Tempest looks great but ... can haptic sensors finally be calibrated?

I have Sky and Air, and Sky’s rain measurements are near useless.

The reason is simple methinks. I mounted it on a 3m aluminum pole attached to antenna mast on the roof of my house. The house has a corrugated iron roof. What I find is consistent over-measurments of rain and the likely reason is vibrations from rainfall on the roof travelling up the mast to the haptic sensor and or wind induced vibration in the mast that are confusing the haptic sensor.

As it happens I work int he field of sonar data analysis and am rather familiar with the possible methods and strategies for extracting signal form noise in such contexts, and this one of course has its challenges because it’ passive sonar not active (so the pulse is not of predefined form and nature but will depend on the size of the raindrop and how it strikes the haptic sensor). But because of my experience in the field and a general engineering mindset (as indeed I suspect many early adopters of such tech have) I have always wanted to:

  1. have access to the actual signal that the sensor produces (pre-processed)

  2. be able to calibrate it, that is if I can measure rainfall over a calibration window (1 or 5 minutes say) with a standard rain gaage beside the sensor and then use the app to say, over this time bracket I measured this much rainfall, either enough of these over different rain intensities the thing could calibrate itself to return more accurate rain measurements. Most definitely the case if the noise is itself linearly related to the actual rain intensity which is fairly likely.

I’d be keen on an upgrade to Tempest if it offered just that!


Remember these weather stations are entry level market and that they already give a very open api to get all the data.
What you ask is a local connection to the sensor data via a local connection. This is again an extra cost and means more support to provide.

since you seem to be very skilled and interested, why not make your own haptic sensor … might be fun for you.

the current system has an accuracy of 20% for accumulated rainfall. The tempest has an accuracy of 10%. If yours it outside this range, ask support to have a look. Note that accuracy isn’t absolute, you still can have some measurements that fall outside the stated values. But 2/3 of every measurement should be within the given range.

I would like to eliminate the rain readings from my app and reported conditions. I have a galvanized steel pipe attached to my metal fence. It is so windy that my rain readings are WAYYYY off . I live in Laughlin and report rainfall, even though we get very little, to COCORAHS with a very accurate manual rain guage. The weather station reports 0.83 inches today from just the wind .

it seems that filling the pipe with sand helps in reducing the vibrations. Give it a try. Is perhaps the fence rattling in the wind and pass on the vibrations to the sky unit?

I temporarily attached one of my telescoping fiberglass flag poles to my metal fence. When the wind would blow just right the SKY would report rain. When I moved the pole to another mount the false rain readings went away. Here is a video of my SKY on the telescoping flag pole mounted on my RV and it is not giving any false rain readings.

Sorry for the long silence. Checked in on account of another thread.

Re: sand filling, interesting but hard. It’s a hollow aluminium tube actually sued for an old antenna on the mountain nearby, repurposed. I have only a small portion of the original, but it’s basically in two bits, so I can lift it up and brig the top down (given then Sky unit is 3m above roof top as recommended) To wit, filling it requires both sealing the two parts, and dealing with that weight or not probably meaning I can’t put it up and pull it down)., If curious they way the work is that I have one tube over the top of my TV aerial post on the roof and it fits snug, slipped inside that snug is another hollow tube (they come that way, snug fits that protrudes 10-15 cm, and the top half fits snug over that with a slot and pin so that we keep regular North orientation, i/e/ there’s only one orientation you can put the top on in).

Still I may go up and check for wobbles and damping options. My best guess is it’s actually wind vibration on the pole yes.

@eric, thanks for the tip, but hey, I don’t have time to hang around much let alone build a new haptic sensor. Whole attraction with Sky was plug and play ad it’s pretty awesome. I don’t buy into the “extra support” line so easily, it’s easy enough to publish interface specs on a no support condition (community support only). Been done before.

Bigger problem is probably technical. I would guess that the raw signal is fully processed in the sky and only a processed summary sent to the hub. Basically because I suspect transmission drains battery far faster than some computation. No idea what’s inside the Sky but presumably a low power processor like the ATmega4809 or ATSAMD21G18 (straight from Arduino specs :wink: ). The transmission to the Hub is a mystery but probably low power RF (AX5043 or CC1050 maybe) with the plugged-in hub handling wifi (which is more power hungry).

To support community contribs to data processing all you’;d need to consider is enabling a raw data transmit mode (with caveat on power consumption and battery drain, so for experimenting only) that sends the hub raw data it can publish in UDP packets, and that the community can as it wishes examine. It needn’t be heavily supported.

Of course I trust you have some really good DP skills in house! But loads of suppliers have found tapping into community contribs very rewarding. And in your case it has special value as your problem is getting the raw data the field users face! But if a user submitted a raw data dump of a minute of rain and another minute rain detected but not actually there, phew the learning … your in house skill base can grow from it, and/or others can chip in.

It’s a whole fruitful field of collaborative work in a sense of open community contribs of data and even possibly ideas and code.

I don’t have any spare time to be making weather systems, a mere side interest and I’m snowed under with competing projects. But these are ideas that IMHO hold some merit for a business.

I don’t know what is inside the unit that listens to the rain. But my guess is that if you want to analyze the raw sound of the rain remotely, you get a data stream comparable with that of a wireless microphone. For sure that would eat up your batteries. But you could do so when using a better power source.
However, I think it is more likely that during the development of the unit, weatherflow did just that, listening to plenty of raindrops, and deduce a practical way to guess the size of each drop. How it is done, I don’t know, but I wouldn’t be surprised if that practical way doesn’t involve analyzing the wave form of the sound at all, so the data you are looking for isn’t even available.
But I don’t know.

“guess the size of each drop?” I don’t think so. It’s classical signal processing essentially. Yes, you need trials to determine what the range of desired signal looks like and characterise that. No doubt about that. Then ideally you look also at classical noise (undesired signal). Then there a number of ways with different computational overhead to increase the signal to noise ratio knowing these patterns and count and characterise what you want. It’s a bigger problem if multiple overlapping pulses are arriving which is likely the case with the Sky but there my expertise runs out as I’m not familiar off hand of the typical density of rain (drops per unit area at a given time) Ideally you’d use a sensor that is about the inverse of that in size (so you expect one drop at any time on average on the sensor, but that may not be possible.

Adding a wireless microphone (or in other words a haptic sensor) in parallel is to defeat the whole purpose and to create masses of work I have no time for. The purpose is to see what the Sky sees, the very signal the Sky is processing (for a period) to draw conclusions form that.

The reason is, that such signals fall into a umber of categories. One is so noisy it’s impressive what a good SP algorithm can draw out of it. The other extreme is signals in which the human eye can definitely see information (patterns) that SP is failing to detect. This latter case is particularly interesting to observe.

Anyhow, it sure would rock if Sky and/or Tempest firmware (along with Hub) had a debug or raw data bode that could be turned on and off to collect raw haptic data. Even if it’s only to send in to WF to asses. “Here’s what my SKy sees” when it thinks its raining and it isn’t. If that were coupled with the wind measures information may well be in that data that helps tune Sky to see rain through that noise.

But there is always the possibility that some vibrational noise mimics raindrops too well to differentiate. if so I’d be asking WF to consider the pole mount design and what they can do to that to deaden, or dampen vibration noise arriving from its mount. That is another design issue alltogether.

I’d even be happy if there were community based solutions to that. The Sky current fits snugly on my aluminum pole. But if there were something between the sky and the pole, something female to fit over top of my pole and male to fit inside the Sky that served to deaden/dampen vibration transmission, wow, that would rock.

Designing and constructing something like this would best be in WF’s camp knowing what kind of noise they need to filter out in the mount (the kind that looks to Sky like raindrops).

guessing the size of a drop is perhaps not the best description, lets call it estimate instead of guessing. But it is probably very, very inaccurate. Only because they collect many drops during a raining period and adding those estimates together, the total becomes somewhat near the real amount of accumulated rain.

There are many community based solutions to prevent the vibration to register as rain. Some suggest sand filled pipes, other have rubber or foam parts in their setup.

As noted sand filled is not suitable or compatible with the recommended 3m elevation from a rooftop say. Unless we had a rooftop secured ladder … ha, ha. The pole has to be something you can lift up and out and lay flat to access the Sky for service.

It’s also a hell of a cumbersome experiment to try even based on a maybe, when the question I’d tables is, where is WF in the equation and why is their pole mount not equipped with a vibration barrier? Comment form WF appreciated?

Rubber and foam parts more feasible but again I haven’t much time for experimentation and would like a solution that is known to work if I’m going to great trouble. To work on a foam barrier I have to create a whole new fitting, I have to but some sort of diameter step down cap on the top of my pole that is lower diameter so that I can wrap foam around it and mount the Sky around the foam say, as an experiment, assuming it has a solid enough fit not to blow off in heavy winds so foam glued to pole and Sky tightened to foam.

Of course yet another possibility is to leave Sky mounted, cut the pole say 5-10cm or so beneath the mounting and then insert a lower diameter coupler that is foam padded. Again, doable, but a lot of effort on a maybe.

Of you know of community based solutions can you share a link or links to discussions? And or trial outcomes? Which is a digression, albeit a welcome one, because it is still the case that signal processing may well overcome it if only WF have enough raw data samples from real setups and a way of collecting them and submitting them.

With their latest tempest design, weatherflow did improve on excluding false rain by adding a rubber ring between the top, where the sensor is, and the rest of the unit. It might not be enough, but it helps. If you live in the USA they enabled something called raincheck, that tries to eliminate the false rain, but that happens only on data from the previous day in the hours after midnight. I’m not a strong fan of that idea, but perhaps it is better than the alternative.
I’m not in the position to recommend one community based solution over the other, but I would recommend to search this forum.
My setup includes a 2 meter solid acrylic mast (because of the looks). In strong winds it sways way more than I expected (which probably isn’t a very good thing), but doesn’t cause false rain reading. Only this year, some bird decided to use it as his favorite singing spot and that does cause false rain. When the batteries run out, I will add some spikes around it to prevent that.

From reading what others have done I would say there is no one solution that fits all. Some have used the sand filled pipe, others rubber gaskets, cut up inner tubes, some thicker wall pipe, some used guy wires. Each install is unique and may require any of those, a number of those or maybe even a new solution.

Me, I made mine as solid as I could. Ran a galvanized conduit pipe 2’ into the ground, ran it up the side of the house and bolted it solid to the eve. From there it is 6’ above the peak. Maybe not the “recommended” height but it is far enough above the roof to minimize its effects and there are no trees within a 100’ that are higher. No false rain other then from birds or when I hit the pipe with the weedwacker.

So, what works for me might not work for you and you may have to try multiple solutions until you find what works best.

Interestingly I did some reading and math. Turns out rain falls from about 0.25 to 50 mm/hour which maps using the Marshall-Palmer distribution from 0 to 7mm sized drops to between 1500 and 4500 raindrops per cubic meter, which for a 7mm (largest drop size) slice is 10 to 30 drops per meter squared, which means a sensor size between 30 and 30cm in diameter recivies no about 1 drop at a time. A smaller sensor (and Sky is smaller) can feel comfortable that it has on average to deal with only one impacting rain drop at a time, Greatly simplifying the signal processing.

I wonder if I got the math right?

I’m curious… is wind driven rain a problem with the haptic sensor? A small drop driven by hard wind, well, F=MA. So in a 50mph gale, small raindrops look much larger to the haptic sensor. Does the tempest calibrate for wind speed, which it measures separately?

Good question! Remember that that rain drop is also going to be going at an angle so won’t have as large of a footprint on the sensor top so maybe it cancels out to some extent. I do know that for the rain reporting WF has several check boxes to indicate if there were high winds so I assume that the signature of a given drop changes as wind speed increases.