A novel swarm intelligence: cuckoo optimization algorithm (COA) and SailFish optimizer (SFO) in landslide susceptibility assessment
Rana Muhammad Adnan Ikram,Atefeh Ahmadi Dehrashid,Binqiao Zhang,Zhihuan Chen,Binh Nguyen Le,Hossein Moayedi
DOI: https://doi.org/10.1007/s00477-022-02361-5
IF: 3.821
2023-01-31
Stochastic Environmental Research and Risk Assessment
Abstract:Inherent hazards such as landslides pose a threat to human life and may inflict significant harm on the surrounding ecosystem. For planning, controlling, and avoiding landslide situations to minimize damages, a landslide susceptibility map is necessary. As a consequence of this, the current research makes use of a methodical approach and upgraded algorithms to identify and forecast locations that are susceptible to landslides. When it comes to problems associated with landslides, standard optimization techniques have been used quite a bit. This study presents a novel approach to the development of an artificial neural network (ANN) in the Iranian region of Kurdistan by using the cuckoo optimization algorithm (COA) and the SailFish optimizer (SFO) as metaheuristic approaches. In order to maximize the computational properties of these algorithms and depict a new kind of swarm intelligence, a multi-layer perceptron (MLP) neural network is used in the synthesis process. The findings of the landslide hazard maps were checked and compared using actual landslide sites. There were 1072 landslides shown on the inventory map. There was a 70:30 split between training and testing locations at random. Model input was narrowed down to 16 different landslide qualifying variables, namely elevation, slope aspect, slope angle, NDVI, distance to fault, plan curvature, profile curvature, rainfall, distance from river, distance to road, SPI, STI, TRI, TWI, land use, and geology. All of these parameters were considered to be important in determining the likelihood of a landslide occurring. The area under the curve (AUC) criterion was used to evaluate the accuracy of the probabilistic models that were put into use. Incidentally, the calculated comparable AUCs were as follows: 0.797, 0.789, 0.784, 0.779, 0.763, 0.758, 0.749, 0.740, 0.725, and 0.716 for COA-MLP, and 0.719, 0.695, 0.682, 0.675, 0.671, 0.670, 0.662, and 0.650 for SFO-MLP. The greatest hybrid model for forecasting landslide detection corresponds to the COA-MLP model, and it has a swarm size of four hundred people. As a consequence, the findings demonstrated that these two models had an effective performance for ANN-MLP optimization. Taking into consideration this evaluation, the hybrid models that were provided are trustworthy for the modeling of landslide susceptibility. As a result, the map of vulnerability that was developed can be utilized for hazardous design and increased planners' knowledge of dangerous locations.
environmental sciences,engineering, environmental,water resources, civil,statistics & probability