A piecewise compensation scheme is proposed to solve the problem that the digital closed-loop fiber-optical gyroscope is sensitive to changes of ambient temperature in its working process, which may lead to temperature drift.By analyzing the causes of temperature drift, a multi-model fitting method is introduced based on a new segmentation strategy and the traditional single-model fitting method.In this method, the full-range test data is divided into several data areas, and the parameters in each interval are estimated through multiple linear regression.In its realization process, the microcontroller collects the temperature signals in the center of the fiber-optical ring in every second, which are used for data processing.The compensation amount is calculated by using the model parameters from corresponding segments.The FPGA reads the compensation data for temperature compensation.Experimental results show that this scheme has better compensation effects than that of the method of single model fitting,and the full temperature error of the compensated gyro is reduced by an order of magnitude, compared with that of the uncompensated gyro.