BPN model for long-range forecast of monsoon rainfall over a very small geographical region and its verification for 2012

Abstract

New operational long range forecasting model of India Meteorological Department (IMD) is statistical in nature, which has many inherent limitations. The correlation between monsoon rainfall and its predictors can never be perfect. They may suffer epochal changes and there may be a cross correlations among predictors. It is almost impossible to identify appropriate predictors of monsoon rainfall over a smaller region like district or division as well. Thus, attempts to forecast monsoon rainfall over a small geographical region like district through this current IMD's operational model become inaccurate. It is found that Back Propagation Neural Network (BPN) is skilled enough to identify the internal dynamics of chaotic data time series and sufficiently suitable to predict future value by past recorded data time series. Thus a BPN model in deterministic forecast has been developed for a long range forecast (LRF) of monsoon rainfall over smaller Indian geographical region. Our study area, Ambikapur is located at 23° 07' 23" N, 83° 11' 39" E, an average elevation of 623 meters (i.e., 2078 feet) and Total Geographical Area (TGA) is 15 733 km2. Performance of the model during the development period (1951-2007) has been found excellent. The performance during the testing period (2008-2011) has also been found good except for the years of 2009 and 2010. The model has also been verified independently and operated for the year 2012. The deviation between actual and predicted monsoon rainfall in Long Period Average (% of LPA) for this year is found 2.7% only. These facts exhibit the efficacy of the proposed model.