Streamflow drought index-based prediction of hydrological drought using M5P and gene expression programming models

Abstract

Hydrological drought is one of the complex natural phenomena affecting the surface water resources. Thus, reliable prediction of hydrological drought is crucial for effective planning and management of available water resources on a basin scale. In the present study, hydrological drought was predicted at Naula and Kedar stations using two data-driven models including the M5P tree and gene expression programming (GEP) based on streamflow drought index (SDI). The nomination of the significant inputs (lags) was done based on partial auto-correlation function (PACF) analysis of SDI6 and SDI9, and the same was supplied to the M5P and GEP models for drought prediction at Naula and Kedar stations, located in Naula watershed, Uttarakhand State (India). The predictive output of the data-driven models (M5P and GEP) was compared with actual values of SDI6 and SDI9 based on statistical indicators including mean absolute error (MAE), coefficient of correlation (COC), root-mean-square error (RMSE), and Nash–Sutcliffe efficiency (NSE). Also, the scattered and temporal-variation diagrams, Taylor diagram, and box plot were created for the visual interpretation. The appraisal of results showed that the M5P model performed better for SDI6 (MAE = 0.235, 0.201; RMSE = 0.345, 0.274; NSE = 0.820, 0.661; COC = 0.911, 0.849) and SDI9 (MAE = 0.175, 0.158; RMSE = 0.265, 0.211; NSE = 0.905, 0.767; COC = 0.953, 0.903) prediction at Kedar and Naula stations, respectively. The findings of this study can be useful for decision-makers and hydrologists in formulating drought mitigation plans and risk management strategies for sustainable water resources management in the study catchment.