Massive landslides over large regions can be triggered by heavy rainfalls or major seismic events. Mapping regional landslides quickly is important for disaster mitigation. In recent years, deep learning methods have been successfully applied in many fields, including landslide automatic identification. In this work, we proposed a deep learning approach, the ResU-Net, to map regional landslides automatically. This method and a baseline model (U-Net) were collectively tested in Tianshui city, Gansu province, where a heavy rainfall triggered more than 10,000 landslides in July 2013. All models were performed on a 3-band (near infrared, red, and green) GeoEye-1 image with a spatial resolution of 0.5 m. At such a fine spatial resolution, the study area is spatially heterogeneous. The tested study area is 128 km², 80% of which was used to train models and the remaining 20% was used to validate accuracy of the models. This proposed ResU-Net achieved higher accuracy than the baseline U-Net model in this mountain region, where F1 improved by 0.09. Compared with the U-Net model, this proposed model (ResU-Net) performs better in discriminating landslides from bare floodplains along river valleys and unplanted terraces. By incorporating environmental information, this ResU-Net may also be applied to other landslide mapping, such as landslide susceptibility and hazard assessment.

The 2008 Wenchuan Earthquake that occurred in a mountainous region of China induced massive landslides and caused numerous casualties and property losses. Analyzing the disturbances on vegetation detected from the abnormal sudden drops of the normalized difference vegetation index (NDVI) within a short period can be used for the purpose of rapid landslide identification. Although much research has confirmed the necessity of high-resolution satellite images in landslides identification, Moderate Resolution Imaging Spectroradiometry (MODIS) products still have their usefulness for high temporal resolution, as investigated by the authors. Using MODIS MOD09Q1 NDVI products at a temporal interval of 8 days during 2008, this letter presents a method that has been developed to identify landslide distribution and evolution patterns. First, to find the optimal threshold, the MODIS NDVI time series are analyzed in a training area by an iteration searching procedure. Second, the chosen threshold is used in a larger validation area. To examine the effectiveness of the proposed method, the results are compared to interpreted landslides using SPOT5 images with a spatial resolution of 2.5 m acquired before and after the main shock. An overall 75% accuracy is achieved, and better consistency is observed for landslides extending over one MODIS pixel. The proposed method has also been applied to the Wenchuan earthquake affected areas with seismic intensity IX and greater, and the similar spatial pattern of landslides distribution is obtained when compared with results by using high-resolution images and field investigation. This technique can be applied practically for rapid landslide assessment at a relatively large region after a major earthquake or other severe disturbance events.