Abstract: Aiming at the issue that traditional slope landslide early warning methods for mining areas fail to effectively consider the influence of terrain characteristics on landslide processes, this study proposes an adaptive early warning threshold method based on dynamic time decay weight (DTDW). First, before processing radar image data, an automatic outlier detection and correction algorithm is applied to reduce data anomalies and enhance monitoring data quality. Next, the processed time-series radar deformation images are analyzed to extract the maximum deformation area as the study region, and multiple monitoring parameters (including velocity, acceleration, and coherence) within this area are calculated. Subsequently, the dynamic time decay weight (DTDW) is introduced, which assigns varying degrees of decay weights to deformation image parameters from historical moments, thereby enabling continuous influence on the parameter estimation at the current moment. Finally, a three-level warning mechanism is established, along with an adaptive warning threshold adjustment method based on a sliding window analysis strategy. This method dynamically monitors parameter trends and optimizes warning thresholds in real time. By continuously updating and calibrating warning criteria for different parameters, the system’s responsiveness to abnormal deformations and its adaptability to environmental changes are improved. Experimental results based on real mining area landslide data show that, compared with traditional fixed-threshold warning methods, the proposed approach can identify landslide signs earlier and issue warning signals more promptly. It demonstrates significant advantages in both timeliness and accuracy, effectively enhancing the monitoring capability for landslide risks.