When we go from 50K to 18K, the sample cools down. Maybe this means that the thermal component of the motion of the atoms (so of the nuclei) in source becomes smaller. This means more of the source nuclei obtain the “correct” speed when put on an oscillator. As more nuclei have the correct speed to be correctly Doppler shifted, there will be more resonant scattering, so the dips in the spectrum will be much deeper at a lower temperature.
I am not entirely sure of this explanation. Maybe there is an effect that causes the magnetic hyperfine field to be stronger at lower temperatures (so deeper dips), but I would not know what could cause it.