|
INTERNATIONAL JOURNAL OF INNOVATIONS IN APPLIED SCIENCES & ENGINEERINGInternational Peer Reviewed (Refereed), Open Access Research Journal (By Aryavart International University, India) E-ISSN:2454-9258 | P-ISSN:2454-809X | Estd Year: 2015 Impact Factor(2021): 5.246 | Impact Factor(2022): 5.605 |
Dr. Rajendra Singh
Vol. 8, Issue 1, Jan-Dec 2022
Page Number: 118 - 126
Abstract:
Deep learning technology, specifically DLWP-CS, has been proposed for weather prediction using cubed spheres in data-driven simulations of global weather fields. For basic fields like temperature and geopotential height, DLWP-CS performs admirably, but for complex, non-linear fields like precipitation, it is computationally demanding. Precipitation precursors are the input for the modified DLWP-CS (MDLWP-CS) technique, which changes the architecture from temporal to spatio-temporal mapping. The technique predicts precipitation using a 2-m surface air temperature as a proof of concept. In comparison to the GFS output with a one-day lag, the hourly ERA-5 reanalysis used to train the MDLWP-CS model outperforms both linear regression and the Global Forecast System (GFS) in daily precipitation prediction with a one-day lag. This provides an effective DT framework for quick, high-fidelity precipitation predictions.