• The mine cannot be within 1.5 km of the rail

This was done by utilizing field delimiters and creating sql statements. The sql statements were used to methodically eliminate mines that didn't match the criteria above.
After the mines were added to the geodatabase, the script analyzed rail terminals to eliminate rail to air terminals.

Script 2 Part 1

Script 2 Part 2

Script 2 Part 3

Script 3:

For lab 8 I wrote a python script to generate a weighted index model based on frac sand mine risk factors. One of the factors was to be multiplied by 1.5, so it would be taken into greater account when the model was being processed. The weighted factor was added with all of the other factors, then saved to the geodatabase.

Script 3

Data Gathering

Introduction:

Before studying the impacts of frac sand mining on western Wisconsin, several data sets are needed. The goal of this lab was to build familiarity with data acquisition from a plethora of sources. It also served to prepare the data for future geoprocessing operations.

Methods:

Data were sourced from several organizations:

• Land Cover and Elevation data were acquired from the U.S. Geologic Survey via The National Map viewer.
• Rail Lines data were acquired from the U.S. Department of Transportation via the National Transportation Atlas Database 2015.
• Cropland data for Wisconsin were acquired from the U.S. Department of Agriculture via the Natural Resources Conservation Service Geospatial Data Gateway.
• Soil data were acquired from the U.S. Department of Agriculture via the Web Soil Survey App.
• All other data for Trempealeau County, Wisconsin were acquired from the county Land Records Office.

Figure 1: Maps of Trempealeau County
made with data sourced from several agencies.

Data Accuracy:

The data were judged on the following criteria:

• Scale - What scale the data were designed to be used.
• Effective Resolution - The ideal pixel size for a specific scale.
• Minimum Mapping Unit - The smallest depictable or plottable object at a certain scale.
• Planimetric Coordinate Accuracy - How close the points are to their real locations on Earth.
• Lineage - Documentation recording how the data were collected and processed, and by whom.
• Temporal Accuracy - How up to date the data are. When the data were published.
• Attribute Accuracy - How accurate the data classifications were when compared to the real world. Recorded as a specific number for metric attributes, and as a accuracy percentage for categorical attributes.

Figure 2: The Accuracy Assessment.

Conclusions:

This exercise will affect how I use certain data sets in future exercises in this class. I will preferentially use the USDA crop data over the USGS land cover data whenever I can, as the USDA validated the accuracy of their data. If I am to use the railroads data to perform any sort of analysis, I should download an updated version of the data, as it is updated incredibly frequently. I should be aware that the USGS land cover data is due for an update next year, and isn't necessarily going to show recent development.

Sand Mining in Western Wisconsin Overview

The western half of the State of Wisconsin sits on the ancient remains of a shallow sea. In the Cambrian Period, layers upon layers of sand collected on the seafloor and slowly, over time, compressed into sandstone. The constant movement of the sand across the seafloor smashed the grains of sand into one another, breaking off any sharp corners, leaving the grains smooth and round. As the sea slowly retreated east, the sand started to become covered with coral, which would form the limestone that composes the bedrock of eastern Wisconsin. If the sea hadn't retreated, western Wisconsin's sandstone formations would have also been covered by limestone layers, making their extraction cost-prohibitive.

Figure 1: The bedrock of  central Wisconsin is defined by sandstone formations. 

The expansive sandstone formations of Wisconsin have been a source of sand for glassblowing and other industries for over 100 years, but in recent years, demand has increased for the incredibly smooth grains of ancient sandstone. This demand has come from the oil and natural gas industries.

Figure 2: Sand with varying grain sizes and shapes.
(https://upload.wikimedia.org/wikipedia/commons/9/91/Sand_Grains.jpg)

Figure 3: Frac Sand
(http://wgnhs.uwex.edu/pubs/fs05/)

Oil and natural gas naturally occur between layers of rock, deep below the ground. When extracted, their removal can cause overlying layers of rock to collapse, stopping the flow of natural gas or oil from the well. Hydraulic Fracturing (fracking), is a process that circumvents this problem by causing it in a controlled way. When a well is drilled deep into a oil or gas deposit, small cracks are created in the rock. Next, the engineers inject water, sand, and some binding chemicals into the cracks, to hold them open. The sand prevents the layers from collapsing, and allows substantially more oil and gas to be extracted from the wells. Sand taken from the ancient sandstone formations has proven to be the best for this process, because its rounded grains are all of the same size.

Figure 4: An explanation of the fracking process.
(https://sites.google.com/a/cornell.edu/the-controversial-sustainable-energy-extraction-method-hydrofracking/home/definition-what-is-hydrofracking)

Figure 5: This map shows sandstone formations with
minimal overburden. 

In order to get the sand from the ground into an active fracking operation, several steps must be done. First, the topsoil or "overburden" must be removed from above the desired formation (Figure 5). Next, if the sandstone formation is loosely cemented, the sand is excavated by front ended loaders or excavators and sent to be processed. If the formation is tightly cemented, blasting can be done to loosen the formation. The chunks of sandstone formation broken off by the blasting process are crushed into smaller pieces, for easier processing. Processing consists of washing and repeatedly sorting the sand, to ensure that all of the grains are the same size. After the sand has been processed, it is transported to the fracking location via rail. When a mine extracts all available frac sand, they transition into a reclamation process involving the stabilization of walls, and refilling the mine site with the overburden.

Figure 6: The increase of frac sand mine operations between 2012 and 2014.
(http://wgnhs.uwex.edu/pubs/fs05/ , http://wcwrpc.org/frac-sand-factsheet.pdf)

Frac sand mining is rapidly expanding throughout western Wisconsin, and this isn't without problems (Figure 6). Frac sand mining produces air pollution from both the dust generated from the extraction process and greenhouse gas emissions from the machinery used to process and transport the sand. The mining process can also have impacts on water as a result of processing sand, mining sand from below the water table, or possibly from sand-laden water flowing into wetlands. There are also concerns from runoff impacting the fisheries by increasing the turbidity of the water, and by possibly suffocating eggs under layers of sediment. There have also been issues with excess light and noise pollution caused by mine operations. The transportation of sand along roadways has also caused increased deterioration as a result of the increased traffic.

Figure 7: Frac sand mining has caused increased rail traffic.
( http://wisconsinwatch.org/2015/04/12-sandy-gifs-an-animated-guide-to-wisconsins-frac-sand-rush/)

Throughout the course of this class, Geographic Information Systems will be used to study the impacts of frac sand mining on Western Wisconsin. We will be focusing on how the mines have increased the amount of traffic on the roadways of western Wisconsin and the resulting costs to the counties who maintain them. We will also design suitability models to determine ideal locations for future frac sand mines to be located in order to maximize the output, and minimize environmental impacts.

Sources:
http://wgnhs.uwex.edu/wisconsin-geology/frac-sand-mining/
http://dnr.wi.gov/topic/Mines/documents/SilicaSandMiningFinal.pdf
http://wisconsinwatch.org/2015/04/12-sandy-gifs-an-animated-guide-to-wisconsins-frac-sand-rush/
http://www.mpm.edu/content/collections/learn/reef/geol-wisc.html