My smartphone has a weather app on it that gives a forecast of the next six days. I glance at it occasionally, but in September, we had our annual “Away Day” at work, which has a significant outdoor component. So I was watching the forecast quite carefully. It said it was going to be sunny. Great! Next day it was predicting cloud, then it changed back to sun, then to rain. Umm. Well, on the day it was fine, fortunately.
But all that made me a bit suspicious of the app. I know longer range forecasts are essentially useless, with the weather being a chaotic system, but I thought 4 or 5 days out was now in the bounds of possibility. So I’ve been keeping data from my app for the last month or so.
|click to see the icons|
Each strip of 6 icons represents the forecast for a given day: the item near the top the forecast six days in the future, the one at the bottom from just the day before. There’s a lot of variation.
Here’s a graph summarising the data.
|click to see the scales|
- #symbols (light green): the number of different forecasts given over the six days: a consistent forecast would use one symbol, with more symbols showing greater indecision. Over the 32 days of data collection, it managed the perfect “one symbol” 4 times (12.5%).
- #changes (mid green): the number of times the forecast changed its mind. This minimum possible value is #symbols – 1; a higher figure indicates vacillation.
- end game (dark green): the number of days the final symbol stayed constant. The maximum possible is six: a perfect six day forecast (managed 12.5% of the time); the least reliable is one day (managed 15 times, or 47% of the time.)
I think I’ll go back to the trusty classic “Seaweed dry, sunny sky. Seaweed wet, rain you'll get.” We can add more states, with the well-known joke: “Seaweed gone: wind so strong!” And to bring it fully up to date we simply add: “raining seaweed, weird indeed”.