A Novel Deep Geospatial Neural Network for Predicting Urban Land Subsidence

The problem of urban land subsidence (depending on the region) poses an increased danger to the systems infrastructure, human life, and the ecosystem in the environment. This paper is based on the premise of proposing a new model called Deep Geospatial Neural Network (DGNN) focused on accuracy in predicting land subsidence in terms of space and time. The model's capability spans across acquiring geo-spatial information which includes satellite based InSAR images, digital elevation models, satellite imagery, developed area, and even hydro-geological factors. Its ability in positioning different levels of hydrology and ‘anthropogenic’ factors relies on a deep learning design that integrates CNN for surface level data interpretation and LSTM for temporal data. The information that is fed into the LSTM is employed for training the model on a 10-year dataset-subsidence prone metropolitan areas. The outcomes indicate that the model is capable of achieving new levels when compared to traditional machine learning algorithms alongside CNN and LSTM in terms of accuracy and the level of details provided. In addition, the model provides crucial information on high-risk regions for efficient planning to avert disaster. This work demonstrates the capabilities of geomapping learning methods concerning infrastructure disturbances in the census region focused on intelligent data processed monitoring systems.