Bali Air Dispatch

§ IWhat this dispatch is

Our last dispatch went out in April, as the dry season began and this register was still small. Since 17 April the network has grown to twenty-plus live sensors across seven networks and has logged roughly two thousand sensor-days of PM2.5 — the island's air, measured every fifteen minutes, in the first dry season under Bali's new waste rules.

We built this network to answer one question: what is the routine open burning of garbage doing to the air Balinese families breathe? Ninety days in, the data has begun to answer — in the numbers, and, more tellingly, in the clock.

§ IIThe season turned, and it is worse than last year

The island-wide picture, month by month (mean of every archived sensor-day; the WHO 24-hour guideline is 15 µg/m³):

A caveat we owe the reader: the sensor mix changed as the network grew, so monthly means partly reflect where coverage was added. The cleaner test holds one sensor constant. At Bali Umalas (Villa Fusion) — a residential neighbourhood sensor archived continuously since November 2024, nowhere near a rice field — the same 1 June–20 July window reads: 2025, mean 26.2 µg/m³, 92% of days over the guideline. 2026, mean 31.4 µg/m³, 100% of days over. At the one sensor able to compare, this season runs about a fifth dirtier than last year's, and not one day since June began has met the WHO guideline.

§ IIIThe clock evidence: smoke on a household schedule

If the island's smoke came mainly from a seasonal agricultural event, the record would show broad daytime plumes where the harvest is. It shows something else: a twice-daily rhythm, everywhere, on the clock of domestic life.

Averaging every 15-minute reading since 17 April by hour of day (WITA), at two kinds of sensor held constant all season:

Two peaks, the same two peaks, at every sensor: a morning maximum at 08:00–09:00 — when household rubbish fires are lit and the night's accumulated smoke still sits trapped under the overnight inversion layer — and an evening climb from 19:00 as the day's second round of burning starts and the air stills again. The cleanest air of the day is mid-afternoon, when burning pauses and the sea breeze and daytime heat stir the boundary layer.

This is not the signature of a harvest. It is the signature of a chore — the daily, island-wide routine of burning mixed household waste in yards, roadsides, ditches and field edges, morning and evening, seven days a week. The rural western sensors show the same schedule at higher amplitude: more fires, bigger piles, stiller air — and, as anyone who has walked a field edge knows, those piles are rarely crop residue alone. Plastic, packaging and household rubbish are burned alongside whatever else is at hand.

The practical reading is blunt. The worst hour of the day is nine in the morning — the school run, the market run, the commute. The "fresh morning air" instinct has it exactly backwards in burn season. The clean window is roughly 14:00–17:00.

§ IVEverywhere, not somewhere

The heaviest validated days since 17 April belong to ten different stations in every corner of the island — city, suburb, and countryside:

Days at 90–165 µg/m³ are six to eleven times the WHO daily guideline. (A community sensor in Tabanan's rice country recorded days higher still, up to 162; its calibration is unverified, so we leave it out of the tables — but its direction agrees with the validated Tabanan and Umadawa stations beside it.)

And the residential burden is constant, not episodic. Since mid-April: Umalas exceeded the WHO daily guideline on 95% of days (mean 27.6), Sidakarya in south Denpasar on 82% (mean 23.0), Kopernik in Mas, Ubud on 74% (mean 24.2) — with instantaneous readings in Ubud as high as 133 µg/m³. These are neighbourhoods of homes, schools and warungs. Nobody is harvesting rice in Umalas.

§ VWhat we can and cannot claim

A PM2.5 sensor weighs smoke; it cannot chemically tell burning plastic from burning straw. The remarkable consistency of the morning and evening peaks across multiple regions suggests widespread combustion sources operating on similar daily schedules.

Determining the relative contribution of household burning, village waste-processing facilities, agricultural burning and other sources is the next stage of the analysis.

§ VIThe network itself: what died watching

Ninety days of measurement cost us machines. The Nafas network in Bali went dark station by station — Pemogan on 13 June, Sanur and Ubud on 16 June, and Tonja, the station that recorded the island's worst day, on 1 July. Nafas stations elsewhere in Indonesia update normally; the withdrawal is Bali-specific, and it removed what was once the island's best consumer network mid-season. Their final readings are preserved here, and co-located failover feeds now cover two of the sites. The Jimbaran Smart Citizen cluster also left the outdoor record in June–July.

Against the losses: hourly records from ten IQAir stations, a second government-adjacent station, and a direct line to the AirGradient community network.

§ VIIWhat the data is telling us

1. The air is worse than last year — about a fifth dirtier at the one sensor that can compare seasons, with residential neighbourhoods over the WHO daily guideline nine days in ten.
2. The dominant pattern is domestic or industrial, not agricultural. The smoke keeps house hours — morning and evening, everywhere at once — the schedule of garbage fires, not of any harvest.
3. Exposure is worst when people are most exposed. The 08:00–09:00 peak coincides with school runs and markets. Shifting outdoor exercise, ventilation and children's travel toward the 14:00–17:00 window is free and cuts the daily dose materially.
4. The record is fragile exactly where it matters. The stations that recorded the worst air kept dying, and a commercial network withdrew mid-season. The next step is more independent sensors in more hands — especially beside the burn sites themselves — with every record preserved when the hardware fails.

§ VIIIWhat would help most now

Answering the question this dispatch leaves open takes more sensors, in more places. Every additional monitor turns an assumption into a measurement, and the blank areas on our map — the north, the east, the mountain villages, and the burn sites themselves — are simply places where nobody has installed one yet. A basic outdoor unit costs about what a weekend away does, publishes automatically to the open networks this site reads, and joins the public record here within minutes of coming online. If you can host one, the setup guide is here.