# Forecasting Climatic Variables Using ARIMA and VAR with Box Cox Pre-Processing

Abstract

Climate variables like temperature, rainfall, and humidity display rich temporal behavior that demands precise handling using appropriate statistical methodology to ensure credible forecasting results. This paper undertakes a comparative study to assess univariate and multivariate time series modeling approaches to these important climate variables, employing Box-Cox transformation as a preprocessing step to enhance homogeneity. This is followed by checking the variables' stationarity and appropriateness after applying the validation test, forming ARIMA on univariate variables and a 9th-order VAR on multiple variables. The model choice is informed by information criterion functions, while forecasting comparison is assessed via error calculation and graphic representation of forecast trajectories. The empirical findings, supported by natural climate variables observed in northern Iraq, authenticate that Box-Cox pre-processing improves homogeneity to better estimate model parameters with better precision within both sets of models. The ARIMA models on each separate variable display outstanding forecasting capability, while within the multiple variables' model VAR, interactions exist but fail to outperform separate specifications uniformly. The results signify that measures of predominance within multiple variables' forecasting are crucial, being sensitive to the degree of correlations between variables. The paper demonstrates pre-processing and comparison aspects within forecasting to emphasize credible results within this work and provides a factual illustration to determine whether to prefer univariate rather than multiple variables' forecasting on natural variables. © 2026 NSP Natural Sciences Publishing Cor.