Performance Evaluation of an ANN based Bias Correction algorithm in Monthly and Daily Precipitation Time Series of La Farge Station, USA

Abstract

Understanding the change of future precipitation over long run is highly necessary in climate change impact studies. Mostly, simulated future precipitation series are found to be biased more with the historically observed precipitation series which need to be corrected before use for any impact studies. Many conventional and data-driven methods are available to correct this bias. In this study, to bias correct the monthly and daily precipitation series, Artificial Neural Network based method is applied and compared with the conventional methods. The normalized root mean squared errors obtained for monthly and daily series are 0.786 and 2.55 respectively. It is found that the performance of ANN-based method is poor in daily series and good only in monthly series. The reason for poor performance in daily series is analysed. In addition, the superiority of ANN based method over conventional method is established in monthly precipitation time series.