Time Series and Machine Learning Methods for Estimating Missing Meteorological Data: The Western Black Sea Basin Case

The development of artificial intelligence applications is rapidly advancing today. It enables making existing data meaningful through a wide range of applications. This increases the importance of data for all artificial intelligence subfields. The accuracy, continuity and meaningfulness of data are very important for training and testing the models. Discontinuities or errors are likely to occur in data obtained from long-term physical measurements. It is inevitable for meteorological data, which are highly influenced by external factors to have gaps. Mentioned situation causes negative effects on the analysis reliability of meteorological data, which plays a significant role in climate change and hydrological modelling. Within the scope of this study, monthly precipitation measurements of meteorological observation stations in the city centers of seven different provinces in the Western Black Sea basin were examined. The gaps in the monthly rainfall data measured between 2000 and 2023 were estimated using time series, statistical and machine learning approaches. In the modelling process, ARIMA, SARIMA, ARIMAX, XGBOOST, and mean imputation methods were employed. The analyses revealed that SARIMA models, which consider seasonal effects, provided more consistent results, as demonstrated by performance metrics. The completed data form the basis for advanced drought analysis. Thus, impact of deviations due to data loss in future drought analyses is minimized.