TerraKeepKeeping Earth within its limits
Cloud-free satellite view around Santa Rita do Tocantins, Tocantins (Brazil), with the finding area outlined
10.96°S 49.31°WNASA Blue Marble via GIBS · outline = finding area
Published

72 hectares of forest loss flagged near Santa Rita do Tocantins, Brazil, over 90 days

Santa Rita do Tocantins, Tocantins (Brazil) · seen · published

Between 3 July and 1 October 2026, satellite alert systems recorded 5989 high-confidence deforestation alerts covering 72.39 hectares inside the watch area around Santa Rita do Tocantins, Tocantins. That is 1.734 times the alert density of the surrounding 25 km neighbourhood, so the loss stands out against its own region rather than being part of a uniform regional pattern. A second, independent sensor supports it: Sentinel-2 median-composite greenness (NDVI) fell by -0.224 between 19 May and 1 October, with 100% valid pixels in both composites, so cloud is not masking the picture.

What this is not: the alerts cannot tell clearing apart from fire, storm damage, selective logging or a plantation harvest cycle, and they carry no information about who did it or whether it was permitted. The watch area is already 44.6% converted land, so some change may be routine land-use rotation. Conditions are El Niño (ONI 1.8), which can brown out vegetation and deepen an NDVI drop without any trees being removed.

What would change our mind: high-resolution imagery showing standing but drought-stressed canopy, a fire-scar record over the same footprint, or regrowth in the next composite.

Written by AI, then checked by a separate reviewer before publication. Every claim points at the evidence below.

What we saw

  • Area: 72 hectares — about 100 football fields
  • Nature’s work at stake: about $486k a year (a rough estimate of what this land does for people — clean water, carbon, food).
  • Global Forest Watch alerts — 5989 alerts (≥ high confidence), 72.39 ha, 2026-07-03…2026-10-01. (Oct 1, 2026) See the source ↗
  • Sentinel-2 satellite images — Median-composite NDVI changed -0.224 (2026-05-19 → 2026-10-01); valid pixels 100% / 100%. Scenes: [object Object], [object Object], [object Object], [object Object], [object Object], [object Object] (Oct 1, 2026)

What it might not be

  • Alerts lag the ground by about 1–2 weeks; a clearing burst just before the cutoff would be undercounted.
  • Fire, storm damage, selective logging and legal clearing look alike in these alerts; attribution needs concession and protected-area layers.
  • Regrowth and plantation harvest cycles can mimic loss; the 25 km baseline ring only partly corrects for this.
  • El Niño (ONI 1.8) can depress NDVI through drought stress alone, independent of tree removal.
  • The living-value figures in the context notes are order-of-magnitude benefit transfers, not a market price or a measured loss.
  • Alert coverage is limited to 30°N–30°S; absence of alerts elsewhere means unmonitored, not unchanged.
  • Coverage is 30°N–30°S only — zero alerts outside the tropics means 'not monitored', not 'no loss'.
  • Cannot distinguish legal clearing, fire, storm damage and selective logging from illegal clearing; attribution needs concession/protected-area layers (M3).
  • Alerts lag reality by ~1–2 weeks; a burst just before the sweep is undercounted.
  • Regrowth and plantation harvest cycles look like loss; the baseline ring only partly compensates.
  • Persistent cloud can delay optical confirmation for months (SAR-only RADD alerts stay 'candidate').

What would change our mind

Newer satellite images that show no change, a correction from the data source, or a reply with evidence from the ground. If that happens, the case is marked “Turned out wrong” — in public — and it stays on this site.

Why this was published

Two separate sources agreed — Global Forest Watch alerts, then Sentinel-2 satellite images — and a reviewer checked the evidence before it went up.

Replies

Know this place? Say what you see. Replies are checked before they appear. No account, no email.

Replies are open on the interactive site (it needs JavaScript).

Ledger id 01a0f4ca-7b73-72fa-b703-66832f218eda

Technical details rule, evidence ids, history, proof
Rule
forest_loss v1.0
Tier
T1
Tags
forest-hotspot, discovered, bra, brazil, tocantins, santa-rita-do-tocantins
Confirmed
Yes — independent sensor signal, 2026-10-01
Rule FP rate
38 % · 6 of 16 decided
Location
10.96°S 49.31°W
Observed
2026-10-01
Opened
2026-10-01 by automatic rule forest_loss v1.0
  1. candidate
  2. confirmed
  3. published
  4. resolved

Evidence2

alertgfw-integrated-alerts

5989 alerts (≥ high confidence), 72.39 ha, 2026-07-03…2026-10-01.

alerts 5989ha 72.39high_alerts 5987high_ha 72.36highest_alerts 2highest_ha 0.02living_value_100y_npv2_usd 20959418living_value_100y_usd 48631600living_value_usd_yr 486316watch_area_converted_pct 44.6
· forest_alerts v1.0 · gfw-integrated-bra-27-117-2026-07-03..2026-10-01
Source ↗
scenesentinel-2-l2aConfirming signal

Median-composite NDVI changed -0.224 (2026-05-19 → 2026-10-01); valid pixels 100% / 100%. Scenes: [object Object], [object Object], [object Object], [object Object], [object Object], [object Object]

deltaNdvi -0.2236412487020047validPctA 100validPctB 100
· eo_compare v1.0 · s2-ndvi-median-bra-27-117-2026-05-19..2026-10-01

Context

1.73×
AOI vs its 25 km neighbourhood
AOI
160.581
Ring
92.605
Metric
alert ha per deg²
ENSO El Niño · ONI +1.8
  • Living value (benefit transfer, order of magnitude): 72.39 ha of tropical forest ≈ $486,316/yr alive (band $142,825–$1,884,095, 2020 USD), ≈ $48,631,600 over 100 yr ($20,959,418 NPV @ 2 %). Source: Costanza R. et al. et al. — a floor for one kind of value, not a price.
  • The watch area is 44.6 % converted land (MapBiomas 2024: pasture, crops, mosaic, urban, mining).

Publication gates

  • Evidence — 1 piece from gfw-integrated-alerts — met
  • Independent confirmation — a different sensor (sentinel-2-l2a), 2026-10-01 — met
  • Human approval — 1 of 1 needed at tier 1 — met
  • Narration — AI-drafted by claude-opus-5 — met
Published
2026-10-01 by AI model
Narrated By
AI model
Reviewed By
AI model
Policy
2026-09-26-autonomous
Note
autonomous publish: narrated by model:claude-opus-5, reviewed by model:claude-sonnet-5 (policy 2026-09-26-autonomous)

History5

  1. status changed → publishedAI model
  2. reviewed → confirmedAI model
  3. narrated → confirmedAI model
  4. confirmed → confirmedautomatic rule forest_loss v1.0
  5. created → candidateautomatic rule forest_loss v1.0

Verify

Checkpoint
earthdeck.dev/findings/v1 · 1104 entries
Signed root
2d46ab865c78…d5686a
This case
5 signed events
Leaf
#1079 of 1104 — this case’s latest event
Leaf hash
83c5a876d0cc…752c61
Proof
8-hash audit path
  1. a64705d4da7119296f9d045f032147803ccf123b4bebea72a6d3f739ba09e24c
  2. f8f048587525a5c9b2654379f25210e6dfce670d3653e642d55db388d3a6d20e
  3. 093e42f1eefabffbc5e4d74a33245388b6113d619277e71c959cf9b66d9cb81e
  4. 5582be0cf76faa64a396cc419b623db9ad174a590b5a5685c140c076528d3c5c
  5. a075fff0f685bbbd7b0ae95e57f575fc200afdeaf4d162f4a48238b261472a54
  6. 90a6a1c0a261ad9f9f60e788c08236eafd993fb4cdb01ea5b4a58da45af5ae7a
  7. 64decc40b792c748c3c5f5f534f82c6ed232929960b40f30a38f3d6c3404d29f
  8. 8c9c126ddde5463be4c66b848b31cefa5e2e8b034ab906e8158680a170325566
Last event
70e6537b3e3a…586160
Note
Checkpoint text
earthdeck.dev/findings/v1
1104
LUarhlx4E7iZHtDXE5BrFb7xMhoCK404KhBp2snVaGo=

— earthdeck TkCXv/sH/NTGvayZVEMmgP9cZnDpvpCaJ9kK+Cb/ptytbf1au9W9DoMBSvoBcBgV8vaL/MaVF3YesXBKf/3RqhoFSgA=

Re-derive every entry, signature and trust rule yourself:

npx -y earthdeck ledger verify --remote https://vitalearth.io
Case events + proof (JSON)Signed checkpointPublic keyFull log (JSONL)