|
Hi Rob,
I've stopped adding hardware and completed my coding on this project for now. I'm now focusing on collecting data to see if I can improve my cloud model further.
The document you found on the Boltwood device shows the format for the one-line file. The column positions and number formatting are important for the ascom driver to work properly. There is some test apps included with the ascom driver that create a dummy file that you can use to see the actual format of the data.
In terms of hardware items I'd suggest with the benefit of hindsight, the MLX is definitely needed. That sensor has ambient and sky temperature outputs and that is sufficient to replicate the Boltwood methodology (this is also described in that PDF you found).
I was trying to improve the prediction of cloud as I observed simple temperature difference was not particularly accurate. My model needs relative humidity in addition to sky and ambient temperature. I'd recommend bypassing the DHT for humidity as they don't have a great reputation and go for a better sensor. I went for an Adafruit SHT31.
If you want light sensing, I'd go straight for something like the TSL2591 as the LDR was much too insensitive apart from detecting daylight vs night.
If you want more info on my approach, I'd be happy to share what I've done.
I've included a couple of plots below. The first one is ambient-sky temp vs observed cloud fraction. This is the Boltwood approach. You select two deltaT's - one for clear and one for partially cloudy. Based on my observations they would be -18 degC and -11 degC. The problem with this approach is there are plenty of data points where the deltaT would suggest there is cloud when in reality it appeared to be clear.
The second plot is my current model of observed cloud vs predicted cloud. Again not perfect but I'm still collecting a data set and each iteration is improving my R^2.
Cheers,
Peter
|