Rainfall data feature extraction and its verification in displacement prediction of Baishuihe landslide in China

Rainfall is one of the main factors that influence the stability of slopes. However, rainfall data have some common features, such as huge data volume and difficulties in direct use. Currently, measurements such as daily rainfall, total rainfall volume and rainfall intensity are widely used for rainfall data feature extraction, which weakens the comprehensive impact of rainfall on slope stability. A feature extraction method for rainfall data is proposed in this paper. Rainfall data is transformed into feature matrices, which have much smaller data volumes. These feature matrices contain lots of useful information and can be used directly in landslide analyses. Based on the statistics of each of the rainfall events, this article applies K-means to classify these events. By dividing rainfall volume into categories of evaporation, infiltration and runoff, feature extraction is conducted. To quantitatively analyze the comprehensive impact of rainfall on landslide stability, Particle Swarm Optimization (PSO) is utilized to search for an array of weight coefficients for evaporation, infiltration and runoff under various rainfall types, which eventually leads to the feature extraction of rainfall data. This feature extraction method is applied to the rainfall data feature analysis of the Baishuihe landslide area. The rationality and validity of the method are verified by the results of landslide displacements predicted by Back-Propagation (BP) neural network. This study provides an effective rainfall data feature extraction method and a new direction for quantitative analysis of landslide monitoring data.