ALTO map Methods

Methods and release notes

Reviewer-confirmed crowns and palm-type crowns

Release
6.2
Published
Previous release
6.1 · 13 Oct 2026
Next release
6.3 · 17 Oct 2026
Overview
How ALTO is made
Dataset ID
metro-2026.10.13-pointcloud-v6.2

This release applies reviewer confirmations to 326 records, shows 16,226 crowns that release 6 left undetermined as palm-type, describes the Change since 2013 layer in the release metadata, and no longer draws the near-canopy polygon under the crown release 6.1 promoted. Tree numbers, positions and crown outlines are those of release 6.1.

What changed

  • Reviewers confirmed 326 records as the trees their crowns show. 317 are on 278 crowns drawn from a record in release 5.2 (236 tree crowns and 42 low-canopy crowns). 9 are on trees release 6.1 changed on a reviewer decision: the 5 crowns whose outline a reviewer drew (C00038, C00040, C00067, C00151, C00172), C00307, C00239, C00320 and C00201. A confirmed crown has a solid outline and normal fill and no longer reads “awaiting confirmation”; its link says a reviewer confirmed it is the record’s tree (2024 aerial imagery and LiDAR).1
  • The palm detector’s display threshold is lowered from 0.5 to 0.2 for crowns release 6 left “growth form not yet determined”. 16,226 of them now read “Likely palm or cabbage tree (palm-type crown seen in 2024 aerial imagery)”, and 118,634 crowns show a palm-type growth form (palm, cabbage tree or tree fern), against 102,408 in release 6.1. Every other crown keeps its release 6 display. This scope is conservative; the palm check gives the reason.2
  • The release metadata describes the Change since 2013 layer (version 1): its validation status, a blind photo check in which 92% of the changes shown agreed with the reference labels (88–95%; 315 changes; labels from machine-learning image-interpretation models, no field check), the figures by class and the layer’s limits.3
  • The near-canopy layer no longer draws the low-canopy polygon under the crown that release 6.1 promoted to a tree crown (C00307, tree T295FVK7). The polygon stays in the layer, marked as promoted.4

KYTE’s crown-check missions leave out every reviewed record.

Palm check

Before release, a blind check tested the lower threshold on the crowns it would newly show as palm-type. The protocol, the scoring code and the sample were hashed before any card was made. To pass, the reference labels had to call at least 0.80 of the newly shown crowns palm-type, and at least 0.80 of those on the central isthmus.5

60 newly shown crowns, 30 on the isthmus and 30 outside it, were mixed with 30 controls: 15 crowns release 6 shows as palms seen by the palm detector and 15 crowns it shows as evergreen broadleaf. The 90 cards were labelled from the 2024 aerial imagery and LiDAR profiles by a machine-learning image-interpretation model (one pass; no independent second labeller yet). The labels were hashed before the key to the sample was opened.

Table 1. Crowns the reference labels call palm-type in the palm check, 13 Oct 2026. The check passed.
CrownsDecidedPalm-typeShare agreeing with the reference labels (95% interval)
Newly shown, central isthmus30260.867 (0.703–0.947)
Newly shown, outside the isthmus30290.967
Newly shown, both areas60550.940 (0.874–0.991)
Control: palms seen by the palm detector in release 61515
Control: evergreen broadleaf in release 6141

Both areas: weighted by the eligible crowns in each area (5,105 on the isthmus, 14,328 outside). Controls: 1 evergreen crown was left undecided; the reference labels call 13 of the 14 decided other than palm-type.

All 5 newly shown crowns the reference labels call other than palm-type were crowns release 6 already shows as another class: 2 tall conifers of the Norfolk pine type at palm-detector scores of 0.20–0.22, 1 broadleaf crown and 2 hedges with a small palm.

Table 2. Newly shown crowns in the check, by their release 6 display.
Release 6 displayDecidedPalm-type
Growth form not yet determined4949
Another growth form116

Release 6.2 therefore lowers the threshold only for crowns release 6 left “growth form not yet determined”. The restriction is conservative and post hoc: it was chosen after the results were known, and it shows fewer crowns than the pre-registered rule, which passed on the full set.

Service estimates

Where a crown has no recorded species, its benefit estimates follow the growth form shown, by release 6’s rule. On the newly shown palm-type crowns this changes the estimates of 16,161 of those crowns; the other 65 have a recorded species and keep its estimate. The confirmations change no estimate.6

Table 3. Change in the release totals, release 6.1 to release 6.2.
TotalChange
Estimated annual benefits−NZ$145,973 a year
Avoided runoff−14,359 m³ a year
Stored carbon+3,162 t CO₂e

Estimated annual benefits, sensitivity scenario: about NZ$295 million a year in both releases.

Marker, schedule and crown counts are unchanged: the map shows 3,574,821 map markers, with 3,570 map markers linked to Schedule 10 (ALTO checks the Notable Trees schedule only; Auckland Council determines whether a tree is protected). Totals change for 15,367 of the 368,540 properties, each holding or listing a changed tree. As in release 6, the estimates remain sensitivity scenarios.

What did not change

  • Tree numbers: no tree, crown or record changes identity, so every number carries over from release 6.1 and none is new. The alias table and data/tree-numbers-v61.json are release 6.1’s.
  • The map: positions, crown outlines and tree numbers are release 6.1’s, as are the crowns and the LiDAR-inferred trees. 16,511 map markers change in their properties only: 285 with a confirmation and 16,226 newly shown as palm-type.
  • Full records: each equals release 6.1’s except the planned fields of 18,907 records (the 330 records of the confirmations and 18,577 records on crowns newly shown as palm-type).
  • Every other crown’s growth form, Council positions, the Change since 2013 file and KYTE’s nearby trees are as in release 6.1.

Limits

  • 248 crowns drawn from a record in release 5.2 still await confirmation; they keep the dashed outline and read “awaiting confirmation”.
  • 3 records on confirmed crowns were not reviewed, so their own link still reads “awaiting confirmation”: 2 markers whose crowns were confirmed through another record on the crown, and 1 record on a confirmed crown.
  • The palm check asked whether a crown is palm-type from above; it did not identify species. Its labels come from a machine-learning image-interpretation model in one pass; no independent second labeller has checked them. The isthmus figure rests on 30 crowns.
  • The 3,402 crowns the pre-registered rule would also have shown as palm-type, where release 6 shows another class, keep their release 6 display until a later check tests them; in this check 6 of 11 such crowns were palm-type.
  • Change since 2013: partial losses are not shown, growth and shrinkage are not validated, some likely changes are hidden, and photo dates and offsets limit the check. The metadata still marks the layer as not release-eligible.
  • Five corrections await review and are not applied.
  • Benefit estimates are not independently validated; they remain sensitivity scenarios.

Notes and sources

  1. Each confirmation is an accepted reviewer decision already in the corrections ledger: the 317 identity links decided on 10 Oct 2026, each with its evidence and SHA-256, which release 6.1 reported as already in the release. No ledger event was added; the ledger is unchanged since release 6.1. The release database’s table tree_seeded_confirmation_v62 gives each of the 330 records affected (326 confirmed) its values before and after, the correction and the evidence. The receipt lists 331 decisions applied, 17 recorded, 1 already in the release and 5 awaiting review: corrections applied in this release. ↑
  2. Threshold, scope (v6_not_determined), the check’s summary and the sentence on training labels: growth-form counts, threshold and palm check, recounted on the release 6.2 display. With the threshold at 0.5 the display rule reproduces release 6 for all 3,352,099 scored crowns. ↑
  3. Release metadata, models.tree_change: the file data/tree_change_v1.pmtiles, validation status photo_checked, the photo check with its figures by class, and the limits. ↑
  4. The near-canopy layer was rebuilt from release 5’s source with release 5’s recipe (tippecanoe 2.79, same settings), which reproduces all 82,805 live tiles byte for byte; the new file differs in 6 tiles, all at the promoted polygon. The polygon keeps every property and gains canopy_class promoted_tree_crown (its earlier class in canopy_class_v5, promoted_to_tree T295FVK7, correction C00307); the map’s near-canopy filter draws low_canopy and sub_canopy only. ↑
  5. Pre-registered on 13 Oct 2026: the protocol, scoring code and sample were hashed before any card existed, and the labels before the key was opened. The newly shown crowns were drawn from the 5,105 eligible on the isthmus and the 14,328 outside, after leaving out every training, tuning and test set and the release 6 audit sample. The cards were made with the release 6 audit’s renderer, unchanged. ↑
  6. Release 6’s rule for crowns without a recorded species, applied on the release 6.2 display; the table tree_current_services_v62 holds the 16,161 changed rows. The rule as re-implemented first reproduced 20,000 sampled release 6 rows exactly. Before publication 386 checks on the release products passed and none failed. ↑

Files

Releases