When constructing a manual localisation matrix, what are the implications of using binary (1/0) values instead of continuous values between 0 and 1? For instance, steady-state heads in layer 1 may strongly inform hydraulic conductivity (K) in layer 1 and (to a lesser degree) K in layer 3 (due to leakage), but will not inform specific yield (Sy) in either layer. If I assign localisation weights of 1, 1, 0, 0, does that indicate that K1 and K3 are influenced equally by heads in layer 1? Or does the algorithm simply exclude the zero-weighted elements and let the covariance determine the relative influence among the remaining parameters?
In my case (and probably most cases) it is difficult to quantify the degree to which each parameter is informed by a given observation. However, it is usually possible to specify, in a more binary sense, which parameters are potentially informed and which are not. I am just looking to confirm that this approach isn’t biasing the parameter upgrades by overweighting the importance of less important parameters.