International Journal of Mathematical, Engineering and Management Sciences Vol. 1, No. 3, 93–106, 2016 https://dx.doi.org/10.33889/IJMEMS.2016.1.3-011 93 Artificial Neural Network (ANN) Based Empirical Interpolation of Precipitation Rajesh Joshi G. B. Pant National Institute of Himalayan Environment and Sustainable Development Kosi-Katarmal, Almora-263643, Uttarakhand, India E-mail: dr.rajeshjoshi@gmail.com (Received June 17, 2016; Accepted August 18, 2016) Abstract Various climate impact studies need to generate estimates of climate variables at a given location based on values from other locations. It is well established fact that there are strong sensible physical linkages between global climate and local scale weather phenomenon. Therefore, empirical interpolation or downscaling has emerged as a prospective tool to relate atmospheric circulation patterns to surface variables for forecasting regional climate from GCM and RCM output dataset. In this paper, application of Artificial Neural Networks (ANNs) based soft computing model for empirical interpolation of precipitation in Himalayan region is attempted. This method uses ANNs to generate precipitation estimates for 11 districts of Uttarakhand state (India) given information from a lattice of surrounding locations. In the present paper, we have used Feed Forward Back Propagation (FFBP) algorithm to develop a Multilayer Perceptron ANN model for empirical downscaling of precipitation in Himalayan region. The model is developed using climate data of Climate Research Unit (CRU) and observed data for past 110 years (1901-2010). The robustness and suitability of the developed ANN model is verified by testing its applicability for 11 districts of Uttarakhand state. 80% of the data are used for training of the model and remain 20% are used for testing of the model. The performance evaluation of the model is tested by RMSE value. The study show that the model works quite well for climatic records of most of the district after bias correction. Keywords - Empirical downscaling, artificial neural networks (ANNs), feed forward back propagation (FBBP) algorithm, precipitation, climate change, Himalaya. 1. Introduction In recent years, considerable efforts have been devoted to investigate the effects of large scale climate change on rainfall variability in different parts of the world. Statistical and multiple nonlinear regression methods have been used for predicting rainfall on regional scale. To assess the impact of climate change, scientists depend heavily on global circulation models (GCMs); spatial resolution of these models remains quite coarse and varies from 2.5 0 x2.5 0 up to 8 0 x 10 0 which is too coarse to assess the impacts of climate change on various ecosystems components at local and regional scale (Giorgi and Mearns, 1991; Clark, 1985). With such a coarse resolution, the regional and local details of the climate, influenced by spatial heterogeneities in the regional physiography particularly in the Himalaya, are lost. Therefore, there is need to convert the GCM outputs into a reliable data set with higher spatial resolution, with daily rainfall and temperature time series at scale of a watershed or region to which impacts of climate are to be investigated. Since, many impacts models require information at scales of 25 km or even less; therefore, to generate regional information on climate, interpolation of GCMs output to finer resolution is required. Even though GCMs can be run at high resolutions, still results from such models need to be downscaled for individual sites or locations for impact studies which enables construction of climate change scenarios for regional level at daily/ monthly time-scales.