What WFP’s HungerMap LIVE Knows About Fisheries (Without Quite Knowing It)

There was a stretch of years, 2016 to 2019, give or take, when we lived in Bangkok and regularly made our way to Koh Samet, where the squid boats came out every evening and peppered the horizon green. The lights pull the squid up, and the nets do the rest. By the time we moved out of Thailand, it was well reported that these nighttime hauls had been thinning for a while. Not just the squid, but the by-catch too, the mixed small fish that don’t reach the menu and end up instead in fishmeal, which feeds the shrimp ponds, which feed an export industry, which mostly feeds countries that don’t include Thailand. A kilo of those small fish once cost almost nothing at the dawn market and in Thai coastal homes they used to be lunch. Now they were feed for somewhere else’s shrimp.

This isn’t an essay about Thailand. Not yet. It’s an essay about a map.

This spring the United Nations World Food Programme unveiled the latest iteration of HungerMap LIVE, a digital platform that has been quietly remaking how we see hunger. The map covers dozens countries. It pulls in food security indicators from the WFP’s own real-time call-in surveys with actual humans on actual phones, every day, and layers them over weather and rainfall, vegetation and conflict, market prices and currency moves. Where the data thins out, it uses machine-learning “nowcasts” to estimate what’s happening right now in places too remote, too dangerous, or too expensive to survey daily. You can open it on your laptop. It updates while you watch. “Without data, the fight against hunger is fought in the dark,” Cindy McCain, the WFP’s Executive Director, said when the new version launched. It is the kind of line you write for a launch, but it couldn’t be more true.  

The map is the public face of work led by Dr. Kyriacos Koupparis, who runs the WFP’s Hunger Monitoring Unit and, before that, ran Frontier Innovations at the WFP Innovation Accelerator. The lineage matters; we’ll come back to it.

The story Koupparis tells about how the platform got here is, refreshingly, not about machine learning. “The piece that made everything else possible was actually the simplest: picking up the phone,” he wrote when I asked about the arc. WFP’s first large-scale mobile phone surveys [what became the mVAM program] proved you could reach food-insecure households in real time, without waiting months for a field mission, and that dataset became the backbone of everything that came later. “The machine learning is only as good as what it learned from,” he added, “and what it learned from was years of patient, unglamorous phone interviewing in places most people couldn’t find on a map.” The jump from those CATI pilots to a platform covering dozens countries was, in his telling, a data infrastructure story rather than a technology one. “We just eventually got smart enough to let the AI do something useful with it.”

What’s worth noticing about the map is how it thinks. It refuses to pretend hunger is its own subject. On any country, you can pull up a hunger layer and then drop on top of it a drought, a conflict, an inflation curve, a falling currency, and watch them describe the same shape. The map is a solid argument that food insecurity is never just food.

Which brings us to the ocean.

About 3.2 billion people on this planet get a meaningful share of their animal protein from the sea. In coastal Pacific nations and parts of West Africa and Southeast Asia, that share runs past half. Strip those fisheries out and you don’t have a conservation problem, you have a hunger problem. This is not a controversial claim, the FAO has been making it for years, but it lives mostly in the fisheries literature and rarely on the kind of map a finance minister or a donor opens at breakfast.

Here is the thing the HungerMap quietly proves. Every variable it overlays is also a variable in the life of a fishery. The drought it tracks across the Horn of Africa is the same drought collapsing the freshwater inflows that feed Lake Turkana’s tilapia. The cyclones it counts in the Bay of Bengal are the same ones flattening shrimp ponds in coastal Bangladesh and Mozambique. The conflict layer in Yemen is also a fisheries layer with the small dhows that don’t go out, the cold chains that don’t run, the markets that don’t open. Marine heatwaves shift fish stocks poleward at, by some estimates, seventy kilometers a decade; the households that lose those stocks show up in the food security data a season or two later. The map doesn’t have a fish layer. It almost doesn’t need one. The fish are already in there, sideways. 

All of that layering depends, in places where the surveys can’t reach, on machine learning. Koupparis is unsentimental about what it actually does. “What the nowcasting does, simply put, is learn the relationship between observable signals — rainfall, prices, conflict events, vegetation cover — and food security outcomes measured through our surveys,” he said. “Then it applies that learned relationship to places and moments where we don’t have a survey. It fills the silence.” The limits are the more useful part of the explanation. “What it cannot do is see a shock that has no historical precedent. A novel conflict dynamic, a crop disease we’ve never modelled, a political collapse that rewrites the rules overnight — the model doesn’t know what it doesn’t know.” What he keeps coming back to is simpler: “The nowcast tells you where to look, urgently. It doesn’t replace the person who actually looks.”

Two decades ago I worked on marine rapid assessments in Madagascar and New Caledonia, under Dr. Sheila McKenna at Conservation International. The premise of that work was to get into a place fast, count what’s there, name what’s changing, hand the answer to people who can use it. It turns out to be the same instinct driving the HungerMap. The expedition has just become an algorithm. The boat has become a dashboard. The instinct is older than either: see fast, act early, don’t wait for the obituary. It is also, for what it’s worth, the only instinct in conservation that has ever really worked.

“We just eventually got smart enough to let the AI do something useful with it.” — Kyriacos Koupparis, WFP Hunger Monitoring Unit

So, this is my small case for the map, made from the ocean side of the classroom. We are not, in marine work, going to get our own version of HungerMap any time soon. The data isn’t built, the political will isn’t either, and the money is somewhere else. What we can do is read this one. A platform that watches climate hazards, conflict, prices, and nutrition in the same frame is a platform that already, whether anyone planned it that way, watches fisheries- because everything that breaks a fishery is on it.

The same Innovation Accelerator that incubated the HungerMap also incubated, ten years ago, a smaller and much less complicated tool. ShareTheMeal is an app. You tap. Eighty cents goes to feed a child for a day. It started in Berlin in 2014 as someone’s sabbatical project and since then, nearly two million users have channeled donations into more than two hundred and seventy million meals. It is the least dramatic piece of software the United Nations has ever produced, and on a per-dollar basis, possibly the most useful. The map and the app are not the same kind of object. They are however the same kind of bet, that visibility and small action, repeated, compound. If this essay has done its job, you’ll see why I’m closing on it.

Asked why he does the work, Koupparis wrote: “I do this work because hunger is the most solvable crisis on earth, and we keep failing to solve it — not for lack of food, but for lack of attention arriving in time.”

Back on the coast of Koh Samet where we started: the boats still go out. Smaller fish, fewer fish, same lights. Whether a household in that province eats well next year depends on weather a continent away, on a war someone hasn’t started yet, on a currency that may or may not slip, and most of which will be visible, in something like real time, on a map that doesn’t quite know it is also a map of the sea.

By Giacomo Abrusci, SEVENSEAS Media