Rain seem off (auto-calibration system not yet deployed)

sky
false-rain
drizzle

#85

rain on the wind sensor can interfere with the wind readings a bit
but also there was one firmware that was causing gusty windspeeds readings to be ignored wrongly, for some
the next firmware update , due out soon, should fix that


#86

Today WF measurement 2.697”. NOAA rain gauge 0.92”. Noticing a 3x error still.


#87

Another update after a period of Rain…

Vue - 8.9mm

Sky - 0.41mm

Tbh i am starting to lose a little faith in the device - humidity has been out and the rain is way out, often either too notably high or too low.

I’m aware is a low cost device with a brave and innovative series of sensors (to be applauded) - however the range of data coming in it feels increasingly random and sadly i now look more towards the Vue for my data. Even my my old Netamo is beginning to seem more reliable.

Andy


#88

Thanks for your help. I’m glad you agree AI may help! A few points here:

  • We’ve collected many station-years worth of both laboratory and field observations, compared them to several “well known instruments” (some old, some new, and all independently and recently calibrated)
  • The haptic rain sensor is a very different beast than a tipping bucket or CoCoRaHs collector tube. For one thing, its basic response to vertically-falling rain is not linear, it’s exponential. So all the linear regression in the world won’t help.
  • We have developed a factory calibration process that works very well for adjusting the basic response of the sensor so that device-to-device variation is very small.

This approach works great most of the time. However, we have learned a few things:

  1. the specific installation conditions can have a significant effect on the rain sensor calibration.
  2. The rain sensor’s response is dependent on more than simply the amount of rain falling on it.

To elaborate on that first point, let’s say you have a particular SKY unit that was calibrated at the factory. If you then install that SKY on a 3 meter PVC pole bolted to a wooden roof, the same rain event will likely register very differently than if you install it on a 10 meter steel pole hammered into the ground.

As a result, it’s not enough to have a factory calibration. Each individual installation needs a custom calibration. That’s what we’re doing now, in a simple way: After recording a few rain events, your SKY’s data, along with other relevant data (from both models and observations) is analyzed and run through a set of algorithms (some that use artificial intelligence and machine learning, some that use fewer buzz-words). And the result of that analysis may lead to an adjustment of your rain sensor’s calibration, if necessary. And this works reasonably well in many cases.

However, thanks to the second point, above, this simple calibration approach does not work all of the time. That’s because the response of the sensor is also sensitive to other conditions like rain rate, wind speed, and temperature. That’s another issue where we’re using AI and ML and all kinds of fancy data techniques to work it out.

The bottom line is that under- and over-reporting is a significant issue for the rain sensor. We are still learning how all the various factors affect the rain sensor. This is a work in progress, but accumulation accuracy will improve as we collect more data and adjust both the calibration parameters and the very shape of the calibration curves.

Keep the faith, Andy! Both of these issues will be resolved soon. We are working on it non-stop (except when we have to stop and read the community forum :slight_smile: ) You’ll be able to retire both your Vue and Netatmo soon!


Weatherflow "Continuous Learning" Algorithm
Very high rainfall amounts. How to calibrate Sky?
Rain sensor and vibrations
#89

I have had my station for about 4 months now and I do not think the rain sensor works at all. When it is clear as a bell out I get It’s Raining and when it rains, quite a bit, I get no indication and nothing showing it actually is or did rain. I may have a lemon.


#90

Sounds like it is time for you to e-mail support@weatherflow.com. They’ll be able to get you back on track.


#91

any chance birds are landing on it (for rain when there is none)?


#92

Is there a limit on the angle of incidence for rainfall?
I’ve had zero rainfall recorded since 11th Aug and it has rained to some extend each day.
Very little of the rain is “straight down” on the sensor often it is drifting in from the side and not necessarily large drops.
I see my hub upgraded to v91 this morning - will this help?


#93

The angle affects calibration but you should get some signal at any angle. If your SKY has never recorded rain and you know it has rained, one of two things could be going on:

  1. The rain may be so light that it doesn’t generate a response from the signal. This would be extremely light (aka “misting”) rain.
  2. It’s possible your rain sensor is bad. You can test this by lightly tapping on the top of the SKY with your fingers for a few seconds.

#94

Pretty sure that is what happened to me this morning. When I got up the road, cars were wet, and when I left the house there was visible rain, but extremely light. Sky had not notified me of any rain. By the time I reached the train station it was raining a bit heavier, and sure enough I had the rain notification when I checked the phone.

That said, it was not what I would refer to as fog or mist (where moisture tends to hang in the air), there was clear rain falling, but what I would call drizzle in its intensity (you know, the horrible stuff which completely soaks you to your bones, despite seeming to hardly be raining!) :slight_smile:


#95

This morning my Sky is not registering the drizzle that is falling. About .01" of rain fell according to conventional rain gauge but nothing on Sky. So it appears the Sky isn’t sensitive enough on low intensity rainfall but too sensitive with higher intensity rainfall giving totals 3 to 4 times higher than actual.

EDIT. As a follow-up, today’s rain event was a warm rain process-tropical type process with lots of small droplets. Sky reported 0.192" vs 0.34" rain gauge. This is the first time for me that Sky actually under reported a significant rain event since it is usually 3 to 4 times over for larger rain drop events.


#96

It must have heard us talking about it! recorded rainfall yesterday and today.
The rain is more than misting though - the kind that gets you soaked quickly without being felt as falling drops.

-Coley.


#97

Just to follow up now it stopped raining. I have no way to measure the rain, but Alexa told me this morning to expect about 8mm of rain (not sure what service they use). Dark Sky said 11.3mm of rain. Sky reported 9.07mm, so for me, so far, seems relatively accurate.

We have only had a couple of rain events so far, and I never even thought to check elsewhere for a comparison, so today was the first time I did that.


#98

Question for the WF staff - it’s still in the super-dry/wildfire season here between Seattle and Tacoma but it’s going to start raining again in about a month (typically) and be the usual ugly for many months thereafter with morning marine layer heavy-mist and general grey consistent light precipitation a ridiculous percentage of the time.

I have a VP2 and a CoCoRaHS gauge that typically match exactly. The VP2 is online at all the usual places (PWS, CWOP, WU).

For your continuous learning tuning, how do you know what actual reality is ?

  • in this area you can’t go by the SeaTac airport readings, as there is too much variation depending on your actual location and elevation

  • there are also of course lots of cheapo stations online nearby that have bogus data online so you can’t try to cook up some aggregate lat/lon-based average or the like.

  • any particular station of course could have errors as well such as blocked weep holes, weak Reed Switches, not-level mounting of the tipper gauge etc. (just to mention three typical Davis issues I’ve seen in almost 10 years using one).

  • the Internet site (WU most noticeably - ugh) can even drop a lot of posts so the online data is likely imcomplete

So how do you tune the units to reflect actual reality ? Just curious…


#99

I can tell you think about this stuff a lot, @vinceskahan! And you ask all the right questions! Are you looking for a job?

Seriously, we use a wide variety of “trusted sources” for the validation portion of our continuous learning system, and we’re always exploring more. They vary by parameter and location and range from networks of gauges to pure numerical models and everywhere in between.

For rain, nothing beats a co-located, high quality 8" tipping bucket that’s been recently calibrated, but that’s hard to find - as you point out. The next best source we’ve found (so far) comes from a small company that specializes in accurate, high-resolution modeling (based on gauge data as well as radar and satellite).


#100

Hi, not sure how this forum works, can’t seem to find a reply button to add to others remarks.
I installed my unit yesterday. It was a very windy day with a little rain. I installed it next to an official rain gauge. I took 1.6mm from the gauge this morning and the weather Flow measured 3.45mm. I will shorten the mounting pole to decrease vibration to see if this helps. I also believe the QNH reading to be a little high. I lowered my site elevation by 7m in the setup menu to compensate. In future would it be possible to have an adjustment in the app to allow for calibration? Something that shows a plus or minus from the factory setting.


Weatherflow "Continuous Learning" Algorithm
#101

Hi. Welcome to the forum! We moved your post to this thread which is mostly a discussion of rain calibration issues. See the long thread above!

Note the pressure sensor is in the AIR device (not the SKY). Be sure to set your “elevation” to the ground’s height above sea level and the “height above ground” for each device (AIR & SKY) to just that: the vertical distance from the surface to the device.

We do intend to add a user calibration feature to the app eventually, but normally your pressure sensor should auto-calibrate (assuming your elevation and height above ground values are set properly). If you suspect it’s out of calibration after a week or so, drop us a note at support@weatherflow.com or http://got.wf/contact


#102

Thanks David, I have readjusted the Air units hight. I’m wondering how the units self calibrate, I could understand if there was a calibration standard in near proximity. Particularly with rainfall. Yesterday we had 11.2mm, the Sky measured 17.2. I’m not complaining but just wondering if there’s any way of helping?

Thanks,

Langdon


#103

Hi Langdon. It’s okay to complain! Your rain sensor’s accuracy will improve, but it may take a little time while we sort through all the data, run it through our algorithms, and repeat, repeat, repeat. See my reply to Vince, above, regarding what data we use to calibrate your rain sensor. Short answer: many.


#104

Reported in another forum but this one is closer to the mark.

SKY has been measuring here for several months. I know you’re still working tons of hours on fixing things, but Sky is still reporting inaccurate rain data despite months of reporting problems by many. Not good.

In past 24 hrs, rain is way off:
… WeatherFlow Sky: 12.1 mm rain
… Davis Vantage Pro 2 on same pole as Sky : 6.6 mm
… NetAtmo: 7.6 mm

Getting a bit discouraged. I would suggest you describe the firmware as BETA until things are fixed. Some measures are working better now (temp and humidity) but the standard deviation and error rates of other data are much too high across the world to be reliable.

In my view, the unit is not ready for prime time. I’m happy to continue being a beta tester as needed, however.