Visibility into & Control of Continuous Learning Processes

Any update on when this feature will be available?

Here is a good example I believe why I have some issues with CL when it comes to humidity calibrations. My station (Dollar Outlet) has CL for humidity turned off whereas the surrounding Tempests has CL for humidity turned on. As you can see, my untouched Sensirion 31 humidity sensor is matching well with the nearby airports, whereas the CL calibrated humidity sensors are all reporting dew points in the upper 70s.

What’s your thoughts @dsj

Are you saying that because of CL all the other tempest stations near you are displaying the wrong value? (a few degree higher dew point)

Why is your station so low? Of course there is some calibration applied to your unit as well. It might not change over time because CL is turned off, but where is that calibration based on? Did you adjust it once to match the lower measurements of the airport?

the stations around you are supposedly calibrated against some model with good data as input. That data might include the airports around you.

I believe the dew point values are too high for the other Tempests.

I haven’t adjusted anything. I had @dsj turn CL off just for humidity. The Sensirion 31 was and still is performing within specifications. It’s reading that low because the sensor is performing within specs and the ASOSs confirm that. Plus the Tempest is in an open area on the roof.

Now keep in mind that when comparing dew point in the FL Panhandle, you have to keep the wind direction in mind. It can heavily influence dew point readings.

True but it appears that some PWSs data is being used by CL and is messing up the humidity calibrations. My PWSs have humidity sensors with wet bias issues. Very few PWSs manufacturers handle and install Sensirion sensors correctly. WeatherFlow is one of the very few that do that. That’s why the sensor performs very well out of the box and lasts for a good while before any calibrations are needed.

As you know humidity is like many other variables very tricky. To be able to say if your nearby PWS are ‘wrong’ you would need to know exactly how they are installed. Moving a PWS within 1 meter of a tree or plant can alter the readings quickly. As you pointed out already, being coast line is enough to play with wind direction.
for ex look at the UV index ! According screenshots most were taken quickly and still the index goes from 0.1 to 8.3 … shadow ? tree ? building ???

If disabling Cl for you is ok, perfect but don’t make generalities from one case. And regarding Metar data, I hope those stations are well maintained (I know a few around me that could need a good maintenance as they have been installed and forgotten for years …)

1 Like

You’re correct. If you’re out between the woods, you’re going to get a lot of humidity spikes due to evapotranspiration. And it is worth mentioning too that some ASOSs aren’t maintained like they are supposed to be either. It is a known fact that Davis doesn’t handle the Sensirion 31 sensor well and has wet bias issues when humidities are below 80%. Other PWSs like Ambient has wet bias issues overtime and from what I’ve heard, they use the SHT-30.

Recently, I thought Davis might’ve fixed the wet bias issues with their humidity sensors but in high heat and humidity, their sensors seem to succumb to a wet bias fairly quickly.

To me, it appears that CL includes the CWOP/PWSs stations in its analysis and it’s causing it to make wrong calibration decisions. Typically CL does better when there is a well maintained ASOS or AWOS nearby.

I know a Tempest owner, Bobby Boyd, having issues with CL RH calibration and he used a psychrometer for comparison and the Tempest DPs were running too high. WeatherFlow used his psychrometer data and fixed his issues.

IMO as I previously stated, if CWOP/PWS data is being used in CL’s formula, it is keeping CL from making the best decisions in calibrating individual sensors (unless there is good data nearby that can help CL see that the PWS RH data isn’t reliable). I do think if the PWS/CWOP stations were removed from the formula, CL would do way better in calibrating the humidity sensor.

But it’s hard to calibrate sensors anyway without having a reference sensor at the station’s site.

BTW: I do know for a fact that the west end PCB Tempest is at a location with lots of evapotranspiration going on, so the DP will generally be higher there than the airport. But not consistently 3° or higher.

I wonder if WeatherFlow would be able to temporarily remove CWOP/PWS as data sources (if CL is using them) out of the equation and test and see how CL does for humidity calibrations.

I would love to hear @dsj and @WFstaff take on this. I’m always learning about microclimates and how certain things affect it. I’m wondering if WeatherFlow can reveal CL reference sources like they do with RC?

Hi Justin. We still do intend to surface more CL information to the user in the future. We have several concepts on the drawing board, none of which has risen high enough in priority to get done, yet. It’s the actual CL processes, which are still evolving and improving, that get our most attention. Anything we put into the API or the apps takes a lot of effort to plan and implement, and we don’t want to start that process until we are pretty confident there won’t be a bunch of changes.

Related, we also intend to give the user more control over field calibration, which is possible now but only with direct support from someone on our team. We’re happy to do that, but it’s not efficient. Ultimately we will offer four levels of calibration support: (1) initial factory calibration, (2) automatic field calibration based on CL, (3) manual field calibration and (4) factory re-calibration. The first two are already performed on every Tempest out there. The third one is available to sophisticated users, as you know, by working with our team. And the fourth option is available to everyone, but only very rarely needed.


Thank you sir! Sounds like some good stuff is coming.

1 Like

Any updates on the CL process for Humidity? Any adjustments made how CL calibrates humidity? @dsj

1 Like

If there isn’t any updates to CL for humidity, that’s fine. I was just curious if there has been any tweaking to the formula like the “RainCheck” feature? @dsj @WFstaff

Are there any plans to tweak the CL formula for humidity in the near future? @dsj @WFstaff?

I would like the CL for pressure to be more accurate. WF seems satisfied with the 1 mb (0.03 inHg) spec value. My Tempest is .02 to .03 inHg low, and I would like it to be more accurate. There is a AWOS 5 miles from my station so I would expect to see better consistency between the two readings. This difference is giving my station a bad reputation, raising questions on its accuracy.

This feature request includes several features some of which are already available through support. In my situation I wanted my automatic calibrations to pressure turned off and simply set to my requested setting. Support were happy to do that for me. I have also had changes made to rain and humidity. And I am surprised this feature request only has 3 votes.

I agree that there is a discrepancy with pressure readings on my tempest. I am located within ~7-10 miles of and in between two Oklahoma mesonet stations and to have my pressure readings in line with them I had to set my station altitude 33 feet lower (1340’) than the actual (~1373’)elevation. As for the humidity readings, I have noticed some issues with that as well recently. As in values less than 100% when it was almost certainly at 100% (all nearby mesonet stations at 100% and dense fog and drizzle occuring). And during the summer months my dew point readings were also reading higher than surrounding mesonet stations as well and my tempest is sited on an open grass area with the grass mowed and brown and dried out from heat and drought and a neighboring plowed wheat field. Given my expected micro climate variables I wouldn’t expect to have higher humidities. Don’t get e started on my tempest’s rainfall readings. They are almost useless. I have a co-located cocorahs station(4" manual gauge) and an electric tipping bucket gauge. The tempest is very rarely even close even though it is sited literally 5 feet away. That being said I have been thoroughly pleased with my temperature readings which have appeared to be highly accurate.

1 Like

I’m also seeing this on my end with CL off and no manual calibration. I’m noticing other stations 98 to 100% with both of my own Davis and multiple AWOS station with my Tempest topping out around 95%; then when drier air comes back into the area it takes a very long time for the humidity to drop on the Tempest unit. I also have a SHT41 that performs very well with the changes in humidity.


If the haptic sensor has any vibrations, it’s going to be off. You basically have to install it ideally in order to get good consistent results.

Double check and be sure that WF has removed all calibrations done by CL for humidity before you had it turned off.

The Sensirion SHT-31 could be out of calibration due to condensation of the sensor or constant 80% + humidity.

My Tempest at home did a similar thing when RH was in the 90 percentile. Only thing that fixed it was several days of very dry air and the sensor returning back into calibration.

I think WeatherFlow should consider an update to the Tempest design that allows you to swap out the temp/hum sensor or update the CL for Humidity formula that would work well without the need of using PWS data from CWOP/WU (if possible). I think the use of PWS data hurts CL’s performance.

Unless having a standard reference on-site when calibrating the humidity sensor, it’s really hard to do a good calibration by only using nearby stations. There is a lot to consider when comparing humidity/dew point data to other stations.