Spatial heterodyne spectroscopy has been used in atmospheric trace gases remote sensing field because of its high signal to noise ratio and spectral resolution, and the adaptive correction is a key link in spectrum pretreatment. Based on the spectral characteristics, two methods: the threshold fitting and improved empirical mode, were used separately to correct baseline in measured near infrared moisture spatial heterodyne spectrum. Results show: both the two methods can realize spectral baseline deduction automatically; in addition, the spectral distortion and similarity are 0.761 and 0.955, respectively, when using threshold fitting, and the spectral distortion and similarity are 0.717 and 0.954 (somewhat better), respectively, when using improved empirical mode. As far as the time-consuming is concerned, improved empirical mode using fewer iterations to attain the final baseline spectrum, which is no more than one-tenth of the threshold fitting.