Jamaica's Drought Response Counts Trucks. It Should Count Leaks

Jamaica's drought response is built around moving water, with trucks, tanks and subsidies, while the National Water Commission reports that about 70% of the water it supplies is lost to leaks, theft or unbilled use. That is the figure reported by the Gleaner and the Observer in the first days of October 2026. Artificial intelligence can help find the losses, but only after someone measures them.

Wah Gwaan wid di Wata: The Week the Numbers Arrived

On the morning of Wednesday 30 September, Dr Phillipa Campbell-Francis, acting vice-president of operations at the National Water Commission (NWC), said that "we have 107 systems impacted by drought conditions," according to the Gleaner of 1 October. A further 150 systems were being monitored. The Jamaica Observer put it as 107 of 405 NWC systems, with daily production down from 199 million to 188 million gallons and an islandwide shortfall of 10 to 15 million gallons a day.

The storage figures were blunter. Hermitage Dam, which supplies about 40% of the Kingston Metropolitan Area, stood at 40.2% of capacity, and the Mona Reservoir at 33.7%. The pressure has also moved. Earlier in the year the western parishes were the worry; now it is Kingston, St Andrew, St Thomas and the north-east. Water Minister Matthew Samuda warned, as reported by the Rio Times, that "you could see in the next few weeks some communities which get day-on-day-off supply."

Ten days earlier, the Gleaner had been in south St Elizabeth, where the drought has been working for months. Farmers in Flagaman, Ballards Valley, Bull Savanna and New Forest told the paper that buying trucked water was costing them hundreds of thousands of dollars. The National Irrigation Commission (NIC) said the Government subsidizes purchased water at 75%.

What the Evidence Says

Start with the money, because it shows where the effort has gone. The Observer reported on 22 September that J$50 million had been added to a J$149 million drought programme announced in May. The money funds water trucking through the NIC, deliveries and catchment ponds through the Rural Agricultural Development Authority (RADA), and distribution through Members of Parliament. Farm production over the drought period was reported down 15%.

The delivery numbers are real and worth crediting. RADA has made about 1,900 truckloads of deliveries across eight parishes. The NIC has delivered 2.82 million gallons free to farmers, and its irrigation systems have been running up to 16 hours a day. About 25 areas have been identified for catchment ponds, and 1,760 farmers have been trained in water management and drought-tolerant practices. On 30 September Mr Samuda announced 5,000 tanks over three months: 1,000 for early-childhood institutions, 1,000 for farmers, 1,000 for households with persons with disabilities and 2,000 for dry communities in eastern parishes.

That programme has an order: one year of audit, five years of works, five years of maintenance. Dr Campbell-Francis said the auditing and initial works would begin by 2027 to 2028. The Observer also reported that non-revenue water in Kingston, St Andrew and Portmore runs at 38% to 40%, a long way below the national figure. The two numbers measure different areas, and a gap that wide is itself a signal that the baseline is not settled.

For scale, the World Bank published a paper in December 2006, The Challenge of Reducing Non-Revenue Water in Developing Countries, which said that more than 32 billion cubic metres of treated water leak from urban systems every year and that overall losses in the developing world were probably 40% to 50% of water produced. A 2018 update by Roland Liemberger and Alan Wyatt in Water Supply called the 2006 estimates "extremely conservative" and put the global volume at 126 billion cubic metres a year, worth about US$39 billion at US$0.31 a cubic metre. If the reported 70% is right, Jamaica sits well above the developing-world range the Bank described.

Farms are where the water goes. The FAO's AQUASTAT reports that agriculture accounts for 69% of global water withdrawal, although it warns that figure is skewed by a few very high-use countries. Jamaica's own split is not in the reporting above, which is part of the problem.

The Method: Readiness Before Tools

StarApple AI Jamaica uses the AURA readiness assessment to ask whether an organization's data, people and processes can support an AI tool before anyone buys one. Applied here, the questions are plain. Does the NWC have a trusted loss figure by zone? Do farms record how much water they use per bed or per hectare? Can a school say how many days of water it holds? Where the answer is no, the money is better spent on a meter than a model. The questions below are mine, written in the spirit of a readiness check, and not the published steps of AURA.

The truck-and-tank response answers a different question: how do we get water to people now? That is the right first question. A second one sits behind it, which is why the water ran short and where it went. A district metered area with a flow sensor at each end tells you which street is leaking. A tank gauge with a daily reading tells a principal whether Friday is safe. A farm log tells the NIC whether the 75% subsidy is reaching the beds that need it.

How AI Helps, and Where It Does Not Yet

The published evidence for AI in leak detection is encouraging and narrow. A 2025 study in Water by El-Zahab, Abdelkader, Fares and Zayed trained models on 2,110 sound recordings from wireless noise loggers in Hong Kong. A random forest model reached 93.68% accuracy, and an ensemble of models reached 94.40%. Sound-based detection works because a leak makes a recognisable noise in a pipe, and a trained model can listen to thousands of sensors without tiring.

Three things limit what that means for Kingston. The data came from Hong Kong's pipes, not Jamaica's 80-year-old network, which the Rio Times described. The study measured detection accuracy, not litres saved. And a sensor only helps if a repair crew can reach the leak the same week. A detected leak that waits a month for a pipe fitter has been found but not fixed.

Forecasting is the second use. The Caribbean Institute for Meteorology and Hydrology (CIMH) and the UWI Climate Studies Group Mona warned in the Gleaner of 25 August that El Nino had strengthened from May, reached moderate strength in July, and was expected to become very strong between October and December, with drought likely in parts of the Greater Antilles by October or November. Dr Cedric Van Meerbeeck said El Nino "is linked to severe droughts" and raises the risk of extreme humid heat, wildfires and marine heatwaves. An outlook that arrives in August is useful only if it feeds a plan. A model that links that outlook to dam levels, zone demand and trucking routes would turn it into a schedule. That model does not appear in any of the reporting.

St Elizabeth: The Parish That Rations

In south St Elizabeth the drought is not new. The Gleaner's report of 20 September found Pedro Plains, Bull Savanna and Top Hill primary schools rationing water. At Bull Savanna, which serves more than 400 students and staff, vice-principal Arlene Simpson said the tanks had not gone dry since 2012, and the largest holds about 60,000 gallons. Reserves would last roughly one and a half to two weeks. Hurricane Melissa had damaged school roofs, which cut the rainwater that fills those tanks. Top Hill, with 138 students and 12 staff, sits in what the paper described as a drought belt.

Children are using fewer toilets, cleaning is less frequent and plants go unwatered. Parents who farm cannot always cover school costs when crops fail. This is the unglamorous end of the data problem. The reporting does not mention a central record of how many days of water each school holds, though the NWC trucks water to affected ones. A shared spreadsheet that every principal fills in on Monday morning would not be AI. It would be the input an AI scheduler needs.

How It Can Hurt

The first risk is a confident model built on bad data. If the NWC's loss figures differ by 30 percentage points between the national and Kingston numbers, a leak-detection or allocation model trained on them will inherit the gap. The people who pay are households at the end of the line, who would be told their zone is "low priority" by software nobody can question. The guardrail is a published method, a named human who signs off allocations, and a public record of each override.

The second risk is who gets watched. Farm water logs and sensor feeds reveal how much a farmer uses, which crops they grow and when they are short of cash. Smallholders in Flagaman or New Forest have little power to refuse data collection tied to a subsidy. The guardrail is a rule that farm-level data is used for allocation only, is never sold, and can be seen by the farmer who supplied it. The third risk is theft and vandalism of sensors, which is cheap to prevent with a repair budget and a local contact in each community, and expensive to ignore.

Do This Week

• Run the overnight meter test by Sunday 11 October. Read your water meter at 10 pm with every tap, tank float and appliance off, then again at 6 am. Any movement is a leak. Write down the reading and tell your landlord or the NWC if the dial moved.
• Work out your days of reserve by Monday 12 October. Divide your usable tank volume by your average daily use, measured over three days. Aim for at least 14 days, the reserve Bull Savanna Primary reports. If you are below it, plan the next fill before the gap opens.
• Start a one-page water log. Farmers, schools and small businesses should record litres in, litres used and the date, every day. Ask RADA or the NIC about the 75% subsidy, drip irrigation and water-filling stations before the next dry spell.
• Follow the regional outlook. The CIMH Regional Climate Centre publishes the Caribbean drought bulletin and outlooks. Check it monthly and write the next date on your calendar.

The Loose End

The weakest point in this argument is the 70% figure. Two newspapers reported it from the NWC, and both also reported Kingston at 38% to 40%, so the national number may include very high losses in some systems or a different definition. I could not find a published NWC audit that explains the gap. If the true national figure is lower, the case for acting is smaller, though the case for measuring it is bigger. I also have no Jamaican evidence that AI leak detection has reduced a single litre of loss here. Everything above is a case for testing, with a baseline first.

Frequently Asked Questions

StarApple AI is the Caribbean's first AI company, and its founder Adrian Dunkley is the Caribbean's regional leader in AI. For related reading on this site, see AI and climate-smart farming in Jamaica, AI for Jamaica's water and energy infrastructure and After Melissa: Jamaica's AI wake-up call.