247,000 satellite alerts, 200 carbon projects: the uneven map of Indonesian REDD+ risk

KarbonLens has cross-referenced 247,000+ high-confidence forest-loss alerts from Global Forest Watch's Integrated Alerts layer against the polygon buffers of every Verra-registered Indonesian carbon project. The headline finding is not that Indonesian projects are bleeding alerts. It is that the bleeding is wildly uneven.

The forest-loss-pressure profile across the country's REDD+ portfolio varies by more than two orders of magnitude. Some projects sit on contested peat and mangrove frontiers where the satellites flag fresh clearing every week. Others are quiet. A buyer treating Indonesian REDD+ as a single asset class is buying noise.

How the dataset was built

The alert layer comes from GFW's Integrated Alerts product, which fuses three independent detection systems: RADD (Sentinel-1 radar, cloud-penetrating, near-real-time), GLAD-S2 (Sentinel-2 optical at 10m resolution), and GLAD-L (Landsat optical, 30m). When two or more systems agree, the alert is flagged high-confidence. GFW reports the integrated layer detects 97% of disturbance events that any single system would catch, with shorter lag than any single sensor.

KarbonLens runs a polygon-with-buffer spatial join: each Verra project's registered boundary, plus a configurable buffer ring to catch edge-leakage. Full details live in /data-sources and /methodology. The 90-day count we report is the count of integrated alerts whose centroid falls inside either the polygon or the buffer.

That distinction matters, and we will come back to it.

The top 10

These are the Indonesian Verra projects with the highest 90-day alert counts as of this week.

*Agroforestry & Reforestation: the 515ha figure is the registered planted area; alerts almost certainly include nearby clearing inside the buffer ring rather than on planted parcels.

The contrast that jumps out: the Livelihoods mangrove project shows 18.5 alerts per hectare in 90 days. Jati Dharma Indah, also in the top 10 by total alerts, shows 0.066. That is a 280-fold difference in alert density between two projects on the same shortlist.

What 18.5 alerts per hectare means

It does not mean the Livelihoods project has failed. It means the satellites are picking up disturbance signals at a rate of one flagged event per 540 square metres of project footprint, every quarter. Mangrove pixels are noisy: tidal flushing, sediment plumes, aquaculture pond rotation, and the silvo-fishery model Livelihoods runs with Yayasan Gajah Sumatera all generate radar and optical signatures that integrated alerts can flag as canopy loss. The same is true of the PURE PROJECT agroforestry footprint: small registered parcels with high-cadence land-use change around them will look like a wildfire on the alert map.

The honest reading is that alert density on small-footprint blue-carbon and agroforestry projects is partly an artifact of the geometry, not a direct proxy for project failure. The KarbonLens reversal-risk sub-score, which contributes 35% to the integrity composite, weights for project type and biome precisely for this reason.

The three Central Kalimantan REDD+ projects on the list — Kapuas Pisau, Gunung Mas, Muara Teweh — are a different story. Their alert densities (0.68, 0.22, 0.037 per hectare) sit in the range where you would expect genuine deforestation pressure, not measurement artifact. Central Kalimantan has the highest concentration of large-area REDD+ in the country and the most consistent recent track record of fire and clearing, much of it palm-oil-adjacent.

Kampar Peninsula, the APRIL-anchored project on Riau peat, is the one whose alert profile maps most cleanly onto the existing public controversy. A March 2026 investigation in The Diplomat argued the project's baseline assumes a deforestation threat that has not materialised since 2014. Verra raised 24 issues in its 2023 Project Review Report and accepted the project's revisions without changing the baseline claim. 10,743 alerts in 90 days on 14,723 hectares is the kind of number that will keep that argument alive.

What the data does not prove

Alerts inside a polygon do not equal project-caused deforestation. The forest could have been lost without the project. The project may not have prevented it. Some alerts are false positives from sediment, fire-recovery, or selective logging the project may itself have authorised. The buffer ring catches edge clearing that is genuinely off-project.

This is the core epistemic constraint of any satellite-to-polygon join, and it is why modern Verra methodologies bake in leakage assessment and buffer-pool insurance. VM0048 requires dynamic baselines and explicit leakage accounting. VM0007, the older methodology under which the majority of Indonesia's issued REDD+ credits were generated, did not.

The methodology layer

The ICVCM decided in late 2024 that VM0007 credits are not eligible for the CCP label. Verra excluded VM0006, VM0007, VM0009, VM0015 and VM0037 from its ICVCM assessment when applying. As ICVCM's own release makes clear, these methodologies "account for the majority of REDD+ credits on the market and around a quarter of all carbon credits retired in 2023." Verra now requires operational unplanned-deforestation projects to transition to VM0048, with requantification.

That requantification is where the satellite data starts to bite. A project transitioning from VM0007 to VM0048 has to demonstrate its baseline against a jurisdictional reference level. If the integrated alerts in the surrounding jurisdiction show declining or flat deforestation, the requantified baseline drops. If the alerts inside the polygon are elevated, the project's net claim shrinks further.

Of the Indonesian top-10 by alert count, three are unplanned-deforestation REDD+ at the scale where VM0048 transition will be material: Kampar, Gunung Mas, and Kapuas Pisau. Muara Teweh and Jati Dharma sit at scales where the same logic applies but the dilution is lower. The hydropower and gas-recovery projects (Sipansihaporas, Bina Pitri, MedcoEnergi) use different methodologies entirely and are not REDD+; their alert counts almost certainly reflect buffer-ring noise from adjacent road clearing or industrial activity, and we would not weight them in a reversal-risk read.

The buffer pool question

Verra holds back a percentage of every land-based project's issuance into a global buffer pool, to be canceled if reversals occur. The deduction is set by the AFOLU Non-Permanence Risk Tool and runs 10-40% for REDD+, with most Indonesian projects sitting in the 10-30% range depending on permanence assessment.

The question worth asking: is a 10-30% buffer calibrated for projects experiencing the alert densities we see at Kampar or Kapuas Pisau? Sylvera's own policy consultation response on Verra's Long-Term Reversal Monitoring System argued the buffer percentages need recalibration against actual reversal observation. The Indonesian alert data is exactly the kind of input that recalibration needs.

The buyer due-diligence implication

If you are a corporate ESG buyer or a foreign Article 6 buyer running diligence on Indonesian REDD+, the alert-density layer changes the shortlist.

First test: is the project transitioning to VM0048 or still issuing under VM0007? If the latter, the credits cannot carry the CCP label and that constrains the buyer pool. The cumulative 22.29 million tCO2e of VCUs issued across Indonesia's Verra portfolio is heavily weighted toward older-methodology vintages.

Second test: what is the project's 90-day alert density, normalised against the local jurisdictional reference level? A REDD+ project showing higher alert density than its surrounding jurisdiction has a leakage problem or an effectiveness problem. A project showing lower density is doing something the satellites can see.

Third test: where do the alerts sit geometrically? Inside the registered polygon, or in the buffer ring? KarbonLens publishes both. The polygon-only count is the harder test.

Fourth test: what does the rater see? Sylvera and BeZero both rate Indonesian REDD+; Sylvera's own framework reports that fewer than a third of REDD+ projects globally rate as high quality, and BeZero rates more conservatively still, with around 8% earning top marks. A project that fails the alert-density test and carries a sub-A rating from one of the two main agencies is, at best, a hedge.

What Indonesian policy did and did not change

The October 2025 Perpres 110/2025 lifted the moratorium on international voluntary carbon credit trading and explicitly enabled Corresponding Adjustment for SPE-GRK transfers. Permenhut 6/2026, the implementing forestry regulation, set the procedural framework for issuance and international trade of forestry carbon units.

What the new framework does not yet do is move the alert-density picture. The 247,000 alerts we have mapped post-date both regulations. Whether the new safeguard mechanisms in Permenhut 6/2026 close the gap between paper performance and satellite-observed disturbance is a 2026-2027 question, not a 2026 answer.

What the follow-up data project needs

This investigation is the first cut. The data project that needs to happen next is project-by-project polygon validation: for each top-10 project, the polygon and buffer geometries pulled from Verra registry filings, the alerts inside polygon-only stripped out from buffer-only, and a sample of the high-density alert clusters ground-truthed against Planet or Sentinel-2 true-colour scenes.

Two things will fall out of that work. One: a defensible polygon-only alert density per project, which is the number that should sit in a buyer's diligence pack. Two: a list of the alert clusters that warrant Mongabay-style on-the-ground reporting. Both Kampar and Kapuas Pisau already have public reporting; Muara Teweh and Gunung Mas do not, in proportion to their alert load.

The dataset that anchors all of this — 247,000 alerts, 200+ projects, 73 cumulative top-10 buffer-zone matches over the rolling year — is public on KarbonLens, project by project. The story those numbers tell is not that Indonesian REDD+ is failing. It is that the variance between projects is so wide that treating the country as one bet makes no analytical sense.

Ringkasan

KarbonLens telah memetakan lebih dari 247.000 peringatan kehilangan tutupan hutan dari Global Forest Watch ke poligon dan zona penyangga setiap proyek karbon terdaftar Verra di Indonesia. Temuan utamanya bukan bahwa proyek-proyek REDD+ Indonesia gagal secara kolektif, melainkan bahwa kepadatan peringatan satelit di antara proyek-proyek tersebut sangat tidak merata, bervariasi lebih dari dua orde besaran.

Proyek restorasi mangrove Livelihoods Fund di Aceh dan Sumatera Utara mencatat 18,5 peringatan per hektare dalam 90 hari, sebagian karena geometri lahan kecil dan dinamika lanskap pesisir. Sebaliknya, proyek Jati Dharma Indah di Papua Tengah mencatat 0,066 peringatan per hektare. Tiga proyek REDD+ di Kalimantan Tengah — Kapuas Pisau, Gunung Mas, dan Muara Teweh — menunjukkan kepadatan peringatan yang mengarah pada tekanan deforestasi nyata, bukan sekadar artefak pengukuran.

Konteks regulasinya: ICVCM telah memutuskan VM0007 tidak memenuhi syarat CCP, sementara VM0048 memenuhi syarat. Verra mewajibkan proyek menghindari deforestasi tak terencana untuk bertransisi ke VM0048 dengan rekuantifikasi baseline. Perpres 110/2025 dan Permenhut 6/2026 membuka pasar internasional, tetapi belum mengubah profil peringatan satelit. Pembeli korporat dan pembeli Article 6 perlu memeriksa kepadatan peringatan, geometri poligon, dan peringkat Sylvera atau BeZero sebelum membuat komitmen.

Sources

1. ICVCM — Three REDD+ methodologies approved 2. Global Forest Watch — Integrated Deforestation Alerts 3. The Diplomat — How an Indonesian pulp giant built a carbon project 4. Livelihoods Funds — Yagasu Indonesia partnership 5. Sylvera — Policy consultation on Verra's reversal monitoring 6. Sylvera — Carbon credit buffer pools 7. IKI Indonesia — Perpres 110/2025 revitalizes the VCM 8. ASEAN Carbon Alliance — Permenhut 6/2026 industry analysis