Key points are not available for this paper at this time.
Rainfall is a major component of the hydrologic cycle and thus requires a comprehensive understanding of its dynamics and variability. This study aims to develop and test the applicability of recurrent models for forecasting rainfall in extremely arid regions on a monthly time scale. Specifically, Neural Auto-regressive Networks (NARs) and Auto-regressive Integrated Moving Average (ARIMA) were utilized for modeling a rainfall dataset from Kuwait City from 1958 to 2018. The study site possesses extreme arid conditions with long-term average annual rainfall of less than 120 mm. This harsh condition imposes challenges on modeling efforts. The results of the modeling showed that the NAR model was more efficient in modeling rainfall dynamics over the study period. A notable bias was encountered in models within abnormal wet seasons. The results of the modeling efforts presented in this study were found to be reasonable, and they qualify the NAR model for making objective rainfall forecasts in the study area and other similar climatic zones. The overall Nash–Sutcliffe (NS) coefficient was found to be 0.206 for the NAR model, with the model showing an even better performance within the medium-to-low rainfall intensity months (<30 mm per month). With the outcome of this study, an operational framework for water managers is presented for arid and hyper-arid zones to aid in developing resilient and efficient water management plans to cope with the adverse impacts of climate change.
Abdullah A. Alsumaiei (Fri,) studied this question.