Weather forecasting using Deep Learning

Abstract

Global human life is significantly impacted by weather forecasts. It takes a lot of computing resources to solve mathematical equations that are based on climatic circumstances. As a result, during the past ten years, deep learning algorithms have been integrated with enormous amounts of weather observation data. At the moment, a lot of data is being consumed. So, we can increase the accuracy of weather forecasts by combining this enormous amount of data with deep learning methods. In this paper, we implement benchmark datasets for autoencoder and linear regression. We are using z500 dataset, temp 2m dataset and t850 data set. As training the linear regression on the full data will take a lot of memory which is why we took every 5th time step that almost give the same result. Using a linear regression approach and an auto encoder model, we trained and obtained day-level predictions using the ERA5 reanalysis dataset (Hersbach et al., 2020) to determine the accuracy of the test data and training data.