Carleton Comps Project: Computing the Weather

Final Results

Computing the Weather

Advisor: Sherri Goings

I. Background

People all over the world love to complain about weather forecasts. How many times have you yourself looked at the current weather on some forecast site to see something like

Today's weather: High Temp: 72 Current Temp: 78

Today's weather: Chance of rain: 0% (as you watch rain hitting your window outside)

Bad enough that many sites can't update on a regular enough basis to even make their forecast correct for the current weather, but how accurate are their predictions for a few hours from now? Tomorrow? These now popular "extended" forecasts 6,10, or even 14 days in the future? Do we have any idea when it seems no one rates weather sites?

I don't know about you, but when I plan on taking the dogs for a run or any other outdoor activity, I don't look at the "30% chance of rain" for the day, but rather check out the radar map. It seems to me that from watching the rain (if any) on the radar move for the last hour, I can reasonably predict how soon it's likely to get to where I am, or to end. Wouldn't it be great to get at least an accurate short-term forecast from a weather site? Say something like "will it be raining in 30 minutes?" or "at what estimated time will the rain I see on the radar reach my house?" Is this sort of accurate, short-term forecast so difficult or is it just that weather sites don't want to update that often? I would like to find out!

II. The Project

The overall goal of this project is to develop an as accurate as possible "VSTF" (Very Short Term Forecast) system, and to design and build a customizable weather application that incorporates this forecast.

Some of the steps that you will take to achieve that goal are:

Research what methods are commonly employed by commercial sites to predict the weather
Discover and obtain available past weather data you can use to train a system to make future predictions. The NOAA has made publicly available data from the last 20 years or so, so there's lots to train on, but you will need to be able to store, parse, and analyze this data. What format is data for different weather parameters provided in? What does raw radar data look like and how can you convert it to something useful?
Choose one or more techniques to create a prediction system. Machine learning? Evolutionary algorithms? Mathematical modeling? Research the chosen method so that you can apply it as effectively as possible to the type of data available.
Develop a customizable weather page (and/or phone app) such that each user can get exactly what information they want in the format they want it. This is not a trivial piece of the project, but rather should be a near-production quality application that I would actually enjoy using. This will involve working with web-design, HCI, and incorporating actual feedback from users.

III. References

In the fall, you'll work with a librarian to do a thorough literature search to find out what others have done in this area (there is lots of work out there on weather prediction). Other references will highly depend on what method you choose to create your prediction system. However here are a few relevant resources.