Early detection of riverine flooding events using the group method of data handling for the Bow River, Alberta, Canada

Abstract

ABSTRACT While numerous studies have investigated physically-based analytical approaches for estimating stream flow probability distributions and occurrences of overbank flow, these types of models are limited by their associated complexity to incorporate a wide range of data from all components of the hydrologic system to model their influence on river flows. Alternatively, the Generalized Structure Group Method of Data Handling (GS-GMDH) is a polynomial network approach used in this study to train and test models for daily and hourly times series flow prediction for riverine flooding using available data from 1990 to 2018 and 1996 to 2018, respectively. The model is found to accurately predict daily flows with R 2, RMSE, MAE, Bias and NSE of 0.6441, 46.884, 6.700, 1.800 and 0.6441, respectively, for nine years of flow data in application to the Bow River in Alberta, Canada. Hourly flow data used to train (70%) and test (30%) the GS-GMDH model results in R 2, RMSE, MAE, Bias and NSE of 0.998, 3.323, 0.997, 0.00438 and 0.998, respectively. The trained hourly model can predict up to 17 h in advance while maintaining R 2 greater than 0.90. Horizontal error highlights a weakness in model performance, contrary to other evaluation statistics, due to presence of imitation error.