|
چکیده
|
Land subsidence is mainly caused by excessive groundwater abstraction from aquifers. This study introduces Dynamic Subsidence Vulnerability Index (DSVI) by estimating possible land subsidence time variations by considering changes in groundwater level based on the ALPRIFT framework in Iran’s Hadishahr Plain, which is summarized in three modules. (i) Module I: mapping Subsidence Vulnerability Index (SVI) utilizing the ALPRIFT framework and optimization its weights by the Multiple Artificial Intelligence Models (MAIM) strategy; (ii) Module II: predicting groundwater level by Group Method of Data Handling (GMDH); and Module III: mapping DSVI by combining the results from Modules I and II. A two-pronged strategy is employed in MAIM: In Level 1, multiple models are derived from Sugeno Fuzzy Logic (SFL) and Support Vector Machin (SVM); and in Level 2, the outcomes of Level 1 models are combined by Artificial Neural Networks (ANN). According to the results: (i) ALPRIFT exhibits a correlation coefficient (r) of about 0.55 with corresponding measurements of land subsidence; (ii) using SVM and SFL to optimize the weights, r is raised to 0.83 and 0.74, respectively; (iii) the use of multiple models at Level 2 results in better performance than that of a single model at Level 1; and (iv) on the DSVI map, the central part of the plain is vulnerable at hotspot areas where groundwater is being improperly withdrawn from the Hadishahr Plain aquifer, increasing the risk of subsidence.
|