A map that predicts the San Francisco fog, using data from 130 personal rooftop weather stations. Live version here.
##GaSiProMo
I'm using this project for GaSiProMo. My goal for the month is to build a machine learning algorithm that does a better job forecasting fog than my current version. I'll be keeping a log of my progress through the month here.
If you have feedback, please share on my GaSiProMo Reddit thread!
###11/1/2015 GaSiProMo begins! This week, I'm going to find a training data set – times when flights were delayed at SFO airport because of fog, and the preceding weather conditions. I'm going to use this to train a forecasting model for fog. Read the full log entry for more info.
Fog or Not makes hourly predictions about how much fog there will be in San Francisco, for each neighborhood.
The predictions are based on micro-local weather data from hundreds of personal weather stations across the city. These are basically small monitoring units that weather geeks like me put on roofs and outside windows.
Fog or Not is built on the Weather Underground API, which aggregates data streams from over 100K+ personal weather stations around the world. (If you're into weather, it's a really neat data stream to explore.)
The prototype in R lives here.
The production back-end is in Python using Pandas, NumPy and Shapely, with Flask architecture running on Heroku. I created the visualization of forecasts by applying GeoJSON layer on top of Google Maps with jQuery/HTML5.