Cloud-Integrated Meteorological Parameter Prediction by Leveraging Multivariate Statistical Time Series and GANs

Abstract

Meteorological parameters are increasingly influenced by unsustainable development and environmental pollution, contributing to the rising frequency of natural calamities. Accurate prediction of these parameters is crucial for prevention. This study uses historical data to predict six key meteorological parameters, such as dew point, cloud cover, precipitation, temperature, wind speed, and wind direction of Bengaluru and Delhi city of India. Using cloud-based services ensures scalability and computational efficiency for data storage, processing, and model deployment in weather parameters. To address data gaps, generative adversarial networks (GANs) are employed for data imputation within those parameters. Hence, the present work predicts six important meteorological parameters using the vector autoregression model (VAR), vector moving average model (VMA), vector autoregression moving average model (VARMA), and cointegrated vector autoregression model (CVAR). The comparative analysis demonstrates the superior performance of the CVAR model over other models, as measured by the normalized mean square error (nMSE) and normalized root mean square error (nRMSE) for a single parameter (2.12, 0.3) and (1.21, 0.24) for both the cities, in forecasting all the specified meteorological parameters.

Keywords

• CVAR model
• data imputation
• multivariate time series forecasting models
• VAR model
• VARMA model
• VMA model