Abstract: Objectives: Under complex geological environmental conditions, due to occlusion and reflection by vegetation and slope bodies, etc., multipath errors will occur in GNSS observation data, thereby reducing the accuracy of landslide monitoring. Unsupervised learning methods have strong adaptability and flexibility. They can directly mine potential patterns from the original GNSS observation data and then separate GNSS multipath signals. To this end, a GNSS observation data quality optimization method based on multi-index thresholds is proposed. Methods: Four prior characteristic parameters, namely signal-to-noise ratio, elevation angle, pseudo-range consistency and carrier-phase consistency, are comprehensively utilized. The high-quality GNSS observation sequences are screened through the Fuzzy C-means (FCM) clustering algorithm, and the minimum value of the 3𝜎 interval is taken as the threshold for eliminating the co-viewing satellite observations. If the observed value is less than the threshold, it will be eliminated due to the influence of multipath error. If it is greater than the threshold, the observed value will be retained for subsequent positioning and calculation. Results: In a normal environment, this method does not have a significant impact on the ambiguity fixation rate and positioning results; In the complex landslide environment, compared with the traditional ADEM method based on terrain space, the ambiguity fixation rate and horizontal positioning result of this method have increased by 5.0% and 34.7% respectively. Moreover, the experimental results based on the continuous observation data for one week show that the threshold determined by this method can stably and effectively weaken the influence of multipath errors. Conclusions: The proposed method can stably and effectively weaken the influence of multipath errors, improve the ambiguity fixation rate and positioning accuracy, avoid the cumbersome calculation process, and make it more suitable for long-term landslide monitoring applications.