↖ CPA Weather Lab

Teleconnections as one system, and what they do to Pennsylvania

Twenty-eight climate indices, computed from the raw NOAA, NCEI, PSL, BOM and CPC records, correlated against each other month by month and lag by lag, then against Pennsylvania temperature and precipitation from the daily scale to the season. Every number on this page was calculated from the source files, 1950–2025, deseasonalized and linearly detrended.

The web of links

Each dot is an index, grouped by the part of the climate system it lives in. Lines connect pairs whose correlation clears the threshold; thicker is stronger, red is in-phase, blue is opposite. Pick a calendar month to see how the wiring changes through the year, or a lead to see which index moves first. Tap an index to isolate its links.

Tap an index

The panel lists its strongest partners at the current month and lead, and its Pennsylvania signature.

Timescales: why some links are same-day and others take a decade

An index can only carry a signal for as long as it remembers its own state. The bar spans from how quickly the index changes to how long it stays correlated with itself (e-folding time). Atmospheric patterns forget in about a week; the ocean holds on for years. That memory is the upper limit on useful forecast lead for Pennsylvania.

Causal pathways that reach Pennsylvania

The correlations are symptoms. These are the physical chains most of the literature agrees on, with the numbers this dataset gives for each link.

Pair explorer

Choose two indices. The bars show how their same-month correlation changes by calendar month (strongest, median and weakest marked). The curve shows correlation when the first index leads (right) or lags (left). The window numbers show how stable the relationship has been across sliding 30-year periods.

Full correlation matrix

Same-month correlation for every pair. Tap any cell to load that pair in the explorer above.

Pennsylvania impacts

Correlation between each index and the monthly temperature or precipitation anomaly (percent of normal) in the region chosen, for target months in the chosen season, with the index leading by the chosen number of months. Tap a bar to see that index's full lead profile and its per-standard-deviation effect.

Tap a bar

The lead profile shows when the signal begins, peaks and fades, per season.

Precipitation is the hard part: no index explains more than about 5% of Pennsylvania's monthly precipitation variance (|r| ≤ 0.22). Temperature is better but still modest, with winter NAO/AO and autumn EP/NP near 15–30% of variance. Treat these as tilts of the odds, not forecasts.

Days to weeks: daily indices and the MJO at Harrisburg

Daily CPC AO, NAO and PNA against Harrisburg International (USW00014711) daily mean temperature anomaly, winter or summer. Right: average Harrisburg winter temperature anomaly following each active MJO phase (RMM amplitude ≥ 1), at leads up to 25 days.

Cells: °F anomaly. Read across a row to see a phase's influence arrive and fade; read down a column to see the phase cycle.

Index reference, most to least widely known

What each index measures, how it is calculated, where the record begins, which time resolutions are published (P) or can be built (C), how it relates to the others, and where to download it.

Less common patterns (described, not computed here)

Compute the Pennsylvania effects yourself

The same pipeline that produced this page, as one Python script. Save it, then run the block below from that folder.

pip install pandas numpy scipy requests && python3 pa_teleconnection_toolkit.py

The steps it performs

  1. Downloads monthly indices (CPC teleconnections, AO, RONI, PDO, AMO, QBO, TNI, MEI.v2, DMI) and NCEI nClimDiv Pennsylvania statewide plus the ten climate divisions.
  2. Removes each calendar month's mean and a linear trend from every series, so warming and seasonality do not masquerade as teleconnection signal. Precipitation becomes percent of the monthly normal.
  3. Standardizes each index by calendar month, so slopes read as "°F or % per one standard deviation of the index".
  4. Correlates index at month t − L with Pennsylvania at month t, for L = 0…6, grouped by the target season, with effective sample size reduced for index autocorrelation.
  5. Compares the index level against its month-to-month change (value versus rate).
  6. Repeats at daily resolution for AO, NAO and PNA against Harrisburg, and composites Harrisburg temperature by MJO phase at 0–30 day leads.

Turning correlations into a Pennsylvania outlook

For a season ahead, add up index anomalies times their per-SD slopes only for indices that are both significant and weakly correlated with each other (for winter temperature: NAO or AO, not both; WP; PDO or NP, not both; EP/NP in autumn). Because the atmospheric patterns forget within about a week, their monthly slopes are only useful once the month's pattern is known or forecast; the oceanic ones (PDO, TNI, AMO) carry across months. Validate by leaving out one winter at a time.

Method and caveats

← CPA Weather Lab · Learn