From Peatlands to National Parks: The 2026 Trail of Fire and the Responsibility Behind the Word “Sorry”

From Peatlands to National Parks: The 2026 Trail of Fire and the Responsibility Behind the Word “Sorry”

MADANI Berkelanjutan’s monitoring recorded 4.00 million hectares of Indicative Burned Area (AIT) in January–October 2026. Read the full analysis and demands for accountability.

MADANI Berkelanjutan’s monitoring recorded 4.00 million hectares of Indicative Burned Area (AIT) in January–October 2026. Read the full analysis and demands for accountability.

No headings found on page

An apology from Manila and unanswered questions

“It’s not our fault, but we apologize.” Those words came during President Prabowo’s speech while attending the 49th ASEAN Leaders' Summit in Manila, the Philippines, on November 15, 2026. The Indonesian President officially apologized to Malaysia and Singapore for the haze from forest and land fires that spread to both countries.

At the ASEAN Summit, Indonesia appeared to be sitting in the “hot seat.” Under the ASEAN Agreement on Transboundary Haze Pollution, Indonesia is obligated to prevent, monitor, and mitigate forest and land fires (karhutla) along with the transboundary haze they generate. According to the President, Indonesia “has done its best.” Indonesia has deployed 71,000 forest firefighters and revoked 17 corporate concession permits. In addition, 11 corporations have been charged, 26 companies are under investigation, 23 companies are slated to be charged, and 104 companies are being investigated by the police.

The President’s statement emphasized response measures taken once fires had already grown large. Yet the questions that need to be examined are far more fundamental: why do landscapes under a protection mandate continue to record vast indicative burned areas? Furthermore, has the protection of citizens—especially the most vulnerable, such as pregnant women, children, older persons, and persons with disabilities—been treated as a benchmark equally important to firefighting and law enforcement?

Monitoring by MADANI Berkelanjutan recorded 4,004,540 hectares of Indicative Burned Area (AIT) across Indonesia from January to October 2026. A total of 1,603,988 hectares of AIT was detected in October alone—the highest monthly figure since January.

Four million hectares: a figure that must be read carefully

MADANI Berkelanjutan’s Indicative Burned Area (AIT) model is based on the delineation of high-confidence hotspots combined with satellite imagery. These figures are indicative, rather than official burn figures verified on the ground. The latest series analysis does not count areas burned more than once: if the same location is detected as burning over multiple months, its area is counted only in the first month.

In this series, 95.3 percent of the January–October AIT was detected within just three months, namely August–October. The monthly area surged from 777,411 hectares in August to 1,435,832 hectares in September and 1,603,988 hectares in October.

Figure 1. Area of AIT first detected each month and its cumulative percentage. Source: MADANI Berkelanjutan, Indicative Burned Area (AIT) Update January–October 2026.

Spatial dynamics reveal a concentration of fires in Kalimantan, Papua, and southern Sumatra. Central Kalimantan recorded the largest indicative burned area at 1,123,030 hectares, followed by South Papua with 582,350 hectares and West Kalimantan with 522,817 hectares. At the regency level, AIT in Merauke reached 411,633 hectares. Although the indicative burned areas were relatively concentrated, the reach of haze exposure extended across national borders (see Figure 2). According to a report by the Centre for Research on Energy and Clean Air (CREA), the number of Indonesian citizens exposed to unhealthy to deadly air in September alone reached 58 million people, not including citizens of neighboring countries such as Singapore, Malaysia, Brunei Darussalam, Timor-Leste, and Australia.

Figure 2. Visualization of carbon emission distribution from forest and land fires in Indonesia and surrounding regions for the October–November 2026 period (Source: Copernicus Atmosphere Monitoring Service (CAMS), Fire Emissions Watch, 2026.)

The trail of fire in areas meant to be protected

A total of 2,932,908 hectares, or 73.2 percent of the indicative burned area, lies within forest areas. Within this, 587,840 hectares were recorded in conservation areas and 386,039 hectares in protection forests. The proportion of indicative burned area within conservation areas rose from 7.1 percent in August to 19.4 percent in October. This demonstrates that legal status alone is insufficient to ensure effective protection, prevention, and control on the ground.

Specifically, National Parks recorded an indicative burned area of 306,579 hectares. Roughly 156 thousand hectares of this occurred in Tanjung Puting National Park, followed by Wasur (±42 thousand), Berbak Sembilang (±40 thousand), Sebangau (±29 thousand), and Lorentz (±16 thousand). A total of 187,561 hectares, or 61.2 percent of the AIT in national parks, was detected in October alone. This pattern underscores the immense pressure placed on conservation areas during the final month of observation.


Figure 2. Five national parks with the largest AIT and distribution by month of detection. Source: MADANI Berkelanjutan, Indicative Burned Area (AIT) Update January–October 2026.

Fires in National Parks are not merely an issue of forests and biodiversity, but also of peatlands that release massive amounts of carbon emissions. In Tanjung Puting, roughly 94 percent of the AIT overlaps with peatlands. In Sebangau, the entire indicative burned area was detected within peatland ecosystems.

Nationally, MADANI’s monitoring identified 1,574,353 hectares of indicative burned area within peatland ecosystems, including 928,828 hectares in peatland protection zones and 640,290 hectares in peat domes, which are meant to receive the highest level of protection. A recent study by World Weather Attribution released on November 13 concluded that decades of peatland clearing and drainage have created vulnerable conditions that trigger severe, difficult-to-control fires when peatlands burn amid a strong El Niño.

Natural forests were not spared from the flames either. Approximately 1,027,106 hectares of AIT occurred within 2024 natural forest cover, including roughly 142 thousand hectares of primary forest. AIT findings in natural forests do not automatically prove permanent deforestation, yet they are specific enough to warrant field verification and post-fire land-cover change monitoring. The government's statement that burned forest and peatland areas must be reforested must be substantiated through transparent public reporting on how much area burned, what burned, and what subsequently emerges on those burned lands. This is critical because the public has long lacked free and open access to detailed forestry and land data, as well as fire data and restoration plans.

Figure 3. Land-cover composition within AIT across five national parks. Source: MADANI Berkelanjutan, Indicative Burned Area (AIT) Update January–October 2026.

Law enforcement is important, but who oversees the permits that burned?

In his speech at the ASEAN Summit, the President outlined the revocation of concession permits and legal proceedings against several companies. MADANI’s AIT analysis shows that 1,647,387 hectares, or 41.1 percent of the AIT during the January–October 2026 period, overlapped with oil palm permits, Forest Utilization Business Permits (PBPH), and mineral, coal, and oil and gas concessions. Roughly 532 thousand hectares of AIT were detected in oil palm plantation permits; 521 thousand hectares in PBPH; 250 thousand hectares in oil and gas concessions; and 127.5 thousand hectares in mineral and coal mining concessions. The detected indicative burned area is even larger when accounting for overlapping permit boundaries.

Geospatial overlap does not automatically prove that permit holders deliberately set fires within their concessions, yet it provides a strong basis for conducting investigations and licensing audits grounded in the principle of strict liability, which obligates permit holders to prevent and control fires within their concession areas. Once their territory burns, permit holders must be held accountable. Furthermore, the authorities that issued these permits must also answer to the public regarding their oversight of permits whose areas have burned. Relevant authorities must explain to the public what they will do differently moving forward to prevent a recurrence of the same crisis.

Transboundary haze and the responsibility behind the word “sorry”

The ASEAN Agreement on Transboundary Haze Pollution provides a regional cooperation framework to prevent, monitor, and address transboundary haze. By mentioning Malaysia and Singapore in his speech, the President acknowledged forest and land fires (karhutla) as an international matter. However, the President insisted that the fires were “not our fault.”

Herein lies the distinction between fault and responsibility. Law enforcement against corporations that ignite or allow fires to burn must proceed. In addition, the government bears the responsibility to evaluate whether systems for prevention, monitoring, health protection, and regulatory enforcement have functioned effectively. Therefore, the statement “not our fault” should not end the conversation. An apology may serve as the starting point for diplomatic accountability, yet the public has the right to know what concrete measures the government is preparing to reduce karhutla risks in the coming years. This is increasingly urgent as fire risks in Indonesia are exacerbated not only by El Niño, but also by a worsening climate crisis.

Behind the acreage figures: who is most vulnerable to the haze?

The President’s speech in Manila noted that 16 people died and many personnel were injured while fighting the fires. However, the humanitarian crisis resulting from karhutla should not be viewed solely through the casualties among those battling flames on the front lines. Exposure to fine particulate matter from smoke, particularly PM2.5, poses severe health risks whose impacts often manifest only later in life, such as heart attacks. Children, older persons, pregnant women, and individuals with cardiovascular or respiratory conditions are far more vulnerable when air quality deteriorates.

Persons with diverse disabilities are frequently left furthest behind when karhutla strikes and haze blankets communities. Early warning systems that rely solely on sirens are inaccessible to Deaf citizens; visual information lacking accessible formats creates barriers for citizens with visual impairments; and inaccessible evacuation sites restrict citizens with mobility impairments. Moreover, persons with psychosocial or mental disabilities often have their specific needs overlooked and unrecognized. Yet the UNDRR emphasizes the importance of involving persons with disabilities in disaster risk reduction so that preparedness does not protect only a fraction of the population.

Gender inequality also widens the impacts experienced by women during karhutla. Women with caregiving responsibilities often shoulder additional burdens during clean water crises, when family members fall ill, or when children’s schooling is disrupted. Outdoor workers (such as street vendors) and low-income households are forced either to halt their livelihood activities or to continue working without adequate protective gear such as masks. Indigenous peoples and local communities whose livelihoods depend on forest, coastal, and aquatic ecosystems also face severe disruptions to their means of subsistence.

During the Women's Voices for Climate Justice Workshop—co-organized by MADANI, the Indonesian Women's Coalition (KPI), the Indonesian Association of Women with Disabilities (HWDI), and the Center for Women's Resources Development (PPSW) on November 16–18—various impacts were documented, including the destruction of thousands of cacao trees that served as the savings and investments of Indigenous women in the Seram Islands, Maluku, consumed by the flames.

The IPCC demonstrates that climate vulnerability is shaped by socioeconomic inequality and marginalization, and that vulnerable groups require disaggregated risk assessments. To date, however, no comprehensive map exists to verify the numbers, locations, and specific needs of vulnerable groups. Indonesia also lacks a system that consistently tracks the multi-year impacts borne by women, pregnant and lactating women, persons with disabilities, children, and older persons as a result of karhutla and haze crises.

Consequently, the measure of success in handling karhutla should not merely be how many personnel are deployed or how many corporate cases are processed, but what percentage of vulnerable groups receive adequate protection and recovery assistance. Vulnerable groups are not merely recipients of aid during disasters, but rights-holders who must be involved from the planning and prevention stages onward.

Five benchmarks to test state accountability

First, disclose consistent data. Publish official maps and progression data on karhutla to enable evaluation and methodological comparison.

Second, focus prevention on protecting high-risk landscapes. To prevent recurring fires, the government must protect and restore all peatland ecosystems, not solely peatlands categorized as protection zones or peat domes. Over 1.5 million hectares of burned peatland ecosystems must be prioritized for restoration by the National Peatland and Mangrove Management Agency (BPGMN). All remaining natural forests must be protected from deforestation and degradation that heighten karhutla risks, including secondary natural forests, natural forests outside designated state forest areas, and natural forests located inside permits or concessions.

Third, disclose law enforcement outcomes transparently and measurably. Release detailed updates on case stages and follow-up actions within burned permit areas, including clear restoration obligations with measurable timelines. Revoke the permits of companies whose concessions burn repeatedly.

Fourth, measure the protection of vulnerable groups. Strengthen PM2.5 monitoring and healthcare services. Provide multi-format early warnings for karhutla and air pollution alongside accessible evacuation systems. Establish protocols to engage Indigenous peoples, women, and persons with disabilities in disaster preparedness and response, and allocate sufficient budgets for regional Disability Service Units.

Fifth, track impacts after the flames are extinguished. Update the damage status of fire-affected peatlands and changes in forest cover. Publish restoration plans and progress statuses so they can be monitored by the public. Track smoke-related health impacts for several years following the fires, including disaggregated data for vulnerable groups.

An apology to neighboring countries, however half-hearted, is part of diplomacy. Yet the scope of accountability extends far beyond that. The state is responsible for protecting remaining natural forest and peatland landscapes, restoring what has burned, strengthening oversight and conducting licensing audits, and protecting the vulnerable groups who bear the heaviest impacts.

The remaining question is not merely who is to blame for the haze, but who survives when the haze arrives.

Download the January–October 2026 Indicative Burned Area infographic (Page 1)

About the Indicative Burned Area Model

Methodology of the Indicative Burned Area (AIT) Model

The Indicative Burned Area (AIT) model estimates the extent and location of forest and land fires (karhutla) in Indonesia using hotspot data, developed by MADANI Berkelanjutan since 2019. The term “indicative” is used to distinguish AIT from field-verified burn scar maps delineated from satellite imagery. AIT serves as a rapid estimate indicating where and how extensively fires are indicated to have occurred. It complements official tabulated karhutla area data from SiPongi (KLHK) by providing stakeholders with a near real-time overview of fire progression.

Analysis stages:

  • Data preparation: Retrieving VIIRS hotspots, filtering by confidence level, and screening out hotspots not originating from fires.

  • Modeling: Converting cleaned hotspot data into indicative burned area footprints.

  • Validation and output: Cross-checking results against Sentinel-2 imagery and SiPongi data, then presenting them through visualizations and analyses.

2. Data Used

The AIT model uses one primary input dataset (VIIRS hotspots) and three supporting datasets for screening and validation.

Data

Role in the Model

Description

VIIRS Hotspots

Primary input

Includes only nominal and high confidence classes. VIIRS pixel resolution is approximately 375 m.

Reference layer of non-karhutla heat source areas

Screens out hotspots not caused by forest and land fires, such as volcanoes, smokestacks, and other sources

Covers urban areas, industrial utilities (factories, stockpiles, construction zones), and active volcanic crater complexes.

Sentinel-2 Imagery

Visual verification of sample AIT locations

Spot-checks AIT model results against visible fire and smoke. With a 10–20 m resolution, it is sufficient to detect burn scars.

Tabulated karhutla area data

Benchmark for AIT acreage by province

Compared over the same period using accumulated AIT versus SiPongi (KLHK) data.

3. AIT Model Workflow

The AIT analysis is conducted in nine stages.

Stage 1. Hotspot data collection

VIIRS hotspot data for the reporting period are collected daily and compiled weekly. Each point carries attributes for coordinates, date and time of detection, and confidence class. Only hotspots with a confidence level above normal are used in the AIT model.

Stage 2. Filtering by confidence level

Only hotspots with a confidence class above normal are retained. Low-confidence hotspots are discarded to prevent false detections from being counted as fires.

Stage 3. Hotspot screening (hotspot filtering)

This stage ensures that only hotspots originating from vegetation fires enter the model. Hotspots are overlaid with a reference layer of non-fire heat source areas, and any hotspot falling within those areas or within a surrounding buffer distance is removed.

Three categories of areas are screened out:

  • Urban areas, as heat from settlements and urban activities can be detected as hotspots.

  • Industrial utilities, namely factories, stockpiles, and construction sites that emit heat from production activities rather than land fires.

  • Active volcanic crater complexes, as volcanic activity generates thermal anomalies similar to fire.

First decision point. The screening results are re-examined: are there still any hotspots falling within the exclusion areas? If so, the screening is repeated, for example by refining the reference layer or buffer distance. If none remain, the data proceed to the next stage.

Stage 4. Clean hotspots

The output of Stage 3 is a set of clean hotspots—thermal points believed to originate from forest and land fires. This dataset serves as the model input.

Stage 5. AIT model input: clustering and polygon generation

This stage converts point data into area footprints.

  • Clustering. Clean hotspots located in close proximity are grouped into a single cluster. The assumption is that adjacent hotspots originate from the same fire event and mutually corroborate the presence of a burned area.

  • Polygon generation. Each cluster is formed into a polygon enclosing the distribution of hotspots within it. This polygon is referred to as the Indicative Burned Area. Based on observations since 2019, this is most effective within a radius of less than 1 km.

  • Area calculation. The area of each polygon is calculated in hectares and then summed by province and reporting period.

Stage 6. Dual validation

Model outputs are validated using two concurrent approaches.

  • Sample verification with Sentinel-2. A number of AIT polygons are selected as samples for spot-checking and matched against Sentinel-2 imagery at the corresponding location and timeframe. A polygon is declared verified if the imagery shows a burn scar within it. This spot-check represents polygon size classes across several hectare brackets.

  • As one of the closest available tabulated karhutla datasets released, SiPongi data are statistically compared against AIT data. Accumulated AIT acreage by province is compared side-by-side with SiPongi’s tabulated karhutla area for the same period. The closeness of fit between the two is measured using the correlation coefficient ($r$) and coefficient of determination ($R^2$) on a scatter plot, with one data point representing each province.

Stage 7. Validation of AIT polygon shapes

Shape adjustments and threshold settings are applied to the resulting AIT model outputs. Any irregular AIT polygon shapes undergo re-verification testing. Furthermore, based on observations since 2019, fragmented polygons with an area of less than 3 hectares are excluded. These areas are categorized as model residuals because the intensity of clean hotspots is too low to warrant inclusion in the final AIT model results.

Stage 8. AIT output

Outputs that pass validation consist of indicative burned area polygons along with their total acreage per province. Outputs are released with a lag of approximately one week from the reporting date.

Stage 9. Visualization and analysis

AIT outputs are presented in distribution maps, temporal acreage trend charts, and inter-provincial comparisons. Additionally, AIT analysis is overlaid with 18 thematic spatial layers to add context and enrich the analytical perspective of MADANI Berkelanjutan’s karhutla reporting. Ultimately, these results serve as a reference for stakeholders to monitor the latest karhutla developments.

4. Validation results

The $R^2$ test between AIT and SiPongi consistently remains above 0.8. This indicates that the model is sufficiently accurate to track karhutla trends.

Period

Correlation (r)

R²

Notes

January to July 2026

0.90

0.82

SIPONGI 202,000 ha vs AIT 229,000 ha for the same period

January to September 2025

0.91

0.75

AIT vs SIPONGI by province

January to October 2024

0.92

0.85

AIT vs SIPONGI by province

Beyond acreage alignment, trend graphs across the periods demonstrate that AIT fluctuations mirror those of SiPongi, including spikes during the peak of the 2023 fire season. In 2026, verification using Sentinel-2 imagery was conducted for February, June, and July. Sample results show AIT polygons overlapping directly with burn scars visible in the imagery.

5. Model limitations

AIT is an estimate of ongoing and past burned areas, rather than a final measurement of burn scars. Several considerations should be noted:

  • Reliance on satellite detection. Fires obscured by thick clouds or smoke, occurring outside satellite overpass times, or burning underground (such as in peatlands) may go undetected as hotspots.

  • Polygon boundaries are indicative. Polygons are generated from point distributions, meaning their boundaries may be broader or narrower than the actual burn scars.

  • Screening quality depends on reference layers. Newly established industrial zones or settlements not yet recorded in the reference layer may pass through as fire hotspots. Consequently, screening reference layers must be continuously updated over time.

  • SiPongi validation is retrospective. Comparisons with SiPongi tabulations are frequently hindered by delays in official data updates. Validating AIT against other thematic datasets has not yet been possible due to limited publicly accessible data.

  • Sentinel-2 validation is sample-based. Due to constraints in access and resources, verification covers only a subset of polygons and can be impeded by cloud cover in satellite imagery.

AIT Validation vs. Sentinel Imagery



Related to read