Read the papers without mixing their claims¶
A paper can explain a model, show a numerical calculation, publish a reconstruction or compare it with measured vegetation. These are different uses of evidence. Before treating a reported success as validation, identify what was compared, on which inputs, in what units, and with what independent reference.
The repository's main-paper inventory identifies nine canonical accounts. Each explains the scientific question and assumptions, works through a source-to-result example, separates the authors' findings from our own calculation, and states which inputs or decisions are missing. The machine inventory records source access and data-release scope. These report references describe the local source revision; a local candidate does not establish public deployment or independent review.
Follow the evidence route¶
Counts are not cover. A sediment sample's pollen fraction divides counted grains. A vegetation fraction divides land cover within a defined spatial area. Productivity and dispersal differ among taxa, so one cannot stand in for the other. REVEALS estimates regional cover; LOVE uses regional background and small-site pollen to infer local cover. Their physical basin, age, taxon, parameter and area assumptions need source-specific admission. A finite calculation can still yield an invalid cover composition.
A code comparison tests calculation. Our selected author-R and Python values agree on declared inputs, as explained in the method correspondence results. This does not authenticate the complete historical publication inputs or measure ecological accuracy.
A pollen prediction score tests pollen. Whole-site or country exclusion reduces training overlap under a declared protocol. The result still needs the taxon/count denominator, attempted and refused bins, outcome exposure and temporal support. Our source scope and prediction results retain those limits.
Vegetation accuracy needs an independent matched measurement. A nearby plot, historical map or another model's cover estimate must match the proposed physical site, time, taxa and spatial estimand. Our current vegetation reference review has no admitted Nordic pair for repository prediction accuracy. Published validation in another study does not supply that missing outcome.
Choose an account for your question¶
| Read | Evidence role | Our current boundary |
|---|---|---|
| Githumbi et al. (2022), LandClim II and the LandClim account | Published regional archive and historical reproduction target | Selected author-example parity and an exact-window archive discrepancy; full historical input identity remains open. |
| Theuerkauf and Couwenberg (2024) and the LOVEoptim account | Local-inference method and published simulation | Printed parameter correspondence and later source fixtures; no reproduction of the original simulation. |
| Sugita (2007a) and Sugita (2007b) | Regional and local theoretical estimands | Accounts disclose abstract-level original source access; later calculation traces do not reproduce original simulations. |
| Hellman et al. (2008a) and Hellman et al. (2008b) | Original regional validation and sampling/parameter sensitivity | Full original input/scenario vectors are unavailable here; original findings remain distinct from our figure-label arithmetic. |
| Gaillard et al. (2010) | Carrier of the figure used in our current validation explanation | Nine rounded selected labels from two regional compositions; no full 26-taxon or repository score. |
| Cui et al. (2014) | Potential historical-map vegetation reference | Map support and pollen anchors replay; numerical class fractions are unavailable for a new score. |
| Tabares et al. (2021) | Modern pollen/vegetation reference observations | Complete source-table join and descriptive distances; out-of-domain development, rather than Nordic validation. |
The canonical reports annotate partial primary-text access and precise unsent source/reviewer questions. They preserve numerical differences, invalid compositions and missing outcomes. Actual collaborator comprehension and qualified scientific review remain outstanding.
The report and presentation downloads provide all nine canonical accounts and three audience briefings as source-bound PDFs. Their main text, tables and diagrams can be read offline; linked external evidence still needs access to its source.
Use a result responsibly¶
Start with the scientific question, then check the claim-to-evidence index for exact rows, source identity and producer. Read the corresponding account's unavailable conclusions before selecting a parameter or using a model output. A source-bound numerical test, development prediction and independent vegetation comparison have different acceptance conditions; the current model-use results explain which choices their evidence supports.