On the morning of Friday, 11 September 2026, three children walking a shortcut to Merry Wood Primary School in Breadnut Walk, north-western St Elizabeth, were struck by a boulder that broke loose above the old railway track they use instead of the parish's own community road. One child remains in intensive care at the University Hospital of the West Indies (UHWI). A second suffered a fractured arm and a skull injury at Black River Hospital. A third was treated and released the same day.
AI cannot flag the one boulder above a specific footpath that is about to break loose. What AI-processed satellite radar and machine learning can do is compress the parish-by-parish landslide mapping Jamaica's Mines and Geology Division has managed to finish for only four parishes since the programme began, into a job measured in months, not years, for St Elizabeth and the seven other parishes still unmapped.
A Shortcut Three Children Should Not Have Needed
The community road connecting Breadnut Walk to Merry Wood Primary runs 3.6 kilometres, and residents describe it as barely passable. "The road turn bush, it is just a donkey track," resident Daisy Gordon said in the days after the incident, adding that she hopes "all efforts will be made to ensure children are safe." Ipswich division councillor Clinton Samuels put the choice residents face in plain terms: "To come through this area is just over one kilometre and then the road is bad," comparing the deteriorated stretch to "a tunnel like it is going to collapse." The railway shortcut cuts the walk to just over a kilometre, which is why generations of children from Breadnut Walk have used it instead of the official road.
Education, Skills, Youth and Information Minister Senator Dr Dana Morris Dixon visited the more seriously injured child at UHWI on Saturday, 12 September. "It's really really sad. It is three children that got injured, and one is more severely injured than the others," she said, adding that the child's injuries "are significant and will take a long time to heal. We are asking all of Jamaica to pray for her." Dr Kelvin Metalor, head of UHWI's Department of Anaesthesia and Intensive Care, has been overseeing that child's treatment. Morris Dixon also confirmed the ministry would provide psychological support for classmates who witnessed the accident, and named the reason this happened at all: "This is a route that has been used for a long time, and a lot of children at the school would have used that route."
Omar Smith, another Breadnut Walk resident, summed up the mood in the community: "Is three children got hurt. Everybody is mournful." St Elizabeth's police divisional commander, Deputy Superintendent Coleridge Minto, has also been involved in the response. In the days after the incident, nobody in Breadnut Walk described it as a freak event. They described it as the outcome of a road nobody had fixed and a hillside nobody had assessed.
Six Finished Maps, Eight Unmapped Parishes
Jamaica's Mines and Geology Division (MGD) produces landslide susceptibility maps that show which slopes are more or less likely to fail, based on factors such as slope angle, soil type, rainfall, and land use. As of this month, complete parish-wide maps exist for only four of Jamaica's fourteen parishes: St Thomas, St Mary, Portland, and St Catherine. Two more locations, Ocho Rios in St Ann and Jacks Hill in St Andrew, have town-level maps rather than full parish coverage. The remaining eight, Kingston, Trelawny, St James, Hanover, Westmoreland, Manchester, Clarendon, and St Elizabeth, have no landslide susceptibility map of any kind. The Division's own guidance on the maps it has finished is direct about their limits: they are not meant for site-specific purposes, and anyone needing that level of detail is told to contact the Division for further advice.
St Elizabeth is not without any hazard planning at all. ODPEM holds a Community Disaster Risk Management Plan for New River, a separate community elsewhere in the parish, and the agency's Jamaica Landslide Risk Reduction Project has already put a documented methodology to work in the country: a US$2.37 million grant from the Japan Social Development Fund, an affiliate of the World Bank, funds slope-stabilisation works under the MoSSaiC approach, Management of Slope Stability in Communities, developed by University of Bristol researchers working in the Eastern Caribbean. That project selects communities based on documented hazard history and existing landslide occurrence. Neither the funding nor the methodology has reached Breadnut Walk, and neither maps hazard ahead of an incident. MoSSaiC stabilises specific slopes a government has already identified as high risk. Breadnut Walk was never on that list, because nobody had mapped St Elizabeth to put it there.
What Satellite Radar and Machine Learning Actually Add
Artificial intelligence changes the arithmetic behind that gap, not by predicting a single falling rock, but by speeding up the mapping that tells a government where to look. Satellite radar interferometry, known as InSAR, measures ground deformation from orbit. The European Space Agency's Sentinel-1 mission provides this radar data free of charge, with a six-day revisit cycle between its two satellites and resolution down to five metres, unaffected by cloud cover or the tropical rain that grounds optical imagery over Jamaica for weeks at a time. Commercial operators such as SKY Perfect JSAT now offer InSAR-based deformation monitoring accurate to under 10 millimetres. Feed that radar data into a machine learning model trained to recognise pre-failure deformation patterns, and the mapping that once took a geologist years of fieldwork to complete for one parish becomes a data-processing job measured in months.
This is not speculative technology. An AI Earth Cloud InSAR processing system, documented in the journal Landslides, tracked deformation ahead of the Zhenxiong landslide in China's Yunnan Province before it failed. Similar AI-enhanced InSAR monitoring is now operating across landslide-prone terrain in Japan, and the Italian firm NHAZCA runs commercial InSAR and photo-monitoring systems for exactly this purpose. A peer-reviewed study of large-landslide early warning using satellite InSAR found accelerated deformation signals ahead of failure in roughly a third of the landslides it examined, with the data proving sufficient for a failure-time estimate in about a quarter of those. That is a meaningful head start for identifying which hillside deserves attention. It is not certainty, and none of the researchers publishing this work claim otherwise.
Jamaica already has an institution built to run this kind of work. The Mona GeoInformatics Institute (MGI) at the University of the West Indies maintains a geospatial database of flood and landslide events across the island, built in part through a systematic review of Jamaica Gleaner archives reaching back to the 1800s, blended with real-time field data. When Hurricane Melissa struck in October 2025, MGI publicly proposed deploying drones fitted with high-resolution cameras to map landslides, flooding, and blocked roadways, feeding the imagery into GIS systems for rapid damage mapping, alongside offline digital survey tools such as ArcGIS Survey123 and ODK Collect for areas without connectivity. That proposal already covers most of what AI-accelerated landslide mapping for St Elizabeth would require. Extending it from post-storm damage assessment to pre-storm hazard mapping is a change of timing, not a change of capability.
The case for AI mapping runs into a genuine limit here, not a hypothetical one. Even a well-built, AI-processed susceptibility map produces a parish-scale or community-scale hazard rating, not a warning about one specific boulder above one specific footpath. Dense hillside vegetation, common across St Elizabeth's cockpit-adjacent terrain, degrades optical satellite readings and complicates some radar signals too, which is a large part of why the Mines and Geology Division's finished maps carry the warning that they are unsuitable for site-specific decisions. Adding machine learning to the process does not remove that warning. A better map would very likely have shown the Breadnut Walk corridor as moderate-to-high hazard terrain. It would not have named the rock.
The Same Failure Mode Plays Out Across the Region
St Elizabeth's unmapped hillsides are a Jamaican story with a Caribbean pattern behind them. Tropical Storm Erika struck Dominica on 26 August 2015, dropping close to 13 inches of rain in twelve hours and triggering landslides that killed more than 30 people in the village of Petite Savanne alone. Total damage came to roughly US$483 million, close to 90 percent of the country's GDP that year. MoSSaiC itself, the methodology ODPEM now applies in Jamaica, was built by University of Bristol researchers working in the Eastern Caribbean precisely because islands from Dominica to St Lucia share the same terrain problem: steep volcanic and limestone slopes, heavy seasonal rainfall, and settlement patterns that predate any hazard map.
CDEMA, the Caribbean Disaster Emergency Management Agency, coordinates regional response once a disaster has already happened. Nothing in its current mandate coordinates a shared, AI-accelerated hazard-mapping push across member states, even though the underlying technology, free satellite radar feeds and machine learning models, does not stop at a border between Jamaica and Dominica. A regionally pooled mapping effort would almost certainly cost less per parish or per island than each government running its own, slower, manual programme in isolation. That coordination does not currently exist, at CDEMA or anywhere else in the region.
The Core Argument
AI cannot tell a government which specific rock is about to fall on which specific footpath. It can tell a government, in months, not years, which of its unmapped hillsides deserve a closer look before the next boulder does the telling instead.
What Would Actually Change This
None of what follows requires new invention. The Mines and Geology Division and the Mona GeoInformatics Institute could formalise a partnership to run AI-processed InSAR analysis and drone photogrammetry across the eight unmapped parishes, prioritised using data the Ministry of Education already holds: school catchment areas, enrolment numbers, and the kind of informal shortcut routes that only get documented after a child is hurt on one. Cross-referencing those two datasets is an afternoon's work for someone with access to both, not a research project.
A cheaper interim step exists too. Drone photogrammetry of specific known problem corridors, the Breadnut Walk rail track among them, costs a fraction of a full satellite mapping contract, and MGI already flies drones for other disaster-response purposes. Pair that with a simple WhatsApp-based reporting channel for residents to submit photos of new cracks or exposed rock, human-reviewed and sorted into a watch list, and a community gets a warning system built mostly from tools that already exist, months before any full parish map is finished. Responsibility for the community road itself, meanwhile, sits with the parish's local authority working alongside the National Works Agency, and no amount of hazard mapping substitutes for fixing it.
Funding precedent already exists too. ODPEM secured US$2.37 million from the Japan Social Development Fund to run MoSSaiC's physical slope-stabilisation work in communities already known to be at risk. A mapping-focused successor project, aimed at the eight parishes with no data at all rather than further stabilising sites already identified, fits the same donor logic and the same World Bank-adjacent funding channel that has already worked once.
Frequently Asked Questions
What is a landslide susceptibility map?
A landslide susceptibility map shows which parts of a parish or region are more or less likely to experience a landslide or rockfall, based on factors such as slope angle, soil type, rainfall, and land use. Jamaica's Mines and Geology Division has produced these for St Thomas, St Mary, Portland, and St Catherine parishes, plus the Ocho Rios and Jacks Hill areas, but the Division itself warns the maps are meant to guide planning and mitigation decisions, not to pinpoint the exact spot where a rock or slope will fail.
Does AI-based landslide mapping apply to a small rural community like Breadnut Walk, or only to large infrastructure projects?
It applies to small communities specifically because they are the ones current mapping has skipped. AI-processed satellite radar and machine learning lower the cost and time of mapping an entire parish instead of a single site, so a place like Breadnut Walk, with no dedicated hazard budget of its own, gets covered as part of a parish-wide effort instead of needing a separate study. The barrier has been mapping speed and cost at the parish level, not project size.
How would a Jamaican parish actually get an AI-generated landslide hazard map built?
In practice, it would mean the Mines and Geology Division commissioning satellite radar data covering the parish, feeding that data and existing geological and rainfall records into a machine learning model trained to flag deformation patterns linked to slope failure, then validating the output against known landslide history, the kind the Mona GeoInformatics Institute already holds in a Gleaner-sourced archive dating to the 1800s. The Division would publish the result as an updated susceptibility map, the same format it already uses for the four parishes covered so far.
How much would AI-based landslide mapping cost compared to Jamaica's current approach?
Neither the Mines and Geology Division nor ODPEM has published a cost figure for AI-accelerated mapping, so no reliable comparison exists yet. What is documented is the funding scale for related work: ODPEM's MoSSaiC slope-stabilisation project runs on a US$2.37 million grant from the Japan Social Development Fund. A mapping-only project, without physical construction, would likely cost less per parish, since satellite data licensing and model training are cheaper than retaining walls and drainage works, but until someone runs and prices a pilot, that remains an estimate, not a fact.
What is the difference between AI-based satellite mapping and the MoSSaiC method ODPEM already uses?
MoSSaiC, Management of Slope Stability in Communities, is a physical intervention: it targets specific slopes already known to be unstable and funds drainage and retaining work to stabilise them. AI-processed satellite radar and machine learning do the step before that: they scan wide areas to identify which slopes are unstable in the first place, including ones nobody has flagged yet. Jamaica currently runs the second step through MoSSaiC in specific, already-identified communities. It does not yet run the first step at scale for parishes such as St Elizabeth that have no hazard map to work from.
Can AI actually predict when a specific boulder or rock will fall?
No, not at that level of precision, and no landslide-monitoring programme anywhere currently claims otherwise. Peer-reviewed research on satellite radar interferometry and machine learning for landslide early warning has found accelerated deformation signals ahead of failure in roughly a third of studied cases, with data sufficient for a failure-time estimate in about a quarter of those. That is useful for flagging which hillside or corridor deserves closer attention. It is not the same as knowing which rock, on which day, above which footpath, is about to give way.
Is landslide and rockfall hazard mapping mandatory in Jamaica?
No single law requires the Mines and Geology Division to complete susceptibility mapping for every parish on a fixed schedule. The Division produces the maps under its general mandate for geological hazard assessment, and development planning bodies are expected to consult them where they exist, but no statutory deadline forces coverage of the eight parishes, St Elizabeth included, that currently have none. ODPEM's landslide risk reduction work operates under disaster risk management policy, not a mapping mandate.
Where is AI-based disaster risk mapping headed in Jamaica and the wider Caribbean over the next few years?
Expect the underlying technology, satellite radar interferometry combined with machine learning, to keep getting cheaper and more accurate, following the trend already visible in commercial services offering sub-10-millimetre deformation measurement. Whether Jamaica and its neighbours adopt it faster depends on institutional decisions nobody has made publicly yet: whether the Mines and Geology Division partners formally with the Mona GeoInformatics Institute, and whether a regional body such as CDEMA ever coordinates a shared mapping push across islands that share the same volcanic and limestone terrain problem. The technology is ready before the institutions are.
What Happens the Next Time a Child Walks That Track
The 3.6 kilometre community road connecting Breadnut Walk to Merry Wood Primary School has not been rebuilt as of this writing, and no agency has announced plans to close the rail track or map the hillside above it. Jamaica's Mines and Geology Division has completed full parish maps for four of the country's fourteen parishes since its landslide susceptibility programme began. St Elizabeth is not one of them, and no date has been set for when it might be.