@article{4764,
  author = {Nguyen Minh Tuan},
  title = {Multiscale Time-Series Analysis of Temperature Variability in Ho Chi Minh City: Spectral, Seasonal, and Wavelet-Based Characterization},
  journal = {Journal of E-Technology},
  year = {2026},
  volume = {17},
  number = {3},
  doi = {https://doi.org/10.6025/jet/2026/17/3/103-121},
  url = {https://www.dline.info/jet/fulltext/v17n3/jetv17n3_1.pdf},
  abstract = {This study investigates the multiscale temporal dynamics of near surface temperature variability in Ho Chi
Minh City, Vietnam, utilizing a comprehensive three-year hourly meteorological dataset comprising 28,508
observations. A unified analytical framework integrating time domain, frequency domain, and time
frequency techniques was applied to robustly isolate periodicities and assess structural stability.
Methodologies included Fast Fourier Transform (FFT), Fisher's g-test, Monte Carlo surrogate modeling,
stationarity diagnostics (ADF and KPSS tests), autocorrelation analysis, classical seasonal decomposition,
and Continuous Wavelet Transform (CWT) coupled with multi resolution decomposition.
Results reveal a highly significant, deterministic 24-hour diurnal cycle governing temperature, relative
humidity, and precipitation, primarily driven by intense solar forcing and regional land sea breeze regimes.
Conversely, mean sea level pressure is dominated by low frequency seasonal fluctuations of approximately
1.08 years. Stationarity tests confirm the dataset's inherent non stationarity, while autocorrelation diagnostics
reveal strong temporal persistence and long memory. Furthermore, wavelet scalograms and multi-resolution
decomposition illustrate that while long term climatic forcing accounts for the overwhelming majority of
statistical variance, short term weather anomalies introduce critical high frequency fluctuations.
These findings underscore the limitations of standalone linear models for environmental forecasting. The
identified hierarchical temporal structure strongly supports the development of hybrid predictive
architectures. Future forecasting systems should integrate wavelet based multi resolution feature extraction,
routing long term trends through deep learning frameworks while optimizing short term residuals via machine
learning ensembles to significantly enhance urban climate and energy demand predictions.},
}