GKB-IWAE: A Weighted Ensemble Imputation Framework with Deep Learning for River Water Level Prediction

¹ Brainware University, Barasat, Kolkata, West Bengal 700125, India
² Manipur University, Canchipur, Manipur 795003, India

^* e-mail: This email address is being protected from spambots. You need JavaScript enabled to view it.
^** e-mail: This email address is being protected from spambots. You need JavaScript enabled to view it.
^*** e-mail: This email address is being protected from spambots. You need JavaScript enabled to view it.
^**** e-mail: This email address is being protected from spambots. You need JavaScript enabled to view it.
^† e-mail: This email address is being protected from spambots. You need JavaScript enabled to view it.
^‡ e-mail: This email address is being protected from spambots. You need JavaScript enabled to view it.

Abstract

Floods are among the most damaging natural events. They require timely and effective early warning. This study presents a data-driven method for predicting river water levels (RWL). It combines an ensemble imputation approach with deep learning (DL) models to tackle missing values in multivariate, multi-station hydrometeorological dataset collected from the Water Resource Department and Directorate of Environment in Manipur. To tackle this issue, we propose an ensemble imputation method called GKB-IWAE. This method combines Gaussian Process Regression (GPR), K-Nearest Neigh-bors (KNN), and Backpropagation Neural Network (BPNN) using an inverse error-based weighted averaging strategy. The reconstructed dataset is used to train long short-term memory (LSTM), artificial neural network (ANN), and one-dimensional convolutional neural network (1D-CNN) models for next-day RWL prediction. We evaluate performance using Nash–Sutcliffe efficiency (NSE), root mean square error (RMSE,) and mean absolute percentage error (MAPE) across different lag settings (3, 5, 7, and 10 days). Results show that the GKB-IWAE dataset with 5-day lag outperforms other configurations and individual models, with LSTM consistently beating the other models. This framework offers a dependable method for RWL prediction, aiding flood early warning efforts.