Conclusion

The logs need a clean up but otherwise I am done here.

My last roof reading was:

• Date (start)           4/Mar/18
• Sunset                  19.0833
• Sunrise                  5.8667
• Assumed altitude       -4.3
• Latitude                  -32.25
• Longitude               115.75
• Error                        <2 km BUT this is for a calibrated altitude.

While the longitude is very close (within a minute of time), the altitude calibration is dependent on the site and varied between -2.5 (inside window) and -4.3 degrees (roof top). The difference in light levels between these two sites could not be much more extreme!

You could use the average sunlight intensity during the day to estimate the site altitude factor using the solver or a look up table. Without that the latitude accuracy is in the order of +/-5 degrees (+/-500 km). Worse if cloud affected. Still did anyone really believe you could get GPS accuracy with a LDR?

Update

I have been running the box for a week while I was on holidays. Upon return I downloaded the results:

As you can see the first  results are about right but the latitude (-32) drifts off. Clearly I have have a coding error and have not updated the Day Number properly.

Fixed but there is something else wrong. The average is right but individual results scatter around the average in a rather predicable way.

Errors

In my mind there are three types of errors:

1. Numerical (very hard to fix)
2. Coding (i.e. mistakes)
3. Conceptual (incorrect model or assumptions)

I check the various modules and but could not find any coding errors. Therefore I have made a bad assumption somewhere. All code has assumptions where we decide whether the model we ae using is fit for purpose.

I created an exact set of sunrises and sunset times and test my code. The results indicate a problem. Here is the longitude oscillating around the true value (115.844):

And the equation:

• Lon = 15*(GMST-RA)

One solution is to take the average of the day and night estimates (the results are accurate as the errors appear to cancel). The error in the longitude is relatively easy to identify. The assumption that the average of the sunrise and sunset times is true noon is accurate enough for this application is false.

Here is the latitude oscillating around the true value (-32.09):

The error in the latitude is more complicated. Yes the average of the day and night estimates gives a very good results (i.e. the errors appear to cancel), feeding the RA and Dec for exact noon is not sufficient.

I need to dig a little deeper for a more exact equation.

AlanX