The startup error of fiber optic gyroscope (FOG) in north-seeking is the error caused by the zero-bias drift of FOG caused by drastic change of the temperature in the starting process. The start-up error significantly increases north-seeking error during the cold startup phase compared to the stable phase, which prolongs the effective north-seeking time. Through the analysis of the factors affecting the temperature drift of FOG, the multi-parameter linear model was established by empirical mode decomposition (EMD), autoregressive-moving average (ARMA) modeling and Kalman filtering to realize a temperature drift compensation method applied to FOG north-seeking. The experimental results show that the method can reduce the north-seeking startup error by nearly 80%, so that the startup north-seeking precision is equivalent to the stable phase and the effective north-seeking time is shortened.