[werk_ag]

...Now I'm not doing any tests, I'm only calculating like our WeatherDuino TX and RX.

Hi

I'm almost sure that when you start to test the readings with a real unit, you will find that the original formula was not so wrong as it seems at first sign, especially if your sensor cover will be a flat and reflective material (like the one I'm using).
I'm not saying that your calculations are wrong. It's more like theoretical data versus practical observations.

I taken the liberty of using your nice spreadsheet, and added some data. The used formula, seems to give good output results at least with my three test units, each one have different sensor cover materials (but all flat, and reflective).

My formula seems to produce over estimated results, especially at the lower part of the scale, but this compensates the losses due to reflexion at high incidence angles (sunset and sunrise).

One more time I say, this is a result of a practical observation, done to obtain readings comparable to the readings from other nearby weather stations.

Real time readings:

Just another thought:
Using your formula increases memory usage in more than 40 bytes.
Also, avoiding complex floating point calculations as much as possible, it's a good rule.
If you gonna use your proposed formula, I'm sure it can be simplified.