IMPROVED STREAMFLOW FORECASTING AND FLOOD WARNING USING VARIATIONAL MODE DECOMPOSITION AND EXTREME GRADIENT BOOSTING

Numerous studies have recently shown the potential for Machine Learning (ML) to enhance the prediction of hydrological processes such as runoff and evaporation (Moosavi et al., 2017; Rezaie-Balf et al., 2019). Studies investigating the prediction of streamflow have explored a wide variety of ML models, along with different preprocessing and decomposition techniques (Wang et al., 2018; Zuo et al., 2020). Extreme gradient boosting (XGBoost) has been shown to be an effective approach for streamflow forecasting (Ni et al., 2020). Many preprocessing techniques, including Gaussian mixture models and various forms of time series decomposition such as wavelet transform and empirical mode decomposition (EMD) have been proposed to improve upon ML streamflow prediction models (Moosavi et al., 2017; Ni et al., 2020), but Variational Mode Decomposition (VMD) was recently proposed (Dragomiretskiy & Zosso, 2014) as an alternative approach to EMD (shown to effectively isolate underlying cyclical patters related to external factors) which improves upon limitations caused by EMD’s lacking mathematical foundation (Zaji et al., 2019). In this study, a VMD-GMDH streamflow forecasting model is developed and assessed. To overcome limitations common among similar models such as accurate hindcasting alongside poor forecasting or misleading performance due to imitation error, varying loss functions are compared in their ability to produce an effective flood forecasting tool. For example, horizontal error (HE, proposed as a measure of a model’s vulnerability to imitation error (Zaji et al., 2019)) is crucial when assessing the performance of a forecasting tool, as it provides a realistic measure of flood warning ability. Mean squared error and mean average error are common metrics used to measure loss when optimizing a model, and thus will be assessed relative to and in combination with horizontal error to minimize imitation error and improve overall flood warning. A HE-based VMD-XGBoost model has shown to accurately predict daily stream flow in the Bow River (Alberta, Canada), with r², RMS MAPE, and HE of 0.83, 33.0, 2.66, and 0.39, respectively. Similar models are assessed using varying timesteps including hourly, 10-day, and monthly flow rates to outline the strength of VMD for preprocessing at varying timesteps.