Hi @Niek
I don’t understand the need of converting Az/El measurement to satellite position with topocentric to ECI transformations.
Converting station based measurements to a satellite position is the purpose of an IOD algorithme is you want an initial guess for your orbit determination.
Because you are working with Az-El, I don’t think that using a position based IOD method like Lambert or Gibbs is a good idea. These two methods are usefull for people working with on-board GNSS measurements.
In my opinion, the best candidate for your need is the Gauss IOD method. If you also have Ranging data, so you can use the Gooding method which gives better results.
In order to understand why Gauss results are bad for the estimated position, what is the time difference between azel0 and azel2?
Also, what is your orbit type? (LEO, polar, heliosynchronous, MEO)
Please find below an interesting topic close to what you want to achieve:
Thank you and best regards,
Bryan