0

weather station selection algorithm

Di Xue

April 12, 2021 13:04

None

Follow

Hello!

Thanks a lot for providing this awesome api.

I am using coordinates to get the current timestamp's weather data.

I noticed both in the documentation and by experimenting myself that your api is not using the closest weather station's data directly. Instead, there are a couple of weather stations being considered. However, I didn't find more descriptions about how those weather stations are selected given a coordinate as input.

I wonder if this selection algorithm is a secret, i.e., closed-source, or I just didn't manage to find the right doc page. If this part is not closed-source, could you please provide some reference/description about the algorithm you use to select the weather stations?

Very appreciate getting an answer about this!

Thank you in advance.

• Facebook
• Twitter
• LinkedIn

4 comments

• 
  info April 12, 2021 18:17 Official comment

  Thank you for your question. The stations are chosen for an observation are chosen based on a combination of the distance from the location to the station and also the quality of the weather station.

  Higher quality weather stations are chosen from further distance than a closer station that reports missing or suspect data. When the quality of all the candidates are the same and there is no missing data, the closer stations are weighted using a decay proportional to the square of the distance - therefore a closer station will be weighted more heavily than others.

  A common case we see is that some weather stations do not report all  the weather elements (for example rain, wind or snow information). 

  In addition, stations are calculated on a row by row basis. This allows us to bring in additional station information if a closer station is offline for maintenance etc. This will manifest itself as a changing set of stations on a row by row level.

  Finally, in some cases, a further station will be chosen because it is the only station in a location that reports a specific weather element. Some automated stations are not equipped to measure some weather data elements such as solar radiation or snow depth.

  If you have a specific query you would like us to investigate, please let us know and we can explain how the stations were chosen.

  Regards

  Visual Crossing Support

   

   

• 0
  
  Di Xue April 12, 2021 19:41

  Thank you very much for your detailed explanation!
  For example I am now running this query:

  response = requests.get('https://weather.visualcrossing.com/VisualCrossingWebServices/rest/services/timeline/57.67417%2C12.00722/today?unitGroup=metric&key=MY_API_KEY&lang=id&include=current')

  One of the keys in the response is as below:

  'stations': {'E2142': {'contribution': 0.0,
                           'distance': 14176.0,
                           'id': 'E2142',
                           'latitude': 57.801,
                           'longitude': 11.993,
                           'name': 'EW2142 Gothenburg SE',
                           'quality': 0,
                           'useCount': 0},
                 'ESGG': {'contribution': 0.0,
                          'distance': 17435.0,
                          'id': 'ESGG',
                          'latitude': 57.67,
                          'longitude': 12.3,
                          'name': 'ESGG',
                          'quality': 50,
                          'useCount': 0}}

  There are two stations included in the response, E2142 is closer but has the "quality" key as 0, while ESGG is further while has the "quality" key as 50. In this case is the E2142 has a weight that is 1/(14176.0)^2 while ESGG has a weight that is 1/(17435.0)^2 ?

  Another thing I wonder is, even though a location has a closer weather station with good data quality, the final weather data will still take in at least one more other weather station to get a weighted averaged value? Is there a case when only one weather station's data is used even there are other somehow further away weather stations with same quality data at surrounding? 

• 1
  
  info April 13, 2021 03:48

  Thank you for the query example. Current condition queries are a little bit different from regular history queries as they optimize for stations that are as close as possible to the requested location that has reported very recently. This helps provide current conditions with the most timely and accurate recent observations. In general the current conditions will only use one or two stations whereas normal history queries will use three or more stations.

  In your case, the E2142 station provides the closest, most recent observation. However this station is a 'personal' weather station (often operated by a private individual, school or local government such as Department of Transportation).

  Whilst these personal weather stations typically produce the most timely and localized conditions for the basic parameters (temperature, precipitation, pressure and wind), they are often not able to record more advanced properties. These advanced parameters include items such as cloud cover and visibility - both of which require sophisticated equipment, such as laser powered 'ceilometers'.

  In these cases, we pair the local weather station with another, full service weather station which in your case is the ESGG (Gothenburg-Landvetter Airport) which is able to provide the more advanced properties.

  The data for the stations  that exists for both stations (such as temperature), will be combined according to the standard rules. 

  For normal historical observations, we will attempt continue to add additional weather stations until each hourly bucket of data has at least the minimum number of weather stations (by default 3) even if we find a station with 100 rating quality. This allows us to perform the interpolation.

  Finally, yes the overall weight of the station when combining records is 1/(distance^2).

  Regards

  Visual Crossing Support

   

• 0
  
  Di Xue April 14, 2021 11:03

  Thank you very much for your explanation! This info is really helpful. Much appreciation!