Screen Shot 2017-06-28 at 6.31.25 PM.png

Thesis Statement

My goal for this final project has been to create a useful tool for Virginia’s policymakers as they seek to maintain a commitment to reducing greenhouse gas emissions in line with the Paris Agreement, despite President Trump’s intention to formally withdraw the U.S. from that global pact. I have created a raster-based analysis highlighting rural areas of Virginia that could be converted from farmland to natural vegetation reserves, to serve as carbon sinks to mitigate Virginian agriculture’s impact on the climate.

Globally, agriculture, including livestock and crop production, accounts for approximately 21% of anthropogenic greenhouse gas emissions (Tubiello, et al. 2014). These emissions include carbon dioxide released by deforestation and other land use change, including conversions to cropland, as well as methane and nitrous oxide from livestock and fertilizer (Lal 2004). Though demand for food is expected to sharply increase as world population grows at least until 2050, crop production on land with low yields represent a waste of resources by unjustifiably releasing CO2 gases without producing proportion caloric and nutritional value. These areas of low productivity could serve the public good more effectively by being converted back to natural vegetation to serve as carbon sinks; meanwhile, yields should be intensified on farmland with high productivity. Actively-managed forests can sequester more carbon than farmland, so meeting food demand on less acreage will reduce land competition and increase overall carbon sequestration (Lamb et al. 2016).

Since President Trump’s June 1st announcement that the United States would renege on its commitment to join the world in combating the greatest existential challenge of our era, thirteen states, including Virginia, have formed a Climate Alliance to adhere, albeit informally, to the goals of the Paris Climate Agreement (Merchant 2017). These states will need actionable geospatial data analysis to target agricultural policies toward efficient and climate-sensitive food production. This project aims to aid Virginia’s policymakers in identifying agricultural lands that could be put to better use as carbon reserves, thereby allowing targeted agricultural subsidies or other policy to move Virginian agriculture one step closer to truly global sustainability.



Project Scope and Data Acquisition

I began my project by determining the best technique to achieve my goal. A site suitability analysis based on raster data, created by reclassifying raster datasets according to significant criteria, then querying each raster to create a combined raster that meets all of the criteria, seemed the most straightforward and effective approach. Most of the agricultural and climate data that I found came in raster form, and I knew that a complicated policy issue like the best way to tackle global warming would require multiple data points, even for a relatively simple initial attempt like this project.

I determined the three criteria for the site suitability analysis based on the journal article that accompanied one of the datasets:

Natural ecosystems are likely to be maintained when crop production needs are met elsewhere and on lands where (i) crop yields are marginal, (ii) the value of carbon outweighs the value of crops, and/or (iii) natural ecosystems provide multiple high-valued services such as water purification, recreation, or biodiversity conservation. (West et al. 2010, “Trading Carbon,” 19646-7)

While I did not find comprehensive data on ecosystem services, I did incorporate the other three criteria presented. In order to be a potential carbon sink in this analysis, land must:

  1.  Be in a rural area outside of an urban buffer zone – urban lands cannot realistically be converted to natural vegetation; additionally, cities require land to grow food for their population.
  2. Produce poor soybean yields – croplands with low yields are not “pulling their weight” in food production relative to the negative impact that annual agriculture has on the public welfare. Soybeans, Virginia’s top field crop in 2016 (USDA/NASS 2016), serves as a proxy for all field crops in this analysis. Though soy is widely grown throughout the state, most Virginia soy yields fall below the national average for the years covered by the data (1997-2003).
  3. Have high carbon storage potential - certain types of land (lowlands, higher precipitation) are better suited to support large amounts of natural vegetation; these areas would be more effective carbon sinks than high-altitude drylands.

I acquired the relevant datasets for analysis from the U.S. Census Bureau and EarthStat. For the first criterion, I downloaded “Urban Areas,” a polygon feature class, from EarthStat provided the rasters “Global Harvested Area and Yield for 175 Crops” (Monfreda et al. 2008) and “Carbon Storage in Potential Vegetation” (West et al. 2010) for criteria 2 and 3. Additionally, I utilized features from Esri Press, Natural Earth, and the U.S. Census Bureau for my basemap. Each of these datasets required processing before they would fit the frame of my project.


Data Transformation

I began my basemap by adding USStates to my map file. USStates is rendered in the geographic coordinate system “GCS_WGS_1984,” which became the on-the-fly coordinate system for my Data Frame. I used “Select by Attribute” to create a new feature class for only the Virginia polygon. I then added USCounties to the map and used “Intersect” with Virginia to get feature class for VirginiaCounties, then removed USCounties. I also used “Clip” to limit the USUrbanAreas feature to only VirginiaUrbanAreas.

After getting to know the raster files by finding their descriptive statistics (range, mean, standard deviation) and adjusting their symbology (switching to “Classified” based on standard deviations), I attempted to limit the extent of the rasters using the “Extract by Mask” tool. Initially, I set the mask using a larger feature class containing polygons for all of the states in the South Atlantic region – from Maryland down to Florida. After setting the analysis mask (Geoprocessing à Environments à Raster Analysis à Mask) to SouthAtlanticStates, I attempted to extract my carbon storage raster layer, CarbonStoragePV. However, the process was unsuccessful; the program was still churning unsuccessfully almost an hour later. As I learned later, ArcGIS had difficulty handling a mask with multiple polygons. Looking for other options, I switched to the “Clip (Data Management)” too instead, which worked well enough.

Later, when I attempted the “Extract by Mask” procedure with my single-polygon Virginia as the mask, I was initially still unsuccessful. I received an error notice, number 010067, “Error in executing grid expression. Failed to execute (ExtractByMask).” I wondered if my Virginia feature layer might have residual issues from being created from a multi-polygon feature class. I made a new Virginia feature class, again, by selecting the single polygon in the layer, and set that as the mask – still no luck. After Googling for some answers, I decided to try resetting the scratch workspace (Geoprocessing à Environments à Workspace) to a new file geodatabase that I had created in ArcCatalog. The next try worked, and I did not encounter the error again.

In order to determine how much acreage I would need for my urban buffer zones, I needed to know the populations of Virginia’s urban areas. Because Virginia has few major urban centers, and just a few more minor ones, I decided to add the population data manually to the Attribute table instead of finding or creating an Excel table, ensuring it was properly formatted with a shared value for each record, and joining it to the feature class. Instead, I added a Pop2017 field, started editing from the Editor toolbar, and manually typed in the population for the 10 most populous urban areas.


Data Analysis

 To create my site suitability analysis, I needed to first create three raster datasets, one for each criterion, coded for suitability as potential carbon sinks. Creating the raster for rural areas, or areas outside the urban buffer zones, took the most steps and the most trial-and-error.

PROJECT 2 Urban Buffers.jpg

For this analysis, I decided to use the proxy that one person needs one acre of farmland to support them (Bradford 2012). Because each urban area in Virginia has a different population, the buffer zones around each polygon need to have different acreages. Because the polygons have irregular shapes, I had to use trial-and-error to determine what buffer radii would produce appropriately large buffer zones for each city. To measure area, I had to change my Data Frame from a geographic coordinate system to a map projection, which I did temporarily. I changed the Data Frame to the NAD 1983 Virginia projection, made my measurements and created my buffers, then changed it back to the original geographic system (which matched all of my raster layers).

The first buffer I created was for the Hampton Roads polygon. I measured the feature and found that it was 345,752 acres. Then, I selected the polygon in the Attribute table and used “Buffer” with a 3-mile buffer. The resulting buffer had 843,150 acres, or a net of only 497,398 acres. With a population of 1.4 million, Hampton Roads would need a larger buffer. A couple more tries led me to settle on a 10-mile buffer for Hampton Roads. I continued in this manner, trying different radii, until I created a 2-mile buffer around the D.C. area and a 1.5-mile buffer around Richmond. I defaulted to a 1 mile buffer for the rest of the urban areas, which based on my experiments would provide more than enough acreage.

I used “Merge” to combine all of the resulting buffers into one feature class, then used “Dissolve” (with the SUM statistic to keep Population accurate) to unite overlapping buffers into a single polygon. To get a raster from this vector-based feature class, I used “Polygon to Raster (Conversion).” Now I had a raster with only one value: 1 = Urban Buffer Zone. Because I needed 1 to equal a potential candidate for a carbon sink, I used “Reclassify” such that 1 = 0 (not suitable), and NoData = 1 (suitable). Unfortunately, the extent of the resulting raster cut off about half of the Eastern Shore, but I was able to rectify that later.


My SoybeanYield and CarbonPotentialPV reclassified rasters took much less time and fewer steps to create. The difficult part of each of these was determining what the “cutoff” value would be to differentiate between suitable and not suitable sites.

 For soybeans, I needed to determine a minimum yield for which it would be “worth” it to continue farming the land rather than convert to a carbon sink. The soybean yield data that I used was the average of census data from 1997 to 2003. So, found the average soybean yield for the entire U.S. during that time, which was 33.17 bu/acre (USDA/NASS 2017), or 2.23 tonnes/ha. I decided that, if a Virginia farm was yielding below the national average, it would be a suitable candidate for a carbon sink, so I set that as my threshold. I used “Reclassify” to create a new SoybeanYieldThreshold raster with 0 = >2.23 (not suitable) and 1 = <2.23 (good candidate).

West et al. (2010) suggest that, for a site to be better used for carbon storage than for farming, the value of the carbon must be greater than that of the crop that is given up. Therefore, for my CarbonStoragePV threshold, I determined that the value of the carbon stored should be greater than the gross value of three years of soy production. The average price per tonne of CO2 over the past year in the European Emissions Trading Scheme has been $5.60 (Investing 2017), which equals $20.53 per tonne of carbon. Virginia’s CarbonPotentialPV range is 98-134 tC/ha, valued at $2012-$2751/ha. With the current price of soybeans at $9.11 per bushel (MacroTrends 2017), a farm with an average yield of 2.23 tonnes/ha can earn $747/ha gross per year. Therefore, with a three-year standard, a farmer would need the CarbonStoragePV value to be at least $2240/ha to match their loss from soybeans. That would equal 109tC/ha. So, I used “Reclassify” such that 0 = <109 (not suitable) and 1 = >109 (good candidate).


To build the final site suitability raster, I used the “Boolean and” tool. First, I queried my CarbonStoragePVThreshold raster against my SoybeanThreshold raster to get an intermediary Boolean_CarbonSoy raster. Then I queried that against my VAUrbanReclass raster. The resulting raster, CarbonSinkSuitable, shows areas that meet all three suitability criteria.

In order to improve the usability of this tool, I decided to increase the granularity of districts highlighted for carbon sink potential. In the raster above, we see that many counties intersect the suitability raster but do so incompletely. I decided to incorporate census tracts to provide more actionable detail for targeted policy development. I used “Raster to Polygon (Conversion)” (with the “Simplify polygons” option checked) to turn my CarbonSinkSuitable raster into a polygon feature class. I then used the Editor toolbar to correct the area covering the Eastern Shore, which had been cut off earlier by the extent of my UrbanBuffer raster. I edited the vertices of one of the polygons to cover the full extent of the shore. Then, I selected the polygons that corresponded with the “1” value of the original raster (“suitable”) and created new feature classes from the selection, which excluded the polygons that had been created from the “0” values. Finally, I used “Select by location” to identify the census tracts that intersected with the CarbonSinkSuitableVector feature class, then created a new feature class from those selections. I symbolized these census tracts to be color coded by county, so that their administrative boundaries could still be seen.

I also created one more version with the “CarbonSinkSuitableVector” class set as transparent and overlaid onto the extracted census tracts, so that you can see the areas of imperfect overlap.

PROJECT 2 Overlaid CarbonSink CensusTracts.jpg


After measuring the CarbonSinkSuitableVector features and comparing the acreage to the whole of the state of Virginia, I found that 39% of Virginia’s land has potential as a carbon sink, according to this initial analysis. These areas are clustered in the Coastal and Piedmont (foothill) regions in the center of the state. The Blue Ridge mountain region is generally unsuitable due to low carbon storage potential, while a large chunk of the southeastern coastal area is excluded due to urban density.

The resulting suitability raster is limited by the granularity of the carbon storage data; the raster’s large cell size leads to ragged edges that I tried to clean up when converting to a polygon. The final analysis would also be improved by including further criteria, which would also eliminate much more land from consideration. These other criteria could include land zoned for residence or small towns, farmland with high yields in other crops besides soy, or lakes, streams and ponds.

The tool could also be made more useful by including tables with relevant data like average yields and carbon storage potential by county and the corresponding dollar values, to help the viewer assess economic trade-offs.


Lessons Learned

Main Failure

I attempted to re-project a raster from a geographic projection with spheroid datum to a map projection with a plane-based datum, but I was unsuccessful. The raster gave an account of the amount of carbon already stored within the soils of the world to a depth of 1 meter (“Global Carbon Storage in Soils”, from WRI, Matthews et al. 2000). It would have been very applicable to this analysis and could have been helpful in identifying soil with low carbon levels to maximize the “CO2 sponge” effect, i.e. to withdraw as much carbon dioxide from the atmosphere as possible.

My main problem, as I understand it, is that there is no built-in geographic transformation from a spheroid datum to a plane-based datum. I attempted to use the “Create Custom Geographic Transformation” tool, setting up a (GCS_Sphere_Arc_Info to GCS_WGS_1984) transformation and calling it “SpheretoPlane.” I then used it when I tried “Project Raster (Data Management).” I was unsuccessful, even after multiple consultations with the Esri Geonet forum.

Challenges Overcome

I struggled throughout the project with different datasets that had different coordinate systems and datums. Though the case I mention above was the worst, I still had to finagle with the other projections to get different tools to work. Even though I set the projection in each of my Data Frames, and I tried “Project Raster” for the rasters, I continued to encounter bugs. Luckily, in all other cases I found workarounds. However, the experience shows me the importance of understanding different ways to project geographical data and how crucial it is for me to become familiar with manipulating projections.

Final Thought

The biggest challenge I had with this project was in defining its scope. I realized that my major block was in understanding the ArcGIS tool itself. I could not get a handle on what kind of project I should aim to produce until I fully understood the tools and techniques at my disposal. Once I sat down and reviewed through the GIS tutorial book and refamiliarized myself with the broad range of manipulations and analyses that GIS can perform, I was able to envision the type of project it could allow me to do. They say that when you have a hammer, every problem looks like a nail. However, in this case I found that I had no idea what to do with my nail until I figured out what the hammer did.




Esri Press derived from Tele Atlas. 2016. US Counties. GIS Tutorial 1: Basic Workbook for ArcGIS 10.3.x, data included with workbook. Redlands, CA: Esri Press. Accessed May 11, 2017. Retrieved from

The feature class, created by Esri for ArcGIS using data derived from Tele Atlas, provides polygons for each county in the 50 states of the U.S.

Esri Press derived from Tele Atlas. 2016. US States. GIS Tutorial 1: Basic Workbook for ArcGIS 10.3.x, data included with workbook. Redlands, CA: Esri Press. Accessed May 11, 2017. Retrieved from

The feature class, created by Esri for ArcGIS using data derived from Tele Atlas, provides polygon outlines for each state in the U.S.

Matthews, Emily, Richard Payne, Mark Rohweder, and Siobhan Murray. 2000. Pilot Analysis of Global Ecosystems: Forest Ecosystems, Global Carbon Storage in Soils. Washington, D.C.: World Resources Institute. October. Retrieved from  

A raster dataset from WRI’s Pilot Analysis of Global Ecosystems (PAGE) project, this data includes values for metric tons of carbon per hectare in the world’s soils. The range extends from 0 to 1,250 metric tons/ha, though the metadata points out that only 2% of the total area contains more than 1,000 metric tons of carbon. Greenland and Antarctica are excluded from the dataset. The data was compiled from a number of sources, including the FAO, published and unpublished articles, the World Inventory of Soil Emission potentials (WISE). The data is based on a 1-meter soil depth, when applicable, because this depth is “believed to be the most directly involved in interactions with the atmosphere, and most sensitive to land use and environmental changes.”

Monfreda, C., N. Ramankutty, and J.A. Foley. 2008. Global Harvested Area and Yield for 175 Crops. Dataset from Farming the Planet: 2. Geographic Distribution of Crop Areas, Yields, Physiological Types, and Net Primary Production in the Year 2000. Global Biogeochem. Cycles, 22, GB1022. doi:10.1029/2007GB002947. Retrieved from

The authors of this dataset expended an extraordinary amount of effort to obtain, in the most detail possible, the yield and distribution data for 175 different crops in almost every country in the world. Each crop is available in a different raster file, making the overall zip file downloaded quite large, and necessitating clarity on which crops will be analyzed. Data is presented at 5-minute resolution, but the authors recommend that the data not be interpreted at that level – instead, the data should be used at the county level or larger. They also warn that, since agricultural data collection tends to focus on major crops, information for minor crops is less reliable. The main source of the underlying data was Agro-MAPS, a “collection of subnational statistics on crop area production and yield for most countries in the world.” Supplemental data came from national agricultural censuses and surveys, the FAO, as well as independent surveys conducted by the authors in certain countries.

Natural Earth. 2017. Ocean. Accessed 27 June 2017. Natural Earth. Retrieved from

Natural Earth provides open source geospatial for building basemaps.

U.S. Census Bureau. 2016. 2016 TIGER/Line® Shapefiles: Census Tracts: Virginia. U.S. Department of Commerce. Accessed 23 June 2017. Retrieved from cgi-bin/geo/shapefiles/index.php?year=2016&layergroup=Census+Tracts

Polygon features for the 2016 census tracts in Virginia.

U.S. Census Bureau. 2017. Cartographic Boundary Shapefiles – Urban Area. U.S. Department of Commerce. 7 April. Accessed 23 June 2017. Retrieved from

Polygon features for every urban area in the United States.

West, P.C., H.K. Gibbs, C. Monfreda, J. Wagner, C.C. Barford, S.R. Carpenter, and J.A. Foley. 2010. Carbon Storage in Potential Vegetation. Dataset from Trading Carbon for Food: Global Comparison of Carbon Stocks vs. Crop Yields on Agricultural Land. Proceedings of the National Academy of Sciences (PNAS) 107(46), 19645-19648. Distributed by St. Paul, MN: EarthStat. Accessed 10 June 2017. Retrieved from data-download/.

 The dataset compares the amount of carbon held within the plant material of agricultural crops versus the potential carbon storage capacity of natural vegetation were it to be allowed to grow on the same area. The authors used methodology described by the Intergovernmental Panel on Climate Change to ensure high standards of accuracy. Crop carbon was measured conservatively (i.e. the authors erred on the side of more carbon) by assuming that the crop’s total yield represented its standing annual carbon stock all year (whereas crops are generally harvested and removed for most or part of the year). The carbon storage potential of natural vegetation was calculated using a “committed carbon flux approach.”


Aspatial and Theoretical Data:

Bradford, Jason. 2012. One Acre Feeds a Person. Farmland LP. 13 January.

Investing. 2017. Carbon Emissions Futures Historical Data. Accessed 26 June 2017.­emissions­historical­data

Lal, R. 2004. Soil Carbon Sequestration to Mitigate Climate Change. Geoderma 123 (1–2): 1–22. doi:10.1016/j.geoderma.2004.01.032

Lamb, Anthony, Rhys Green, Ian Bateman, Mark Broadmeadow, Toby Bruce, Jennifer Burney, Pete Carey, et al. 2016. The Potential for Land Sparing to Offset Greenhouse Gas Emissions from Agriculture. Nature Climate Change, January. doi:10.1038/nclimate2910

MacroTrends. 2017. Soybean Prices – 45 Year Historical Chart. MacroTrends. Accessed 26 June 2017.­prices­historical­chart­data

Merchant, Emma Foehringer. 2017. Briefly: We’ll Always Have Paris. Grist. 5 June.

Romm, Joe. 2008. The Biggest Source of Mistakes: C vs. CO2. ThinkProgress. 25 March.­biggest­source­of­mistakes­c­vs­co2­c0b077313b

Tubiello, F. N., M. Salvatore, R. D. Cóndor Golec, A. Ferrara, S. Rossi, R. Biancalani, S. Federici, H. Jacobs, and A. Flammini. 2014. Agriculture, Forestry and Other Land Use Emissions by Sources and Removals by Sinks. Statistics Division, Food and Agriculture Organization, Rome.

USDA/NASS. 2016. 2016 State Agriculture Overview: Virginia. United States Department of Agriculture: National Agricultural Statistics Service. Accessed 23 June 2017.

USDA/NASS. 2017. Quick Stats. United States Department of Agriculture: National Agricultural Statistics Service. Accessed 26 June 2017.

West, P.C., H.K. Gibbs, C. Monfreda, J. Wagner, C.C. Barford, S.R. Carpenter, and J.A. Foley. 2010. Trading Carbon for Food: Global Comparison of Carbon Stocks vs. Crop Yields on Agricultural Land. Proceedings of the National Academy of Sciences (PNAS) 107(46), 19645-19648. Accessed 10 June 2017. doi:10.1073/pnas.1011078107

WPR. 2017. Population of Cities in Virginia (2017). World Population Review. Accessed 26 June 2017.­population/cities/