Observed PNA did not remove the 2020-2025 mean PDO mismatch in a short-memory model

We tested whether observed Pacific–North American (PNA) atmospheric values remove a sustained Pacific Decadal Oscillation (PDO) mismatch. Models fitted before 2020 used month intercepts, one-month PDO memory and lagged Nino 3.4. Adding current PNA gave a 2020-2025 simulated mean of -0.251, compared with the observed -1.995. Its central 95% model range was -0.896 to 0.374. None of 3,999 coefficient and residual block draws was as low as the observation. Seasonal slopes, a later training start, trends and longer residual blocks did not remove the mismatch. Yet adding PNA reduced monthly reconstruction error by 11.66% in an earlier 1980-2019 test, with a conditional 95% interval of 2.40% to 20.96%. Earlier six-year checks often missed observed means. Thus added monthly information did not validate ranges for sustained departures. This is a test of one atmospheric proxy in a limited model, not a causal test or a forecast. A selected interval, revised indices, incomplete slow memory and conditional uncertainty limit inference. The package gives code, source hashes, a retrieval script, model evidence, a figure and separate numerical checks.

PDO · PNA · ENSO · North Pacific · atmospheric circulation · conditional reconstruction · sustained anomaly · model check