Weather Forecasting using Deep Learning Algorithm for the Ethiopian Context

No Thumbnail Available

Date

2018-10

Journal Title

Journal ISSN

Volume Title

Publisher

Addis Ababa University

Abstract

Weather forecasting is the application of science and technology to predict the state of the atmosphere for a future time and a given location. Now days, forecasting of accurate atmospheric conditions is the major challenge for the meteorologist and poor forecasting has significant impact on our daily lives. This brings the necessity to make research works on forecasting of the weather events with respect to Ethiopia. A number of algorithms have been proposed for forecasting of atmospheric condition such as support vector machine, neural network, numerical, and statistical models. However, in this research the design and implementation of weather prediction for the Ethiopian context based on the forecasting ranges using deep neural network, support vector machine for regression, and numerical based regression is presented. Four and half years’ time series daily and hourly, temperature, precipitation, humidity, visibility, dew point, air pressure, and wind historical recorded data is used from National Oceanic and Atmospheric Administrator (NOAA) to implement the system. Since making discussion on all-weather variables makes the report to long, forecasting of temperature and precipitation weather variables for Addis Ababa are only considered to be discussed and evaluated as a sample. And their results are examine based on percentage of Root Mean Square Error and time consumption. The same data records are applying for all algorithms; and the experimental result shows that, in forecasting of a big data, DBN provides a better performance relative to SVM and numerical regressions. In short range experiment we have achieved a forecasting accuracy of temperature 88.6%, 79.6%, and 52.5% using DBN, SVM, and Numerical algorithms respectively. However, if we apply a small dimension dataset as input values SVM and numerical regressions completely outperforms the DBN due to shallow training.

Description

Keywords

Weather, Deep Belief Network, Support Vector Machine, Restricted Boltzman Machine, Meteorology

Citation