A early warning model of regional landslide in Qingchuan County, Sichuan Province based on logistic regression

Abstract

In Qingchuan County of Sichuan Province, landslide disasters occur in a large number of places and cover a wide range of areas. Early warning of regional landslide disaster is an important means of effective disaster prevention and mitigation, and an early warning model is the core of successful early warning. The traditional regional geological disaster warning model is limited by the lack of big data and analysis methods of the complicated investigation and monitoring mechanism of the landslide in the study areas, and it has some problems, such as limited warning precision and insufficient refinement. In this paper, the training sample set of landslide disaster in Qingchuan County is constructed on the basis of the integrated collation and data cleaning of the results of geological disaster investigation and monitoring and precipitation monitoring. The sample set includes 27 input feature attributes such as geological environment rainfall and 1 output feature attribute, covering the total number of the samples in Qingchuan County in the past 9 years (2010—2018) up to 1826 (613 positive samples, 1213 negative samples). Based on the logistic regression algorithm, the study and training of the sample set is carried out with a 50%-fold cross validation. The Bayesian optimization algorithm is used for model optimization, and the accuracy and model generalization ability of the model are verified by such indicators as accuracy, ROC curve and AUC value. The ROC curve is also known as the “Receiver Operating Characteristic” curve. AUC value represents the area under the ROC curve. The verification results show that the training result model based on logistic regression algorithm is of good accuracy and generalization ability (accuracy 94.3% and AUC 0.980). Finally, it is proposed that in the actual warning of regional landslide, 27 characteristic attributes of each warning unit in the research area are input according to the format of characteristic attributes of training samples, and the pre-learned and trained model is called to output the probability of occurrence of landslide disaster, and the warning level of landslide disaster is segmented according to the output probability. A yellow alert is issued when the output probability P is greater than or equal to 40% and P is less than 60%. An orange alert is issued when the output probability P is greater than or equal to 60% and P is less than 80%. A red alert is issued when the output probability P is greater than or equal to 80%. In the next step, the accuracy of the model will be further verified in the landslide disaster early warning business in Qingchuan county.

Keywords

• landslide hazard
• early warning model
• logistic regression
• model building
• warning level