How Much to Trust a Seven-Day Forecast
The seven-day forecast is a fixture on every city and province page here and the piece of information trip planners lean on most. Yet within a single forecast, day one and day seven differ enormously in how far they can be trusted, and temperature, rain and wind are reliable to very different degrees. This article explains how forecasts are made, how error grows with lead time, which kinds of weather are hardest to call, and how to use a seven-day forecast for decisions without being misled by it.
1. Where the numbers come from
Modern forecasting rests on numerical models: the atmosphere is divided into a three-dimensional grid, today’s observations from surface stations, radiosondes, satellites and radar are fed in, and the equations of physics are stepped forward in time. Global models use grid spacings of roughly 10–25 km; regional models go down to 1–3 km. National weather services blend several models and let forecasters adjust the output before publishing city forecasts. The Chinese city forecasts shown on this site are the public output of the National Meteorological Centre, the Japanese ones come from the Japan Meteorological Agency; the figures on our pages organise those official forecasts and are not predictions of our own.
2. How error grows with lead time
The atmosphere is a chaotic system: tiny errors in the initial observations grow exponentially as the model steps forward. The working numbers in the trade run roughly like this: the mean error of a maximum-temperature forecast is 1–2°C within 24 hours, 2–3°C at three days, and 3–5°C at five to seven days, with individual misses beyond 8°C. In one sentence: the first three days can be treated as things that will happen and used to fix specific plans; days four and five are a probable trend; days six and seven show only a direction — “a clear cool-down mid next week” — never a number. The outlook line on our city pages picks out only changes of 5°C or more within the week for this reason: at a week’s range, smaller differences are essentially noise.
3. Which elements are reliable and which are hardest
The daily high is the most reliable element in a seven-day forecast because it is governed by large air masses and changes smoothly. The low is somewhat worse, being sensitive to night-time cloud and local terrain; valley and basin cities often run colder at night than forecast. Precipitation is the hardest: whether it rains, in which hours, and how much are three questions of steeply rising difficulty. “Rain tomorrow” is usually right; “rain at three in the afternoon” is close to meaningless beyond three days, and rainfall amounts more so. Thunderstorms and local convection are the hardest of all — the afternoon storms of the southern and northern Chinese summer can only be given as probabilities. Wind matters most on coasts and plateaus, but is equally sensitive to terrain, and real gusts commonly exceed the forecast mean wind by a wide margin.
4. The situations most likely to go wrong
A few patterns can upset even a two-day forecast. First, the arrival time of cold air: a front six hours early or late shifts a day’s high by 6–8°C, most often in North China and the Yangtze region in autumn and winter. Second, the position of the plum-rain or autumn-rain belt: a 100 km north–south wobble decides whether Shanghai has days of drizzle or muggy sunshine. Third, typhoon tracks: landfall shifting by 100–200 km is routine, and typhoon impact is all-or-nothing, so coastal cities in typhoon season should be read alongside the meteorological service’s track-probability charts. Fourth, plateau and mountain cities: the model grid is too coarse for valley terrain, and Lhasa, Lijiang or Jiuzhaigou often run more extreme than forecast. Fifth, sea fog and inland haze, which barely affect temperature but hit visibility and flights far harder than any forecast can express.
5. Why different sources disagree for the same day
Readers often ask why this site, the phone’s built-in weather and an international site give tomorrow’s high two or three degrees apart. There are three reasons. Different models: official Chinese forecasts lean on the National Meteorological Centre’s blended product, while international sites often show raw output from the European Centre (ECMWF) or the American GFS. Different reference points: a city spans thousands of square kilometres, the official forecast refers to a designated station, and an international site may take the grid point at the city centre. Different update times: forecasts are issued several times a day, and a morning view and an evening view may be different editions. A two- or three-degree spread is normal and does not mean anyone is wrong; when several sources agree on the direction — all say cooler — the confidence is much higher.
6. Six principles for using a seven-day forecast well
One: fix plans on the first three days, set direction on the last four. Two: read trends, not single-day numbers — if the highs step from 30°C to 22°C across the week, a clear mid-week cool-down is all but certain, and whether one day lands at 24 or 26 hardly matters. Three: judge rain by runs of wet days, not by one day’s symbol; three consecutive rain days is a reliable forecast, an isolated raindrop icon deserves a discount. Four: shade the forecast low down by 1–2°C before deciding what to pack, especially inland, in the mountains and on the plateau. Five: in typhoon and cold-wave seasons, look again every day; forecasts converge quickly as an event approaches. Six: for severe weather rely on official warnings — every figure on this site is a compilation for reference, not a warning.
7. Checking a forecast’s track record yourself
The best test is hindsight. The year-to-date block on every city page records each day’s actual high over the past months, and the monthly reviews collect each month’s extremes and sharpest one-day drops. If you spend a week in one city, screenshot the seven-day forecast each morning and compare it with what happened a week later; you will see directly how error grows with lead time. That kind of personal calibration builds a better feel for forecasts than any statistic.
8. How forecasts are marked on this site
The seven-day cards on city and province pages show each day’s high, low and conditions; Japan pages add the chance of rain. The date buttons on the home page switch the map to a future day, and the colours, labels and ranking update to that day’s forecast values — remember that what you then see is a forecast, not observation, and that the later the day, the more you should read only the relative pattern the colours show rather than the numbers. Forecasts refresh three times a day, pages are rebuilt daily, and the date at the top of the page is the base date of the data.