VOACAP Predictions at Low Frequencies
I find that VOACAP underpredicts contacts at the lower part of the HF spectrum, say 1.8 or 3.5 MHz.
George Lane: Poor George Haydon and John Lloyd (the creators of IONCAP) pulled their hair out when I said IONCAP had to go down to 2 MHz. I knew there are real life situations where 2 MHz is a needed frequency such as in northern Germany, Norway, Alaska, etc. in the winter time. Army radio links were always failing and it wasn't until we got chirp sounders on those links that we found the MUF was going below 3 MHz at night on short paths. George Haydon said there was very little data below 4 MHz but there was some for short paths that did go down to 2 MHz. So they modeled a fit to those cases. Risky, but it has proven to give good results for Near Vertical Incidence Skywave (NVIS) situations.
Radio amateurs and SWLs (who like to DX on 1.8 MHz band or pick up signals from very distant MW stations at night) are finding VOACAP does not predict such contacts at 2 MHz whereas they know that they are making such contacts. What is happening is that VOACAP is a ray hop model for normal HF skywave propagation and it does not consider the 'wave guide' sort of propagation that occurs with the residual E layer on all dark paths for the medium wave band.
John Wang at the FCC has just had his worldwide medium wave prediction model adopted by the ITU-R. I think John Wang's model is excellent and it should be used for medium wave coverage predictions. I would not try mixing it with VOACAP unless you run both models at 1.8 MHz and select the higher signal power. VOACAP is quite good for NVIS and Wang's model is good for long distance MW paths in full darkness.
Since ITS delivered IONCAP to the Army in 1978, I have fought to keep IONCAP theory unmodified. VOACAP is a cleaned up and corrected version of IONCAP and it retains all of the theory as put forth by Lloyd, Haydon and Lucas. I don't recommend taking portions of Wang's model to be put into VOACAP. The reason is that both programs are semi-empirical in that the theory has been forced to fit certain measured data but not the same set. From a purely statistical point of view, one should not mix two dissimilar models without redoing the correlation analysis.