Planning a shallow stratigraphic drilling campaign when only a few wells are allowed: which sites to drill, what their cores can date, and what the first cores change.
Where older units come up beneath the seabed, short cored holes give the first hard data on source rock, reservoir and seal in a basin with no wells. My part as the geologist: planning and supervision of the first shallow stratigraphic drilling campaign on the Russian Arctic shelf (Malyshev et al., 2023, which I co-authored): 10 wells, 300 m of core, 13 stratigraphic units; those figures are published. The campaign was planned for five wells and drilled ten in one season. The example below is made up; the last line under each finding says what the published paper reports.
AWhy not stop at five: what each next well buys
In most drilling, fewer wells is the aim. Here it is the other way round: the product is core, and the scarce resource is the weather window and the vessel's time, so each extra hole in the same season is another try at the targets. A curve like this is the case for a number of wells.
In the published campaign. Sites were chosen as principal and reserve ones, aiming at the widest stratigraphic range with wells shallower than 100 m. One of the ten wells was a reserve site, drilled during a storm mooring; it penetrated the same Upper Devonian unit as one of the principal wells.
BThe weak link is dating, not coring
Not to scale. Each hole is drilled through the cover without coring, then cored. Hole 1 sits in the middle of a unit. Hole 2 crosses a boundary and can date it, if the pick puts the boundary inside the core; the bar is the pick error.
A core in the middle of a unit is safe on a rough pick. A core across a boundary is worth more, but needs a sharp pick and rock that gives ages on both sides. The same choice comes up when core points are set in an exploration or appraisal well from a seismic depth with an error.
In the published campaign. Ages came from conodonts, fish remains, spores and other fossils; in well SSD 23 no macrofauna was encountered and the age came from palynology. The fossil ages moved two boundaries on the seismic: the Ordovician–Silurian boundary to an unconformity earlier taken to separate Silurian and Devonian, and the Silurian–Devonian boundary to near a horizon earlier read as lying inside the Lower Devonian.
CThe first cores change the plan
In the published campaign. The later sites were adjusted from new seismic rather than from cores: while one vessel drilled, a second shot ultrahigh-resolution seismic across the planned sites. Within a well, total depth and coring intervals were decided from the rock description sent from the site. Overall, the first results tend to make the section older than the base model implied.
Try your own section
The plan in full
score: points for units cored and boundaries dated, on average
target units cored, on average
boundaries dated, on average
cored, of metres drilled
The plan, in drilling order
#
Site
Seismic
Core as planned
If the hole delivers, chance of each target
Adds
Reserve sites and stand-ins
Site
Seismic
Core as planned
If the hole delivers, chance of each target
Adds
More answers: where each core lands, high-res lines, how firm the choice is
Section by well: the chance of each unit and boundary, well by well
Chance that each planned well cores at least the set length of a unit, or dates a boundary, holes that fail included. The Plan column combines the wells; on a narrow screen the table scrolls sideways. The cores are stitched mainly through the horizons between them; where two cores also overlap in the section, they tie directly.
Limits
The example is made up. At each site the page takes one dip, no faults between the core and its anchor horizon, and a flat base to the cover.
Errors are spread evenly over their ± ranges, so "always" and "never" mean within the ranges entered. Picking the wrong reflector lies outside any range: the page sees it only after drilling, as a core that does not tie.
After drilling the page knows only the unit identified and the depth of one horizon at each site. It cannot re-pick a horizon: that takes the seismic and the whole core description.
More limits
Faults repeat or cut out section; anchor on a horizon on the same side of any fault.
What the picks of one horizon share is a rough model. Only the part of the range carries over from a core to the other sites, not the size of the miss: a 1 m miss and a 25 m miss narrow them alike. Sites on a high-res line and sites on 2D only do not inform each other, and an error common to all horizons, one velocity model for the whole area, is not in it.
A core that does not tie may put the neighbouring horizons in doubt too; the page holds only the sites on that one.
A hole either delivers or not. In practice holes also stop early on boulders or refusal, deeper targets and thick glacial cover carry more of that risk, and core is lost most at contacts, where dating needs rock; lower the age chances where that matters.
Every site gets a high-resolution site survey for shallow hazards before drilling: use 2D picks to rank sites for the survey, then re-run with the new picks. The made-up cores under finding C keep the 2D picks so that the numbers are large enough to see; on site-survey picks the same logic runs on a few metres.
Unit thicknesses come from the regional map. Rock 30–100 m below the seabed can be weathered, glacially deformed or a rafted block, and its porosity, seal quality and maturity differ from the same unit at depth.
The ages that tie a core to the section come from the laboratory, often months after the campaign; on the vessel the unit is first told from the rock itself. The page checks the unit against the pick, not the dating.
The dip, cover and thickness errors are stepped on a grid that stops a few percent short of the ends of their ranges, so "nothing within the errors" can miss the outermost corner.
Each site must clear a shallow gas and seabed hazard check and keep off mapped closures; steep dips can also push a hole off vertical. Nothing here on cost, vessels, weather or permits.
How it works
At each site the beds dip at one angle under a flat cover. A point z metres below the seabed lies (zH − z)·cos θ above the anchor horizon in the section, zH being the horizon's depth at the site and θ the dip. So the core, from the base of the cover q to the bottom of the hole D, spans (D − q)·cos θ of section, placed by the pick.
For each site the page takes every combination of the errors: the pick error is integrated exactly, and the dip, the cover and the thickness are stepped on a grid of 21 × 21 × 11. It counts how often the core holds the set length of each unit and the set rock on both sides of each horizon, then multiplies by the share of holes that deliver.
A horizon is dated only if the rock on both sides gives an age: the chance of crossing it is multiplied by the two units' chances of giving one. Below the deepest unit, the older rock is taken to give ages like that unit.
Sites anchored on one horizon and picked on the same kind of seismic share part of the pick error. The ± range is cut into equal parts, two when they share the side: which part the true depth lies in is the same for all of them, where it lies inside the part is each site's own. Each site's error stays spread evenly, but two such wells hit and miss together more often than two wells on different horizons.
The score of a set of wells adds, for each unit, its priority (high 3, medium 2, low 1) times the chance that at least one well cores it, and for each horizon its value (key 2, useful 1) times the chance that at least one well dates it. Wells on different horizons count as independent; holes fail and rocks give ages independently everywhere.
Rules it uses
Best set: every set of the chosen size is scored and the best kept; on equal scores, the set that cores more metres, then the first found. With up to 12 sites that is at most 4,095 sets.
Order: each next well is the one that adds most to the wells already drilled. The note under the plan then says whether stopping after any well leaves the best set for that number, or how close it comes.
Stand-ins and reserves: for each planned site, the stand-in is the site that best keeps the score if that one cannot be drilled; reserves are ranked by what each would add to the full plan.
Verdict: sure when every high-priority unit and key boundary is at least 90 % likely; thin when one of them is under 90 %; short when a high-priority unit is under 60 %, a medium one under 30 %, or a key boundary under 30 %. One hole alone gives a target at most the share of holes that deliver, so 90 % usually takes two holes on it.
After drilling: a unit identified in a core fits the plan if the errors can put it in the core at least one time in five; unlikely if less; no fit if never. A horizon met in the core fits if its depth lies within the pick error. The unit is the one the rock or the age matches in the model you planned with, which is the model being tested.
What to drill next: where the errors allow what a core showed, each part of the pick range is weighted by how well it allows it; that narrows where the horizon can lie at the other sites that share its pick error. An unlikely core is still used, and marked. Where nothing allows it, the horizon does not tie: its sites still to drill are held until it is re-picked, and the drilled well counts the unit identified only. The wells still to drill are then chosen again on top of the wells drilled; a drilled hole has delivered, and the unit identified in it counts as cored.
Checks against known answers:
Each check compares the page with an answer worked out another way: by hand from the geometry, or by an independent count that draws two million random combinations of the errors for the sites on each horizon, searches every set of wells, and chooses the next wells again after the made-up cores (sites_ref.py in the repository). The checks test the arithmetic, not the geology or the error ranges. Change the code and break a formula, and a row turns FAIL.