Have done this already and yes it “works”, but there is a big questionmark about what you are really measuring by this.
The photodiodes used by different people have different characteristics and responsiveness to wavelength so the different users data can actually be hardly compared.
I pointed in another thread (actually hijacked my own thread a bit by the subjetc, but see here http://discourse.weather-watch.com/p/379374) that the biggest problem is that temperature sensor that you are removing is nonlinear and the unit is compensation for this. Replacing it by photodiode with nearly perfectly linear responsiveness gives you very biased readings. As a result, your unit is very sensitive at low light conditions , but at full day it is very little difference between 1000 and 800 W/m2. WD can only interpret the figures you feed from the unit so the temperature is converted solar values that are far from what they should be.
After little tweaking you can use it reasonably well to tell your current cloud cover level, but remember the solar values obtained by this are purely fictious.
I did this mod and decided not to use. Whatever I store should be more or less accurate, it is impossible to have a good data from such mod IMHO.
It is far better to go ahead with temp in jar solution, your readings will be more close to what they should be. Plus - you don’t need any photocell 
my 2c :roll: