Comparison of SARIMA and SARIMAX for Long-Term Drought Prediction

Long-term drought prediction is of considerable importance in water resources management. Time series models are appropriate tools to predict climatic events. In this study, the Reconnaissance Drought Index (RDI), which is based on precipitation and potential evapotranspiration, was applied to calculate droughts of Yazd synoptic station at time scales of one, three, and six months from 1961 to 2018. The period 2006-2018 was selected as the forecast period. Drought data of the forecast period were not considered to the applied models. Results showed that the pattern of drought data at 1, 3 and 6-month time scales in the Yazd synoptic station is seasonal. The SARIMA is a univariate time series model created by adding a seasonal component to the ARIMA model. The SARIMAX multivariate model is created by adding parameter covariate variable (exogenous variable) to SARIMA. In the present study, the efficiency of the seasonal univariate model (SARIMA) and seasonal multivariate model (SARIMAX) in predicting drought in arid regions were compared. To implement the SARIMAX model to predict droughts, precipitation and potential evapotranspiration were provided to the model as covariates, separately. The results of the coefficient of determination (R2) between observed RDI and predicted RDI values by SARIMA showed that the model offers higher performance on 3 and 6-month time scales with 0.66 and 0.71, respectively. The results of the SARIMAX model showed that in one-month time scale, the SARIMAX model based on potential evapotranspiration with 0.60, and for 3-month time scale, the SARIMAX model based on precipitation with 0.79 performed better. In 6-month time scale, the performance of the model for the both covariates were almost the same with the coefficient of determination of 0.79. Comparison of the results of the two models showed that the performance of the SARIMAX model is higher than the SARIMA model for drought predicting in arid regions.