AI & Machine Learning based early stage Landslide detection using GIS

Abstract

Number of population and areas affected by natural disasters are increasing day by day due to lack of proper planning and response/regularity authorities. Better practices and models must be adopted in order to reduce human and economic loss due to such disastrous events. Landslide is one of natural disaster happens due to downward and outward movement of rock debris and earth materials resulting in vibrations which blocks drainage and roads. In Pakistan, northern areas comprises of regions susceptible to landslides due to extensive mountains and rugged terrain. Landslides disaster can lead to enormous casualties and loss of economy. Landslide hazard mitigation can be done effectively with the help of new methodologies, that can develop better landslide hazard understanding and help to make rational decisions for management of landslide risk. The primary objective of this study is i)To identify factors influencing occurrence of landslides, through a quantitative methodology ii)To identify Artificial Intelligence and Machine Learning based models that are effective in detecting landslides, iii) Cre ation of landslide inventory map data for landslide modeling using open source resources. In our research work, we are proposing to identify landslide inventories using Satellite imagery and field data, calculating susceptibility analysis using Geographical Information Systems (GIS) tools for Muzaffarabad area and then based on that data we will detect landslide prone areas in other regions using Machine Learning tools. For that purpose we will be using Analytical Hierarchy Process (AHP) for landslide parameters determination and Support Vector Machine (SVM), Linear regression, Decision Tree, K nearest neighbor (KNN) classifiers and Neuro Evolutionary Algorithm named as Carte sian Genetic Programming Artificial Neural Network for landslide susceptibility. The results shows that almost 90% accuracy when correlate with the landslide inventories. In this project we focused on susceptibility of landslides using Geographical Information Systems (GIS) tools, machine learning techniques and Artificial Intelligence algorithm.Using Analytical Hierarchy Process we identified landslide prone parameters to be used in susceptibility analysis which were later divided into four categories, i.e. low , Moderate , High and very High landslide prone areas. The analysis show that almost 30% of the area comes under high and very high prone areas of landslides. Barren land and grassland in land cover, fault lines and Muzaffarabad formation, Hazara formation and Holocene in geology are found to be most susceptible factors in Muzaffarabad areas which contributes to landslides.The classification results show the model performance. In the given analysis, Cartesian Genetic Programming Evolved Artificial Neural Network (CGPANN) shows better performance as compared to others, Support Vector Machine, K Nearest Neighbours (KNN) and Logistic regression performance is also good.The per formance score shows 0.81 for knn, 0.83 for Decision Trees, 0.85 for Support Vector Machine and 0.87 for Logistic Regression. Cartesian Genetic Programming Evolved Ar tificial Neural Network outperformed other techniques like SVM & Logistic Regression with 0.96 accuracy. Our proposed methodology will help the government to improve the landslide prediction system and utilize available professional resources efficiently in order to deal with the situation of increasing occurrence of landslides in Pakistan. In future we are going to extend this work by installing sensors and cameras, developing heterogeneous sensor network along with Artificial Intelligence algorithms that are effective in developing new landslide reduction services to predict landslides beforehand to save the community and infrastructure from big losses.