Freezing of gait (FOG) in the Parkinson’s disease has a complex mechanism and is closely related brain activities. Timely prediction of FOG is crucial to fall prevention and injury avoidance. Traditional electroencephalogram (EEG) processing methods extract time, spatial, frequency, or phase information separately and use them or their combinations, which fragment the connections among these heterogeneous features and cannot completely characterize the whole brain dynamics during the occurrence of FOG. In this study, dynamic spatiotemporal coherent modes during the FOG were studied and the associated FOG detection and prediction were proposed. For capturing the changes of the brain, dynamic mode decomposition (DMD) method was applied. Dynamic changes of the spatiotemporal modes in both amplitude and phase of motor-related frequency bands were analyzed and an analytic common spatial patterns (ACSP) was used as a spatial filter to extract the essential differences among the normal, freezing and transitional gaits.  The proposed method was verified in practical clinical data. Results showed that in the detection task, the DMD-ACSP achieved an accuracy of 89.1 ± 3.6% and sensitivity of 83.5 ± 4.3%, respectively. In the prediction task, an 83.5 ± 3.2% accuracy and 86.7 ± 7.8% sensitivity were achieved.  Comparative studies showed that the DMD-ACSP method significantly improves the FOG detection and prediction performance. Moreover, the DMD-ACSP reveals the spatial patterns of dynamic brain functional connectivity which best discriminate the different gaits.  The spatiotemporal coherent modes may provide useful indication for transcranial magnetic stimulation neuromodulation in medical practices.