Skip to contents

Key concepts of cross-referencing

J.R. Ferrer-Paris

Source: vignettes/articles/concepts-xref.qmd

Work in Progress

This package is still work in progress, but it is ready for testing! Please provide feedback on any issues, and check back regularly for updates.

The need to cross-reference different classification systems arises frequently in ecology and conservation applications. The conceptual foundations for cross-referencing classifications are discussed in:

IUCN (2025). Standards, methods and guidelines for cross-referencing ecosystem classifications and maps to the IUCN Global Ecosystem Typology. Version 1.0. Keith, D. A., Dayaram, A., Driver, A., Ferrer-Paris, J. R., Francis, R., Hay, S., Murray, N. J., Nicholson, E., Santos, J., Skowno, A. L. & Tozer, M. G. (Eds.). Gland, Switzerland: IUCN. ISBN: 978-2-8317-2348-8. DOI:10.2305/CPPC6126.

Here we try to apply consistent methods and standards for cross-referencing in order to make the relationships between units more explicit and enhance the reproducibility of the process. This will enable efficient update when more information becomes available, and minimise divergent or erroneous interpretations of relationships between classifications.

The process of cross-referencing involves a source classification and a target classification, where the target classification might have a broader scope than the source. Elements in the source classification can be interpreted as directly related to one or more elements in the target classification. This relationship can be quantified as membership values, where the source element assigns memberships between 0 and 100% to each element of the target classification, summing to 100% total.

Membership estimates from multiple interpreters can be combined to produce more robust cross-referencing relationships. Uncertainty due to disagreements between interpreters can be quantified and documented for maximum transparency.

Overview of stepwise cross-referencing process

A stepwise diagnostic process ensures that the cross-referencing exercise aligns with four key principles: alignment of classifications, membership based on entity properties, using weight of evidence, and transparent representation of uncertainty (IUCN 2025).

flowchart TB
    subgraph CLASS1 ["Diagnostic tools,<br> replicate interpreters"]
        STEP3[["3: Eliminate implausible candidates"]]:::MainSteps
        STEP4[["4: Estimate (non-zero) memberships"]]:::MainSteps
    end

    STEP1[["1: Assess alignment between classifications"]]:::MainSteps
    STEP2[["2: Assemble information on properties of units"]]:::MainSteps
    STEP5[["5: Synthesis: compile, document methods and uncertainty"]]:::MainSteps

    STEP1 ==> STEP2 ==> STEP3 ==> STEP4 ==> STEP5
    STEP1 -->|No alignment| NO["Units not eligible for cross-referencing "]:::OutSteps
    STEP2 -->|Insufficient info?| RE["Gather more information"]:::OutSteps -->STEP2
    STEP3 --> ZERO["Zero membership value"]:::OutSteps
    CLASS1:::Grp
    classDef MainSteps fill:#f8f9fa,stroke:#1A1A1A,color:#1A1A1A, stroke-width:2px
    classDef OutSteps fill:#eee,stroke:#6C757D,color:#6C757D, stroke-width:2px
    classDef Grp fill:#fff,stroke:#040,color:#040, stroke-width:2px

The functions in this package focus on step 5 in this diagram: compiling membership matrix from multiple interpreters, linking to spatial data for visualisation, and documenting and analysing sources of uncertainties.

Membership tables

The following diagram illustrates a minimal example consisting of two classifications with two concepts each, and lines joining the source and target concepts with membership values.

flowchart LR
    CLASS1:::Clas
    CLASS2:::Clas
    subgraph CLASS1 ["Source classification"]
        ECO1(["Concept A"]):::Ecos
        ECO2(["Concept B"]):::Ecos
    end
    subgraph CLASS2 ["Target classification"]
        ECO3(["Concept Z"]):::Ecos
        ECO4(["Concept W"]):::Ecos
    end


    ECO1 ==>|100%| ECO3
    ECO1 -.->|0%| ECO4
    ECO2 -->|60%| ECO3
    ECO2 -->|40%| ECO4

    classDef Ecos fill:#dfd,stroke:#040,color:#040, stroke-width:2px
    classDef Clas fill:#fff,stroke:#040,color:#040, stroke-width:2px

These relationships can be represented in different tabular formats:

Membership matrix

A membership matrix will have rows for the source classificaton and columns for the target classification:

Concept Z Concept W
Concept A 100% 0%
Concept B 40% 60%

The size of the matrix will increase with the number of concepts in each classification (nn rows for the source and mm columns for the target).

These matrices will expose the information transparently, and can be useful for data entry purposes, but they can become very clunky when the number of concepts is both classifications is large, and are not optimal to store multiple interpretations (see next section).

Long-format table

An alternative representation of this information is a membership table in “long” format:

Source Classification Target Classification Membership
Concept A Concept Z 100%
Concept B Concept W 60%
Concept B Concept Z 40%

The size of the table will increase with the number of links between classifications (n×mn \times m rows, but only three columns). If we ignore or filter out the zero membership values, then the size can be reduced considerably.

Paired column format

A membership table can also be represented in wide format if we can assume that each source concept matches maximum of qq target concepts:

Source Concept Target Concept 1 Membership Target Concept 2 Membership
Concept A Concept Z 100%
Concept B Concept W 60% Concept Z 40%

This table will have nn rows for all source concept, but a fixed number of columns (1+(q×2)1 + (q \times 2))

Multiple interpreters

Tables in long and wide formats can incorportate multiple interpreters by adding columns to identify the interpreter.

For maximum transparency, these tables could also include columns for date of interpretation and for documentation of methods. This would allow better filtering and comparison of multiple interpretations.

An example with a long format table:

Source Target Membership Interpreter Date Methods
Concept A Concept Z 100% Interpreter A 25 May 2011 Match descriptions
Concept B Concept W 60% Interpreter A 25 May 2011 Match descriptions
Concept B Concept Z 40% Interpreter A 25 May 2011 Match descriptions
Concept A Concept Z 80% Interpreter B 25 May 2025 Compare attributes
Concept A Concept W 20% Interpreter B 25 May 2025 Compare attributes
Concept B Concept W 80% Interpreter B 25 May 2025 Compare attributes
Concept B Concept Z 20% Interpreter B 25 May 2025 Compare attributes
Concept A Concept Z 100% Interpreter A 25 May 2026 Revised opinion
Concept B Concept W 60% Interpreter A 25 May 2026 Revised opinion
Concept B Concept Z 40% Interpreter A 25 May 2026 Revised opinion