loading page

Analyzing the impact of bias correction of ensemble rainfall forecasts on streamflow prediction skill of a hydrodynamic model
  • Amina Khatun,
  • Bhabagrahi Sahoo,
  • Chandranath Chatterjee
Amina Khatun
PhD Research Scholar, IIT Kharagpur, India

Corresponding Author:aminakhatun9286@gmail.com

Author Profile
Bhabagrahi Sahoo
Associate Professor, IIT Kharagpur,India
Author Profile
Chandranath Chatterjee
Professor, IIT Kharagpur, India
Author Profile


Use of ensemble rainfall forecasts has gained popularity in providing detailed uncertainty information and improving hydrologic prediction skill. India Meteorological Department (IMD) provides medium-range multi-model ensemble (MME) forecasts for all over India with 1- to 5-day lead time. In this study, we bias correct the IMD MME rainfall forecasts using a modified version of Kohonen Self-Organizing Maps (KSOM) and analyse the effect of bias correction on the streamflow prediction skill of MIKE 11 Hydrodynamic (HD) model for the years 2012-2014. We have selected the upper region of the Mahanadi River basin as the test bed. The results indicate improvement of rainfall forecasts after bias correction. Subsequently, use of bias corrected rainfall forecasts as input forcing to the MIKE 11 HD model provides better streamflow forecasts at all the lead times (Nash Sutcliffe Efficiency, NSE ranging from 0.89 to 0.41) compared to the use of raw forecasts (NSE ranging from 0.89 to -0.51). To further improve the streamflow forecasts, we have applied the recently developed robust wavelet-based non-linear autoregressive with exogenous inputs dynamic neural network model, WNARX. This post-processing operation tends to improve the streamflow forecast quality to an acceptable range (NSE = 0.92-0.71). The results encourage us to conclude that the IMD MME forecasts has great potential to improve streamflow prediction skill of a hydrodynamic model.