How Weather Forecasts Are Made, and Why They Sometimes Go Wrong

It’s a cloudy afternoon and you’re deciding whether to carry an umbrella. Your weather app says there’s a 60% chance of rain at 4 p.m. Where does a number like that come from, and why is it sometimes so wrong?

Weather forecasting is one of the great achievements of modern science. It combines the laws of physics, a constant flow of measurements from around the planet, and some of the most powerful computers in India.

Step 1: measuring the atmosphere, right now

Every forecast begins with a snapshot of the atmosphere as it is now. The India Meteorological Department (IMD), working with weather services worldwide, collects observations from:

• Surface weather stations measuring temperature, pressure, humidity, wind and rain
• Weather balloons that carry instruments high into the atmosphere twice a day
• Doppler weather radars that track rain and storms nearby
• Satellites, including India’s INSAT series, which watch clouds, moisture and temperature from space
• Ships, buoys and aircraft, especially over the oceans, where there are no land stations

No set of measurements is complete. There are gaps between stations, and every instrument has small errors. Forecasters combine all these readings into the best possible estimate of the current state of the atmosphere.

Step 2: letting the computer run the physics

The atmosphere follows the laws of physics: how air moves, warms, cools and carries moisture. Scientists write these laws as equations in a numerical weather prediction model.

The model divides the atmosphere into a three-dimensional grid of boxes. Starting from the current snapshot, a supercomputer calculates how conditions in every box will change over the next few minutes, then repeats the calculation again and again to step forward in time, hour by hour and day by day.

The smaller the boxes, the more detail the model can show, but the more computing power it needs. In May 2025, IMD began using the Bharat Forecast System, which works with grid squares about 6 km across, down from about 12 km, so it can capture local rain far better. It runs on new supercomputers, named Arka and Arunika, at the Indian Institute of Tropical Meteorology in Pune and the National Centre for Medium Range Weather Forecasting in Noida.

Tiny errors in today’s measurements grow into big differences a week later.

Step 3: running it many times

Because the starting snapshot is never perfect, forecasters often run the model many times with slightly different starting conditions. This is called an ensemble forecast.

If most runs agree that it will rain, forecasters can be confident. If the runs spread out widely, the forecast is uncertain. This is where many “chance of rain” numbers come from.

Step 4: human forecasters

Computer output isn’t the end of the story. IMD’s forecasters compare different models, look at the latest satellite and radar images, and use their knowledge of local patterns, such as how the Western Ghats affect rain, before issuing forecasts and colour-coded warnings.

Why forecasts go wrong

The atmosphere is chaotic

In the 1960s, the meteorologist Edward Lorenz showed that in systems like the atmosphere, very small differences at the start can grow into completely different outcomes later. This is often called the “butterfly effect”. It means there is a natural limit to how far ahead detailed forecasts can be accurate. Forecasts for the next day or two are usually reliable; beyond about a week, they become much less so.

Some weather is very local

A thunderstorm may soak one neighbourhood and leave the next one dry. Even a 6 km grid can’t pin down exactly which street will get rain.

Apps use different models

Many phone apps use global models rather than IMD’s forecasts, and update at different times, which is why two apps can disagree.

What “60% chance of rain” means

It doesn’t mean it will rain for 60% of the time, or over 60% of your city. It means that, in conditions like these, rain is expected at your location about six times out of ten. Carrying an umbrella is sensible.