Linking records to crowns

What changed

3,088 of the 37,409 records that release 5.1 left without a crown are now linked to one: 1,043 linked and 1,456 probable links to existing release 5 crowns, and 589 records on 534 crowns drawn from the record.¹

• 1,655 records joined 1,456 crowns that were LiDAR-detected trees. One record per crown now represents the tree, drawn where the crown’s marker was; the detection stays in the tree’s list of records.
• 844 records joined 664 crowns that a source record already represented. They are listed among that tree’s records and do not get a second marker. The exception is a protected tree (Notable Trees schedule), which never loses its marker: on 1 crown a protected record now represents the crown in place of an ordinary record, which stays in the tree’s list of records; 14 protected records whose crown is already represented by another protected tree keep their own marker where their source recorded them, and the crown’s estimates are still counted once.
• 457 crowns drawn from a record are 3 m or taller. They are drawn faintly, counted with a flag in the totals and benefit estimates, and await confirmation. The other 77 are 2–3 m tall and follow the existing low-canopy rule: they are shown and not counted.

As a result the map shows 2,540 fewer separate trees (3,574,825 in all) and 457 more counted crowns. Estimated annual benefits rise by NZ$47,218: NZ$15,926 from crowns drawn from a record, and NZ$31,291 from crowns now represented by a source record, recalculated for its recorded taxon. The map shows 3,570 protected trees, as in release 5.1.

Four further crowns drawn from a record were left out because each enclosed a small existing release 5 crown; their records stay without a crown, unchanged on the map, and in review.

How records were linked

For each record without a crown, nearby crowns were scored on:

• the distance from the record to the crown’s edge, weighed with an error model learned by comparing the crowns around unlinked records with those around decoy points nearby;
• whether the crown’s height and area suit the recorded taxon;
• conflicts with other records’ taxa on the same crown;
• how many records a crown of that size usually carries.

Neighbouring records were assigned together, so a small crown cannot take several records when one has its own. A record was linked only when its crown was clearly better than every alternative, including no crown at all. Shifting whole digitised rows of records was tested and is not used.²

Crowns drawn from a record

Where LiDAR returns of tree height stand at a record and no release 5 crown covers them, a crown was drawn with the release 5 recipe, including the unclassified returns of palms that release 5 could not see, where the imagery is green.

Link precision

The 95% interval is for the blind audit. –: not measurable by the hold-out test.

The hold-out test hides 31,119 trusted links, moves each record as far from its crown as unlinked records of its source typically are, adds copies whose crown is deleted, and finds them again (0.919 for linked and probable when every unlinked record without a crown is assumed crown-missing). The blind audit labelled cards of fresh records under the final rules, from 2024 imagery, near-infrared and LiDAR heights, without the link’s status or score. Records that could not be decided are left out; counting them as wrong gives floors of 0.76, 0.51 and 0.69. The hold-out test cannot measure drawn crowns, so their figure comes from the audit alone.

Council recorded positions

The Council recorded positions switch in the Layers tab (off by default) shows, for every joined record, where its source recorded the tree and a dashed line to the top of the crown it was joined to. The record’s own position is unchanged in its full record; its marker on the map sits on the crown.

Records sent for review

20,704 records went to the researcher workbench and appear on the map as before: 14,760 with competing crowns to choose between, 4,783 with nothing of tree height within 3 m, 1,021 on tree-height returns no crown could be drawn on, and 140 whose position cannot be matched (for example on a roof). The workbench also asks for confirmation of every crown drawn from a record.

Records that stay without a crown

13,613 records lie where there are no release 5 crowns and no 2024 point cloud: 12,925 Ruru kauri survey records in the Waitākere Ranges and other forest, 687 Notable Trees schedule records and one OpenStreetMap record. Their full records say so. “No crown linked” does not mean a tree is absent.

What a link means

A link is spatial evidence that a record describes the tree on that crown. It does not prove that every record on a shared canopy is the same organism, and a drawn crown is a proposal until it is confirmed.