Science1 publisher2 min readPublished
Richer countries get more accurate weather forecasts, 35 years of ECMWF data show
Two researchers scored 35 years of ECMWF forecasts against the observed weather and found accuracy rises with national income at every lead time out to a week. The reasons are mostly geographic and only partly a matter of spending.
The Scientist · Science desk
What happened
- Manuel Linsenmeier of Goethe University Frankfurt and Jeffrey Shrader of Columbia University compared ECMWF global model forecasts issued between 1985 and 2020 with observed weather, reporting in Nature Communications.
- The gradient was steep enough that, on average, a seven-day forecast in a high-income country was more accurate than a one-day forecast in a low-income country.
- Forecasts became substantially more accurate worldwide after the mid-1980s and countries at all income levels gained, but the gap between richer and poorer countries narrowed only modestly.
- The authors attribute roughly two-thirds of the worldwide variation in accuracy to geography alone, because short-range prediction is inherently harder in the tropics, where many low-income countries lie.
- Low-income countries typically have fewer surface stations and radiosondes, and in some countries the instruments that do exist report observations less frequently.
Compiled by The ScientistSomething wrong?How this is made
Why it matters
- cost Money spent on warning delivery protects people only in proportion to the accuracy of the forecast behind it. An equal per-person budget in a low-income country therefore buys a less accurate warning than the same budget in a high-income one.
- constraint Buying equipment can only close the smaller share of the accuracy gap. A fully funded observation network in a tropical country would still sit behind a temperate one on short-range temperature skill.
- capability Most national services already refine output from a handful of international centres instead of running their own global model. So an upgrade at those centres reaches many countries at once, and a country can get usable forecast skill with staff and computing.
- decision A ministry splitting an adaptation grant between instruments and forecaster salaries has to make that call on other evidence. This analysis names both observational coverage and service capacity as contributors and does not say which matters more.
Accuracy in this analysis means one thing: the distance between what the ECMWF model predicted and what instruments later recorded, at lead times of one day to seven days, across 35 years of forecasts [2][4][23]. Temperature error is easy to score. It is also not the variable most weather-sensitive decisions turn on. The paper compares model output with observations and does not track whether a warning was issued or acted on [2]. Manuel Linsenmeier and Jeffrey Shrader also examined precipitation and surface pressure, and the income pattern they report is the temperature one [3].
About a third of the worldwide variation is left once geography is accounted for [16]. Part of that remainder is attributable to observational coverage, on the authors' statistical analysis [25]. Part sits with national meteorological services, where turning an international centre's output into a local forecast takes skilled staff, reliable data and substantial computing [12].
To gauge that capacity, Linsenmeier and Shrader counted which countries file official forecasts for their capital cities with the World Meteorological Organization, a measure they call rough [13]. Around 76% of high-income countries do; 19% of low-income countries do [13]. The difference is 57 percentage points, and the high-income group files at four times the rate [14][15].
The comparison is between countries as they stand, and the global distribution of wealth overlaps with the geographic distribution of weather predictability [24]. So the design cannot show that installing radiosondes in a particular country would cut its forecast error by a particular amount. The link between observation gaps and accuracy comes from statistical analysis of existing differences [25].
The value of a forecast is not evenly distributed either. Weather information "is particularly valuable in places where livelihoods depend on weather-sensitive activities or where people are especially vulnerable to extreme weather," Linsenmeier said [21].
He is plain about which part of the gap is fixed. "We cannot easily change the fact that weather is harder to predict in some parts of the world than in others," Linsenmeier said [17]. He named strengthening observation networks, improving models and building the capacity of national meteorological services as concrete ways to improve forecasts, and said international cooperation is particularly important [18][22].
"Weather observations and forecasts are a form of global infrastructure," he said [19]. Better observations in underserved regions help local communities and also improve forecasts across national borders, he said, because forecasting models draw on data from around the world [20].
What to watch
- Whether the paper's precipitation and surface-pressure results show the same income gradient the temperature results do.
- Per-country counts of stations and radiosonde launches. Those would put a number on the non-geographic third of the variation.
- A before-and-after measurement of forecast error where an observation network is actually expanded. The current design cannot supply one.