ERA5 2 m 대기 온도 데이터를 활용한 확률적 모델링 및 분석

Climate data, characterized by seasonal cycle and variability, is often classified as cyclostationary time series. However, analyzing such data poses challenges due to frequency redundancy, where overlapping cycles obscure distinct periodicities. This study presents a methodology to analyze cyclostationary time series while mitigating frequency redundancy. Utilizing ERA5 reanalysis data of 2 m air temperature, we conducted a statistical analysis of seasonal stability a(t), weather-related noise magnitude N(t), and long-term forcing f(τ), and developed a corresponding statistical model. The analysis of a(t) elucidates global sensitivity to seasonal climate variations, with late-summer polar instability driven by sea ice albedo feedback and Arctic amplification. This instability accumulates, resulting in a “memory effect”, where a(t) exhibits maximum variance during transitions to stability. Key climate phenomena such as ENSO, Atlantic Niño, and the Indian Ocean Dipole were also identified. N(t), representing weather-related noise, peaks in winter due to pronounced temperature gradients and reveals storm tracks near East Asia. The long-term forcing f(τ) captures gradual changes, such as oceanic variations and global warming, facilitating the analysis of El Niño and La Niña events. The developed stochastic model accurately reflects the statistical properties of climate data and demonstrates strong performance, particularly in the unstable Antarctic region, even when excluding long-term forcing.