The daily bike counter model is built using the following variables:
Max TemperatureF
Min TemperatureF
Max Humidity
Max Wind SpeedMPH
PrecipitationIn
weekday
holiday
day_hrs
snowtotal
Fog
year
sunsethour
CloudCover
bike_events
I'd be interested in getting your feedback. I'm also interested in learning about any other bike events besides Bike to Work day that would impact traffic on the Custis Trail. Thanks!
I really need to log off the internet and go for a ride.
Join Date
Dec 2010
Location
Columbia Heights, DC
Posts
3,963
Post Thanks / Like
Likes (Given)
3118
Likes (Received)
3883
Dislikes (Given)
2
Dislikes (Received)
34
ELITE (Given)
140
ELITE (Received)
405
So you're a Python coder, huh? You should definitely be involved in maintaining the Freezing Saddles website.
I'm curious how much the weather forecast being wrong will affect your predictions. Maybe it's just me, but I feel like this spring the forecasts have been all over the place and haven't had much to do with the actual weather.
Just wanted to chime in and agree with jrenaut - I think accuracy could potentially be improved by looking not at what the weather actually was, but what the weather was forecast to be. People make decisions on whether to ride or not based on the forecast. So if it says 70% chance of rain, they decide not to ride and so the counts are lower, whether it actually ends up raining or not.
I really need to log off the internet and go for a ride.
Join Date
Dec 2010
Location
Columbia Heights, DC
Posts
3,963
Post Thanks / Like
Likes (Given)
3118
Likes (Received)
3883
Dislikes (Given)
2
Dislikes (Received)
34
ELITE (Given)
140
ELITE (Received)
405
Originally Posted by chris_s
Just wanted to chime in and agree with jrenaut - I think accuracy could potentially be improved by looking not at what the weather actually was, but what the weather was forecast to be. People make decisions on whether to ride or not based on the forecast. So if it says 70% chance of rain, they decide not to ride and so the counts are lower, whether it actually ends up raining or not.
I appreciate the credit for an idea I didn't even have - I wasn't thinking of it that way. It's probably hard to get historical forecast data, but it would be interesting to compare whether forecast data or actual data is a better predictor of riding.
My short city commute (5 mile roundtrip to the kids' school and back home) is probably less influenced by weather than someone with a 20 mile commute (who also has to make a decision about riding a lot earlier than I do).
I really need to log off the internet and go for a ride.
Join Date
Sep 2015
Location
Arlington
Posts
2,588
Post Thanks / Like
Likes (Given)
2576
Likes (Received)
3791
Dislikes (Given)
39
Dislikes (Received)
72
ELITE (Given)
311
ELITE (Received)
258
Originally Posted by dasgeh
I wonder whether it's forecast or actual whether in the morning (or when people leave the house). There's a cool econ paper in there
I think it's probably a moving target. I ride every day no matter what, but I adjust the timing based on weather. For commuting, I don't have much choice except leaving a bit later or earlier. On weekends with rain forecasts, the forecast will often change dramatically throughout the day and I adjust timing and duration accordingly.
I really need to log off the internet and go for a ride.
Join Date
Dec 2010
Location
Columbia Heights, DC
Posts
3,963
Post Thanks / Like
Likes (Given)
3118
Likes (Received)
3883
Dislikes (Given)
2
Dislikes (Received)
34
ELITE (Given)
140
ELITE (Received)
405
Originally Posted by Judd
I think it's probably a moving target. I ride every day no matter what, but I adjust the timing based on weather. For commuting, I don't have much choice except leaving a bit later or earlier. On weekends with rain forecasts, the forecast will often change dramatically throughout the day and I adjust timing and duration accordingly.
Just wanted to chime in and agree with jrenaut - I think accuracy could potentially be improved by looking not at what the weather actually was, but what the weather was forecast to be. People make decisions on whether to ride or not based on the forecast. So if it says 70% chance of rain, they decide not to ride and so the counts are lower, whether it actually ends up raining or not.
I wish I had access to Arlington weather forecast data back to 2009, but most of the sources just keep actual historical weather. Do any of you know of a data source for historical forecasts? As an economist, I'd love to "borrow" your research idea testing whether forecast or actual morning weather impacts the number of bike rides taken.
On a different note, I noticed that there is a higher than expected numbers of trips taken on Wednesdays versus other weekdays. Does anyone know why?
Last edited by georges083; 05-18-2017 at 05:08 PM.
Reason: typo
I wish I had access to Arlington weather forecast data back to 2009, but most of the sources just keep actual historical weather. Do any of you know of a data source for historical forecasts? As an economist, I'd love to "borrow" your research idea testing whether forecast or actual morning weather impacts the number of bike rides taken.
On a different note, I noticed that there is a higher than expected numbers of trips taken on Wednesdays versus other weekdays. Does anyone know why?
Wednesday has the fewest teleworkers/flex dayers. If seen a study of federal government workers that indicated as much...
One nice thing about using actual weather rather than forecast is that there are a lot fewer variables to consider. For instance: Whose forecast? How far in advance?
I'd be surprised if any forecast source turned out to be a conclusively better predictor of ridership than actual weather measurements.
Bookmarks