The reserves of Yuba Back Reef Strands Reservoir in Changxin group is large but the thickness of reservoir is thin, therefore, identification of gas and water is one of the most difficult problems. Although the analysis of low frequency shadow is a good way to identify liquid, it hardly identify liquid in thin reservoir. To solve this problem, in this thesis we introduce Deconvolution Short-time Fourier transform. Compared with traditional problem, it provides higher time-frequency resolution. Because STFT is applied in this method, the window adopted in the transform will affect identification result heavily. In this paper, we set up relationship between seismic dominant frequency and time window to obtain best time-frequency resolution by establishing model and analyzing seismic data. Based on the above, we can make low frequency shadow analysis with high resolution. Practical application shows it can improve the result.