Each Site Watch entry in Lilliewatch Onshore is anchored on at least one publicly-documented incident with a specific named location, a specific date, and a quantified operational outcome. Three of the eleven are described in detail below as the foundational case studies. The full inventory of eleven sites spans five South-South states (Lagos, Bayelsa, Rivers, Cross River, Akwa Ibom) with sources from Vanguard, The Guardian Nigeria, Daily Trust, ThisDay, the Nigerian Tribune, Channels TV, SweetCrudeReports, the Bayelsa State Oil and Environmental Commission, the Nigerian Navy, the Nigerian Upstream Petroleum Regulatory Commission, and the Lagos State Government. The complete provenance log is the file docs/evidence/site_watch_sources.md in the public repository.
SAR
Confirmed
Thermal
Confirmed
NDVI
Confirmed
SO₂
Confirmed
After the March raid we kept monitoring. Site Watch tracks six weeks of weekly imagery across SAR, thermal, SO₂ and NDVI for the C12 centroid. Pre-raid, all four sensors agree on active combustion. Raid week shows a coordinated collapse on thermal, SAR and SO₂. The week after, those three sensors stay below 25% of their pre-raid means; NDVI scarring is still visible but the site reads as quiet.
This is the operational follow-up: a re-emergence watch that catches bunkers being rebuilt within two to six weeks of a strike, before they are back at full production. View Site Watch on the dashboard →
Each detection above came directly from the production Lilliewatch pipeline running against the dates and bounding boxes listed. Module 1 ingests Sentinel-1 SAR, VIIRS/FIRMS thermal, TROPOMI SO₂, and Sentinel-2 NDVI imagery via Google Earth Engine. Module 2 applies feature engineering and DBSCAN spatial clustering. Module 3 runs the XGBoost risk classifier trained on Niger Delta weak labels.
Public records were sourced independently through search queries against major Nigerian newsrooms (Vanguard, Punch, THISDAY, Daily Post) and government documentation (Bayelsa State Oil and Environmental Commission, Nigerian Navy press releases). Cross-referencing happened after the detections. We did not run the pipeline against pre-selected incident coordinates.
Two methodological caveats are worth stating directly. First, the Bonny case is an LGA-level geographic match, not an exact coordinate match. The Navy's March 2026 raid was reported as occurring in "Allison Community, Bonny LGA" without published GPS coordinates. Our C12 cluster sits in the same LGA with a HIGH-confidence classification. The match is consistent but not absolute. Second, our case-study AOIs are deliberately small (~30 km × 30 km for Aiteo; ~100 km for TNP). Production deployments would expand bounding boxes to capture full pipeline corridors and downstream creek networks, increasing detection volumes substantially.
The same methodology runs continuously across eleven Nigerian AOIs. The dashboard reflects whatever the GitHub Actions cron has produced in its most recent six-hour cycle.