Visibility into & Control of Continuous Learning Processes

I personally would love to be able to customize the reference stations that CL uses for each measured parameter. Another feature would be to cut off CL for a particular parameter and the a ability to do your own calibrations if needed.

probably there is a lot more involved than just matching some nearby station’s value. I’m guessing they use trusted sources with some fancy modelling to get the expected value at your station.Support can turn of CL for some parameters (mine is turned off for UV until they figure out what has happened). Personally I would prefer to have the station only calibrated say once a year, assuming the drift in values is small.

Sure. I just personally wouldn’t use CWOP and other PWS stations in the formula. IMO, CWOP and PWS stations are not very reliable and I wouldn’t trust MADIS to help either.

I have two Davis stations and I wouldn’t want CL to use my station as a reference for humidity. Now for rainfall, I wouldn’t mind since Davis fixed their rain gauge issues with their new tipping spoon. I would rather CL use ASOS/AWOS stations, COOP, CoCoRAHS, & Campbell Scientific stations that universities use.

IMO, using CWOP/WU stations would put a damper in the effectiveness of the CL system. But I could totally be wrong here.

2 Likes

Yes, this is true. Multiple data sources are used for each analysis and the specifics can change with each new analysis based on the quality control processes.

We do have a project on our roadmap to give the users more visibility into, and control over, the CL process. But it’s not high on the priority list at the moment - we’re more focused on making sure the process works as well as possible!

I’m going to change the name of this feature request from “Customizing Continuous Learning” to a more general “Visibility into & Control of Continuous Learning Processes”. This can be a good thread to capture input on what’s most important to station owners. Can’t promise we’ll implement all of the ideas, of course, but we’ll do our best. Thanks for starting it off @jgentry!

4 Likes

If you have votes left, please vote on this. If you don’t have any votes, just comment on here. That lets WF know that there is high interest and should be raised up in the priority list. Thanks!

I wish this thread would get more votes. I would love to be able to dictate what references that CL can and cannot use for a particular variable. For me personally, I don’t want CL to choose a PWS as a reference source when it comes to humidity and pressure, but I wouldn’t mind for rainfall (as long as it’s a Davis tipping spoon, RainWise, or CoCoRaHS.

1 Like

Could you give more insight into what processes that users will be able to control?

We have not firmed anything up, but the basic idea is to indicate when a calibration change is made and allow the user to disable and/or override the automatic calibration process.

3 Likes

Good deal. Can CL show you what stations it chosen to be the reference sources during a particular calibration session?

how do you expect those multiple data sources to be made visible to you?

Where my Tempests are located, I doubt there is a bunch of data sources it’s using…

think of your national weather forecast… it will also get some value of rain, even if not measured at your location, based on “reliable” sources. Surely weatherflow can do the same.

Never said they can’t…There isn’t that many reliable weather stations nearby my Tempest for CL to use. I figured CL would only use around 5 nearby weather stations as reference sources.

I don’t know (not at all) but my guess is that it isn’t going for weather stations around you, but instead all the reliable sources go into one big, possibly world-wide, model. Give a computer more time and it can do a calculation on a 5 or even 1km grid. That should be good enough. If your station is always lower than this model calculates at your position, you are in for a calibration… or something along those lines.

Hmmm….So it’s basically extrapolate the data from reliable sources and uses that as the “standard” reference?

interpolate. But as I said, I don’t know.
Every couple of hours lots of data is measured by official stations, satellites, ships data etc. That data could be augmented with non-official stations like tempest. It goes through a process to remove the unreliable data and then a calculation is started that calculates the state of the atmosphere at the time of the measurements. (that state goes into the forecasting models). This initial state can be calculated at a pretty high resolution grid level, like 1 or 5 km. To get the values at your station, you do a little interpolation between those grid points. Let me state again, I don’t know what weatherflow does, but to me this seems a pretty sensible way to approach the problem.

Here is what WF says about RainCheck and they have said similar things about the other sensors.

It could be quite a long list of sources used to calculate the value for your location in addition to how those sources were used so I doubt there will be much user tweaking to the CL process other than maybe a use/don’t use option.

2 Likes

Gotcha. Thanks for correcting me. For some reason or another, I had extrapolate on my mind.

For humidity, I think WeatherFlow should consider using the SHT-31 datasheet as the main guide to make calibration adjustments. I.e. do a -3% calibration if the sensor has been exposed to humidities at or above 60hrs straight and then take the -3% out when the sensor goes back into calibration. And whatever the yearly drift is, make the offset based off of that. Then I would just use the official sources and mesonet networks that uses Campbell Scientific stations like Oklahoma Mesonet as a double check. I would not use any PWSs like CWOP, WeatherUnderground etc. But I wonder if CL can use geographical information about your station’s location and use humidity data when the Tempest is first installed (since the sensor is calibrated) to learn about the station’s microclimate so it can produce a solid reference number?

Maybe @dsj or @WFstaff or @WFsupport can weigh in.

1 Like

Thanks for the input, @jgentry. Unfortunately the spec sheet does not prescribe, nor do I think Sensiron has come up with, a specific formula to apply to correct any calibration drift. Hence, calibration using external data is still required. Our CL process works pretty darn well, though, and will continue better over time.

1 Like

The spec sheet does have a drift # but I’m sure that assumes the humidities stays below 80% majority of the time.

I know I’ve said it before, I think CL done fairly well at my PCB site but I personally didn’t think it performed well for central Alabama. So far, the Sensirion sensor has done well without any calibrations. But I would love to see what that sensor is made of at my PCB location. If it can last for about a year without me asking you to turn CL for humidity back on, I’ll be impressed.