Analyzing and Forecasting the Weather Conditions in Jordan using Machine Learning Techniques

Abstract

Weather forecasting is an important research field due to its impact on a wide variety of life aspects. The traditional way of weather forecasting is based on complex physical models that describe the hydrodynamic behavior of the atmosphere. This way is costly, time consuming, often inaccurate and requires supercomputers to make predictions. In this paper, we investigated the performance of machine learning algorithms in predicting the weather conditions in Jordan for a short period. We start by analyzing the used dataset of the weather conditions of the 12 Jordanian governorates over past 13 years, where the long-term trend shows 0.3−°C rise in the average temperature and 10-mm decrease in the average annual precipitation. We propose a prediction model based on encoder-decoder architecture and bidirectional long short-term memory cells (ED-BiLSTM). We carefully tune and train this model and show the importance of integrating the data of nearby locations to the target location's data to improve the model accuracy. Also, we show that the model accuracy improves significantly when adding training instances of other locations. The proposed tuned model trained on the train data of 16 locations and accepting regional weather conditions at the input has very low mean squared error of 1.78×10−6 in predicting Amman's weather for the next 24 hours.