Data Cataloging based SINDy Approach for Weather Forecasting

In the present era, climatic change is a crucial phenomenon that poses a great threat to human and animal life. Accurately simulating and forecasting the temperature of a particular geographical location have been a great conundrum throughout history particularly in tropical regions due to the sophisticated and multi-faceted oceanic transactions. Nonetheless, a smattering number of machine learning techniques such as random forests, deep learning algorithms, neural networks, support vector machines, stacked autoencoders have been exploited to forecast the weather and other volatile field variables with encouraging prospects. These algorithms require dense training data and also consumes large CPU time to perform future weather predictions. This article proposes a data cataloging based Sparse regression of Nonlinear Dynamics (SINDy) method for weather prediction. It differs from the existing prediction techniques by formulating an elevated time series data from the actual measured dataset. By adopting this empirical prediction method, the governing relationship between the input and output data is attained in simple mathematical form rather than complex functions. The MATLAB simulations of temperature and relative humidity confirm the superiority of the proposed approach in aspects of performance criteria (Mean square error).