Antagonistc Thermopile Pair (Horizon Sensor) - Subtract before or after amplification?

The thermopile pair I am referring to is meant to be used for horizon sensing on an axis of aircraft. You can find an example of the system in the attached schematic.

In the schematic, each pair of thermopiles has its outputs subtracted from each other and then the difference is amplified.

I'd like to get all y'all's opinions on whether you should subtract the output of two antagonistic thermopiles from each before or after you amplify. It seems to me that subtracting first would be most accurate (best SNR) if only it did not reduce an already small signal to an even smaller signal which is more vulnerable to noise.

Subtracting after amplification (i.e. using two op-amps, one for each thermopile and then subtracting the outputs from each other using an MCU ADC) seems like the safer option if you're not sure if the signal will be too small if the thermopile outputs are directly subtracted from each other, although it introduces mismatches in differences between offset voltages between the two op-amps used. I am unsure what this would do to the SNR or not, but my gut feels says it would lower the SNR because two -op amps are used.

If you use a differential amplifier (as shown), the subtraction and amplification occur at the same time.

Oh, I guess. But I believe that's more of a conceptual block diagram. The actual schematic is just a regular non-inverting amp two back-to-back series connected thermopiles. Let me see if I can find it again...

I could be wrong though. I don't think for this product they would go that far out of their way to use a diff-amp if they did not have to.

Ah, yeah they do use a diff-amp. That clears a lot of stuff up.