Devendra KumarRawat, Devendra Singh2019-08-072019-08-072019-07http://krishikosh.egranth.ac.in/handle/1/5810119851Artificial Neural Networks model using different inputs were developed for monthly rainfall modelling of Almora. The 70% of monthly rainfall data from 1964 to 2018 was used for calibration and 30% for validation. Different topologies of models were constructed with change in number of hidden layers, processing elements and activation functions. The numbers of hidden layers varied from 1 to 3 and numbers of neurons from 1 to 10. Log-Sigmoid Axon & Tanh Axon transfer functions with back propagation algorithm and Levenberg-Marquardt learning rule were used. The performance of the models was evaluated qualitatively using rainfall time series graphs and scatter plots and quantitatively by employing Correlation coefficient, Root mean square error, Coefficient of efficiency, Integral square error and Pbias indices. The models having higher value of Coefficient of efficiency, Correlation coefficient and lower values of Root mean square error, Integral square error and Pbias were considered to be the best fit model. Based on the selected criteria, the performance of ANN model with 6-8-8-1 architecture with Log-SigmoidAxon transfer function with back propagation and Levenberg-Marquardt learning rule having past 1, 2, 3, 4, 5 and 6 months rainfall as inputs was found better than the other models.ennullThesis