The dataset includes subsurface stratigraphic picks of the tops of the Horseshoe Canyon/Wapiti and Battle formations in the west-central Alberta Plains (Townships 15 to 67, Ranges 16W4 to 3W6) made from wireline geophysical well logs.
The dataset supplements Alberta Geological Survey Open File Report 2011-08, which describes the methodology. We screened the well data to detect errors resulting from deviated wells, as well as incorrect well-header ground and kelly bushing elevation data. Statistical methods identified local and regional statistical outliers, which were examined individually.
A stratigraphic pick in a well is a point defined in three dimensions (X, Y and Z).
The accuracy of the pick depth, either in measured depth from the kelly bushing or with respect to sea level, is difficult to quantify and includes (but is not necessarily limited to) errors in
- well surface or bottom-hole latitude and longitude (X and Y);
- well ground elevation (Z);
- well kelly bushing elevation (Z);
- geological or human error resulting from errors in picking the incorrect stratigraphic top (Z);
- data entry or data transfer (X, Y and/or Z); and
- incorrect well log depth calibration (Z).
The data are tabular (point data with X, Y and Z values).
The author generated all stratigraphic picks. All picks are ranked the same in quality.
As the dataset includes only vertical wells, all location data and well-identifier data (UWI and UWI_MODIFIED) are unique for each stratigraphic formation. In non-vertical wells, surface and bottom-hole latitude and longitude may be different, and several wells may share a common surface location but have different bottom-hole locations. By choosing only vertical wells, this problem was avoided.
The author collected the data from the Alberta Plains where deformation of the Cretaceous sedimentary succession is relatively minor. All points are east of the deformation front at a given latitude; thus, rocks should not be thrusted or structurally duplicated. Therefore, the tops of the Horseshoe Canyon/Wapiti and Battle formations should only occur once in any given vertical well.
No data are missing.
Attribute values were checked to ensure reasonable values. For instance, the author plotted the well locations on a map and observed no obvious anomalous locations. A query checked for any deviations from the vertical of the well surface location compared with the bottom-hole location. These wells were removed from the dataset. If a well is deviated, its surface and bottom-hole co-ordinates should be different. As all remaining wells should be vertical if the surface and bottom-hole co-ordinates are correct, measured depth and true vertical depth should be equal.
In vertical wells, the subsurface depth of a pick in a well, measured with respect to sea level, is calculated by taking the elevation of the kelly bushing (on the drilling platform) and subtracting the measured depth of the pick on the geophysical well log.
Some uncertainty in the vertical depth of the pick will result if the borehole is not entirely vertical. The author compared the bottom-hole latitude and longitude of each well location with the surface latitude and longitude for each well to ensure they were the same. If either the surface or bottom-hole latitude and longitude are incorrect, some degree of vertical error may result. In general, the amount of vertical depth due to deviations from the vertical in boreholes is deemed negligible with respect to other potential sources of vertical error in this study.
Perhaps the greatest source of vertical uncertainty in this study is potential error in the elevation of the kelly bushing (KB). Any errors in surveying the ground elevation of the well site can result in vertical error. In addition, once the ground elevation is determined, the site is usually prepared for the drilling rig. If the original survey marker is disturbed or moved, this can result in potential vertical errors. The KB elevation is usually derived from adding the height of the drilling platform above the ground surface to the survey ground elevation. If this is not done correctly, it can introduce vertical error in the KB elevation, which is then propagated in the measured depth to the pick and the subsea pick depth.
Although incorrect KB elevation data can be difficult to detect, the data were screened by comparing the ground elevation and the KB elevation (derrick height) for each well. An acceptable range of derrick height (calculated by subtracting ground elevation from KB elevation) of two to six metres was used. Wells with derrick heights outside this range were excluded.
To check for potential gross errors in the ground elevation for wells, ground elevations were compared with shuttle-radar digital elevation model (DEM) elevations extracted for well surface locations. If the difference between the ground elevation and the elevation derived from the DEM data was more than 2 ± 9 metres (i.e., -7 to 11 metres; approximately the mean of this difference plus or minus three standard deviations for all wells in the Alberta Plains), the data from those wells were excluded. This method potentially excluded wells for which well ground elevation values are correct, but for which the DEM data for that well location are incorrect. It also may have not detected relatively small errors in either ground or KB elevation data for a well, as long as those values met the screening criteria. However, it did detect large errors in well KB or ground elevation data.
Vertical error in the pick subsea elevation can also result from human or geological error resulting from uncertainty or incorrect placement of the pick on the well logs. The occurrence and magnitude of this error is difficult to identify, but checks for internal consistency (such as identification of global and local outliers using statistical methods and gridding data while picking) minimized this source of error as much as possible.
Prior to making picks for a given surface, we studied the published geological literature with emphasis on representative sections. If available, we examined outcrop sections and drillcore (with associated geophysical well logs) to provide a link between the rock and downhole geophysical signatures.
Geophysical well logs (both digital and raster format) were examined using Petra and Accumap software and picks were recorded in a database. If well density and log availability were sufficient, the author selected wells according to the following criteria:
- vertical wells only;
- wells with a spud date between 1975 and the present; and
- wells with downhole geophysical well-log suites that include gamma-ray, neutron, density or sonic, and resistivity logs. This requirement was relaxed in areas where the Battle Formation is close to surface and logs were obtained through casing.
A minimum well density of one well per township was aimed for, although well density greatly exceeded that number in most areas. Data tended to be sparser where the Battle Formation is shallow, and in those areas all available wells were picked. We picked about 655 townships, resulting in an average density of about 13.5 wells per township for the top of the Horseshoe Canyon and Wapiti formations.
To facilitate correlation, wells were never picked in isolation, but always on cross-section, with a maximum well spacing of 13 km. In most areas, we used a much smaller well spacing. During the process, picks were gridded using the triangulation method to identify and check outliers, which appeared as ‘bull’s eyes’ on a structure contour map.
After making picks and prior to modelling the surface, we eliminated or minimized errors resulting from incorrect
- depth data (well deviation);
- well-header kelly bushing (KB) elevation data;
- well-header ground elevation data; and
- pick depth (due to human error).
Picks and well-header information, including KB elevation, ground elevation, surface location (longitude and latitude in decimal format) and bottom-hole location (longitude and latitude in decimal format), were exported from Petra (IHS) software into a comma-separated value format. The datum for the well location is NAD83. The picks are in metres, given as measured depth relative to KB elevation. Pick elevations relative to sea level were calculated by subtracting measured depth (MD) from the KB elevation.
A query of the well surface location compared with the bottom-hole location checked for any deviations from the vertical. If a well was deviated, its surface and bottom-hole co-ordinates should be different; therefore, these wells were removed from the dataset. As all remaining wells should be vertical if the surface and bottom-hole co-ordinates are correct, measured depth and true vertical depth should be equal.
Although incorrect KB elevation data can be difficult to detect, we screened the data by comparing the ground elevation and the KB elevation (derrick height) for each well. An acceptable range of derrick height - calculated by subtracting ground elevation from KB elevation - of 2 to 6 m was used. We excluded wells with derrick heights outside this range.
To check for potentially gross errors in the ground elevation of wells, we compared well-header ground elevations with shuttle-radar digital elevation model elevation data extracted for well surface locations. If the difference obtained by subtracting the well-header ground elevation from the elevation derived from the DEM was more than 2 ± 9 m (i.e., -7 to 11 m; about the mean of this difference plus or minus three standard deviations for all wells in the Alberta plains), that well was excluded. This method potentially excluded wells for which well-header ground elevation values were correct, but for which the DEM data for that well location were incorrect. It also may not have detected relatively small errors in either ground or KB elevation data for a well, as long as those values met the screening criteria. It did, however, detect large errors in well-header KB or ground elevation data.
Data were then screened for both global and local outliers. Outliers are values outside a specified normal range compared with the entire dataset (global outliers) or within a local area (local outliers). If they are caused by errors, outliers can have detrimental effects on an interpolated modelled surface, and should be either corrected or removed before the final surface is created. Outliers may result from one or more of the following factors:
- incorrect ground elevation and/or KB elevation data not detected during the initial screening;
- incorrect location data for a well;
- deviated wells that are not marked as such and have either incorrect surface or bottom-hole location data;
- incorrect stratigraphic pick data due to human error; or
- geological structure.
A variety of geostatistical methods was used to identify outliers, including examination of neighbourhood statistics, inverse distance weighting interpolation and Voronoi maps. Outliers were flagged and the well data and geophysical logs examined to determine whether each outlier was the result of geological variability or bad data. In cases where no error could be identified, additional data were gathered to refine the definition of local structure. In these cases, if a stratigraphic surface anomaly caused by a single outlier remained and no geological evidence was present to corroborate structure, then the outlier was removed.
Once initial outliers were either removed or confirmed, the outlier screening process was repeated at least three times. This iterative process identified increasingly subtle outliers. As each pick was made during this project and all statistical outliers were examined and some removed, the largest source of error and uncertainty in the elevation of the stratigraphic surfaces is likely related to the surveyed KB (and ground elevation) for a given well.