I am currently working on the calibration/tuning of radio propagation models for wireless network planning and optimization. I want to know how can I appraise the accuracy of the tuned models.
One way is to find the difference between experimental values (measured) and the values generated by your model. To do it you can use RMSE. To find reliable model you have to add the effect of height of receiver as well as transmitter, center frequency, frequency bandwidth, distance between transmitter and receiver, the type of channel and so on. For more details you can refer to HATA or Okumura Models.
RMSE and standard deviation are just statistical parameters of your random data. Being scalars, they allow for convenient comparison. As an alternative you could consider plotting the PDF/CDF of some parameter, for your model and the ones you want relate them to, and compare them visually. This is used in 3GPP to calibrate system level simulations, see TR 36.814 for some examples.
You can also use channel estimation using system identification approach, which should be adaptive, once the MSE is acceptable , you can have the frequency and time domain responses, whether the response of channel is flat or frequency selective.