Search

Machine learning models have been widely used in landslide susceptibility prediction. However, landslide multidimensional feature extraction, model generalization ability, and prediction quantification of deep learning are still challenging. This paper proposes a new approach, the landslide density-based bidirectional long short-term memory (LD-BiLSTM) model with multichannel input and an optimized sampling strategy to predict and map landslide susceptibility in active seismic mountainous areas of Sichuan Province, China. First, to ensure the generalization ability of the LD-BiLSTM model, other regions in Sichuan were selected as the model training area independent of the prediction area (Luding County). Multichannel landslide datasets were constructed to extract high-dimensional geospatial features of landslides. Subsequently, the landslide density of each grid cell was utilized as the label for the corresponding input sample. The LD-BiLSTM model was improved by using transfer learning and sampling optimization strategies, which makes our method attenuate the impact of historical landslide inventory deviation on the spatial susceptibility mode compared with the existing DL model, which usually uses landslide objects (LO) as input sample labels. Model performance evaluation results show that the LD-BiLSTM model (precision = 0.903, recall = 0.899, F1-score = 0.901, Area under receiver operating characteristic curve (AUC) = 0.940) outperformed the LO-BiLSTM model (precision = 0.812, recall = 0.815. F1-score = 0.813, AUC = 0.910) in the case areas. Meanwhile, the performance of the LD-BiLSTM model (AUC = 0.9407) significantly outperformed both the information value (IV) (AUC = 0.7207) model and the random forest (RF) (AUC = 0.8116) models in the landslide prediction area (Luding), which confirms that the proposed LD-based method is superior to traditional LO-based methods. Significantly, our approach can effectively extract the spatial distribution of landslides and predict potential landslides in complex high-mountain environments. [ABSTRACT FROM AUTHOR]

The environmental factors of landslide susceptibility are generally uncorrelated or non-linearly correlated, resulting in the limited prediction performances of conventional machine learning methods for landslide susceptibility prediction (LSP). Deep learning methods can exploit low-level features and high-level representations of information from environmental factors. In this paper, a novel deep learning–based algorithm, the fully connected spare autoencoder (FC-SAE), is proposed for LSP. The FC-SAE consists of four steps: raw feature dropout in input layers, a sparse feature encoder in hidden layers, sparse feature extraction in output layers, and classification and prediction. The Sinan County of Guizhou Province in China, with a total of 23,195 landslide grid cells (306 recorded landslides) and 23,195 randomly selected non-landslide grid cells, was used as study case. The frequency ratio values of 27 environmental factors were taken as the input variables of FC-SAE. All 46,390 landslide and non-landslide grid cells were randomly divided into a training dataset (70%) and a test dataset (30%). By analyzing real landslide/non-landslide data, the performances of the FC-SAE and two other conventional machine learning methods, support vector machine (SVM) and back-propagation neural network (BPNN), were compared. The results show that the prediction rate and total accuracies of the FC-SAE are 0.854 and 85.2% which are higher than those of the SVM-only (0.827 and 81.56%) and BPNN (0.819 and 80.86%), respectively. In conclusion, the asymmetric and unsupervised FC-SAE can extract optimal non-linear features from environmental factors successfully, outperforms some conventional machine learning methods, and is promising for LSP. [ABSTRACT FROM AUTHOR]