Go Down

Topic: How to predict weather from temperature, humidity and pressure data (Read 4567 times) previous topic - next topic

wvmarle

The more information, and the more detailed the information, the more accurate the weather forecast will be. I'm already quite surprised that wind direction & pressure trend can predict that accurate, even if just half a day ahead. A neural network will not bring much improvement without much more information.
Quality of answers is related to the quality of questions. Good questions will get good answers. Useless answers are a sign of a poor question.

Johan_Ha

This is perhaps not the right forum to discuss how to make good predictions, only how to make good readings from the sensors.
Anyway, I contribute with one thing. If the barometer says high pressure and it hasn't changed for the last hours, it will continue. If it starts to drop, there will be a change. If the barometer says low pressure, it may change sooner, whether or not it has stayed low for hours.
I would add wind speed and direction sensors. And time. It might also be good to switch to a Raspberry Pi to perform more data logging and heavier calculus. With only the given parameters, it could actually work, if you record long time data lists and search for patterns. They might be specific for your very geographical spot. Instead of adding lots of sensors and lots of theories for your particular kind of environment (coast, inland, mountains, latitude, whatnot), your recorded data will tell you the patterns.
____________________

If you ask for help and write 'u' instead of 'you' because you think it's convenient, I will write 'no' instead of 'yes'. For same reasons.

zoomx

Maybe another MCU is better than a RaspberryPi, a Teensy 4.0, for example.

About forecast: maybe a neural network?

wvmarle

Maybe another MCU is better than a RaspberryPi, a Teensy 4.0, for example.
Why?

Quote
About forecast: maybe a neural network?
Gonna need a lot more info than the data of a single weather station to be able to make a difference.
Quality of answers is related to the quality of questions. Good questions will get good answers. Useless answers are a sign of a poor question.

zoomx

Why?
It's cheaper and uses less power. Maybe also an STM32F7. Maybe also an Atmega 328 is enough.

jb63

You might want to read a bit on Prof. Edward Lorenz of MIT, who 'played' with non-linear coupled differential equations to predict some aspects of the weather ... and that of course led him to Chaos, and the conclusion that weather is 'difficult' to predict reliably. Interesting, I've had a similar discussion with other PhD's ... who believe that, with more 'advanced/powerful' computers, one day we'd be able to come up with better weather predictions ... It is not the computer power that's the problem .. it is the physics that are intractable.

On a different note, the cheap weather stations some of us have in our homes, are able to predict tomorrow's weather based on today's information. Might be worthwhile to investigate how that's done there. My understanding is that it needs not only today's info, but also info from the past few days as well. This tells me the prediction is done with 'dynamics equations'.

The RPI put out a weather station kit few years ago, maybe that's a lead you need to investigate and ask users there for some info. www.raspberrypi.org/forums

jb63

Except in New England:

"If you don't like the weather, wait 5 minutes!"
Absolutely! ... though that's true in many areas of the US where the topography is rather flat, whether it's New Jersey or the MidWest. We had 24C here in Chicago 2 days ago ... today it's only 1C.

Go Up