Pacific Data Viz Challenge 2026

The Pacific can describe its rain
after the fact. It cannot
see it coming.

In 2015 a severe drought hit Papua New Guinea's highlands, but the country lacked the observations needed to see its local impact coming. Forecasting rain depends on layers of observation, and across the Pacific those layers are thinning. The upper-air network that provides early warning is almost gone. As rainfall patterns become more variable, we are losing the ability to see those changes as they arrive.

1 · The cost

When the drought came, we couldn't see it

In 2015, El Niño brought severe drought to Papua New Guinea's highlands, where the rainfall observations needed to see its local impact were largely missing.

The rain began falling away in April 2015. Between mid-July and mid-August, at the one high-altitude research station recording it, temperatures fell to minus two degrees before dawn. The frosts destroyed sweet potato, the staple crop, in gardens above 2,200 metres, where a replanted crop takes eight to twelve months to mature.

27
of 271 rural districts with food very or extremely scarce
770k
people in those districts, an upper limit the authors stress
88%
of them in the highlands
0
official death toll
How much of the country was actually looked at
Each square is one of Papua New Guinea's 271 rural districts.
food supply reported (141) estimated from neighbouring areas and from 1997 (130)
The only national account was assembled from more than two hundred reports: surveys, emails, phone calls, newspaper stories, photographs posted on Facebook. 130 of the 271 districts were never observed, only inferred. Its authors asked for the numbers to be checked in the field. That was not done.
Source: Bourke, Allen & Lowe (2016), Estimated impact of drought and frost on food supply in rural PNG in 2015, Development Policy Centre Policy Brief 11, ANU.
The 27 districts
Hover any district. Larger dot = rating 5, no food available at all.
very high altitude, 2200–2800 m highlands and highlands fringe inland lowland Western Province
Source: Bourke, Allen & Lowe (2016), Estimated impact of drought and frost on food supply in rural PNG in 2015, Development Policy Centre Policy Brief 11, ANU, Table 2: the rural districts rated 4 or 5 for food scarcity, with altitude zone and province. Population is the authors' 2015 estimate for each district.
An El Niño declaration is a statement about the ocean. A rainfall forecast for a district is built from what rainfall at that place has done before, so it can only ever be as local as the record behind it.

The ocean was watched. By early April 2015, forecasters including the Australian Bureau of Meteorology were reporting a severe El Niño moving west. What no one could say was how much rain would fall in the Kandep basin, or when it would stop.

The highlands are where most of Papua New Guinea's people live, between roughly 1,500 and 2,800 metres. They are also where almost none of the measurement is.

Timeline and mortality: DFAT Office of Development Effectiveness, Evaluation of Australia's response to the PNG El Niño drought and frosts 2015–17.
—
metres, the highest rain gauge in the country
—
gauges still reporting, of — ever
0
have ever been placed above 1,000 m
Gauge elevation against where the drought bit
Internationally exchanged rain gauges. The band is 1,500–2,800 m, where the frosts hit.
still reporting stopped the drought zone
Peaks above 4,000 metres; nothing in the internationally exchanged rainfall record above — metres. These gauges are on the coast because that is where the airports are. The WMO register tells a different story.
Source: NOAA GHCN-Daily inventory, PRCP element, stations exchanged internationally. PNG may hold highland records that never leave the country, but the forecast centres and global models draw on the exchanged set.

2 · Seeing rain

What it takes to see rain coming

Forecasting rainfall depends on observations at three levels: balloons in the atmosphere, buoys at sea, and gauges on the ground.

Rainfall is not one measurement. It takes three: the ocean below, the ground itself, and the air above. Each does a different job. Scroll to see what each one contributes, and what the forecast loses without it.

UPPER AIR radiosondes, twice a day OCEAN SURFACE moorings, ships, drifters GROUND rain gauges nothing up here
The ocean surface

It fuels the rain

Moored buoys along the equator, ships on the shipping lanes, drifters in between. They show where the warm water and the moisture sit, weeks to seasons ahead. This is the layer the El Niño forecast is built on.

— platforms reported in the tropical Pacific in July 2026, and — sat within 1 hPa of the model, against — globally.

working
The ground

It records what arrived

Rain gauges measure what actually fell. They are what every satellite estimate is calibrated against, and what any local forecast has to be built from. You cannot say what an El Niño means for a district without decades of rainfall measured in that district.

Papua New Guinea's highest gauge sits at — metres, in a country with peaks above 4,000.

thin, misplaced, half-silent
The upper air

It steers the rain

Radiosondes rise twice a day and read temperature, humidity and wind through the whole depth of the atmosphere. That is what tells a forecaster where the convection is going over the next few days. That is the difference between knowing an El Niño is running and knowing which valley gets the rain.

Across the Pacific there were — sounding sites at the 1979 peak. There are — now.

almost gone

The ocean layer works, but on its own it is not enough

—
platforms reporting in the tropical Pacific, July 2026
—
within 1 hPa of the forecast model
—
the same figure globally
Where the ocean is watched
Each dot is a marine platform. Colour is how closely it agrees with the model.
within 1 hPa 1–2 hPa over 2 hPa
The ocean around these islands is better observed than the ocean on average, and almost none of it is paid for or run from within the region. That matters for how long that lasts, not for whether the data exists today. It does. But an ocean reading describes the season ahead for half a hemisphere, so the ocean layer on its own cannot tell a farmer in Kandep when the rain will stop.
Source: WMO WIGOS Data Quality Monitoring System, marine surface quality, July 2026. wdqms.wmo.int

The ground layer is the one that has to answer the farmer's question, and it is the one asked to cover the most water. Rainfall varies over kilometres, and a gauge only ever speaks for the ground around it. So what matters is not how many gauges a country has. It is how much of the country is close to one.

How many gauges, against how much ground they cover
Across: share of land within 50 km of a reporting gauge. Up: how many gauges. Hover for a locator.
Pacific island country or territory Papua New Guinea no gauge reporting internationally
Counting gauges tells you what a country owns, not what it can see. Palau and Papua New Guinea both have five. Palau is within reach of 99% of its land, Papua New Guinea of 5%. Six countries have none at all.
Sources: NOAA GHCN-Daily inventory, PRCP element, stations with an observation in 2021 or later; Marine Regions EEZ-land union v4. Land sampled on a 5 km lattice, each sample measured to the nearest reporting gauge. Working: gauge-coverage.R.

3 · The gap

Counted, but not reporting

Across the Pacific, stations exist on paper but many are silent, while the upper-air network has almost disappeared.

A station in the register is not a station sending data. WMO monitors both, and the gap between them is where the network quietly stopped. Under GBON a station counts as compliant when it delivered 80% of required reports on at least 80% of days in the month.

The highland stations do exist. They are simply not switched on.

—
of Papua New Guinea's expected surface observations arrived in July 2026
—
of its — registered surface stations sent nothing at all
—
was reporting normally
Registered, counted, and silent
Every Papua New Guinea surface station on the WMO register, and what it sent in July 2026.
reporting partial nothing received
Goroka and Mendi North are both in the highlands provinces the drought hit hardest. Goroka returned 0 of 403 expected observations that month; Mendi North, 0 of 248. Not missing. Registered, counted, and included in every tally of how well observed the country is. A silent station still adds to the national station count the region is assessed on, and it has not recorded a millimetre of rain.
Source: WMO WIGOS Data Quality Monitoring System, land surface availability, July 2026. wdqms.wmo.int

Count the stations instead of listening to them, and the region looks adequately covered. Against the international standard most of it is, which says more about the standard than about the network.

Against the international standard, the ground layer mostly passes
Stations per million km² of exclusive economic zone. Dashed line is the GBON requirement for small island states.
Tuvalu is 26 km² of land, so one gauge satisfies 500 km spacing thousands of times over. Compliant, and almost meaningless. The spacing was set for the pressure field, which is smooth and continental. Rainfall is neither. Countries that sit high here do so because their exclusive economic zone is small, not because their network is large.
Standard: WMO GBON Baseline 2023, INF 6.2 to the Sixth SOFF Steering Committee: 200 km surface spacing, relaxed to 500 km for Small Island States, i.e. 4 stations per million km². Areas: Marine Regions EEZ v4.

The layer that steers the forecast has gone

—
radiosonde sites across the Pacific at the 1979 peak
—
still operating in 2026
2015
Papua New Guinea's last sounding, the year of the drought
Pacific radiosonde sites, 1950–2026
Sites with soundings in each year. Hover for a count.
Every launch costs a balloon, gas and a sonde. Upper-air observation is not installed. It is paid for twice a day, forever, with trained staff. So it is the first thing cut when a programme ends, and its ending produces no visible event: the forecast keeps arriving, interpolated from whatever is nearest.
Source: NOAA Integrated Global Radiosonde Archive (IGRA) v2 station list.

— never had one at all.

Who pays decides which layer works

Papua New Guinea is registered for upper-air observation and received — of — expected soundings in July 2026. Solomon Islands and Tuvalu are also registered, and also sent nothing. The obligation does not go beyond the books. The balloons are not going up.

The two layers also come apart inside the same country, which points at what is actually driving it.

The layer somebody else pays for is the layer that works
Observations received against expected, July 2026. Hover any country.
both layers equal Pacific island country or territory
The radiosonde programme in the freely associated states is funded through the US National Weather Service. The surface network is national. Where an outside programme carries the recurring cost, the data arrives. Papua New Guinea funds both itself, and sits at the bottom of both axes.
Source: WMO WIGOS Data Quality Monitoring System, land surface and upper-air availability, July 2026. Metadata-fault categories (grey, yellow) excluded from every rate. A handful of countries exceed 100% because their stations delivered more reports than the register expects of them, so both axes run past 100 rather than clipping those countries off the chart.

This is also what makes the gap easy to miss. Every silent station is still counted: in the register, in the network totals, and in any assessment built on how many stations a country has. On paper the region is better observed than it was thirty years ago.

The same logic decides what is watched. Tide gauges and rain gauges belong to different programmes, on different funding cycles, answering to different agencies.

One record comes from orbit. The other has to be paid for on the ground.
Left: the satellite sea level record, the same 31 years for every country. Right: weather stations reporting internationally.
satellite sea level record, 1993 to 2023 stations reporting internationally none
Every green bar is identical because it is one satellite reading every country from orbit. Every blue bar is different because it is a different national budget. Nauru and Niue hold the same complete sea level record as everyone else and, on this indicator, have never had a weather station reporting internationally. Which makes this fixable. It is a question of which line items have a funder, not a fact about small islands.
Sources: SPC Pacific Data Hub SEA_LVL and METEO_MONITOR_NET. SEA_LVL is satellite altimetry: annual anomaly in metres, 1993 to 2023, complete for all 21 countries and territories, which is why the left column is the same on every row. The station series is annual and the earliest of them begins in 1889. Four more sea level countries (Guam, American Samoa, Northern Mariana Islands, Wallis and Futuna) are absent from the station indicator.

4 · Going blind

Bigger swings, fewer eyes

As rainfall patterns become more extreme and uncertain, the observation network needed to anticipate them is shrinking.

Historically, ships logged sea temperature on their way to somewhere else, and satellites later read the same surface from orbit. Both cover the Pacific because the Pacific is in the way, and neither was built to watch these islands. A line drawn back to 1850 is not the same thing as a country measuring its own weather, and a record built for one purpose cannot answer what is being asked of it now: the regional warming rate roughly tripled in the mid-1980s, and it is rainfall, not temperature, that has to be planned around.

Regional sea temperature anomaly, 1850–2025
Countries averaged. Fitted line has breakpoints at 1915 and 1985.
observed fitted, three eras
—
°C / decade before 1915
—
°C / decade 1915–1985
—
°C / decade since 1985
The warming roughly tripled in the mid-1980s. It is detectable only because the same places were measured the same way for a very long time, which is exactly what rainfall never got.
Source: SPC Pacific Data Hub, SST_ANOM, 21 countries and territories.
Find yourself in the record
Pick a birth year and a place. The line is the sea temperature anomaly measured there, against the 175-year record above.
—
warming since your first decade
—
years of record for this place
Source: SPC Pacific Data Hub, SST_ANOM. Anomaly in °C.

That is the part we can measure. The same ships and satellites have read the same patch of ocean for 175 years, so the warming shows up clearly, and it is the number every summary of the Pacific reports. But rain is the part people plan around, and it is the part nobody could see coming.

Temperature is the thing the long record measures well, and it is the one every summary of the Pacific reports. Rainfall is the thing people plan around, and it behaves differently. It does not climb. It lurches, and it lurches far harder than the trend ever moves.

—
times further rainfall moves year-to-year than the trend moves in a decade
—
mm, typical year-to-year movement (median country SD)
—
mm, typical trend per decade (median absolute)

Rainfall anomalies 1985–2025, the same window as the temperature trend. Trends are taken country by country and sized without regard to direction, so a wetting country and a drying one cannot cancel each other out.

Regional rainfall anomaly, 1979–2025
Hover for a year. Switch to a single country to see how much wilder one place is.
Flat-looking averages hide it. Individual countries move by tens of millimetres from one year to the next, in both directions, with no trend to speak of.
Source: SPC Pacific Data Hub, RAIN_ANOM, 22 countries and territories.

The temperature trend was only measurable because the same places were measured the same way for decades. Rainfall asks for more than that: it moves faster and over shorter distances, so seeing it takes more instruments in more places. It is also the thing the region has fewest instruments for.

The swings are not random, which is what makes them forecastable in principle. In an El Niño, warm water slides east and the rain follows it, so one half of the Pacific is wet while the other is dry.

The South Pacific Convergence Zone shifts north-east toward the equator and loses its diagonal shape, so the convection migrates. Islands near the equator gain rain. Islands to the south-west lose it.

Two halves of one ocean
Mean rainfall anomaly for each side. Shaded band is the 2015–16 El Niño.
wetter in El Niño (Kiribati, Tuvalu, Nauru, Tokelau) drier in El Niño (8 south-west countries)
In 2015 the two sides move in opposite directions at once: the equatorial group +38 mm above normal, the south-west group −9 mm below it. One ocean, two seasons, and no single regional average describes either.
Grouping: Brown et al. (2020) Nature Reviews Earth & Environment; BoM/CSIRO PACCSAP country brochures; Météo-France; NIWA Island Climate Update. Every per-country statement checked against source. Sides were fixed from published climatology before the rainfall data was examined, so the split is tested rather than assumed. Ten of the 22 countries are excluded for a lagged signal, a sign that flips inside one country, or no published assessment.
Relative Oceanic Niño Index, 1950–2026
Above +0.5 is El Niño; below −0.5 is La Niña. Hover for a year.
Source: NOAA Climate Prediction Center, relative ONI (RONI) annual mean.
So the mechanism is known and the pattern holds. But a farmer does not plant against a pattern for half an ocean, and Papua New Guinea in 2015 is what that costs. What has to be forecast is how much rain falls in one district and when it stops. That can only ever be answered from the record of that district.
Everything that would show the rain coming has been switched off, one station at a time, and much of what remains is registered, counted, and silent.

You cannot prepare for what you cannot watch arriving. The next El Niño will be forecast from the ocean, accurately, months ahead. Nobody will be able to say what it means for a valley in Enga.