Assessment of the reliability of a statistical model for developing a scenario of possible air temperature changes in the spring season over the territory of Azerbaijan

Abstract

Problem Statement. This study validates statistical estimates of Azerbaijan’s mean spring temperatures up to 2022. Although precise long-term forecasts remain difficult, analysis of meteorological data from 1968–2022 demonstrates that statistical methods effectively capture regional spring temperature trends and provide a reliable basis for future climate research and scenario development.

Purpose. The study aims to develop a statistical spectral model that reliably identifies climate-induced changes in meteorological elements, particularly air temperature. The model employs optimized analytical methods whose adequacy is verified through comparisons of observed and calculated temperature series using smoothing techniques, Schuster tests, and correlation analyses. Establishing this model and its optimal application enhances the development of regional climate models and scenarios, improving the reliability of projected climatic trends.

Methods. The methodology is based on analyzing long-term air temperature series, classified into periodic and non-periodic variability indicators, the latter being useful for studying future temperature changes. Given the non-stationary nature of meteorological variables, trends in long-term series are treated as reliable, but series must first be transformed into stationary sequences for analysis using statistical techniques. Extrapolating observed trends beyond a few years is physically unreliable due to natural variability; thus, only anthropogenic trends linked to CO₂ increases are suitable for climate scenario development. Additionally, the length of meteorological records is critical, as classical climatology assumes stationarity and estimation accuracy improves with longer observational series.

Results. Spectral analysis revealed dominant 3-, 5-, and 7–9-year temperature cycles across most of Azerbaijan, while Nakhchivan exhibited only 3- and 9-year cycles. The model showed strong agreement between observed and simulated spring temperatures, with R² values of 0.906 (1998–2000), 0.953 (2001–2005), and 0.933 (2006–2010). Error assessment showed that 36–56% of cases had ΔT ≤ 1°C, 76–88% had ΔT ≤ 2°C, and nearly all cases had ΔT ≤ 3–4°C. For ΔT ≤ 2°C, the forecast accuracy reached 76–94%, confirming the reliability of the proposed spring temperature scenario.

Conclusions. Using a statistical spectral model, a new scenario of spring temperature change for Azerbaijan was developed. Projections recalculated from linear trend equations for 1968–2010 demonstrate higher accuracy compared with earlier approaches, and the post-2023 estimates can be treated as updated climate scenarios, improving the reliability of future assessments. A five-year (2026–2030) averaged spring temperature map for the country’s physiographic zones was also compiled. Overall, the model offers a robust and scientifically grounded basis for future spring temperature projections and broader climate-related analyses.