ML Blog

I took up sailing for the second time in my life about mid-way through grad school. Madison was a great place for it, because (in case you didn't know), the city is basically surrounded by lakes. The biggest of those is Mendota, which houses the Memorial Union and Terrace on its southern shore. The weather on Lake Mendota is monitored by a tower weather station (on-shore) and a buoy (in the middle of the lake, nicknamed "David Buoy"), so interested sailors can check current conditions and trends to plan their sails. These data are continuous, arriving as frequently as every minute (seriously, we've gotten advance notice of thunderstorms based on these readings), and archived online.

I was curious what single week (starting on a Monday) would have been the best week to sail. I decided to use the public data from the three most recent summers to find when I should have been sailing rather than working. You can see the full notebook at this link, but I'll summarize below.

The data can be accessed through a public API in CSV form, and so I downloaded temperature and wind data from the AOSS Tower and the Mendota Buoy. On the left are the measurements from 2018-2020.

I noticed that sometimes only one of the buoy and tower stations are active. I decided to aggregate the temperature and sustained wind measurements between those two sources: when both are active, you take the average, and when only one is active, you believe that measurement. Averaging wind directions requires circular statistics, which is basically just averaging the sines and cosines of the wind angles. The figure on the left shows precipitation and gusts, too, but both are only available on one source, so I elected not to use them.

Since I prefer to sail in sunny, warm weather and light to moderate winds (10-18 mph); and since we're not allowed to sail after dark, I had to find the weeks with the largest fraction of daylight hours that satisfy those criteria. I first used a library called suntime to find the local sunrise and sunset times for Madison, and filtered the measurements by whether they happened during daylight hours. Next, I applied simple cuts to temperature (60-90 F) and wind speed (10-18 mph). I also want to exclude weeks with a lot of south-wind days (which are frustrating on a southern lakeshore), but that's handled separately.

I then grouped all measurements by their year & week, and computed the fraction that satisfied my temperature and wind criteria. I separately computed the fraction of measurements reporting a southern-quadrant wind (145-225 deg) and excluded them. The best week, the week of September 23, 2019, had a whopping 51% of daytime measurements reflecting good conditions! Even more interestingly, the data indicate that the best sailing weather tends to come in the fall (mid-September through early November). This squares with my experiences in summer, where winds are calm. The old-timers in my sailing club tell me the summers have gotten worse over the past 10 years, and the falls have gotten a little better, but I don't have access to historical data that could speak to that.

Thanks for reading! Once again, the full notebook describing my process is here.