Even if the cable has a large height above ground plane, it will still be in the vicinity of low 100s ohms, common mode. That is, driving one line, alone or even with multiple remaining lines grounded, will end up with the common mode driving over GND, while the differential mode(s) within the cable decay more rapidly due to the higher capacitance between them.
So, you have a situation that varies with frequency and position/length. High frequencies are absorbed quickly, so that only the nearest length behaves as a differential transmission line, and the remaining fraction of the input signal (which divides between the driven line and ground line(s) according to their relative impedances) follows the common mode path, which similarly decreases over distance, but should be slower than DM as mentioned, so you should observe two cutoff frequencies for DM and CM. And since DM decays faster, you can measure the CM on any wire at the far end, driven or otherwise.
At low frequencies, at some point, DM will be perceptible at the far end, and CM will be less significant. Obviously, at DC it's straight through, GND is GND and driven is driven.
A single pole RC should not give a good fit, but a chain of arbitrary length, with equal values of R and C, should give a good match with some adjustment. It will diverge at high enough frequencies, as a lumped-element RC is only order N, while your cable is distributed, giving I think a cutoff probably something like exponential with frequency, the roll-off point depending on length.
Remember to model capacitance between wires in the cable, and from the cable (from each wire) to GND. GND can be assumed ideal. Set R from the DC measurement (resistance of the length, divided by the number of segments being modeled), and adjust the C's until the frequency response matches observation.
Your plots look rather noisy, which suggests using too long a length and too high a test frequency range. s12 should be sufficient, or you can do a 4-port with ports 1 and 2 as +/- source and 3 and 4 as +/- load, measuring s13, s14, s23, s24.
Tim