[A-AS04] Machine Learning Techniques in Weather, Climate, Hydrology and Disease Predictions

[AAS04-10] Forecasting Standardized Precipitation Evapotranspiration Index (SPEI) Using Multilayer Perceptron Artificial Neural Network Model for Central India

*Sourabh Shrivastava^1,2, R. Uday Kiran¹, P K Bal³, K K Singh⁴, Tomohito Yamada² (1.University of Aizu, 2.Hokkaido University , 3.Qatar Meteorology Department, 4.India Meteorological Department)

Keywords:Artificial Neural Network, SPEI, Drought, ROC Curve

This study presents an Artificial Neural Network approach to estimate the drought events in an Indian state of Madhya Pradesh. Sea Surface Temperatures data obtained from Centennial Observation-Based Estimates (COBE) and seasonal rainfall data (June-August) obtained from India Meteorological Department for the period 1971 to 2013 are used in this study to develop two different ANN models, aiming to forecast the standardized precipitation evapotranspiration index at four different stations in Madhya Pradesh. Preliminary results from the first ANN model (derived from SST only) showed that correlation between observed SPEI and SST ranged between 0.585 to 0.773 where as in case of the second ANN model (derived from both rainfall and SST), the correlation remains between 0.960 to 0.979. Further, coefficient of determination remains in the range 0.312 to 0.591 and 0.919 to 0.958 for the 1st ANN and second ANN model respectively. Furthermore, probabilistic forecasts of SPEI are made and the relative operating characteristics scores are examined and the results indicate that the forecast results from the SPEI values are suitable for various meteorological application and ANNs offer a framework for forecasting the SPEI drought index in the absence of observed rainfall data.