Impact of water vapor on localization metrics. The purity exhibits a relative change of about l%, while the completeness decreases significantly by 4%. The model does not identify many FPs as the noise levels increase, but its capability to accurately detect existing sources diminishes rapidly. Notably, even with the highest noise levels, the completeness score outperforms the previous scores on the test set with the low water vapor noise. The dashed line is at 1.796 PWV units, which corresponds to the training data.