Weather interpolation

[Arcadia197]

Hello dear community of wetterdienst.

I recently became aware of your project and am interested in its development and would possibly want to contribute (as far as I will have spare time avaiable for more projects).

I didn't fully explore all capabilities of the package yet, but as far as I noticed, you mainly focus on weather data avaiable from the stations at specific locations. Have you already thought about implementing an interpolation of nearby stations to approximate weather results at a desired location (for example your own position)? This may be of high interest for users wanting to extract high quality weather data for their positions.

Of course for some weather parameters it doesnt really make sense, but especially for steady values (temperature, pressure) this should be easily achieveable with simple interpolation models.

Greetings,
Julius

[gutzbenj]

Dear @Arcadia197 ,

thanks for reaching out to us! You are right that we currently only allow for acquisition of actually existing stations. I absolutely agree with you that interpolation of those rather large scale parameters such as temperature should be farely easy.

At the moment we have strictly separated existing stations (e.g. DWDObservationStations) and Values of those stations (e.g. DWDObservationValues), but for the future we have in mind to implement kind of an interception system for the transition from stations to values. This is also the place where I would see the interpolation e.g.

stations = DWDObservationStations(parameter, ...)
stations.interpolate(latlon)  # calls stations.nearby_number(), returns InterpolationResult

For this to work we will have to define parameters that can be interpolated. Also we have to make sure that the requested parameter is not actually a parameter set (which could contain parameters which can't be interpolated).

Another problem that would remain is: How do we cope with missing values? Let's say we choose a standard of 5 neighbouring stations for the interpolation (I think it shouldn't be more then that) but 2 of the 5 stations don't have values at a given time...

[Arcadia197]

Hey Benajmain,

thanks for your reply. Nice to hear, that features like this are taken in consideration for future improvements of the library.

Not only would the avaiability of data be something to take into account, but also the actual position of the stations. This could result in some sort of spatial (from which and how many stations do we interpolate the data?) and temporal accuracy (do we want to compute the parameter at a time, where some stations do not have data avaiable and we want to interpolate this as well?)

I guess this can be resolved with some testing, using different ideas and methods (interpolation order, -method) to find something, which fits expectations of standard users.

Greetings,
Julius

[wetterfrosch]

Hi there, just a little thought which came up when reading this issue.

The very local environment (soil, vegetation, topology) has enough effect that one shouldn't mistake an "interpolated" reading as a reading. It might lead one wrong. To get into a hell with no end; one could try to integrate the stated parameters ... or simply assume that our data-provider DWD does this already in order to create the baseline for his forecasts.

Concluding this I would assume that the "current-hour forecast"-data of the grid-wise generated forecast-model (ICON(-EPS)) might be more realistic/accurate than a self-created "interpolation". The "german" grid (ICON-D2(-EPS)) goes down to a resolution of 1.2km per rectangle-side, if I recall correctly.

ps.: Julius, you didn't specify the tense. It would be even more "funnier" when this request would concern the point-station forecast-data (MOSMIX); these are derived from the grid-forecast -- and then integrated ("mixed"/optimized/corrected) with some "statistical knowledge" of the recorded behaviour at that particular station.

So when one creates an interpolation between two MOSMIX-station-forecasts, one would try to "de-localize" the "especially localized" data to get most likely less than on what the localized data is already derived from. (The part I regard as funny ;)

[gutzbenj]

Hey there,

@Arcadia197
I think that the method for interpolation is rather resarched already as you can find lots of different research papers about this topic. However as you said it will be rather hard to work parts of missing data and how to convey that to the user (e.g. what you said, interpolation accuracy, distance, number of stations, etc).

As for the general concept I will be working on the API in the next couple of weeks to tie together Stations and Values and I think this will allow us to also create more easily interpolations - besides the actual way we conceive interpolations. Do you already have an idea about the general concept (with the tools that already exist in this library)?

@wetterfrosch
Actually when looking at the desciption of Mosmix you find

Datenverfügbarkeit
Es werden MOSMIX-Vorhersagen für ca. 5400 Stationen weltweit berechnet. Aufgrund ständiger Qualitätssicherung kann sich die Anzahl der Stationen jedoch in unregelmäßigen Abständen ändern. Circa 2800 der weltweit berechneten Stationen sind sogenannte Entwicklungsstationen (auch "Hauptstationen genannt"), d.h. MOS-Koeffizienten werden für jede einzelne dieser Stationen auf Basis von historischen Beobachtungen und Modelldaten explizit für eine stations-spezifische Anzahl von Parametern, je nach Verfügbarkeit der Beobachtungen, berechnet. Die restlichen ca. 2600 Vorhersagepunkte sind sogenannte Interpolationsstationen, deren Vorhersagen mittels Interpolation (horizontal, vertikal, zeitlich) von benachbarten Stationen, für die explizite Werte vorliegen, berechnet werden.

So basically already half of those Mosmix stations are already interpolated from neighbouring Mosmix stations. Maybe we can take a look at how they proceed with Mosmix to create those interpolations