Catastrophic impact of extreme 2019 Indonesian peatland fires on urban air quality and health

Landscape fires are common in Southeast Asia, often used for land clearing. In Indonesia, particularly Kalimantan and Sumatra, vast areas of peatland underlie cleared forests, creating highly flammable and carbon-rich soils. Peat combustion significantly increases the fuel consumed and particulate matter (PM) released, resulting in severe air quality impacts, especially during El Niño-driven droughts. The resulting haze, thick with toxic PM2.5, affects local and distant populations, sometimes impacting air quality across Southeast Asia. The 1997 El Niño-exacerbated fires highlighted this hazard, leading to the ASEAN Agreement on Transboundary Air Pollution. However, air quality modeling using satellite data indicates that some of the world's worst air quality still results from these peatland fires. This study focuses on Palangka Raya, Central Kalimantan, a region frequently experiencing severe haze due to its location within a peatland area. Limited in-situ air quality data often necessitates reliance on air quality models like CAMS, whose accuracy under extreme conditions is uncertain. The research aims to improve understanding of modeled PM2.5 concentrations using a network of low-cost sensors to assess air quality variability across Palangka Raya during the extreme 2019 fire season and evaluate CAMS model performance under such conditions. The ultimate goal is to improve the accuracy of health impact assessments.

Existing literature extensively documents the health impacts of biomass burning, with studies focusing on PM2.5 from various sources, such as cigarettes and urban areas. Recent research, however, indicates that PM2.5 from biomass burning, particularly peat fires, is even more toxic than that from other sources. Peat fires, characterized by smoldering combustion, produce PM2.5 dominated by organic carbon, with a secondary black/elemental carbon component. Global studies using CAMS outputs estimate significant premature deaths annually due to landscape fire-sourced PM2.5. Studies combining air quality modeling data with epidemiological data provide estimates of excess child deaths in Indonesia, mostly related to tropical peatland burning. However, these studies mainly rely on large-scale modeled PM2.5 concentrations, and the accuracy of these models during extreme haze events remains largely unknown. Previous evaluations of CAMS data primarily focused on areas with far lower PM2.5 concentration ranges than those experienced in Kalimantan during extreme fire events. This study addresses the lack of evaluation under extreme fire-generated PM2.5 conditions by using a densely placed sensor network within a fire-affected region. This allows for a more representative assessment of the CAMS model's performance compared to single point measurements.

A network of ten low-cost Purple Air (PA) sensors was deployed across Palangka Raya during the 2019 fire season (August-October). PA sensors were chosen for their cost-effectiveness and demonstrated performance in previous intercomparison studies. Each PA sensor contains two Plantower PMS5003 particulate detectors, providing PM1, PM2.5, and PM10 data. Pre-deployment co-location tests were conducted at Palangka Raya Airport to assess sensor consistency and compare PA data with gravimetric PM measurements from MiniVol filter samplers. An adjustment factor (AF) was derived to correct for the difference in particulate density between the default assumptions of PA and the actual peatland fire smoke. High-concentration calibration tests were also performed in a combustion chamber to understand sensor performance at PM2.5 levels exceeding design limits. Sensor data were processed to adjust for temperature, humidity, and detector performance differences. The resulting PM2.5 data were used to calculate the US EPA's Air Quality Index (AQI). The measured PM2.5 concentrations from the PA network were compared to modeled concentrations from the CAMS EAC4 reanalysis dataset and the CAMS Near Real Time (NRT) service. To estimate excess mortality due to PM2.5 exposure, the method of Johnston et al. (2012) was applied using population data and all-cause mortality estimates. An alternative methodology by Crippa et al. (2016) was also used for comparison. Spatial data included land cover maps and active fire hotspot detections from VIIRS data. Data from hospital records provided additional insights into potential health impacts.

The PA sensor network captured the extreme air pollution event of the 2019 fire season in Palangka Raya, revealing substantial inter-site variability in PM2.5 concentrations. Mean PM2.5 mass concentration between August 20th and October 24th was 137 µg/m³, far exceeding WHO guidelines. The AQI analysis revealed that a significant portion of days had hazardous air quality. Comparison with CAMS data showed generally good agreement between measured and modeled PM2.5 concentrations, particularly outside the peak pollution period. During the peak period of September, the CAMS model overestimated PM2.5 by 35%. The NRT model had a lower correlation with the measured data. The study estimated that over 1200 excess deaths occurred in the Palangka Raya region in 2019 due to PM2.5 exposure, with a much higher number for Central Kalimantan and the whole of Kalimantan and Sumatra. Analysis of diurnal variation showed that PM2.5 concentrations were highest in the early morning. Hospital data indicated a correlation between birth weight and PM2.5 exposure, with lower birth weights associated with higher exposure levels during pregnancy. The most severe air quality was observed in areas furthest from the city center, suggesting potential protective effects from vegetation in less degraded areas.

The findings demonstrate the catastrophic impact of the 2019 Indonesian peatland fires on urban air quality and health. The relatively good agreement between the low-cost sensor network data and the CAMS model, despite the limitations of the model at extreme PM concentrations, strengthens confidence in the overall estimations of excess mortality. The observed spatial variability in air quality highlights the limitations of relying solely on large-scale models for local health impact assessments. The study demonstrates the value of using low-cost sensor networks to improve the accuracy and resolution of air quality monitoring and health impact assessments. The substantial excess mortality attributed to fire-induced PM2.5 emphasizes the urgent need for effective fire prevention and landscape management strategies to mitigate future events. The diurnal variation in PM2.5 concentrations provides valuable insights for public health interventions.

This study demonstrates the significant and catastrophic impact of extreme peatland fires on urban air quality and health in Palangka Raya, Indonesia. The use of a low-cost sensor network provided high-resolution data that validated the performance of the CAMS model, while highlighting its limitations during extreme haze events. The findings underscore the need for robust mitigation strategies, including fire prevention, landscape remediation, and improved air quality monitoring to reduce future health risks. Future research should focus on evaluating other low-cost sensors for their capability in measuring accurately in similar environments, and expanding the use of this model to other regions in Indonesia and beyond. Further investigation into the long-term health effects of PM2.5 exposure, particularly on vulnerable populations like pregnant women and children, is also crucial.

The study's findings are based on data from a single year (2019), limiting its generalizability. The accuracy of the PM2.5 measurements from low-cost sensors may be affected by various factors including temperature, humidity, and sensor degradation. The study primarily focused on PM2.5, neglecting other air pollutants that could have contributed to health impacts. The use of large-scale models like CAMS inevitably leads to some uncertainty in assessing local impacts. The excess mortality calculation depended on available mortality data for the region and some assumptions about non-fire season PM2.5, potentially influencing the accuracy of the final estimates.