Curve Aliasing and Mnemonics

Purpose

Every later calculation asks for a curve by its role: the bulk density, the deep resistivity, the compressional slowness. A well file offers curves by name, and the name depends on who ran the tool, when, and how the data was processed. Aliasing is the step that maps the names in each file onto a fixed set of roles, chooses one curve for each role when several qualify, and merges curves from several runs into one continuous curve.

When aliasing goes wrong, nothing fails loudly. A workflow quietly reads a raw curve instead of the edited one, a shallow resistivity as the deep one, or a density in the wrong units. The errors are systematic across the whole well and are hard to see in the final answer.

Position in the workflow

Upstream. Aliasing starts from the loaded files: LAS, DLIS or tabular data. It needs the header information (units, null value, service company, tool names) and the curve statistics. Files from different sources, years and vendors are all aliased the same way.

Downstream. Everything that reads a curve by role depends on it:

Log Curve Cleanup, where units are converted according to the family of each curve,
normalization and washout flagging, which need to know which curve is the caliper, the gamma ray or the deep resistivity,
every calculation in Stage 2, from clay volume to saturation.

Error propagation. An alias error is not a random error. A wrong choice shifts a whole curve by a constant, a scale or a different physical meaning, and the result looks plausible until it is compared with core or another well.

Key concepts

Mnemonic. The short name of a curve, a Curve mnemonic. It is a convention and not a definition. See Mnemonic Conventions and Curve Families.

Curve family. A Curve family is the set of curves that measure one quantity, whatever their names or tools. The alias dictionary maps each family's names to one canonical name.

Why mnemonics vary. Each service company named its curves independently, and each tool generation changed the names; wireline and LWD tools use different schemes; processing houses add suffixes for edited, raw or filtered versions; and merged files carry the history of every run. Era matters: an old file may carry a short name from a tool that no longer exists, a recent one a long name that encodes a spacing or a frequency.

Duplicates. The same family can occur several times in one file: raw and edited versions, two passes, different resolutions, a high-resolution and a standard curve from the same tool, or curves from different tools. Choosing between them is a judgement that should be rule-based and recorded.

Array and multi-channel tools. Array tools produce several curves at different depths of investigation, and the names encode that. See Vendor and Tool-Specific Mnemonics.

Runs and splicing. A well is logged in several runs, and the curves overlap. Combining them needs a depth match and a decision about the overlap. See Curve Splicing and Merging.

Method selection guide

Method	Inputs	Use when	Strengths	Weaknesses
Mnemonic conventions	Curve names, units, value statistics; alias dictionary	Always, as the first pass over every file	Fast, transparent, repeatable; works on any vendor if the dictionary is good	A name is only a clue; fails on unknown names and reused mnemonics
Vendor and tool-specific rules	Header (company, tool), names, descriptions	The file holds array, LWD or multi-resolution channels	Gets the depth-of-investigation order and the processing level right	Needs a pattern set per vendor and generation; can be silently wrong on the wrong vendor
Curve splicing	Two or more runs, an overlap, a depth reference	The same family comes from several runs or passes	Produces one continuous curve, depth-matched and documented	Needs a feature-rich overlap; a forced level match can hide a real difference

Decision guidance

Alias first, always. Only curves in the same family and from the same kind of tool are candidates for splicing.
If a file contains array or LWD channels, run the vendor rules before the generic dictionary, so that the channels are not treated as duplicates of each other.
If the duplicates are raw and edited versions of one curve, choose; do not splice.
If they are the same measurement from different runs covering different intervals, splice.
If they are different tool types over the same interval, keep both and choose one by rule for each interval. Averaging them removes information.
If a tie cannot be broken by a rule, ask a person. A guess hidden in a workflow cannot be audited.

Shared parameter picking

Alias dictionary and priority. One dictionary and one priority policy for the project, kept under version control. The policy says which version (edited over raw), which resolution and which tool wins a duplicate.

Unit list and plausible range per family. Used by the match test here and again by the unit detection in Log Curve Cleanup. Keep them in one place.

Minimum coverage. The fraction of valid samples in the interval of interest below which a curve is not chosen as primary. It is a project decision; a curve used only to patch another may have low coverage.

Depth reference run, search window and blend length. Shared by all splices in a well: one reference run, one shift search window, one blend length. See the splicing page.

Review rule. What happens to an unresolved curve or a tie: reported and held for a person.

Recommended default approach

Absent other information, a careful generalist would:

Read the header: units, null value, service company, tool names, run information. Keep it with the data.
Normalise the names and resolve them against the alias dictionary. List what matched, what is ambiguous and what is unknown.
Apply the vendor rules to array and LWD channels and assign shallow, medium and deep resistivity.
Resolve duplicates by rule: edited over raw, higher coverage, correct units, and the same pass for neutron and density. Keep the alternates.
Where the same family comes from several runs, depth-match them to a reference run and splice with a blend in a quiet zone. Do not force a level offset between different tool types.
Write out a mapping table (family, chosen curve, reason, alternates, shifts) and have a person read it once per well.

Combining methods

The three methods are used in sequence, not as alternatives: the generic dictionary, then vendor rules for what it cannot classify, then splicing for what comes in pieces. The order matters. Splicing curves that were never confirmed to be in the same family merges different physics. Aliasing without checking units means the later unit step has to guess from values. The mapping table from the first two steps is the input to the third, and the shifts from the third are applied to every curve of the run, not only to the one used to find them.

QC of results

A good result:

has one chosen curve per family needed, with the reason and the alternates written down,
has chosen curves whose units and values are plausible for their family,
uses neutron and density from the same pass and the same depth reference,
has no unexplained unresolved curves, and
has spliced curves with no visible step, no repeated or missing section and a recorded shift for each run.

Signs of a bad result: a deep resistivity that reads lower than the shallow one in a clean, non-invaded bed, density porosity that is negative over thick intervals, gamma ray that jumps at a run boundary, and a different answer from the same file loaded twice.

Common pitfalls

Trusting a name without checking units and values.
Treating array channels as duplicates, so that a shallow channel is chosen as the deep resistivity.
Choosing a raw curve over an edited one, or the reverse, without a rule.
Resolving ties silently, in an order set by the file.
Splicing curves from different tool types and forcing them to the same level.
Shifting only the curve used for the correlation, and not the rest of the run.
Splicing at a bed boundary, a washout or a casing shoe.
Letting different wells in one project use different dictionaries.

Going Deeper

Aliasing is a data-management problem more than a petrophysical one, and it persists because the industry has no single enforced naming scheme. Standard file formats carry units and descriptions, which makes automatic checking possible, and some operators and data vendors maintain their own curated dictionaries. Machine-assisted matching, using names, descriptions, units and value distributions together, improves on name matching alone, but is still a proposal for a person to confirm. The deeper question for splicing is how much of the difference between runs is error and how much is signal, and that cannot be answered from the logs alone: it needs the tool and run information.