Rainfall Forecasting Using Artificial Neural Networks, Chaos Theory and Statistical Down Scaling of the Data Set
Abstract
Rain is a deciding factor for the income of various sectors that depend solely upon the rain and so it critically affects the economy of the state which is why it is necessary to predict the rainfall accurately to take necessary precautions in case of flood and drought. Rainfall is a naturally occurring phenomenon which cannot be predicted just on the basis of previous data as the parameters causing the rainfall could be minutely different which can completely change the results generated by the neural network, and so it is very important to take in consideration the real-time data and cross analyse it with the output generated by the neural network to accurately predict the rainfall. Chaotic patterns are found in rainfall and a small change in the data can enormously affect the output. In the current study, a model has been developed which uses the daily data of southwest Asia, of months from January to December, available from ECMWF (European Centre for Medium-Range Weather Forecasts) of years 1900 - 2010. To avoid the problem of overfitting in the current study, 10-11 variables have been selected to construct an Artificial Neural Network. The model development consists of following processes 1) As the data is of a larger region, Relevance Vector Model (RVM) has been used for the classification and regression using the probabilistic Bayesian learning framework to downscale the data to a smaller region of Gujarat state of India 2) Multi-layer feed forward Neural Network trained with first order back propagation algorithm with log sigmoid transfer function in the hidden neuron and output neuron. During training a learning coefficient of 0.08 and momentum coefficient of 0.02 has been considered during the training process along with cross-validation approach in order to avoid over-training of ANN. which gave almost accurate results than the other neural networks 3) To overcome the problem of complex behavior of rainfall patterns, chaos theory along with the results of the artificial neural network has been adopted. The goal of this paper is to accurately predict the rainfall and take requisite precautions if necessary in case of flood and drought and to explain the impacts of rainfall on the economy of the state.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2018
- Bibcode:
- 2018AGUFM.H41L2261S
- Keywords:
-
- 1821 Floods;
- HYDROLOGYDE: 1869 Stochastic hydrology;
- HYDROLOGYDE: 4313 Extreme events;
- NATURAL HAZARDSDE: 4315 Monitoring;
- forecasting;
- prediction;
- NATURAL HAZARDS