70% chance of rain: what it means and how to compute it

Why a weather forecast is a math problem

A weather forecast usually gives you two things: a temperature and a percentage chance of rain. The temperature is clear; the percentage less so. Does it mean 70% of the city gets wet? Rain for 70% of the day? Neither. It is a probability, and unpacking it is a great way to practice ideas that show up on every exam: relative frequency, complementary events, independence, and summary measures like the mean and the range.

What “70% chance of rain” measures

The most useful reading is relative frequency. Picture the forecast model finding many past days with conditions like tomorrow’s: same humidity, same pressure, same front approaching. If measurable rain fell at that spot on about 70 of every 100 such days, the forecast says 70%.

So the forecast is not “wrong” when a 70% day stays dry. If it is well calibrated, 3 out of 10 such days should stay dry. You can only judge a forecast over many days, never one.

In set language, let $C$ be the event “it rains tomorrow”. The forecast says $P(C) = 0.7$ .

Out of 100 days with similar conditions, about 70 get rain and 30 stay dry: that is what a “70% chance” measures.

The complementary event

The opposite of rain is no rain. The two cannot happen together and one of them always happens, so their probabilities add up to 1.

P(\overline{C}) = 1 - P(C)

Complement rule: staying dry is 1 minus the chance of rain.

With a 70% chance of rain, the chance of staying dry is 1 − 0.7 = 0.3, or 30%. It looks trivial, but the complement is exactly the tool that cracks the next problem.

Rain on at least one day of the trip

Suppose you are traveling for three days and the (hypothetical) forecast reads: 30% chance of rain Friday, 40% Saturday and 20% Sunday. What is the chance of rain on at least one of those days?

The classic mistake is to add: 30% + 40% + 20% = 90%. Adding only works for events that cannot happen together, and here it can rain both Friday and Saturday. With larger numbers the sum would even pass 100%, which is impossible.

The right move is to look at the complement of “at least one rainy day”, which is “no rainy day”.

P(\text{at least one}) = 1 - P(\text{none})

Complement rule for “at least one”.

Worked example: a three-day forecast

$\displaystyle 1-0.3=0.7 \quad 1-0.4=0.6 \quad 1-0.2=0.8$ Write the chance of no rain on each day using the complement.
$\displaystyle P(\text{none}) = 0.7 \cdot 0.6 \cdot 0.8 = 0.336$ Assuming the days are independent, the chance that none of them gets rain is the product.
$\displaystyle P(\text{at least one}) = 1 - 0.336 = 0.664$ Take the complement to get the chance of at least one rainy day.
Check: the result, 66.4%, is larger than the biggest single-day chance (40%), as it must be, and below 100%. The naive sum of 90% was far off.

The naive sum gives 90%, but the real chance of rain on at least one of the three days is 66.4%, the complement of the 33.6% chance of no rain at all.

A handy special case: if the chance is the same $p$ on each of $n$ independent days, the chance of rain on at least one is $1-(1-p)^n$ . At 20% a day for five days, that is 1 − 0.8⁵ ≈ 0.672, about 67%. Five “mostly dry” days add up to a high chance of getting caught in the rain.

P(\text{at least one}) = 1 - (1-p)^n

Valid when the n days are independent and share the same probability p.

Why independence is a shaky assumption

Multiplying probabilities only works if what happens on one day does not change the chances on another. Real weather almost never behaves that way: a cold front arriving Friday often lingers into Saturday, and a dry air mass can keep the whole week rain-free. The days are correlated.

With positive correlation (rain tends to follow rain), rainy days cluster, so the chance of no rainy day is higher than the product suggests. As a result, the independence calculation tends to overestimate the chance of rain on at least one day. Treat the 66.4% as a teaching approximation, not a forecast.

Mean and temperature range for a week

The other half of a forecast is temperature. Suppose these hypothetical highs from Monday to Sunday, in degrees Fahrenheit: 64, 70, 75, 81, 77, 68 and 66.

\bar{x} = \frac{x_1 + x_2 + \cdots + x_n}{n}

Arithmetic mean of n readings.

Worked example: mean and range

$\displaystyle 64+70+75+81+77+68+66 = 501$ Add the seven temperatures.
$\displaystyle \bar{x} = \frac{501}{7} \approx 71.6$ Divide by the number of days.
$\displaystyle R = 81 - 64 = 17$ Find the range: largest value minus smallest value.
Check: the mean of about 71.6 °F sits between the low (64 °F) and the high (81 °F). The deviations from the exact mean always add up to zero, which is a quick sanity test for the arithmetic mean.

Highs climb from 64 °F to 81 °F and fall back; the mean line sits at about 71.6 °F and the range is 81 − 64 = 17.

The mean summarizes the week’s level; the range shows how much it swung. Two weeks with the same mean can feel very different: one steady within a couple of degrees, another bouncing across twenty. That is why you should read both numbers together. Forecasters also use “temperature range” for the gap between a single day’s high and low; the computation is the same.

Common mistakes

Adding probabilities for different days. That only works for mutually exclusive events; rain Friday and rain Saturday can both happen.
Reading 70% as the share of the area or of the hours with rain. It is a probability for the location in question, not a slice of the map or the clock.
Saying the forecast failed because it stayed dry. One day cannot judge a probability; only the frequency over many similar days can.
Multiplying without questioning independence. The product is an approximation; neighboring days are usually correlated.
Confusing range with mean. The range measures spread, not a typical value, and it depends only on the two extremes.

Frequently asked questions

Is a 50% forecast the same as “we don't know”?

No. It says that in similar situations it rains half the time. That is precise information about uncertainty; “we don't know” would mean having no basis for an estimate.

Can I just add the chances when they are small?

The sum is an upper bound: the chance of at least one event never exceeds it. With small values it lands close to the right answer, but the complement method is the exact one under independence.

Why isn't the mean enough to describe the week?

Very different data sets can share the same mean. The range, or measures like the standard deviation, shows how spread out the values are.

Does the range use all the data?

No, only the largest and smallest values. That makes it easy to compute but sensitive to a single unusual day.