Modeling the impact of landslide processes on agricultural ecology

The impact of landslides on agriculture and the environment is considered. Models for predicting landslides are constructed based on specified multimodal data in the form of numerical, categorical, and masked images. Various cases of landslide classification using machine learning algorithms specified in the form of .csv files and images for a specific area are also considered. Based on geomorphological, climatic, and geological factors, models are constructed using machine learning algorithms for landslide classification. Neural modeling models are built based on transfer learning using various neural network optimizers. Machine learning models are implemented using logistic regression as the base method, random forest from the class of ensemble methods, and a modern approach to landslide classification problems, gradient boosting (XGBoost). In many cases, the DenseNet neural network and other deep learning architectures are used for multimodal data. A DenseNet model for landslide classification is developed based on the provided dataset in the form of tabular data. The construction of a transfer learning model for landslide prediction is based on an extended image database with hyperparameter tuning for processing HDF5 (.h5) data. The resulting models are trained using transfer learning and utilize pretrained models of various architectures, such as EfficientNetV2 with additional training for landslide segmentation. Advanced models based on the UNet model for landslide prediction are built using .tif images. An extended UNet model is presented for solving the problem of image segmentation for landslide prediction. High-level abstractions in the data with quadratic nonlinear relationships are found in the feature space. Models are created for predicting the impact of landslides on the environment and damage to agricultural land.