Aiming at the nonlinear characteristics of temperature drift in fiber optic gyroscopes (FOGs), an adaptive modeling method based on the time scale characteristics of temperature drift experimental data is proposed in this paper. Firstly, the empirical mode decomposition (EMD) to decompose the temperature drift data into highfrequency oscillation sequence and monotonic trend term, and according to its time scale characteristics, the surface fitting regression and adaptive fuzzy reasoning are used for the joint modeling, and then the signal synthesis is carried out. Compared with the singleused surface fitting regression or adaptive fuzzy reasoning modeling, the experimental results of the proposed method have obvious advantages and significant compensation effect. The results of the verification test further confirm the effectiveness of the proposed method in dealing with similar problems.