A Study of bankline shifting of the selected reach of Jamuna river left bank using numerical models

Abstract

The present study deals with the development of a computational procedure to predict the morphological behavior of the Jamunariver such as bed level changesand bankline shifting. A reach of 80 kms long left bank of the Jamuna river from Bahadurabad to Sirajganjwas selected for thisstudy .Numerical models named RIVERFLOW 2D, HEC-RAS 1D and multi-variate regression modeling tools have been used. This study also proposes a machine learning technique for the prediction of bank line shifting of river Jamuna based on the results of numerical models.Accordingly, models have been set for the hydro-morphological computation.RIVERFLOW 2D hydro-morphological model has been used for the assessment of erosion-deposition and bed level changes. This model wascalibrated utilizing the water level data collected from BWDB for the year 2018 and then validated for the year 2019. For calibration and validation, the R2 was found to be about 0.86 and 0.93 respectively. In addition, the HEC-RAS 1D hydrodynamic model has been setup, calibrated and validated using the same data. In broad category, two Scenarios have been carried outin the present computational procedure. These are Scenarioi) use of the output of HEC-RAS 1D and regression tools and Scenarioii) use of results of HEC-RAS 2D and regression tools.In Scenario-i, five variables such as maximum velocity, maximum water level, maximum left over bank discharge, maximum slope and minimum waterlevel were taken from the validated 1D hydro-dynamic model.Performance of the prediction of bankline shifting obtained from the computation have been evaluated using statistical parameters, such as RMSE, R2, MSE, and MAE. These performance evaluation have also been carried outfor various types of regression models such as linear regression, tree type regression, boosted, Gaussian, and SVM type regression models. Performance evaluation showed that that the Boosted regression model outperforms the others. For the boosted regression model, the values of RMSE, R2, MSE, and MAE were 25, 0.80, 700, and 10 respectively. On the other hand, in the Scenario- ii, use of results of HEC-RAS 2D and regression tools, three variables such as the minimum water depth, maximum water depth and maximum velocity have been extracted for the purpose of various regression models run. It is found that theprediction of bankline shifting obtained from calibrated and validated boosted regression model performed satisfactorily in whichfor the calibration, the RMSE, R2, MSE, and MAE values were found 38, 0.60, 1444, and 13 respectively. Computational performance of all the models used in the study has also been assessed in terms of discrepancy ratio. Assessment of bed level changes in terms of net erosion and deposition has been analyzed using RIVER FLOW 2D.Model results showed thatmaximum erosion was 17.7 m/year and maximum depositions have been found 7.3 m/year near Bhuapur hard point area.In order to verify the efficacy of prediction of bank lineerosion, Jamuna river left bank data from Google satellite images for the year 2018-2021were also compared.It is found that boosted regression model performed better when compared with others regression models.Based on the statistical parameters (RMSE, MSE, R2, MAE) and discrepancy ratio,Scenario one (HECRAS 1D and regression model) performed comparatively better (RMSE=56, MSE=3200, R2=0.60, MAE=23). On the other hand, Scenariotwo (HECRAS 2D and regression model) performed relatively better (RMSE=50, MSE=2500, R2=0.65, MAE=20) than that of Scenario one. It is hoped that computational procedure suggested in this study based on river models and multi-variate regression tools can be helpful for the analysis of river bank shifting of alluvial rivers.