Data & power can be sent on 2 wires without too much extra effort. This allows a microcontroller or other electronics to be placed near the sensor, as Olin suggests.
There are various ways of doing this.
You can occasionally stop powering the remote device and "deep modulate the power circuit digitally. This could be achieved by as simple a means as feeding the MC via a diode and using a reservoir cap. When you turning off power feed it detects loss of power feed and outputs signal on the line.
Or you can modulate the current draw incrementally by a small amount of DC.
Or you can have the DC feed as high impedance to AC and inject AC signal on top of the power feed.
You may be able to scavenge power from the motor feed lines and use the final 2 wires for signalling only.
BUT you mention using a thermistor. Unless your requirements are very special you can almost certainly use a 2 wire sensor that draws a current proportional to tempeature which you then convert to voltage remotely. This is rlatively immune to line resistance variations.
As an example the AD592 - datasheet here provides temperature to current conversion, +/- 0.5K uncalibrated initial accuracy, typically +/- 0.15 K linearity (0-70C but similar elsewhere), -25 to 105C operation, relatively low self heating and 4-30V operation. Possibly the greatest issue is whether the thermal time constant meets your requirement. If this is for general ROV environment temperature measurement then this is liable to be acceptable (see data sheet). It can be used in a variety of configuration (see data sheet) with two examples shown below. If desired the top circuit (4-20 MA system) could have all the shown electronics implemented in the ROV and provide a standard 4-20 MA sensor remote signal. This signal is able to be handled by many off the shelf instruments or converters.
Even if using a thermistor, the line resistance can easily be made very small relative to the thermistor resistance so thatthe error is small. Extraneous signals induced in the line can be filtered - easily so if you thermal time constant is more than a second or so.