## Starting Point: Data and Task

Windfinder kindly provided weather data (wind speed, wind direction, air temperature, ...) and the page views for 7 different surfing spots. Our task was to investigate if page views can be used as a proxy for good surfing conditions, what such a model could look like, and what other insights can be found in the data.

## Team

Three Kiel-based students eager to find interesting patterns in data (and playing geotastic along the way) - Emil, Benjamin, and Jan.

## Challenges

Optimal conditions Optimal wind surfing conditions varied substantially between the surfing spots. While for one surfing spot the optimal wind direction would be 270 degrees for example, for another surfing spot the optimal wind direction would be 180 degrees. Thus, it is necessary to transform those values into a scoring. Additionally, the wind speed suffers from similar issues, since a faster wind speed doesn't mean the surfing conditions are more optimal.

Statistical approach Also, if you'd like to use a classical statistical approach (like var(p)) to determine the influence of wind direction and wind speed on page views, not all of the given time series are easily made stationary. Especially the wind speed time series is highly volatile and hard to decompose.

Pageviews as an indicator for good surfing conditions or more visits from windsurfers Pageviews is probably not a a good indicator for good surfing conditions, since people tend to look for good weather conditions in advance.

Lack of data Not really given, how many surfers were there on a certain spot. Also aspects like which characteristics the surf spots have, was to be explored.

## Learnings

Optimal weather conditions are tricky to identify, and while page views seem to be correlated, they are far from being a perfect predictor for good surfing conditions.

If you want to force page views as proxy for weather conditions, a cluster-based model seems like a promising approach to rate wind conditions. For more details, check the presentation on github (link is somewhere below).

Fun finding: people in Denmark (at least parts of Denmark) are interested in the weather and wind data in late December - maybe crazy wind surfers or sailors who want to spend new years eve swimming in the Baltic Sea (or they just got new surfing equipment for Christmas and are very eager to try it out...).

If you lack the theoretical background for a method and haven't yet really build a method before, it's quite challenging to finish this within a 24 hours time window.

You can meet a lot of same minded people in a Hackathon, a lot more fun than to work alone or in usual work atmosphere.