Artificial Neural Network Technique for Statistical Downscaling of Global Climate Model

Abstract

The nineteenth century, an era of industrialization, witnessed huge utilization of natural resources extensively for the construction activities besides emission of nitrous oxide and methane (green house gases). Rises in the GHG play an important role in the impact on climate change. Rapid and dynamic changes of earth climate have affected human life physically, psychologically, and emotionally. General circulation models (GCMs) are the numerical models developed using a set of linear and nonlinear partial differential equations. Climate and weather forecasting reliability can be possible with the use of GCMs. Statistical downscaling is popular among the research community on an account of building a strong and accurate relationship between GCM and local level information. Among the wide range of regression models, artificial neural network (ANN), a multilinear regression method, is the most popular approach, which works on the basis of transfer function statistically relating predictors and predictands. ANN is developed particularly to address present requirements in global environmental change research and necessitate for more detailed temporal and spatial information from GCM. For the present work, CanCM4 at grid size 2.8° × 2.8° developed by Canadian Centre for Climate Modelling and Analysis has been chosen based on the skill score. It gives historical data from 1971 to 2005 as well as the data consisted of future simulations by emission scenarios RCP4.5 from 2006 to 2035. The data were extracted to cover the entire Bhima basin with nine grid points. Observed meteorological data provided by Indian Meteorological Department (IMD) are used for calibration and validation of ANN model. Karl Pearson’s coefficient is taken as guideline to test the sensitiveness of predictors. To evaluate the performance of downscaling model, Nash–Sutcliffe coefficient and root-mean-square error performance indices have been adopted. With the good predictor and predictand correlation, the work moves on to investigate the application of the transfer function in the form of statistical downscaling. The downscaled results can be applied to watershed management such as flood and drought management studies of the Bhima basin.