The Program at a Glance
From March 2022 through April 2023, Aerbits conducted a 13-month aerial detection pilot in Bayview-Hunters Point — one of San Francisco's most environmentally burdened communities. The program was operated in partnership with the Bayview Hill Neighborhood Association (BHNA), a 501(c)(3) nonprofit that has served the neighborhood since 1990.
Using a DJI Mavic 3 Enterprise drone, each 30-minute flight covered approximately one square mile — the entire target area. Custom-trained AI models automatically identified illegal dumping from the aerial imagery. Detections were converted into 311 service requests with GPS coordinates and timestamped aerial photographs, filed automatically through the city's 311 API.
The cleanup crews on the ground didn't need to change anything about how they worked. They received 311 tickets exactly as they always had — just with better data, more precise locations, and photographic evidence attached.
The Controlled Experiment
The most significant finding from the Bayview pilot came from a 52-day period in May and June 2022, when a natural controlled experiment produced what may be the strongest evidence in the illegal dumping technology space.
The experiment followed an A-B-A withdrawal design — a recognized quasi-experimental methodology where a treatment is applied, removed, and reapplied to test whether observed effects are caused by the treatment or by external factors.
Phase 1: The Baseline (Day 0)
On May 9, 2022, an initial aerial census of the coverage area identified 118 active illegal dumpsites. This was not a sample — it was a complete enumeration. Every square foot of the target area was photographed and analyzed. This count became the baseline against which all subsequent measurements were compared.
Phase 2: Daily Monitoring (26 Days)
Over the next 26 days, daily drone flights surveyed the area and the AI system filed 444 service requests to the city's 311 system. Each request included GPS coordinates, aerial photographs, and AI-generated classifications of the waste type.
The result: active dumpsites dropped from 118 to 7 — a 94% reduction. The city's existing cleanup crews, receiving a consistent stream of precisely located and documented reports, cleared sites faster than new dumping could accumulate.
Phase 3: The Break (14 Days)
For the next 14 days, monitoring and reporting were paused completely. No flights. No 311 reports. The city's cleanup operations continued as normal, responding to whatever complaints came in through regular channels.
Active dumpsites surged from 7 to 73 — a 10× rebound, recovering to 62% of the original baseline in just two weeks. The speed of the rebound was a critical finding: it demonstrated that the reduction was being driven by the monitoring and reporting, not by some independent factor like seasonal variation or a change in city staffing.
Why the Break Matters
The 14-day pause is what elevates this from a simple before-and-after comparison to a quasi-experimental study. If the reduction had been caused by something other than the aerial monitoring — warmer weather, a city cleanup initiative, random variation — dumpsites would not have rebounded the moment monitoring stopped. The tight temporal coupling between monitoring status and dumpsite count argues strongly for a causal relationship.
Phase 4: Monitoring Resumed (24 Days)
When flights resumed, the system filed another 410 service requests over 24 days. Dumpsites dropped from 73 to 5 — even lower than the Phase 2 endpoint of 7.
The improvement from 7 to 5 across treatment cycles is consistent with a cumulative deterrence effect: repeated monitoring may have discouraged some fraction of repeat dumpers, producing progressively lower steady-state dumpsite counts over time.
What the AI Found That Humans Missed
One of the most significant operational findings was the gap between what residents reported and what aerial monitoring detected. Across the full 13-month program:
The Detection Gap
30–50% of dumpsites identified by the aerial system on public roads and sidewalks had never been reported through traditional 311 channels — despite being on public rights of way that residents pass every day.
This finding challenges a core assumption of complaint-driven systems: that residents will report the problems they see. In practice, many dumpsites go unreported because they're not visible from commonly traveled paths, because residents don't know about 311, face language barriers, or have learned from experience that reports don't lead to action.
The operational efficiency difference was dramatic:
| Metric | Traditional Methods | Aerial Detection |
|---|---|---|
| Coverage per hour | ~2 city blocks | 1 square mile |
| Detection source | Resident complaints only | 100% aerial coverage |
| Unreported sites found | 0% | 30–50% of total |
| Detection-to-report time | Days to weeks | < 4 hours |
| Evidence quality | Ground photos, witness statements | GPS + aerial imagery + AI classification |
The Partnership Model
The Bayview pilot was not a top-down government technology deployment. It was a community-led initiative.
The Bayview Hill Neighborhood Association — founded in 1990 with a mission "to combat neighborhood deterioration" — brought 30+ years of local knowledge, community trust, and advocacy experience. BHNA identified the priority areas, engaged with residents, and ensured the technology served community goals rather than the other way around.
This partnership model proved essential. Community buy-in meant the program had legitimacy and local support from day one. BHNA's relationships with city agencies helped ensure that the 311 reports generated by the system received appropriate attention. And the community's existing advocacy infrastructure meant that results could be translated into policy conversations — not just cleanup tickets.
The Broader Context
San Francisco spends $47.8 million per year on street cleaning — roughly $60 per capita, twice the rate of Los Angeles. Despite a 63% increase in street cleaning spending over six years, the city's own data shows that dumping and street cleanliness have not meaningfully improved. A September 2024 report from the SF Controller's office, titled "Down in the Dumps," detailed the ongoing challenges across 54 pages.
The city has tried other technology approaches with mixed results:
- Fixed camera/LPR pilot ($300K+): Deployed at approximately 10 hotspots. A 2025 budget analyst audit found "minimal progress" — no staff had been trained to review the footage.
- Smart bin sensors (multi-year, millions): 900+ sensors installed in ~3,000 public cans. City audit later called the data "misleading" — sensor readings were conflated with overflow complaints in the 311 system.
- Manual investigators (6 citywide): Declining citation rates and approximately $3 million in uncollected fines.
Against this backdrop, the Bayview pilot's results stand out not just for the magnitude of the effect, but for the strength of the evidence. No competing technology vendor in the illegal dumping space has published comparable quasi-experimental data.
What Happens Next
The methodology validated in Bayview is now being proposed for deployment in Oakland, where the Public Works and Transportation Committee is considering a six-month, $150K pilot program. Oakland faces its own $24 million annual dumping problem — and the same structural challenges that make complaint-driven approaches insufficient at scale.
The Bayview pilot demonstrated three things that matter for any city considering this approach:
- Consistent monitoring produces consistent results. 94% reduction in 26 days — using the city's existing cleanup infrastructure.
- Stopping monitoring reverses the gains. 14 days without flights returned dumpsites to 62% of baseline. This isn't a one-time fix — it requires sustained operation.
- The effect is cumulative. The second monitoring cycle produced even lower dumpsite counts than the first, suggesting deterrence builds over time.
For cities spending millions on a reactive approach that isn't working, the question from Bayview is simple: what would happen if you knew about every dumpsite, every day?