Land subsidence susceptibility assessment using random forest machine learning algorithm

The mechanism of land subsidence and soil deformation deals with the dissipation of excess pore water pressure and the compaction of soil skeleton under the effect of natural or man-made factors, which can lead to serious disasters in the process of urbanization. The negative effects of land subsidence include structural and fundamental damages to underground and aboveground infrastructures such as pipelines and buildings, changes in land surface morphology, and creation of earth fissures. Arid and semi-arid countries like Iran are highly prone to land subsidence phenomenon. In these regions, precipitation rate and natural recharges are relatively lower than those of the global average showing the importance of ground waters for agricultural and industrial activities. Land subsidence has already occurred in more than 300 plains in Iran. Semnan Plain is one of the most important areas facing this phenomenon. The purpose of this research was to assess land subsidence susceptibility using random forest machine learning theory. At first, prioritization of conditioning factors was done using random forest method. Results showed that distance from fault, elevation, slope angle, land use, and water table have the greatest impacts on subsidence occurrence. Then land subsidence susceptibility map was prepared in GIS and R environment. The receiver operating characteristic curve was applied to assess the accuracy of random forest algorithm. The area under the curve by value of 0.77 showed that random forest is an acceptable model for land subsidence susceptibility mapping in the study area. The research results can provide a basis for the protection of environment and also promote the sustainable development of economy and society.