Daily Flood Forecasts with Intelligent Data Analytic Models: Multivariate Empirical Mode Decomposition-Based Modeling Methods

Flood causes massive damages to infrastructure, agriculture, livelihood and leads to loss of life. This chapter designs M5 tree-based machine learning model integrated with advanced multivariate empirical mode decomposition (i.e., MEMD-M5 Tree) for daily flood index (FI) forecasting for Lockyer Valley in southeast Queensland, Australia, using data from January 01, 1950, to December 31, 2012. The MEMD-M5 tree is evaluated against MEMD-RF, standalone M5 tree, and RF models via statistical metrics, diagnostic plots with error distributions between simulated and observed daily flood index. The results indicate that MEMD-M5 tree outperforms the comparative models by attaining maximum values of r = 0.990, WI = 0.992, ENS = 0.979, and L = 0.920. The MEMD-M5 tree outperforms other models by registering the least value of RMSE and MAE and can precisely emulate 97.94% of daily FI value. Graphical diagnostic analysis and forecast error histograms further reveal that the MEMD-M5 tree has a greater resemblance to that of the observed data supporting the outcomes of statistical evaluation. Such advancements in flood prediction models, attained through data intelligent analytical methods, are very vital and effective in ensuring better mitigation and civil protection in emergency providing an early warning system, disaster risk reduction, disaster policy suggestions, and reduction of the property damage.