And the snow was circling and falling ...
The MyClime project ( http://myclime.info or http://mycli.me ) continues to evolve as planned and today I will tell you which new cities appeared on the service, ask your advice about precipitation and tell you how to use this service :) MyClime is a service for analyzing weather forecasts from various sources and checking their accuracy. For each day, a history of forecasts is maintained and, upon the occurrence of that day, a confidence rating is calculated. Forecasts are downloaded from source sites, as well as actual weather.
So, we have information in the system for all the cities with the population of millions in Russia:
The second major innovation is precipitation forecasts for all sources of forecasts. Now you can see who predicted how much rain and when. Actually, my need to listen to advice is connected with precipitation. Now precipitation is not used in any way when evaluating forecasts, but it seems to me that they must be taken into account. And then the following questions arise - what weights should be at the temperature and precipitation when calculating the rating? For example, if the temperature accuracy is 70% per day, and the precipitation is only 20%, then it seems to me not very logical to sum them up with an equal coefficient of 0.5 - well, what’s the difference of +2 degrees or -2 degrees if I got rained. On the other hand, it is also impossible to make a large coefficient for precipitation due to the fact that the difference between “light rain” and “occasional rain” or “cloudy” and “cloudy” is quite subtle and the main plausibility criterion will be “there was rain — there was no rain” . That's why I wanted to ask for advice on how it would be more convenient for you as users. Maybe, for example, several ratings are separate for precipitation, separately for temperature, but here again, what to do, for example, with the wind is another number? It would be interesting to hear your opinion.
')
And finally, the third important point is that many questions from users have accumulated during the work, and the site usability is at the level of the old, Soviet ZIL refrigerator, so I would like to answer all with a brief instruction manual :) (By the way, if anyone knows an adequate designer who will be able to draw well and the site and sets of icons and is still set to long-term cooperation, throw contacts please)
So let's get started - on the main page there is a list of available sources and the latest forecasts that they give. You can select the city on the right in the drop-down list. If there is a desire, then on the right you can change the display units - Fahrenheit or Celsius.
Opposite each source is given its rating in percent. In order to calculate the rating of each source, I use such indicator as reliability. In this case, the accuracy of the forecast for any date X for N days is how long the forecast for N days corresponded to the actual weather on that day. Those. if the forecasts for today are “today minus 5 days”, “today is minus 4 days”, “today is minus 3 days”, “today is minus 2 days”, “today is minus 1 day” were +1, +2, 0, -1 , +3, with actual weather of 0, the accuracy of the forecast from the source for today is 20%. (In reality, the formula is much more complicated, but if in the first approximation, then everything looks that way).
And the rating in this case is the average confidence for 5 days. The rating value can be used to understand whose forecast was the most reliable lately, and whose value can be guided in the next few days.
If you click on the "Statistics for the Month" link, a page with information on the reliability for the last month will open.
The pie chart shows the number of the most reliable forecasts for the last month for each source for a particular city. The histogram shows when and by whom the most reliable forecasts were made. This information can be used to make a decision on the use of the source in the long term (show the forecast on the site or use it as a source) - for several months or more.
If you return to the main page and click on the value of temperature or precipitation, a graph with the history of the forecast change will open.
The right vertical axis corresponds to the date for which the history was made, and the graph shows how the value of the predicted temperature changed as it approached the day on which the forecast was made. For example, in the picture, the forecast made on March 3rd on March 10th showed -12 degrees, the forecast on March 5th on March 10th showed -7, etc. The blue area highlights the actual temperature on March 10th. Thus, visually, the reliability of the forecast is how many times the orange graph has been inside the blue area.
Well, just answers to the most frequently asked questions:
1. Please add a list of sources rp5.ru
rp5.ru does not prepare its forecasts, but takes them from the British service MetOffice, whose forecasts are used by the BBC. So the rp5 forecasts are the same as the BBC forecasts.
2. Where do you get the actual weather values
The actual values ​​of weather sites are published from various meteorological stations. This data I use to make a picture of the actual weather.
So, we have information in the system for all the cities with the population of millions in Russia:
- Yekaterinburg
- Kazan
- Moscow
- Nizhny Novgorod
- Novosibirsk
- Omsk
- Rostov-on-Don
- Samara
- St. Petersburg
- Ufa
- Chelyabinsk
The second major innovation is precipitation forecasts for all sources of forecasts. Now you can see who predicted how much rain and when. Actually, my need to listen to advice is connected with precipitation. Now precipitation is not used in any way when evaluating forecasts, but it seems to me that they must be taken into account. And then the following questions arise - what weights should be at the temperature and precipitation when calculating the rating? For example, if the temperature accuracy is 70% per day, and the precipitation is only 20%, then it seems to me not very logical to sum them up with an equal coefficient of 0.5 - well, what’s the difference of +2 degrees or -2 degrees if I got rained. On the other hand, it is also impossible to make a large coefficient for precipitation due to the fact that the difference between “light rain” and “occasional rain” or “cloudy” and “cloudy” is quite subtle and the main plausibility criterion will be “there was rain — there was no rain” . That's why I wanted to ask for advice on how it would be more convenient for you as users. Maybe, for example, several ratings are separate for precipitation, separately for temperature, but here again, what to do, for example, with the wind is another number? It would be interesting to hear your opinion.
')
And finally, the third important point is that many questions from users have accumulated during the work, and the site usability is at the level of the old, Soviet ZIL refrigerator, so I would like to answer all with a brief instruction manual :) (By the way, if anyone knows an adequate designer who will be able to draw well and the site and sets of icons and is still set to long-term cooperation, throw contacts please)
So let's get started - on the main page there is a list of available sources and the latest forecasts that they give. You can select the city on the right in the drop-down list. If there is a desire, then on the right you can change the display units - Fahrenheit or Celsius.
Opposite each source is given its rating in percent. In order to calculate the rating of each source, I use such indicator as reliability. In this case, the accuracy of the forecast for any date X for N days is how long the forecast for N days corresponded to the actual weather on that day. Those. if the forecasts for today are “today minus 5 days”, “today is minus 4 days”, “today is minus 3 days”, “today is minus 2 days”, “today is minus 1 day” were +1, +2, 0, -1 , +3, with actual weather of 0, the accuracy of the forecast from the source for today is 20%. (In reality, the formula is much more complicated, but if in the first approximation, then everything looks that way).
And the rating in this case is the average confidence for 5 days. The rating value can be used to understand whose forecast was the most reliable lately, and whose value can be guided in the next few days.
If you click on the "Statistics for the Month" link, a page with information on the reliability for the last month will open.
The pie chart shows the number of the most reliable forecasts for the last month for each source for a particular city. The histogram shows when and by whom the most reliable forecasts were made. This information can be used to make a decision on the use of the source in the long term (show the forecast on the site or use it as a source) - for several months or more.
If you return to the main page and click on the value of temperature or precipitation, a graph with the history of the forecast change will open.
The right vertical axis corresponds to the date for which the history was made, and the graph shows how the value of the predicted temperature changed as it approached the day on which the forecast was made. For example, in the picture, the forecast made on March 3rd on March 10th showed -12 degrees, the forecast on March 5th on March 10th showed -7, etc. The blue area highlights the actual temperature on March 10th. Thus, visually, the reliability of the forecast is how many times the orange graph has been inside the blue area.
Well, just answers to the most frequently asked questions:
1. Please add a list of sources rp5.ru
rp5.ru does not prepare its forecasts, but takes them from the British service MetOffice, whose forecasts are used by the BBC. So the rp5 forecasts are the same as the BBC forecasts.
2. Where do you get the actual weather values
The actual values ​​of weather sites are published from various meteorological stations. This data I use to make a picture of the actual weather.
Source: https://habr.com/ru/post/54026/
All Articles