Rainfall Prediction in Lakshadweep Using Deep Learning-Based Time Series Model

Abstract

Prediction of weather patterns are important factors in many areas including agriculture, disaster management, coastal planning, and in fisheries. It is very important to forecast and provide usable information about these events, as many of the coastal areas of island regions such as Lakshadweep, typically have a lack of available weather monitoring instruments.The vast amount of ever-changing, highly non-linear parameters and wide geographic range of these islands make traditional statistical forecasting methods unreliable because they cannot effectively capture these parameters at a time scale that matches the rapid changes in atmospheric conditions. Therefore, during this research study an alternative approach using deep learning was pursued as a means to accurately forecast future weather conditions for smaller isolated island.Recurrent neural network(RNN) model was developed to accurately learn the long-term temporal dependencies between the meteorological variables to enable a trained model to provide usable forecast predictions. Performance of the developed model was examined with standard metrics Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and multiple accuracy metrics. Model attained a classification accuracy of 76.64%, having low RMSE of 0.045 and MAE of 0.031. Experimental findings indicated that the model used in this research is able to discover the seasonal changes over time in the Lakshadweep, weather variables better than any of the traditional weather forecasting techniques. The final conclusion of this research is that deep learning models can effectively be applied to make reliable weather predictions on a smaller scale, such as in small island regions.