Simulation of Land Subsidence Using InSAR Time Series and Aquifer Parameters in a Neurocomputational Approach
Ashrafianfar, Nazemeh1; Busch, Wolfgang1; Dehghani, Maryam2; Mohammad Rezapour Tabari, Mahmoud3
1IGMC, GERMANY; 2Department of Civil and Environmental Engineering, School of Engineering, Shiraz University, Shiraz, IRAN, ISLAMIC REPUBLIC OF; 3Institute of Geotechnical Engineering, Shahrkord University of Technology, Shahrekord, IRAN, ISLAMIC REPUBLIC OF
This research investigated the nonlinear relationships between InSAR-derived land subsidence and hydro-geological and geophysical parameters of an aquifer system. To this aim, an inverse problem approach was applied using the performance of the nonlinear neural networks. Pumping-induced land subsidence is a complex phenomenon, which happens by changes and interactions among the aquifer system constituents. The ability of a multilayer static neural network was applied for data mining and generalization of nonlinear relation between the subsidence and the aquifers' parameters over the Hashtgerd plain of the north-western Iran. A proper multilayer nonlinear static network was designed by defining the input-output patterns, topology, supervised learning algorithm and the other system trajectories. In this model, the aquifer parameters and the ENVISAT time series results were used as the input and the output patterns, respectively. This model simulated the Hashtgerd land subsidence between the years 2003 and 2008. In fact, the InSAR-derived land subsidence was used as an inverse solution in the implemented network, and the optimized backpropagation learning algorithm of the network used it successfully in the learning procedures. A sensitivity analysis was implemented for the inputs-output patterns of the network to estimate the most important parameter on the subsidence variations. This research results gave us a more clear perspective about the subsidence occurrence over the Hashtgerd plain and can be helpful in the water resources programs for the area.