Application of ensemble fuzzy weights of evidence-support vector machine (Fuzzy WofE-SVM) for urban flood modeling and coupled risk (CR) index for ward prioritization in NCT Delhi, India
Syed Irtiza Majid,Manish Kumar,Netrananda Sahu,Pankaj Kumar,Dinesh Kumar Tripathi
DOI: https://doi.org/10.1007/s10668-024-04926-6
IF: 4.08
2024-05-11
Environment Development and Sustainability
Abstract:NCT Delhi, the heart of India, remains vulnerable to urban flooding from July to October, when southwest monsoon is active over its area of 1483 km 2 . To address the paucity of a comprehensive susceptibility map, this study employs urban flood modeling to quantify the spatial sensitivity of NCT Delhi to water logging. Fifteen flood related variables, including elevation, slope, slope aspect, profile curvature, plan curvature, Topographic Wetness Index (TWI), Stream Power Index (SPI), Topographic Roughness Index (TRI), geology, soil type, land use/land cover (LU/LC), Modified Fournier Index (MFI), water level depth, distance from storm drain, and distance from the Yamuna were analyzed. The study applies the data-driven form of ensemble fuzzy weights of evidence-support vector machine (Fuzzy WofE-SVM). The first-level flood susceptibility map was generated using Fuzzy WofE technique. Using this map and secondary data on water logging locations identified by the Delhi Traffic Police, flood inventory datasets were created. To execute the ensemble methodology, SVM was sequentially integrated with the Fuzzy WofE technique. Urban flood susceptibility zonation maps of NCT Delhi were created using four SVM kernel functions: linear (LN), polynomial (PL), radial basis function (RBF), and sigmoid (SIG). The kernel parameters i.e., regularization parameter ( C ), kernel width ( γ ) and degree ( d ) were optimized using python-mediated k -fold cross validation method. Area under the receiver operating characteristic curve (AUROC) analysis validated that all the SVM kernels yield excellent success and prediction rates. However, in terms of success rate, RBF (AUROC = 0.976) outperforms PL (AUROC = 0.968), LN (AUROC = 0.966), and SIG (AUROC = 0.956). With an AUROC of 0.966, RBF again outperforms SIG (AUROC = 0.964), LN (AUROC = 0.955), and PL (AUROC = 0.952) in terms of predictive performance. The novel Coupled Risk (CR) Index has been developed and presented in this study to increase the applicability of flood modeling by translating macro-hazard perspectives into a polished vulnerability scenario of administrative convenience. Spatial analysis of hazard intensity and endangered population using this novel tool would benefit take specific actions for disaster management by prioritizing wards for mitigation strategy planning and implementation.
environmental sciences,green & sustainable science & technology