I have an Air/Sky myself, but several of my friends from college (all of us are meteorologists) want to purchase WeatherFlow units, however they want actual in-situ measurements of temperature. With the Tempest being placed in potentially full sun (with no fan) that doesn’t seem possible. Yes we realize that the measurement is modified via an algorithm to improve accuracy, however us meteorologists want any measurement inaccuracies to be a function of the actual sensor’s measurement error, not because of where we placed it and some “black box” math being done in the background to bring it back to reality. Thanks!
1 Like
That is a downside of the one-sensor-suite architecture for sure.
Seems to me that you want a calibrated LAN-only instrument in a Stevenson Screen if you want really accurate readings that aren’t “helped” by the vendor changing the tunings or doing magic math on the readings.
I’d suggest WF might not the gear you should be looking for in any event, as they aren’t really site-then-forget kinds of things you’re probably looking for. They focus on a different market.
If you+friends are suitably techy, you can build your own with a pi or arduino, a breadboard, and some ds18b20 (or the like) sensors. Heck, put two or three in there and see if identical sensors read identically (hint - they don’t). Spend your money on the Stevenson screen.
Alternately, you could go Ecowitt for $12/sensor and $35 for a gateway and tune the sensors in their app versus a known-good reference. I calibrated my Ecowitt inside+outside sensors vs an arduino rig I built that had multiple ds18b20 sensors to an arduino and used them as a reference since they all ran +/- 0.5F which was well within the sensor published accuracy.
Really depends on how much you want to spend, how accurate you want to be, and how LAN-only you want to be.
2 Likes
WeatherFlow will come out with a commercial lite weather station in the future. How soon? That I don’t know.
These are good ideas that I’ve also thought about. I guess it’s unfortunate because I placed by Air in full shade and get great readings, but now that’s no longer an option. And a $1,000+ Davis really isn’t in the cards for us on a budget 
1 Like
I have an Air/Sky combo with the add-on solar for sale. I keep thinking about putting it on eBay, but I will gladly sell it here. PM me or email gmm@ t i k c u f dot com
1 Like
Yup. Maybe you can pick up a used station cheaply. Perhaps do a ‘looking to buy’ thread with exactly what you want
1 Like
“black box math” isn’t the same as physics. Do you really care how the temperature is measured? as long as it is correct, it should be fine.
Because the physics model also needs data from solar radiation sensor and wind sensor, it might be slightly less accurate (in case errors in those sensor readings carry significantly through to the temperature reading, which I don’t know).
Why don’t you test it out yourself, take a tempest, put it next to a validated meteo hut with a stevenson screen and compare. You should give the tempest a couple of days to calibrate itself.
Call it what you will, but unless the equations and algorithms are spelled out, a scientist sees it as “black box math”. There are hundreds of meteorological equations in existence, but if they are not explained, any scientist would be skeptical. At the same time, WeatherFlow is a public company that needs to make money, so I don’t necessarily blame for not spelling out what they are doing.
I don’t plan on purchasing both a Tempest AND a Davis just to run the experiment that you suggest, I’m on a budget haha
2 Likes
there is a difference between physics and black box math. Both can be very scientific. I wouldn’t be surprised, if they made a thermal model of the unit and applied the relationship between solar radiation and the measured temperature vs the real temperature. Then do some experiments to get some calibration parameters. It will be something along those lines. If they spelled it out for you, would that help?
You don’t have other temperature sensors available to you or one of your meteorologists friends? too bad, would have been a nice experiment.
But a good starting point without extra instruments, is, to assume the meteorologist and engineers of weatherflow know what they are doing and produce an instrument that measures the temperature correctly.
(I have an old air unit that currently measures 19.3 and the tempest measures 19.4… close enough for me)
btw, never buy two weather stations, one of them will always be “wrong” (saves you some money too)
4 Likes
I did have many ds18b20 and dht22 sensors to compare my Air/Tempest units with, as well as a Davis VP2 and Ecowitt units. So I did that experiment.
The Tempest was slow to warm up, ran hot once it stabilized, and was slow to cool down as the sun set. That tells me bad thermal design of the case.
Add in the software trying to ‘help’ based on solar and wind readings that in themselves were questionable, and you have far too many moving parts trying to fix bad hardware implementation.
- my solar always reported impossible readings for my lat/lon
- I don’t know if the wind was accurate or not, as I didn’t have a colocated reference I trusted
- the Tempest temperature readings when in total shade mid-night were very accurate vs. all 5 references I had here
- the Air also historically was very accurate vs. all 5 references I had here
- but the Tempests in the sun were ‘all bets are off’
Back before I sold/gifted all my WF gear to good homes last summer/fall, I asked there there be a switch in the software for us to be able to turn off the software trying to ‘help’. I also asked that there be a way for the Tempest to report how much it altered the sensor reading. No joy.
My opinion is you can’t expect accuracy when you tweak your reported temperatures based on wind+solar readings that are themselves probably ‘less’ accurate than the temperature sensor.
I only did engineering for a living for 42 years, but to me it’s a bad approach.
well, if your solar is way off, than you can’t expect the model that is used to get to the right temperature to be working correctly. Garbage in, garbage out.
You made me look at my data. Tempest vs air. There sure is some difference during day time, and I think it is more than I would like.
My UV readings seem to be too low, and I expect the solar radiation to be too low as well. Tempest probably became dirty due to a bird taking it as its favorite spot to sing. Now I do have to install the spikes.
If the solar readings are lower, I would expect the reported temperature to be higher. I’ll clean it today and see how it behaves.
You’d need to see the unaltered temperature data with no wind/solar compensation in order to do a true comparison. So to me that means put both sensors in a dark location with no wind for 48 hours or so. Hopefully your Tempest magic charger mods will keep it running that long.
1 Like
well it kind of matches during the night. At night the tempest is kind of like an Air unit, wind has more or less access to the temperature sensor, but shouldn’t effect the temperature.
In the graph you can already see it matches during the night, just as you own experiment showed. I’m not sure what to gain by putting them both inside.
But I already kind of did exactly that because I have an indoor Air unit and the tempest with a bigger battery mod is currently also inside. Those temperatures are at the moment 20.4 vs 19.9. But I need to put them really next to each other.
Unless WF have changed the algorithm since last summer (perfectly possible!), the maximum temperature correction they make is 1.5 deg F, so I don’t think the difference you are seeing is due to the algorithmic correction. It is probably more related to the different thermal properties of the two devices.
1 Like
o, good to know. That is a really tiny adjustment. Almost nothing. Nothing to worry about.
This will cost you $10: two ds18b20 and one Wemos D1. Data is logged via wifi to a RPi or PC. I have this in my home-made Stevenson screen and compare it with Davis readings. The two ds18b20 have an offset of about 0.25K, otherwise they match perfectly.
3 Likes
Yup - I used a nodeMCU and a little LED screen on mine. It connects to wifi and periodically post to MQTT which I process into influxdb and display vs. other stuff with grafana (all in Docker containers)
I have 3 such rigs, all match to +/- 0.5 degF which is well within the ds18b20 specs.
And yes, I need to dust.
The display is nice. It identifies the unit by the ESP serial number as well as which ip address it received, and shows when it is connecting to wifi and then MQTT for nice feedback.
4 Likes