Sweden Lake Priorities¶
The Sweden lake priority surface ranks 6,763 SMHI SVAR registry lakes that have at least one human ancient-DNA locality within 50 km. It asks where the current collection offers the richest combination of direct human evidence, pollen context, archaeology context, animal context, and basic lake suitability. It does not select a coring site.
Every candidate uses a representative point derived from the official lake polygon. Pollen-site coordinates never substitute for lake identity. Registry names that clearly describe engineered water bodies or wetlands are excluded from the shortlist, while duplicate lake names and coordinate ambiguity remain visible as required review actions.
Candidate And Scoring Pipeline¶
flowchart LR
A[40,565 SVAR lake records] --> B[Official polygon representative points]
B --> C[Exclude non-lake identity classes]
C --> D[Require human aDNA within 50 km]
D --> E[6,763 ranked candidates]
E --> F[Score 10, 20, 30, 40, and 50 km bands]
F --> G[Weighted aggregate rank]
F --> H[Cross-scenario consensus]
G --> I[Fieldwork-preparation screen]
H --> I
I --> J[Identity, limnology, access, and permit review]The public atlas exposes aggregate and consensus top-40 layers, top-40 layers for each radius, and a fieldwork-preparation top 20. The overlays are disabled by default because they are interpretations over the base evidence layers.
Evidence Weights Within A Radius¶
| Signal | Weight | Interpretation |
|---|---|---|
| Human aDNA | 0.59 | locality and sample coverage near the lake |
| Direct pollen | 0.14 | pollen records placed on or very near the official lake |
| Nearby pollen | 0.07 | broader pollen context, with chronology-aware credit where supported |
| Lake sampling fit | 0.07 | area- and identity-based screening, not bathymetric suitability |
| Archaeology | 0.07 | SEAD point context and coarse RAÄ density |
| Domesticated animal aDNA | 0.04 | secondary direct-evidence context |
| Evidence diversity | 0.02 | number of represented evidence families |
Within each band, human aDNA locality and sample coverage determine ordering first. Direct pollen breaks the next tie, followed by broader pollen and archaeology context. Sampling fit and the blended score resolve later ties.
Temporal credit is conditional. Neotoma and LandClim records gain stronger chronology contribution only when numeric BP intervals overlap nearby human locality windows. The current Sweden-facing SEAD capture is a site inventory without numeric chronology rows, so it contributes spatial archaeology context but not same-period evidence.
Score And Rank Are Separate Contracts¶
For each radius, the displayed band score is the weighted sum of normalized signals:
0.59 human aDNA
+ 0.14 direct pollen
+ 0.07 nearby pollen
+ 0.07 lake sampling fit
+ 0.07 archaeology
+ 0.04 domesticated-animal aDNA
+ 0.02 evidence diversity
The component weights sum to 1.00, but the score is not the first sort key. Band and aggregate ordering first compare human aDNA locality and sample coverage, then direct pollen support, then broader pollen and archaeology context. Sampling fit and the blended score resolve later ties. This preserves the model's stated priority instead of letting a dense contextual inventory outvote direct human evidence.
Signals are normalized within the governed candidate population. A score is therefore comparable inside the named product version and scenario; it is not an absolute probability, a cross-version scientific measurement, or a field success estimate.
Combining Distance Bands¶
| Radius | Aggregate weight |
|---|---|
| 10 km | 0.35 |
| 20 km | 0.27 |
| 30 km | 0.18 |
| 40 km | 0.12 |
| 50 km | 0.08 |
The aggregate rank favors close evidence while retaining broader regional context. The consensus rank instead rewards recurrence across top scenario slices, then uses mean scenario rank and aggregate rank as tie-breakers. A lake that is consistently strong across radii can therefore differ from the lake with the highest weighted aggregate score.
Explain Rank Movement By Cause¶
Ordinal position can change even when a lake's own evidence does not. A reproducible comparison classifies the cause before interpreting the movement:
| Cause | What changed | Appropriate reading |
|---|---|---|
| source refresh | nearby governed members or their evidence fields | scientific input change |
| candidate-population change | lakes became eligible, ineligible, merged, or separated | denominator and normalization change |
| identity correction | registry match, polygon, name, or representative point | candidate-definition change |
| model change | signal, weight, tie-break, radius, or missingness treatment | decision-policy change |
| precision change | chronology or coordinate posture strengthened or weakened | comparison-rights change |
| unchanged score, changed rank | other candidates moved around this lake | relative ordering change only |
For this reason, “rose five places” is incomplete without the prior and current candidate populations, model identities, component values, and member-level evidence diff. Rank movement is not itself evidence that a lake became more suitable.
Rank, Stability, And Readiness¶
The ranking exposes three different signals that must not be collapsed into a single recommendation:
| Signal | Meaning | Appropriate use |
|---|---|---|
| aggregate rank | weighted evidence richness across all distance bands | identify candidates favored by the declared distance weighting |
| consensus rank | recurrence near the top across scenario slices | identify candidates less dependent on one radius |
| fieldwork-preparation posture | evidence, identity, and lake-screening readiness | order the next review actions |
flowchart LR
Evidence["governed nearby evidence"] --> Scenarios["radius scenarios"]
Scenarios --> Aggregate["aggregate ordering"]
Scenarios --> Consensus["cross-scenario stability"]
Aggregate --> Screen["fieldwork-preparation screen"]
Consensus --> Screen
Screen --> Review{"expert review"}
Review -->|evidence sufficient| Candidate["candidate for field assessment"]
Review -->|gap remains| Deferred["defer with required action"]A high aggregate rank with weak cross-scenario recurrence is sensitive to the chosen radius. A strong consensus rank indicates stability within the tested scenarios, not robustness to missing source families or unmodeled field conditions. The fieldwork-preparation posture can therefore reorder or defer a high-scoring lake without contradicting the ranking.
A Concrete Reordering¶
The current fieldwork-preparation screen places Sjötorpasjön first even though it is third in the aggregate evidence ranking. The screen retains both facts:
| Field | Current value |
|---|---|
| fieldwork rank | 1 |
| aggregate rank | 3 |
| aggregate score | 0.5862 |
| scenario consistency | high; present in six tested top-20 slices |
| sampling posture | sampling_lake_candidate |
| preparation posture | identity_resolution_required |
| identity issue | duplicate Swedish lake name |
| required review | confirm the exact registry match and inspect linked SEAD records |
This is not a contradiction or a hidden override. Aggregate rank answers the weighted evidence-richness question. Fieldwork rank applies a separate human-context, sampling, scenario-consistency, and identity-review contract. The unresolved name prevents the top fieldwork row from becoming a sampling instruction.
Reading Candidate Fields¶
Each ranked row preserves:
- lake registry ID, UUID, water identity, and representative source URL;
- official coordinate-resolution method and mapped area;
- duplicate-name, name-status, and coordinate-spread diagnostics;
- per-radius counts, signals, score, and rank;
- aggregate score and rank plus scenario-presence statistics;
- sampling posture, sampling fit, and the limitations behind that posture;
- direct pollen sources and time-aware pollen counts;
- nearby human, animal, SEAD, and RAÄ context metrics.
Sampling postures are screening labels. small_lake_review flags a micro-basin
that needs validation; compact_lake_candidate marks a small mapped surface;
and sampling_lake_candidate indicates a more plausible area-based posture.
None asserts sufficient depth, intact sediment, access, or coring feasibility.
palaeopen_alignment_posture is also a local screening field. It summarizes
whether a candidate already has at least two direct pollen sources and four
evidence families within 20 km. It adds no score, source record, network
membership, or PalaeOpen endorsement.
Current Aggregate Leaders¶
| Rank | Lake | Score | Area km² | Sampling posture |
|---|---|---|---|---|
| 1 | Bergsjön | 0.5947 | 0.063346 | compact_lake_candidate |
| 2 | Hulesjön | 0.5875 | 0.037617 | small_lake_review |
| 3 | Sjötorpasjön | 0.5862 | 0.603122 | sampling_lake_candidate |
| 4 | Hornborgasjön | 0.5037 | 27.925549 | sampling_lake_candidate |
| 5 | Skårsjön | 0.4818 | 0.021492 | small_lake_review |
| 6 | Rösjön | 0.4651 | 0.956929 | sampling_lake_candidate |
| 7 | Bjärsjön | 0.4573 | 0.132579 | compact_lake_candidate |
| 8 | Tresjö | 0.4433 | 0.104225 | compact_lake_candidate |
Aggregate rank is evidence-richness ordering. The fieldwork-preparation screen reorders candidates by near-lake human evidence, sampling posture, scenario consistency, and identity risk. It also emits required actions such as resolving duplicate registry names or inspecting SEAD context before narrowing an interpretation.
Evidence Still Required Before Fieldwork¶
The public ranking does not contain governed bathymetry, basin depth, sediment preservation, shoreline access, permits, landowner logistics, or field-confirmed coring conditions. Those are blocking inputs for a sampling recommendation, not optional refinements to the score.
A responsible progression is therefore:
- confirm the exact SVAR lake identity and polygon;
- inspect the direct human and pollen records behind the score;
- separate temporally comparable evidence from spatial context;
- acquire bathymetry and sediment-basin information;
- assess access, permissions, conservation constraints, and field safety;
- record the expert decision independently of the ranking score.
That final separation preserves auditability. The model score remains the answer to a reproducible evidence-richness question, while the expert decision records whether the unmodeled practical and scientific requirements were met. If the decision differs from rank order, the reason belongs in the field review rather than in an altered score.
Candidate Decision Dossier¶
Before a ranked row becomes a field-assessment candidate, assemble one dossier containing:
- the exact SVAR identity, polygon, representative-point method, and name-risk review;
- contributing evidence members partitioned by family, role, distance band, and temporal comparability;
- aggregate, consensus, and sensitivity results under the governing model;
- bathymetry, basin morphology, sediment expectations, access, permissions, conservation constraints, logistics, and safety evidence;
- the expert disposition—advance, defer, or reject—with its reason and date.
The dossier does not need to agree with rank order. Its purpose is to preserve why a decision was made after adding evidence the ranking intentionally does not model.
Preserve Candidate State Transitions¶
A candidate moves through new evidence states; it is not edited from “ranked” into “field ready”:
flowchart LR
Ranked["ranked under model and data revision"] --> Desk["identity and evidence desk review"]
Desk --> Dossier["candidate decision dossier"]
Dossier -->|advance| Visit["dated field observation"]
Dossier -->|defer or reject| Decision["reason and recovery condition"]
Visit --> Assessment["separate sampling assessment"]
Assessment -->|supported| Protocol["site-specific protocol and permissions"]
Assessment -->|unsupported| DecisionEach node retains its own date, inputs, method, and disposition. A later visit does not rewrite the historical ranking, and a strong historical rank does not pre-authorize the visit or sampling assessment. This makes disagreement useful: the evidence shows whether the model, identity review, field conditions, or operational constraints caused the decision to change.
Reusing A Ranked Result¶
A defensible reference to a candidate includes its SVAR identity, ranking surface, scenario or aggregate definition, model inputs and weights, active geographic scope, and known required actions. Quoting only the ordinal rank removes the assumptions that give the number meaning.
The portable ranking packet includes the ranked registry, per-radius scenario rows, evidence bands, aggregate definition, model weights, ranking-engine manifest, sensitivity output, and fieldwork-preparation screen. Quoting a row without the engine manifest and scenario identity makes its rank impossible to reproduce or interpret after a source refresh.