Integrating Climate Change Variables in Relative Humidity Prediction with Multivariate ARIMA and RNN Models

Abstract

Abstract The study of relative humidity has gained significant attention in recent years due to its influence on climate change along with the global warming phenomenon. The precise prediction of these phenomena is crucial in various fields, encompassing meteorology, climate simulation, industrial production, agriculture, public health, and epidemiology. Nevertheless, the task of predicting relative humidity remains a persistent issue owing to its vulnerability to various climate-related influences. The current research employs two techniques, specifically Multivariate ARIMA and RNN models, in order to predict the monthly relative humidity in Chapainawabganj, Bangladesh. This study examines the interconnections of climate change, global warming, and Relative Humidity by incorporating many associated factors such as precipitation, wind speed, palmer drought severity index, and temperature. We employ data from the period spanning 1981 to 2011 for the purpose of training our model, whereas data from the years 2012 to 2021 is used exclusively for testing. In the study area, the proposed model had the lowest root mean squared error (5.10–5.65), the lowest mean absolute error (3.76-4.22), the highest correlation coefficients (0.95-0.96), the highest coefficients of determination (0.90-0.92), and the highest Willmott's index of agreement (0.98) for predicting relative humidity. The study concludes that the Multivariate RNN model (a non-linear model) exhibits superior performance in predicting relative humidity compared to the Multivariate ARIMA model (a statistical model). Our model could help to predict relative humidity across the world incorporating the effects of climate change.