The weather has been good in Massachusetts the last few days, but when I checked this morning PAN006 had not recorded any data for 4 days. It appears that the cloud sensor is reporting too low a difference between the sky and ambient temperatures (e.g. about -4.1C aloft and 16.4C at the sensor last night at 2am).
We don’t make use of the AAG software (it runs on Windows), and instead have software for estimating from those temperatures whether it is cloudy, using a very simple model related to the difference between those two temperatures: > 25C is clear, 25C to 15C is cloudy, and less than 15C is very cloudy. In the example above, at 2am the difference was 20.5C, but the skies were clear, as you can see here:
Wilfred tells me that he adjusted the thresholds by modifying POCS/conf_files/peas_local.yaml so that >20C is clear. That would have worked last night, but would be very close to failing.
It seems that we need a better model than a simple linear model. The AAG software uses a linear model by default, but they include the option of providing settings for a non-linear model.
If anyone has pointers to background on this problem, I’d appreciate learning more about how to solve this.