In any case, there will be a strong correlation between the tropospheric delay and ZTD, because the TD is modelized as a product of ZTD and the mapping function.

I will say that it is possible to estimate the coefficient of the mapping function if you can eliminate all other sources of errors. That is if you have a dual frequency receiver to suppress ionospheric delay, and you already know the poisition of the receiver downto a few milimeter, to suppress the geometric errors and phase incertitude. In that case, you're left with the clock errors, the satellite position errors and the tropospheric errors. The first 2 errors are well known and can be mitigate accuratly. So you're left with only the tropospheric errors. If you're able to define an consistent estimator of the coefficients, you can then estimate them.

"strong correlation between the tropospheric delay and ZTD"

Dear Arnaud, in this case, it does not matter. But the correlation between the mapping function and the derivatives of the function with respect to coefficients is important.

"So you're left with only the tropospheric errors. If you're able to define an consistent estimator of the coefficients, you can then estimate them."

This is not a fact. High correlation between the mapping function and its derivatives may not allow for the separation of ZTD and coefficients. This may be so even if on the left side of the measurement equations there will be only slant tropospheric delays.

The error model on the pseudo-distances isn't linear ?

of course, mapping function has a form of continued fraction of Marirni and Murray...

of course*