Prediction of Relative Humidity in Fez Using Recurrent Neural Networks: Comparison of the NAR and NARX Models

Abstract

Predicting relative humidity is essential for environmental management, particularly in meteorology, water resource management and agriculture. Implementing artificial neural networks (ANNs) allows for improved accuracy in relative humidity forecasting. This facilitates more accurate forecasting of extreme weather events, such as droughts, water shortages, forest fires or heavy precipitation. ANNs represents a robust approach to climate data analysis, due to their ability to model complex, non-linear relationships between meteorological parameters. This contributes to ensure the sustainability of ecosystems, improved decision-making and more efficient management of natural resources. The objective of this study is to implement a neural network uses a set of recorded meteorological parameters to forecast relative humidity in Fez, based on several observed climate parameters. To this end, two recurrent neural networks were examined: non-linear autoregressive with exogenous inputs neural network (NARX) and non-linear autoregressive neural network (NAR) methods. The results obtained from both algorithms were used to choose the best model. Moreover, multiple training algorithms were applied in propounding the best model. A comparison of the NAR and NARX models confirmed that the superiority of the letter. With the Levenberg Marquardt algorithm, this model necessitates a reduced number of iterations and exhibits the lowest mean square error (0.0344) and the highest correlation coefficient (0.8932).