Estimation of urban flood volume using Low-Impact Development methods and machine learning approach.

Abstract

In recent years, the growth of urbanization has increased the impermeable levels and has caused an increase in the volume of floods and peak flood discharges. Many types of research have been done in the field of implementing environmentally friendly methods to make the urban environment more natural and at the same time be effective in controlling urban floods. The use of Low Impact Development (LID) methods is one such study. But choosing the best designs from among these strategies has always been a vital problem for urban water system designers. One of the goals researchers in the field of urban hydrology is to find methods to determine the required volume reduction. In the present study, the hydraulic behavior of the surface water collection system in the Golestan town of Semnan has been simulated using Storm Water Management Model (SWMM). In the following, the performance of implementing different proposed designs of three types of LID_ namely rain barrel (RB), infiltration trench (IT), and permeable pavement (PP) _ was investigated. These plans include seven general scenarios, each with ten different LID combinations. These plans include seven general scenarios, each with ten different LID combinations. The results of hydraulic studies indicate the effectiveness of the PP-RB scenario with an average reduction of 90% of peak discharge and an average reduction of 80% of total flood volume. Also, the weakest performance is related to the IT scenario with an average reduction of 60% of peak discharge and 47% of total flow volume. In this regard, the considering the importance of estimation of flood reduction, present paper introduces a novel approach for urban flood mitigation estimation. In this method, intelligent algorithms are used to perform flood calculation operations taking into account the percentage of LIDs (Low-Impact Developments) proposed. In this research, SVM (support vector machines), LSSVM (Least square support vector machines) and LSSVM-GOA (Least square support vector machines-grasshopper optimization algorithm) algorithms have been used. The allocated percentage area of ​​different combinations of the used LIDs, and the reduced peak flow coefficient in each combination were considered as input data; and the reduced flood volume corresponding to each LID combination was used as the output data. This research has been conducted in Golestan town of Semnan city in Iran. The results obtained in this study indicate the success of these algorithms in predicting reduced flood volume. In the test period, the value of R2 index for LSSVM-GOA model (0.9896) compared to LSSVM (0.9266) and SVM (0.8990) intelligent models, indicates the high accuracy of this model in this period. Also, the values of R2 index for the other two algorithms indicate the adequacy of these algorithms in predicting the amount of reduced flood. Also in the training course, LSSVM - GOA model showed that with values of 0.0101 and 0.0185 for MAE and RMSE indices, respectively, has a higher predictive power. The values of these indices are 0.0268 and 0.0361 for LSSVM model and 0.0318 and 0.0434 for SVM model, respectively. According to the results, the use of intelligent algorithms can be introduced as an accurate tool in estimating and predicting reduced flood volume in urban basins.