Developing intuitive probabilistic meteograms: Increasing public confidence in weather forecasts

[milankl]

Problem A 7-day weather forecasts is now as accurate as a 5-day forecast 20 years ago, thanks to faster supercomputers and decades of scientific weather forecast model development. However, there is still a widespread public misbelief that forecasts are not reliable. Traditional meteograms (i.e. a plot summarizing the weather for a certain location throughout the next days or week) do not contain any information about the uncertainty of a weather forecast. A paradigm exists where a layperson is not expected to understand the uncertainty of a forecast. Especially weather apps tend to present forecasts in an overly simplistic way that may support the aforementioned little public confidence in weather forecasting.

Example: Every forecast of an ensemble forecast might propose a little bit of rain in the mornings or evenings over the coming weekend. The overly simplistic way of communicating this forecast as "Rain throughout Saturday and Sunday" gives one the impression of an unsuitable weekend for camping. In contrast, the information that in each forecast it actually rained <5% of the time is not well communicated and could have been presented in an intuitive way, understandable for a layperson.

Proposed solution

The solution I propose is an inuitive way of visualizing the probabilistic information of an ensemble forecast in the form of a meteogram (as the public is mostly interested in the weather at a given location). By "intuitive" I mean a meteogram that is easy to understand for everyone without a statistics background. It is supposed to present the future weather at a given location in a very condensed way, such that 10 seconds are enough to understand what you are likely to expect and what the possible extremes are. I believe that the additional information about the uncertainty of a forecast will increase public confidence in weather forecasting, important for a wider outreach of publicly funded data.

Data (what data do you need? what format?)

A set of historic 10-day probabilistic forecasts (i.e. all ensemble members, the high resolution deterministic forecast and the analysis for comparison) for a only few different cities and start dates around the globe, preferably also including some severe weather events. Typical variables such as Temperature, Wind, Cloudiness, Precipitation. NetCDF is fine.

Team

• Milan Klöwer - role: meteorologist, statistician, data scientist, and graphics.
• You - role: being motivated and enthusiastic
• You
• and you!

Is your team incomplete and you need extra help?

Yes. Get involved.

[kaedonkers]

Interesting proposal! Do you need a Python programmer?

[OpenDataHack2018]

This challenge was selected by team: Meteogram