Looking for feedback on new Google-finance-style graph-based weather site
7 comments
There's such a thing as too much data. I'm looking for information.
I want to be able to set my trip range and make a list of destinations. I want to be able to set options like "I prefer to travel to Hawaii when there is less rain". I want to travel to Paris when the weather is between 60 and 70 and never below 55 or above 85 regularly. I want to avoid clouds and rain at all cost.
The data will turn geeks on, it won't sell to non-geeks. Shoot for information.
I want to be able to set my trip range and make a list of destinations. I want to be able to set options like "I prefer to travel to Hawaii when there is less rain". I want to travel to Paris when the weather is between 60 and 70 and never below 55 or above 85 regularly. I want to avoid clouds and rain at all cost.
The data will turn geeks on, it won't sell to non-geeks. Shoot for information.
Great point. We are 100% on board with this (in fact, your Hawaii use case is one of the actual reasons why we started this project)! As a two-man shop we had to prioritize and decided to create the geek tool first and then start expanding the tabs to cover more focused use cases like yours.
Good luck with the project. Remember less is more; especially when handling copious amounts of data. There's nothing wrong with making a tool that has a touch span of 45 seconds, provided the user leaves with everything or more than expected.
I spent some more time with WeatherSpark this morning, wanted to offer some thoughts on how this would prove to be very useful for me and others like me.
I would like very much to have a trip or travel plotting function. Origin point, traffic, wind, weather and a summary for distinct points in the journey.
For example; departing by car from Richmond, CA at 10:30AM on Saturday and heading to Healdsburg, CA, remaining 5 hours and returning by the fastest route.
Departure time, travel time, weather summary along the way combined with traffic or potential traffic delays. The ability to set how long I will be at each point and a summary of weather impact at such a location. For me, less about the data and more about whether I need foul weather gear, extra water in the car due to delays, or snow chains.
Another example is planning a business trip, or any other kind of trip. Knowing what to expect within a set range of times is much more useful and practical than knowing when the best weather conditions MAY be to support pleasure or optional travel. I do 90% more business travel on demand than I do pleasure.
Example: departing Richmond, CA to San Francisco International Airport. Depart on UNITED 3474 to Cleveland, OH on a specific date and time. Travel by car from Cleveland airport to Wooster, OH, remain for 4 days, return by car to Cleveland Airport, return on UNITED 6761 flight and time from Cleveland, through Denver on UNITED 423 to San Francisco, then obviously, the drive home.
Being able to enter or, even better, upload or automatically link to an itinerary, and understand weather, wind/chill, rain/snow, delays, in a simple and predictive manner is really helpful. As long as it's fast and incorporates into the other tools I use for travel planning.
I would like very much to have a trip or travel plotting function. Origin point, traffic, wind, weather and a summary for distinct points in the journey.
For example; departing by car from Richmond, CA at 10:30AM on Saturday and heading to Healdsburg, CA, remaining 5 hours and returning by the fastest route.
Departure time, travel time, weather summary along the way combined with traffic or potential traffic delays. The ability to set how long I will be at each point and a summary of weather impact at such a location. For me, less about the data and more about whether I need foul weather gear, extra water in the car due to delays, or snow chains.
Another example is planning a business trip, or any other kind of trip. Knowing what to expect within a set range of times is much more useful and practical than knowing when the best weather conditions MAY be to support pleasure or optional travel. I do 90% more business travel on demand than I do pleasure.
Example: departing Richmond, CA to San Francisco International Airport. Depart on UNITED 3474 to Cleveland, OH on a specific date and time. Travel by car from Cleveland airport to Wooster, OH, remain for 4 days, return by car to Cleveland Airport, return on UNITED 6761 flight and time from Cleveland, through Denver on UNITED 423 to San Francisco, then obviously, the drive home.
Being able to enter or, even better, upload or automatically link to an itinerary, and understand weather, wind/chill, rain/snow, delays, in a simple and predictive manner is really helpful. As long as it's fast and incorporates into the other tools I use for travel planning.
Thank you for the suggestion (and for all your comments)! We really appreciate the feedback. Travel is a direction that we are very excited about. I've added your comment to our ticket tracker, so we'll revisit it when we get the travel features.
Very nice interface! I just have to comment since I've been building for a fee weather applications in my spare time for a little while. I know what a mess it can become, especially since you rely on different sources.
I really like how you show the mean and distribution in the background.
I am surprised by the absence of precipitation accumulation, especially since that is often a valuable information for many people who look at long time weather.
Also I don't think the horizontal marker adds information since you show the value for each curve at the cursor, I feel it just clutters the interface.
The meaning of the icons is not always obvious, I am unsure how to interpret two snowflakes out of three or two raindrops out of four etc. It doesn't seem to be tied to the PoP. The intensity maybe?
I really like how you show the mean and distribution in the background.
I am surprised by the absence of precipitation accumulation, especially since that is often a valuable information for many people who look at long time weather.
Also I don't think the horizontal marker adds information since you show the value for each curve at the cursor, I feel it just clutters the interface.
The meaning of the icons is not always obvious, I am unsure how to interpret two snowflakes out of three or two raindrops out of four etc. It doesn't seem to be tied to the PoP. The intensity maybe?
Thanks!
We're working on precipitation accumulation - there are some data fidelity problems that we're trying to solve, then we'll get that in.
The horizontal marker is intended for comparing across the series - i.e. does this peak extend further up than this other peak over there. There's a lot of action going on around the mouse cursor, we'll see if we can pare that down without losing the key information.
Icons: 1 out of 2 snowflakes etc is the expected intensity of the precipitation (i.e. light snow, heavy snow, etc). The probability is reported just next to the intensity in the icon series if you're at a zoom level where the labels fit.
We don't have a legend for the icons yet, need to add that.
We're working on precipitation accumulation - there are some data fidelity problems that we're trying to solve, then we'll get that in.
The horizontal marker is intended for comparing across the series - i.e. does this peak extend further up than this other peak over there. There's a lot of action going on around the mouse cursor, we'll see if we can pare that down without losing the key information.
Icons: 1 out of 2 snowflakes etc is the expected intensity of the precipitation (i.e. light snow, heavy snow, etc). The probability is reported just next to the intensity in the icon series if you're at a zoom level where the labels fit.
We don't have a legend for the icons yet, need to add that.
What do you mean by data fidelity, you mean about how to represent the data so that it is comparable? I can imagine it is a problem especially since your scale can change. I know I am not showing accumulations graphically for a number of reasons myself. For a while I was showing snow accumulation on a graph where it would literally accumulate.
Or something about your data sources? I work with data from Environment Canada and the only annoying thing with the data was making meaningful unit conversions when the nature of the accumulation is snow.
Or something about your data sources? I work with data from Environment Canada and the only annoying thing with the data was making meaningful unit conversions when the nature of the accumulation is snow.
Hi, by "data fidelity" we mean that there are a lot of errors and irregularities in the liquid-equivalent precipitation measurements reported by the stations. Some stations don't report it at all, or not reliably, and some report it but with internal inconsistencies. For example, stations may report in 1-hour, 6-hour, or 12-hour intervals, but they frequently report 1-hour measurements that don't even come close to adding up to the 6-hour measurements. We just want to make sure we take some time to understand these issues with the data feeds before we start publishing them.
The snow depth feed is also a really cool graph. We have it more or less ready, but it is perhaps the least reliable measurement.
For example, during that recent snow storm in Chicago it reported some pretty crazy big snow depth numbers for a couple reports and then without correcting the old reports just dropped the snow depth by more than a meter in the next report. This is fine if you have a human reading the report, but it makes the graph look really weird.
The snow depth feed is also a really cool graph. We have it more or less ready, but it is perhaps the least reliable measurement.
For example, during that recent snow storm in Chicago it reported some pretty crazy big snow depth numbers for a couple reports and then without correcting the old reports just dropped the snow depth by more than a meter in the next report. This is fine if you have a human reading the report, but it makes the graph look really weird.
I absolutely love it, but I'm a total data geek. Just as one anecdotal thing that I got out of the app: 2 weekends ago was ridiculously nice in San Francisco. If you live in SF it was a weird 2 days of perfect weather in the midst of an otherwise fairly cold winter. If you look on the charts there it pops out immediately. There's a huge spike for Feb 5 and 6. Suddenly it jumped from low 60s/high 50s to a nice 72 degrees and then immediately right back down again after the weekend. It was really cool to see my personal experience of the weather quantified like that.
One quick suggestion: show the day of week in the tooltip where you show the date (when zoomed into showing single days at a time obviously). That would make it easier for me to relate the data to my memory of the weather.
One quick suggestion: show the day of week in the tooltip where you show the date (when zoomed into showing single days at a time obviously). That would make it easier for me to relate the data to my memory of the weather.
Thanks - that's exactly the type of observations we're hoping to enable!
Tooltip: Added & deployed, hit reload to get the new app.
Tooltip: Added & deployed, hit reload to get the new app.
Very nice. Climate is what you expect, weather is what you get.
The interface is daunting, though. I'm not quite sure what it's trying to tell me.
http://weatherspark.com/#app;a=USA/CA/San_Francisco;t0=02/16...
It says 2012 on the bottom bar. I'm uncertain whether this is a bug, a forecast for 365 days from now, or an average since 1948 or... what, exactly.
I also don't get a feel for the accuracy of the predictions, especially at what time range. Sure, I can see that in SF winters are wet and summers are dry, but I knew that already. Can it tell me something interesting about a shorter period, like late April?
The interface is daunting, though. I'm not quite sure what it's trying to tell me.
http://weatherspark.com/#app;a=USA/CA/San_Francisco;t0=02/16...
It says 2012 on the bottom bar. I'm uncertain whether this is a bug, a forecast for 365 days from now, or an average since 1948 or... what, exactly.
I also don't get a feel for the accuracy of the predictions, especially at what time range. Sure, I can see that in SF winters are wet and summers are dry, but I knew that already. Can it tell me something interesting about a shorter period, like late April?
The far future is the averages (light blue background) - the forecast is only in the "Forecast"-zone (light purple), but perhaps that can be made more clear (e.g. changing 2012 to "Averages" in the x-axis or something similar).
Accuracy of predictions: the forecasts are as accurate as forecasts are, and we don't influence that.
The averages (mean, 10th percentile, 90th percentile, etc) aren't accurate in any predictive sense - they're just the stats for the recorded history for the station. They tell you what you can reasonably expect, but YMMV.
For SF in late April you can see that the average temperature is steadily rising and the chance of precipitation steadily declining: 62F high on average in early April, 65F towards the end. 8% precipitation vs 5% means almost half as much rain towards the end - I'd hold off until the last week of April if it's an option ;)
Accuracy of predictions: the forecasts are as accurate as forecasts are, and we don't influence that.
The averages (mean, 10th percentile, 90th percentile, etc) aren't accurate in any predictive sense - they're just the stats for the recorded history for the station. They tell you what you can reasonably expect, but YMMV.
For SF in late April you can see that the average temperature is steadily rising and the chance of precipitation steadily declining: 62F high on average in early April, 65F towards the end. 8% precipitation vs 5% means almost half as much rain towards the end - I'd hold off until the last week of April if it's an option ;)
For the far-future accuracy, look at the percentiles. Compare:
http://weatherspark.com/#app;a=USA/CA/San_Francisco;t0=01/01...
To http://weatherspark.com/#app;a=USA/HI/Honolulu;t0=01/01;t1=1...
The temperature spread (10th to 90th percentile) in late September for SF is 64F to 82F for a delta of 18F - it can be very nice, but it can also be kind of crappy.
In Honolulu the temp spread in August is 86F to 91F for a delta of only 5F - it's pretty sure it's going to be nice.
http://weatherspark.com/#app;a=USA/CA/San_Francisco;t0=01/01...
To http://weatherspark.com/#app;a=USA/HI/Honolulu;t0=01/01;t1=1...
The temperature spread (10th to 90th percentile) in late September for SF is 64F to 82F for a delta of 18F - it can be very nice, but it can also be kind of crappy.
In Honolulu the temp spread in August is 86F to 91F for a delta of only 5F - it's pretty sure it's going to be nice.
Totally geeking out on this at the moment. This is absolutely awesome!! The only thing I would say is that you might consider adding an additional simple interface (like a mobile interface) for the less savy viewer. I think the ability to visualize the forecast is really powerful, I'm just picturing my mom looking at this and seeing too many options.
Options are great for us data geeks, but a very simple city forecast page that harnessed your forecast visualization technique could be great for a much wider audience as well.
Options are great for us data geeks, but a very simple city forecast page that harnessed your forecast visualization technique could be great for a much wider audience as well.
Great suggestion! Yes, we are excited to start playing around with ways to make it more accessible to a wider audience. But we thought it would be fun to start with the geek tool :).
Overall I really like it, I don't know whether it's intentional but for today in Oakland CA the area to the left of the yellow "now" bar in cloud and precipitation is blocky columns that don't really tell much. To the right is the great graph of prediction vs history and I'd expect this to look similar to the left but reflect actual data instead of prediction.
For precipitation the prediction & averages are in percent, but the past is in actual hours of precipitation - hence the blockiness for the past (the vertical axis is from 0-1 hours when zoomed in).
For clouds it's percent cloud cover, so a full block means fully overcast.
For clouds it's percent cloud cover, so a full block means fully overcast.
Wow -- this is great. Canadians love to talk about the weather and this provides endless conversation topics!
Looking at the recent data for Ottawa, ON there appears to be a glitch on Jan 17, 2011. When zooming in and out the high line jumps up to 19C at some scales and falls back to normal at others. It wasn't that hot :(
Looking at the recent data for Ottawa, ON there appears to be a glitch on Jan 17, 2011. When zooming in and out the high line jumps up to 19C at some scales and falls back to normal at others. It wasn't that hot :(
Yes, the raw data feeds have occasional blips like that (more so in some stations than in others).
It is a top priority to write some good automatic quality control code, but we didn't want to do a bad job and end up accidentally filtering out real data (we've discovered, for example, that some of NOAA's quality control filters have very high rates of incorrectly deleting measurements that are actually extremely plausible).
We want to take a proper statistical approach, which requires labeling a lot of examples of these kinds of blips, and we just haven't had time to do that yet.
It is a top priority to write some good automatic quality control code, but we didn't want to do a bad job and end up accidentally filtering out real data (we've discovered, for example, that some of NOAA's quality control filters have very high rates of incorrectly deleting measurements that are actually extremely plausible).
We want to take a proper statistical approach, which requires labeling a lot of examples of these kinds of blips, and we just haven't had time to do that yet.
I'm not sure this is a raw data blip, since it seems to change based on the zoom level.
19C appears to be the historical high for that day, but is sometimes included in the band indicating the actual range.
19C appears to be the historical high for that day, but is sometimes included in the band indicating the actual range.
It is definitely a blip. The reason it doesn't appear in all the zoom levels is that the four-hourly zoom level takes the average temperature of the hourly data over the four-hour period. Once you go up to the daily zoom level we switch to showing the high and the low, and you can see the blip again. That historical high is due to that point.
Actually, I'm just sitting down today to start the process of writing quality control filters to get rid of this sort of blip. So hopefully it will be gone soon! :)
Actually, I'm just sitting down today to start the process of writing quality control filters to get rid of this sort of blip. So hopefully it will be gone soon! :)
The site contains all of the worldwide weather station data available back to 1948 (down to the hour!), and makes it available in an interactive graphing interface. In addition to all the historical data, there are also integrated forecasts from several sources, and detailed records and averages (also down to the hour!) that are shown in the background in order to provide context.
The site is intended to serve your daily weather needs, to help inform long-range travel decisions, and to allow geeking out on the raw data.
A few cool random views:
An average day near the end of July in San Francisco (it is cloudy in the morning, clear in the afternoon, and almost never rains): http://weatherspark.com/#app;a=USA/CA/San_Francisco;t0=07/23;t1=07/28
The monsoon in Mumbai, India (the rain is so dominant that it cools down the air for three months, creating a neat multi-modal yearly temperature curve): http://weatherspark.com/#app;a=India/Mumbai;t0=01/01;t1=12/31
That cold snap in Dallas around the Super Bowl this year that was all over the news (turn on the wind speed and direction graphs and you'll see the wind change from southerly to northerly just as the cold snap starts): http://weatherspark.com#app;a=USA/TX/Dallas;t=360209;mspp=900000
Average late July in Houston (the thunderstorms roll in in the afternoon around 4pm typically): http://weatherspark.com/#app;a=USA/TX/Houston;t0=07/23;t1=07/28
The full year of 2010 in my home town (Madison, WI): http://weatherspark.com/#app;a=USA/WI/Madison;t0=2010/1/1;t1=2010/12/31
Please take a look and explore the data. Don't forget to pan & zoom, and to turn on other graphs! :)
Cheers, James