If you cant have any direct contact with electricity , then you're left with a few other options :
First , you can use ultrasonic emitter/receiver to verify the water lever, but that would have to be extremely sensitive. And it will produce notisable noise. I've mounted a couple of those , one in a small project of lenght meter through ultrasonic measurement ( a big pain ) and one in a small automated bot (self guidance system). Not my piece of PI.
Second, you can use several IR leds/receivers to bounce the IR light at the surface of the water, but turbulence, bubbles, foam, etc, distorts the signal. Or, you can use them in several levels, in opposite sides of the reservoir. When the level drops, the IR led gets the light across to the opposite sensor. Several levels can be added in a custom reservoir.
Third, a mechanical device of some sort, with a fluctuator attached to a variable resistance through a wire with a balanced weight, or something similar. Attached to a circuit, you can have a pretty good measure through a simple resistor divider or even a wheastone bridge, for something more sensitive.
Forth, you can use another type of mechanical device, using a magnetic/metalic (it can be anodized, or covered with a plastic filament) with a fluctuator, within a tube connected to the reservoir and rolled with a copper wire around its lenght. It would operate in a similar way like a Hall sensor, and it can measure the deviation of current done by the fluctuator moving up and down.
These are just some ideias so far. I'm more inclined to those last two. I've seen the third working alreay (in a lab experiment), and the fourth is an adaptation of a Bourdon pressure indicator sensor (LVDT). There are alot more of sensors , but most are for industrial applications, or are too expensive. We are looking for the DIY stuff right? Righhht
.
BTW: A closed loop with a radiador doesnt really need a level indicator, IMO. However, in a bong/other evaporative cooling system (just like volenti's) , can make alot of difference.