I had my station in a low (aprox 6ft height) location at my home. I have moved it to it’s permanent location on a 20 ft mast. I’ve noticed the wind speed is consistently a few MPH lower than my trusted Davis at the same height and 15 feet away.
Should I expect the continuous learning to adjust the wind speed closer to the Davis reading? If so, about how long should that take?
You don’t have to do anything as the process is ongoing. Not sure how long it will take but it should adjust itself pretty quickly if you have wind coming from all directions in pretty short time as the 4 sensors will calibrate according wind.
But don’t expect exact same figures as that is impossible. Already the 2 use different techniques and as you know, wind is very strange and even with 50 cm between stations, the readings can differ. I have 2 sky’s in that situation and nope, readings are different but when you average, then they are pretty close. Sampling interval, little interference from nearby obstacles etc … it all counts.
Keep an eye on it for 24 hours and then compare, if still way of, time to ask WF to have a closer look.
Hi @tisawyer The sonic anemometer goes through a sensor balancing routine on restart which tests and assures that the sonic transducer signals are in exact alignment (crucial for timing of sonic signal timing). As part of this balancing, the SKY needs to experience winds from the four cardinal directions. To be clear, this is a completely separate process than our forthcoming applied machine learning / continuous learning (CL)system. Across our entire network, we have found the sonic anemometers to be incredibly accurate once balancing has completed, save a few conditional anomalies attributed to filtering/qc algorithms which can be corrected in the firmware.
The WeatherFlow CL system compares your sensor data to an array of trusted data sources dozen of times daily and tunes each sensor accordingly = ongoing auto-calibration. Wind is the one parameter where trusted comparison data is simply impossible as the wind field in one location, even 15 feet away, can be very very different (read: wind values are not part of the ongoing auto-calibration system at this time). In your specific case, we would recommend waiting a few days and watching the data to see if there is a discernible pattern over time. The most significant factor is siting. Unless your Davis is both professionally calibrated and sited immediately next to the SKY, it really isn’t a clean comparison. If you do see continued discrepancy, please let us know and we’d be happy to dig deeper into the debug from your SKY.
We also take the massive data set and run it through machine learning algorithms = improved sensor tuning and site-specific forecasting across a wide variety of conditions (ie. to understand the influence of temp and wind on rain impact signatures through data science).
Thank you for the detailed response, very educational and interesting. The last few afternoons have good wind at Sunset and Seal. My house isn’t too far from our local kiting/winsurfing spot at the end of Warner Ave in HB. It’s been 6.5 to 6.0 meter sails and 12 to 10 meter kites. Just thought I’d share what my stations are showing this afternoon.
So normally, it does this sensor balancing only once. The question is if after moving the sky and mounting it in its final position, does it need to go through this balancing again, and how does one force this restart of the balancing?
Calibration is not a one time thing, it will continue over time. If you open the battery door, it will be just like a restart for a small period. Routines will pick it up and just restart the process
Before the routines, it already happened with the 4 wind sensors, each time you open the battery it had to get wind from 4 sides to re balance itself …
Remember, you have nothing to do, it just does it’s work for you.
I can’t say how many time a day this re calibration will happen but more then once most probably if I remember David’s writings about it.
Thanks for the share. I windsurfed at SB and HB back in the days before I started kiting (grad school at UC Irvine in the early '90s) - great spot!
We’ve looked a bit more closely at the wind speed data coming from your station (see below) and it actually looks very good. Note, the 3-second wind speed values are exhibiting a pretty wide spread (from gust to lull). This is most likely due to the turbulence created by the wind passing over land and across buildings and trees upwind of your SKY. You would likely see a tighter spread between lull & gust (and a higher average) by raising the height of your SKY a bit, if that’s possible.
Note, this lull/gust spread would not cause your wind speed average to be any lower than your Davis, were it located in the same spot - if the Davis is calibrated properly, the average should be the same. But the Davis (and any spinning-cup anemometer) isn’t able to capture the higher-resolution gusts (and lulls) due to inertia in the cups - it simply can’t accelerate/decelerate fast enough. This is an advantage to the sonic anemometer in your SKY.
Just to clarify this process: the ultrasonic sensor balancing step happens as soon as the SKY collects enough data from each direction. Those balancing values are stored on the SKY and used from that point onward, even if the SKY loses power. However, when you power-cycle the SKY (open and close the battery door), the SKY will go through this step again. It will continue to use the previously stored balancing values, however, until it collects enough data to update them.
Moving your SKY won’t affect the balance of the ultrasonic sensors, so it normally doesn’t need to go through balancing again. But it won’t hurt either.
That plot was generated using the rapid_wind data from your SKY. We don’t store that data long term (except for debugging), but you should be able to collect that data locally (and plot it) using one of the nifty tools #owners:integrations
I’m wondering why this is now suddenly an issue. But if you read the last response from David, connecting it to power (after it has been disconnected) will initiate the balancing after it has collected enough data.
so @jgentry the answer is no, @vidguy7 the answer is yes.