Rain calibration corrections

Had our first real rain event since install. Sky 3.63
Ambient WS-2902 1.42
Acurite 5N1 1.27
These are in inches, and it was a monsoon!
Will give it some time👍

1 Like

Absolutely. Rain hitting the SKY is a very stochastic process. Splash is just one of several factors that can affect the haptic sensor’s signal. We’re working through all of them as quickly as we can.

Look for a new firmware update that will enable another improved rain sensor calibration coming soon…


I’m having trouble buying this. Yesterday, we had our first little thunderstorm since I installed the Weather Flow Sky. Results:

Rainwise tipping bucket #1 - .15"
Rainwise tipping bucket #2 - .14"
Cocorahs gauge at home - .15"
Cocorahs at school (6 miles away) - .15

Weather Flow - .35"

My question is this, the Weather Flow is supposed to self calibrate, but based on what? How does it know if it’s wrong in the first place? My Station is # 4711 if that matters.

I should mention that in the two weeks the station has been installed I have had zero false rain reports.

Expecting 3” on Sunday. How many feet will my gauge report?

And what about hail?

How does the Sky respond on impact of hail.

We had a shower yesterday with some hail between the rain.
At that moment the Sky reported extreme precipitation for a verry short while. But I have the feeling my rainsensor does it verry well.

3 Days with good rain events. But every time it gives almost the same amount of rain as other stations. :+1::grinning:


This should shed some light on the CL algorithm. Not especially confidence-inspiring, but it’s good information.

Could the ML algorithm be better trained using a Disdrometer? This would provide a quantitative description of precipitation for the learning data.


I thought i’d post this here…

I just had a resident jackdaw try to land on Sky, and whilst it has bird spikes, he managed to get one leg on to it. I had the notification on my phone to say it was raining, but thankfully nothing was actually registered. Could this be part of the algorithms differentiating birds feet from raindrops? :ok_hand:


Interesting idea! I sure would like to have one of those instruments . . . but the $5000-$6000 (USD) price tag puts it out of my reach.


The same happened to me the other day with a magpie, also same result -Push App sent notification but no rain recorded in WF app however, interestingly, weather display did register the event.(including the eclipsing of the UV/light sensor.):sunglasses:

True, they are expensive. I was thinking about the WeatherFlow engineers/scientists that are developing the calibration algorithm. Instead of purchasing one they could install systems here; www.arm.gov. At the SGP research site for inter-comparison. They have all manner of precipitation measurements - including the Parsivel.

(Used Car Salesman Mode)

If I could put you in one of those 24 Ghz devices for under $500, would you drive it off the lot today?

(/Used Car Salesman Mode)

(FOB China, of course.)

I was looking at buying one a few months ago, but the published accuracy specs of the ones that I was looking at (including big $ players like Lufft IIRC) were not even on par with a $50 tipping bucket for 50x more…


Had .25 inches of rain overnight. It was a light rain over a long duration but the sky reported no rain at all. I tapped the unit to make sure it was still register in and it reported rainfall.

OK, so first the obvious problem. As stated, serious overestimation of rain.

Just to illustrate take a look at August 14th on my Rain graph:

Then look at August 14th on the local weather station:


The weather station measured no rain and the Sky, well as much as 2m (yes meters!) of rain in an hour. Hmmmm.

Go figure. Now I know that it’s a haptic sensor, and I know that I have a scheduled check for and fix of any loose bits and sources of vibration. Got that. Even went up recently and did half that and decided it needed a fix. For context here is the station:


Basically there’s a TV aerial (A), and I slid an aluminum post over the top of it taller than the aerial by about 300mm (B). Snuggly inside that is another aluminum post, that is about 600m long ©, that sits snuggly on top of the aerial and finally there’s a 3m length (D) snuggly around that and atop that is the Sky.

To be clear B is a little loose on A, so at base has two orthogonally situated clap screws through it holding it tight and no wobble. At the top because the aerial is attached to the top of the aerial pole (A) with a U-bolt, that’s now around B and A. B has two vertical slots cut into it so it compresses tight and snug on A.

That junction between B and A at the top of A, has a wobble in it I found. I have to tighten the U bolt and/or dismount and cut another slot so that aluminum post (B) can close tightly around the aerial post (A).

The whole rig is tall and aluminum so sways in the wind (has no guy wires) but it is aluminum tubing form an old aerial structure and I am indeed still waiting for the strongest gusts of the year to see if it survives well without bending. If I trust Sky the peak gusts since installation have been around 80 kph (on Aug 15, surprise surprise around the day of that extreme rain measure above though the max win gust agrees well with local weather station that recorded a peak of 85 kph that day, certainly within microclimate tolerance).

All that aside, I know that on my list is to remove that wobble from the A-B connection at top of A.

BUT, my real question is this: Clearly if there’s a wobble or vibration in that joint that is travelling back up the post to the sky and tricking the haptic sensor into thinking raindrops be falling I’d be interested in the salient question of … wait for it … calibration. I’m curious what kind of signal is counted as a drop and how (actually as an engineer with a background in signal processing I’d be curious how it worked full stop, imagining a running time series correlation against a typical raindrop signal, though the different sizes of drop and angles would provide a family of candidate signatures methinks and my experience and recollection runs a bit dry). But I am curious what the haptic tract looks like and would love to see a graph of that, annotated with detected drops as a kind of insight into Sky’s innards.

So my question is in a sense, if nay such thing is possible in the field, or if you (devs) would agree that some kind of drill down into this data might have appeal and use to a sector of your market?

Not least because often the human eye remains a good deal better at interpreting such signals and differentiating signal from noise, and it would be nice to see clear noise. In a case like this as a before/after look at the sensor readings, before and after I fix this observed vibration between the poles.

Because I remain concerned that once that’s fixed, that there is still an issue present. We shall see.

Your Sky battery says it needs replacing. Could that be the issue?

Not likely, it was fine until a few days ago. Funny how the battery charge nosedived in two days … rain overestimation has been an issue the whole while I’ve only now found time to sit down and describe it well. But I do have go and tighten up that pole a little and see if it improves. That does leave the question if there’s away, or might be in future, to see the actual haptic sensor output before it’s interpreted to produce rainfall estimates. For reasons I mentioned in the OP that’s very interesting.

My WeatherFlow is overestimating rain as well. I’m going to try to isolate it from the fence it is mounted to… I have 10’ section of 1.25" pole mounted to a iron (more like steel) fence.

It is off by at least 3x on heavy rainfall events.

mine is just…incorrect…

0.23 inches here in last two days - VP2 and CoCoRAHS manual gauge agree perfectly.

I can’t figure out the blasted WF rain gui in the browser but I ‘think’ it thinks I got 0.18 inches those days. Maybe.

weewx running the UDP listener thinks we got 0.15 inches. Grafana measuring and summing up the distinct measurements also calculates 0.15 inches over two days.

Just for the record (so I don’t forget) - the math I did was to add up all the non-zero accumulated rain measurements and convert to inches. Got the same result in Excel and Grafana (whew)…


As a reminder, the WF auto-calibration system for rain accumulation has not yet been deployed.


Do we have any even speculative target date to expect this ?