Seismic lacks low frequencies, so for an absolute seismic inversion (which allows us to be quantitative) a so-called Low Frequency Model (LFM) is required. 

Starting from an empty LFM, we would of course like to post Sand Vp values where there is Sand, Shale Vp values where there is Shale, etc. (for a minimum of 3 impedances and for all facies expected).

But unless we have very favourable circumstances where facies can be interpreted directly from the seismic, we don’t know in any great detail where the facies are located in the subsurface (in fact, understanding the facies distribution is one of the main aims of seismic inversion). So populating a LFM as outlined in the previous paragraph is not possible in most cases.

The LFMs constructed to date are therefore compromised to a smaller or larger degree (for instance, using well log interpolation guided by picked horizons leads to averaging). Importantly during the inversion the seismic cannot ‘fix’ a compromised LFM as – we come full circle here – seismic lacks low frequencies!