The pessimistic NF is SNR_in/SNR_I=SNR_in/SNR_Q. More accurate when modulation scheme is known and the combined output of the I and Q channels are used, however becasue of non-linear detection it can be very difficult in the design phase to calculate with combined signals of certain modulation schemes. Normally the receiver noise ends at the output of 1st IF filter, and calculated/simulated NF too, so quadrature mixing shouldn't affect the final result. Nobody will be angry for you if you use the pessimistic calculation.