The workflow follows the conceptual model outlined in synthetic_SST_zijie.pdf.
Key steps:
- Derive a seasonal climatology using harmonic regression and remove a linear trend to obtain anomalies.
- Estimate the seasonal cycle of variance via log-harmonic regression, yielding daily scale factors.
- Standardise anomalies by those scale factors and fit an ARMA(1,1) process that captures high-frequency
variability.
- Calibrate a slow AR(1) component against the observed 10–60 day autocorrelation, then mix the two
processes.
- Simulate the mixed process for the requested number of years, rescale by the seasonal variance, and add
back the harmonic seasonal mean.
- Diagnostics compare observed and synthetic anomalies, daily variance, and the lag-0–60 autocorrelation.
The resulting series preserves the seasonal mean, variance structure, and autocorrelation characteristics of
the input record while extending it synthetically.