Evaluation of uncertainty in stochastic modeling methods of precipitation data

The analysis of precipitation data holds significant importance in various scientific and engineering fields. These data, which exhibit irregular and variable characteristics, possess a stochastic nature, necessitating the development of appropriate models and the evaluation of their accuracy. Such models are widely used in both the planning of hydrological processes and future forecasting. In this study, the applicability of the Transinformation method in time series modeling was investigated in comparison with traditional model selection criteria, namely the Akaike Information Criterion (AIC) and the Bayesian Information Criterion (BIC). The analysis was conducted using both synthetic data and 51 years of monthly precipitation data from the Aegean Region, as well as the corresponding SPI (Standardized Precipitation Index) series derived from these observations. The study consists of three stages: first, model selection methods were compared using synthetic time series; second, monthly precipitation series from nine meteorological stations in the region were analyzed; and finally, SPI-1, SPI-3, SPI-6, SPI-9, and SPI-12 indices were calculated, and synthetic SPI data for the next 10 years were generated through modeling based on SPI-12 series. According to the results, the Transinformation method demonstrated the highest and most consistent accuracy rates across both short and long datasets, at varying levels of autocorrelation and model orders. While the BIC method performed well especially with increasing data length and autocorrelation strength, the AIC method showed relatively improved performance only in longer and strongly autocorrelated series. These findings reveal that the Transinformation method is an effective and reliable model selection criterion in time series modeling.