Ground Water Contamination at Bemidji, MN

The use of GIS 3D tools to tackle ground water contamination at Bemidji, MN Younis Altobi and Danny Bailey Introduction

As societies expand, groundwater resources are becoming more and more of an important asset. As a downside of expansion, groundwater resources are experiencing many negative anthropogenic affects. One of the most prominent anthropogenic affects is hydrocarbon contamination. Pipelines circulating hydrocarbons, such as oil, stretch across the world either above or below the surface. These pipelines leak at least once in their life span (http://www.epa.gov). More than 14,000 oil-spills are reported to the US Environmental Protection Agency each year (http://www.epa.gov). Oil-spills severely impact the environment and different habitat and resources. For example, vegetation, marine and human species are primary causalities of oil-spills along with natural and man-made resources, i.e. ground water. However, the magnitude of their impact depends on the oil characteristics (Hydrocarbon type; light vs. heavy) and the spreading rate at the spill site. The clean-up process following an oil-spill can be challenging in terms of time needed, cost in terms of training personals and equipment needed, and accurate understanding of hydrocarbon distribution. The severity of the oil- spills’ impact on resources increases with increasing clean-up time. If the hydrocarbon front percolate through the overlying sediment and reach the water table, then large scale contamination treatment and distribution modeling is required. However, successful modeling of hydrocarbons plume within the ground water is a complex and challenging process. This study attempts to develop new monitoring and modeling techniques of oil-spill distribution with time by utilizing ArcGIS tools.

Study site

The focus of this project will be a site investigation of a nationally famous oil- contamination site northwest of Bemidji, Minnesota. In 1979, a crude-oil pipeline burst, spilling approximately 10,700 barrels of hydrocarbon contaminants onto a glacial outwash deposit. Since, the crude-oil has contaminated both the subsurface sediment and groundwater resources in that region. Figure 1 is a location map of the site and a detailed aerial photograph of the site and the north oil pool.

Figure 1a A map showing location of the study site with respect to Bemidji and the state of Minnesota (http://mn.water.usgs.gov/bemidji/). Figure 1b A detailed aerial photograph of the study site showing the well locations and the different oil pools. The outlined box and the map to the right highlight the extent of the north oil pool and the surrounding wells. X is the location where the pipeline burst in 1979 (http://mn.water.usgs.gov/bemidji).

Objectives

As a novice to the use and application of ArcGIS, I was amazed by its capabilities and the usefulness of its applications. After choosing the study site, we wanted to investigate surface and groundwater interaction and their effect of the hydrocarbon plume migration. Using GIS techniques and utilizing Arc Hydro Groundwater Data Models devised by Gil Strassberg and Suzanne Pierce (CRWR), we proposed: • 2-Dimentional surface-ground water interaction data model using aquifer, stream networks, aquifer recharge and contaminant transport. • 3D geological framework of the aquifer using wells, coring and test data. • Model groundwater movement through out the aquifer. • Model hydrocarbon migration since the 1979 oil spill through out the aquifer. • Model the aquifer hydrologic unit to determine the effects of lithology and porosity on water and hydrocarbon movements. • Produce a 3D distribution profile of the hydrocarbon plume across the site.

However, as with any study, within the course of the investigation we had to change and modify our goals to fir with the data we had and timeframe. We also started to learn about the current limitations of ArcGIS and where it can be improved in order to apply it to ground water contamination monitoring and modeling (discussed later). Therefore, this study accomplished the following goals: • Produce a 3D geological framework of the aquifer using water wells and coring. • Model hydrocarbon migration since the 1979 oil spill through out the aquifer. • Produce a 3D distribution profile of the hydrocarbon plume across the site. • Produce 3D surface topography of the site. • Model ground water level changes through time. • Model oil level, thickness, and concentration through time. The goals were achieved using several ArcGIS programs. For example, all the 3D modeling and surface mapping was done using ArcScene. Data layout and projection were done using ArcMap, ArcToolbox and ArcCatalog. Microsoft Excel and Access were used to build the attribute tables needed for the applications above.

All the data were acquired from the USGS Bemidji website (http://mn.water.usgs.gov/bemidji/) (Fig.2). The site contained information regarding maps, current and past research conducted in the oil-spill site, list of contact information of researchers working in the site, data, and results from various projects. The data contained well information (more than 200 wells) regarding well number, (X, Y) locations, elevation, water level, oil level, oil thickness and oil concentrations. All the wells have well number and (X, Y) locations. However, the rest of the wells contained some of the remaining data with few exception that had all the above. Coring wells were also listed in the USGS site and those wells were primarily used to construct the geological framework. Regarding oil concentration, this study focused on modeling the Benzene and Toluene concentrations only.

Data Analysis After selecting wells from and around the north oil pool, surface elevation data from a set of wells scattered across the area were collected in order to interpolate a land surface for the study area. From the coring wells, 6 different lithologies were identified based on the core description provided with each well. Porosity and permeability values were then estimated from text-book values of similar lithologies. For each well, tables with top and bottom depth of each lithology along with lithoID were created. This is going to be used to build a 3D projection of wells into the subsurface. As the study is focused on modeling hydrocarbon distribution through time, we selected time frames for the study. The time frames have a 5 year time span as changes in oil level take place at very low rate. Wells were then organized into the 3 time frames and the 3 different categories, water level, oil level and oil thickness, and oil concentration (Benzene and Toluene).

For each table, time frame, wellID , Latitude, Longitude, category elevation form surface elevation at that well, or oil concentration subdivisions were created.

WellID LATITUDE LONGITUDE Elevation 306 47.57391 -95.09020 430.168 418 47.57386 -95.08899 432.201 509 47.57438 -95.08899 431.871 603A 47.57338 -95.09154 430.816 604A 47.57376 -95.09058 430.081 706A 47.57426 -95.08984 432.215 815B 47.57389 -95.09163 431.470 911 47.57433 -95.09103 431.677 9001 47.57358 -95.08914 432.105

Attribute table of the elevation wells and a sketch showing the location of the wells relative to the north oil pool (see fig.1b). WELLID LATITUDE LONGITUDE LITHOID SURFACE_EL TOP_Z BOTTOM_Z 302 47.57391 -95.09021 S1 430.268 419.905 418.076 LithoID302 47.57391 -95.09021Core DescriptionS2 430.268 Porosity417.466 Permeability415.028 (md) 523S1 47.57361 -95.09099Sand tightS2 430.198 423.7972% 422.5781 530S2 47.57422 -95.08903Silt to Fine SandS2 432.893 421.0068% 420.7011 533S3 47.57398 -95.08967Sand Fine to mediumS2 432.266 428.27615% 428.2464 604S4 47.57376 -95.09058Sand medimum toS2 course 430.081 422.76620% 420.0818 606S5 47.57353Sand -95.09142Course to very courseS2 with gravel430.569 422.41925% 421.61010 707S6 47.57360 -95.09101 Gravel S2 430.187 425.61730% 423.66710 812 47.57356 -95.09026 S2 431.645 423.475 422.195 815 47.57389 -95.09163 S2 431.470 425.310 424.700 302 47.57391 -95.09021 S3 430.268 430.268 419.905 302 47.57391 -95.09021 S3 430.268 418.076 417.466 303 47.57398 -95.09040 S3 430.356 430.356 422.401 411 47.57389 -95.09024 S3 429.841 429.841 424.050 421 47.57387 -95.09014 S3 429.970 423.965 423.874 423 47.57390 -95.08988 S3 431.866 431.866 423.941 522 47.57394 -95.08981 S3 432.283 428.321 427.283 523 47.57361 -95.09099 S3 430.198 425.321 423.797 533 47.57398 -95.08967 S3 432.266 428.246 426.536 533 47.57398 -95.08967 S3 432.266 425.346 424.616 533 47.57398 -95.08967 S3 432.266 424.126 424.006 533 47.57398 -95.08967 S3 432.266 423.916 423.886 606 47.57353 -95.09142 S3 430.569 407.829 406.759 707 47.57360 -95.09101 S3 430.187 423.327 422.907

Part of the attribute table that was used to construct the 3D projection of cores within the study site. Top and bottom depth of each Lithology is calculated with respect to surface elevation. Second table is core description and its interpreted LithoID, porosity and permeability.

Period 1 83-88 Period 2 89-94 Period 3 94-99 PERIOD_1 WELLID LATITUDE LONGITUDE ELEVATION TOP_OIL BOTTOM_OIL Period 1 306 47.57391 -95.09020 430.610 423.810 422.710 Period 1 315 47.57381 -95.09016 430.593 423.793 422.793 Period 1 319 47.57385 -95.09008 430.872 423.872 422.772 Period 1 411 47.57389 -95.09024 430.792 423.842 422.742

Oil level and thickness attribute table for period 1. Notice the number of wells and their wellID.

PERIOD_2 WELLID LATITUDE LONGITUDE ELEVATION WATER_ELEV Period 2 303 47.57398 -95.09040 430.640 423.640 Period 2 306 47.57391 -95.09020 430.610 422.610 Period 2 315 47.57381 -95.09016 430.593 422.693 Period 2 319 47.57385 -95.09008 430.872 422.472 Period 2 411 47.57389 -95.09024 430.792 422.892 Period 2 423 47.57390 -95.08988 432.759 422.609 Period 2 506 47.57438 -95.08934 432.791 423.391 Period 2 507 47.57426 -95.08956 433.050 423.300 Period 2 512 47.57402 -95.08909 432.833 423.233 Period 2 518 47.57405 -95.08943 432.968 423.368 Period 2 521 47.57404 -95.08998 432.142 423.192 Period 2 523 47.57361 -95.09099 430.707 423.607 Period 2 531 47.57410 -95.08931 433.238 423.338 Period 2 533 47.57398 -95.08967 432.875 423.425 Period 2 534 47.57392 -95.08983 432.764 422.714 Period 2 603 47.57338 -95.09154 430.936 423.736 Period 2 604 47.57376 -95.09058 430.291 423.071 Period 2 606 47.57353 -95.09142 431.169 423.619 Part of the wells used to construct the attribute table for water level during period 2. Period_3 WellID LATITUDE LONGITUDE Benzene Toluene Period_3 604b 47.57375 -95.09058 0.4833 0.0028 Period_3 421a 47.57387 -95.09014 14.3050 0.0014 Period_3 9015 47.57386 -95.09010 6.4858 0.0011 Period_3 533b 47.57397 -95.08970 3.9391 0.0007 Period_3 533c 47.57396 -95.08970 0.0552 0.0013 Period_3 532a 47.57401 -95.08956 5.4584 0.0008 Period_3 532b 47.57401 -95.08957 4.8155 0.0035 Period_3 532c 47.57400 -95.08958 1.8207 0.0003 Period_3 532d 47.57402 -95.08954 0.0005 0.0003 Period_3 518a 47.57405 -95.08943 4.1254 0.0007 Notice the number of wells and wellID.

Benzene and Toluene concentration attribute table for period 3. Notice the number of wells and wellID. The number of wells varies for each period depending on the data available. That’s why different wells had to be used in order to have a large enough dataset for each time frame.

Data interpretation 1. All the attribute tables were projected using geographic coordinate (NAD 1983) in ArcMap. 2. The personal geodatabases for each of the time frame (Bemidji 1-3) were then created in ArcCatalog. 3. Each personal geodatabases contained several feature data sets (oil thickness and thickness, water level, oil concentration, elevation and lithology). 4. Each feature dataset was then projected using ArcToolbox to projected coordinates, MN Central State Plane. 5. Within each dataset, the relevant feature class was added from ArcMap and projected using MN Central State Plane. 6. Each dataset was then added into ArcScene and the data were extruded from the elevation surface (layer properties/baseheight/elevation surface). 7. All surfaces were created as raster using Inverse Distance interpolation.

A screen capture of ArcCatalog showing the different personal geodatabases for each of the time frame (Bemidji 1-3) and the ArcToolbox projection wizard.

Results Surface elevation Figure shows elevation Surface interpolated from selected surface elevation points distributed across the study site (green points indicate the location of the elevation points within the site.

3D projection of core lithologies

A 3D Surface elevation of the site with different wells and projected core lithologies. The different colors of the projected cores represent different lithologies.

Water level

Figure shows 3D water level surface for the 3 different time frames. Overall trend shows that the water level rises in period 2 and then drops again during period 3. The extent of the changes and its relation to precipitation data can be further investigated. Combing these surfaces with the oil level and thickness (see below) will highlight these 1 changes. Oil level and Thickness

3 Figure shows oil level (top surface) and thickness in the 3 time frames. Oil level drops with time and oil thickness decreases away from the spill site. The dark color indicates the high elevation of the oil and it lies directly below the spill site. The changes in oil level with time need to be further investigated to address the effects if any of water fluctuation.

Oil concentration (Benzene and Toluene)

Figure shows oil concentrations (Benzene; pink, and Toluene, purple) through the different time frames. The high peak in benzene concentration lies right below the spill site. With time, Toluene concentration is deceasing at spill site and increase away (towards NE). The relation of this changes and water transport has not been investigated in this study. However, assuming NE ground movement direction from these diagrams can be that far off. The peak of Benzene can be explained by assuming that benzene compounds don’t migrate down surface as easily as other compounds and they are concentrated close to the surface.

3D cube of the site

Figure shows 3D cube of the site with land elevation surface, different wells across the site, projected cores, and oil level and thickness. The diagram is for period 2 which had better data in terms of the number of wells. Other cubes were also created for period 1 and 3. Looking at the cube, everyone will agree that plotting the subsurface geology across the site and relating lithological characteristics (porosity and permeability) will help in understanding the oil migration and distribution with time. Conclusions

ArcGIS tools such as ArcMap and ArcScene can play a very important role in monitoring and modeling oil-spill distribution through time. The study results show that even for a small site, if you have the data required, then ArcGIS will be a vital tool in understanding hydrocarbon plume migration and can help in site treatment and remediation. Linking surface waters to ground water can be done if you have the high resolution stream data required for small spill sites. Building geological models and combing them with porosity and permeability distributions across the site are not yet available in ArcGIS tools, but can be developed in the future. They will immensely help in providing the 3D view of the site with different surfaces and lithologies. Tracking analysis and forward modeling of plume migration will be the next step. The combined results will provide the EPA with distribution profiles for spill containment and treatment which will result in better resource management and huge time and money saving.

Limitations

During the course of conducting this study we had to overcome obstacles caused either by data or ArcGIS program limitations. The limitations that influenced this study are:

• Time • Sorting numerous attribute tables; different wells had different data and different monitoring periods. We had to sort these attribute tables with wells that had all the required data for that field and for the different time periods. Most times these data had to be collected from different tables from the USGS website. • Projection; changing the projection for each feature dataset from Geographic to projected using ArcToolbox and repeating the whole process for each feature set for the specified period of time. • Area too small for surface/subsurface water interaction. • Difficulty to link stream network to ground water as most of the surface stream network data is too large for the our study site (200x100 m). The size of our site did limit the extent of the study, but the resolution of the stream data is also too small. Developing high resolution stream and surface data will be extremely useful for future studies concerning small sites. • Geological framework. • Linking different lithologies across the area using ArcScene is still not possible. This is a program limitation as specific layers can be plotted and extruded to a common depth in order to have any type of subsurface configuration. ArcScene is not yet capable to plot real geological cross section with different lithological changes between wells. • Interpreting the spatial lithological distribution between wells is also an ArcScene limitation. This is normally done by plotting fence diagrams and liking varying lithologies together. • Tracking analysis in ArcScene is not yet possible. Because of this we had to do three separate timeframes.

Future Work

The study addressed the large scale problem and produced valid interpretations. However, a few questions need to be addressed in order to quantify the success of applying ArcGIS techniques in monitoring and modeling oil-spill contamination sites. These include; • Investigating the porosity and permeability effects on hydrocarbon distribution and migration. This will address the lithological variation impact on plume transport which is needed to model movement direction and extent with time. • Creating polygons for geological framework might help in viewing the subsurface geology of the site and combined with the porosity and permeability data will help in understanding 3D transport geometries. • Determining the cause of oil level variations (i.e. induced by water table fluctuations, or actual downward migration of oil level).

Acknowledgment

We would like to thank Gil Strassberg (CRWR) for his time and help with ArcScene and projection techniques. Gil also gave us some valuable feedback on our ideas and data analysis methods. Dr David Maidment for his enthusiasm towards this project and his ideas. Suzanne Pierce (UTDoGS) for useful discussion concerning ArcScene applications. Geoff Delin and Todd Anderson from the Bemidji research site for useful help with explaining some parts of the data and where to find more.