Data Gaps: A Critical Examination of Oil and Gas Well Incidents in Ohio

To identify the location and total number of incidents in Ohio, significant data scrubbing was undertaken. For many of the records sent by ODNR, the location data was insufficient. The “location” provided ranged from latitudinal and longitudinal coordinates, partial postal addresses, cross streets, or sometimes only an indication that a violation site was a few miles in any given direction of a city. Street names were incorrectly spelled or lacked specification (e.g., one such record indicated an address of 161 High Street, Warren, OH. In Warren, Ohio, there exists both 161 High Street NW and 161 High Street NE. It is unclear which street ODNR was referencing in this record). Street numbers were missing digits entirely. Postal addresses that had correct spelling and formatting sometimes did not point to a given land structure.

To some degree, any data collection process is prone to error. Mistakes in spelling and digits are expected, because errors while typing occur in any data collection process. In addition, postal addresses that are seemingly detached from any given landmass or building are somewhat normal in rural areas, where there exists less data. You may have noticed something similar if you’ve ever used Google Maps to navigate to a rural community. Sometimes you need additional specifications to supplement the route a GPS creates. However, the extent to which the location data was fragmented was highly unusual for a government body, even for data that does not face public scrutiny. Ideally, a latitude and longitude would be provided for every site, especially those that require an ODNR employee to dispatch to the site of a violation.

For FracTracker Alliance, cleaning data involved much more oversight than it should have necessitated. Geocoding, or assigning a physical location to a locational description like a postal address, is something many GIS software are equipped to handle. However, because the ODNR violation data was so rife with errors and incomplete addresses, geocoding had to be done manually. To best map the data, it was necessary to assign each record a latitude and longitude, but not necessarily a postal address.

Location data was split into two categories: those with latitude and longitude coordinates, and those with postal addresses or references to streets. The latter records were then checked for accuracy and completeness. For mostly complete records, any typos were corrected, and from the corrected addresses a latitude and longitude were determined and added to the record. For records with no locational data, if the “Brief Description” field indicated the incident occurred on the well pad, we matched the record’s American Petroleum Institute (API) number to an API in the ODNR well database and used the reported well coordinates as the incident coordinates. Overly vague records (i.e., records with only a street name or records that indicated an overly broad area of interest) or those that contained no location information were omitted from this map and analysis. Finally, to maintain clarity within the map provided below, some attributes were filtered from the data.

It cannot be overstated how inadequate locational data was in this dataset. While other datasets may require this level of data cleansing for one or two records, a considerable amount of records had to be altered or added to, as evidenced by the roughly 100 records that were omitted entirely. Relatedly, while the original data sets had between 50-60 attributes used to describe a given record, many were not useful or left incomplete. (Table 3 lists attributes included in the map that required explanation.) The “type of incident” attribute, while fairly robust in breadth, fails to allow for a record to be typed by multiple fields. For example, one record’s “Brief Description” states: