One front door to NOAA PSL's monthly WRIT maps and time series, the NCEP/NCAR daily and 6-hourly composites, and the 20th Century Reanalysis daily and 6-hourly plotters. Start from your question; the switchboard picks the tool, checks what's possible, and builds the request.
Pick the question closest to yours. Everything below it adapts: which datasets are offered, which dates are allowed, and which NOAA tool receives the request.
NOAA's Physical Sciences Laboratory runs a family of free plotting tools that sit on top of decades of reanalysis and observational archives. They are powerful and overlapping, and each one grew its own form. The switchboard doesn't replace them. It asks what you are trying to find out, picks the PSL tool that can answer it, refuses combinations the archives can't support, and hands you a finished request. The plot is drawn by NOAA, exactly as if you had filled in their form yourself.
If you only read one thing: start on the Build tab with the question closest to yours, watch the request slip on the right (below the form on a phone), and fix anything marked ✕ before pressing Open plot at NOAA.
Every PSL tool here is a web form that sends its settings to a script on psl.noaa.gov. The switchboard reproduces those forms' exact field names and values, which were read from PSL's own pages and then tested against the live servers. When you press Open plot at NOAA, your browser opens a new tab on psl.noaa.gov with every setting filled in, and NOAA draws the map, cross-section or time series, along with its download links (PNG, PostScript, and usually NetCDF or CSV/TXT).
The plot opens in a new tab rather than inside this page for a concrete reason: psl.noaa.gov tells browsers not to display its pages inside other sites and doesn't let other sites read its responses. Nothing on this page can get around that without a server in between, and this page deliberately has none. The upside is that what you see is always NOAA's own output, with NOAA's own labels and citations.
What the switchboard adds is judgement before the request is sent. It knows each dataset's period of record (probed from PSL on ), which variables each dataset carries in WRIT, which daily sources exist for which eras, which levels each daily tool accepts, the 20-year and 50-year limits of the WRIT forms, and the 00/06/12/18 UTC rule of the 6-hourly tools. When you break one of those rules, it tells you which one and how to fix it.
Use this for a spatial pattern averaged over whole months: July 2012's 500 hPa height anomaly, the 1991–2020 JJA precipitation climatology, the average of several winters. It goes to WRIT's monthly map tool, which holds modern reanalyses (ERA5, JRA-3Q, MERRA-2, CFSR, R1, R2, CORe), the century-scale reanalyses (20CRv3, ERA-20C, CERA-20C) and many observational analyses. You can pick one year, list up to 20 years for a composite, or give a run of up to 50 years to average. Pressure-level variables also offer height–latitude and height–longitude cross-sections.
Two ways to ask “do these agree?”. A difference map shows where dataset A minus dataset B is large for the same months. Area-averaged series show when the two disagree for a region, with the same box, months and base period applied to both. Only variables both datasets carry are offered.
This is the only workflow that resolves individual days. Enter specific dates (up to 20, averaged into a composite) or a continuous period inside one calendar year, choose daily means or 6-hourly snapshots, and the switchboard picks the source: NCEP/NCAR R1 for 1948 to 17 March 2026, the 20th Century Reanalysis V3 for 1806 to 2015. Where both cover your dates you can override the choice, and the slip explains the trade-off.
One number per month or per season, averaged over a box, from any WRIT dataset. You choose the base period for anomalies here, which the map tools don't let you do. Besides a line plot you can ask for the distribution of values, a month-by-year heatmap, a wavelet spectrum or the autocorrelation.
Two routes. Correlate pairs your regional series with a pre-built climate index (NAO, PNA, Niño 3.4, PDO, AMM and others) and reports the correlation, as a lagged cross-correlation or a scatterplot. Composite lets PSL pick the years when an index was extreme, by count, percentile, raw value or standardized value, and maps the average of those years.
Three questions: does the question hinge on particular days, how many datasets do you need, and should the answer be a map, a line, or a link to a climate index. The recommendation explains itself, and one tap opens that workflow.
A reanalysis blends observations with a fixed weather model to produce complete, gridded fields. They differ in which observations they ingest, the model and its resolution, and how far back they go. The right one depends on the era and on what you need to trust.
| Dataset | Record in these tools | What it assimilates | Reach for it when |
|---|---|---|---|
| ERA5 | 1940 – present, monthly | Surface, radiosonde, aircraft and satellite observations through ECMWF's 4D-Var | You want the best current estimate for the modern era. Here it is monthly only. |
| JRA-3Q | 1948 – present, monthly | Conventional and satellite data, JMA system | Checking an ERA5 result against an independent modern reanalysis. |
| JRA-55 | 1958 – Jan 2024 | JMA's previous system | Consistency with older published work; otherwise prefer JRA-3Q. |
| MERRA-2 | 1980 – present | NASA GMAO, satellite era | A third opinion in the satellite era. |
| NCEP/CFSR | 1979 – early 2026 | NCEP coupled atmosphere–ocean | Comparisons with NCEP products. |
| CORe | 1950 – present | Conventional (non-satellite-radiance) observations only | Trends and long comparisons where a steady observing system matters more than peak accuracy. |
| NCEP/NCAR R1 | 1948 – 17 Mar 2026 | 1995-era model on a 2.5° grid | Daily and 6-hourly maps since 1948, which none of the monthly-only datasets offer here. Production has ended. |
| NCEP/DOE R2 | 1979 – early 2026 | R1 with fixes | Rarely the first choice; useful for continuity. |
| 20CRv3 | 1806 – 2015 (monthly, daily, 6-hourly) | Surface pressure only, over prescribed sea-surface temperature and sea ice; an 80-member ensemble | Anything before 1940, and like-for-like comparisons across two centuries. Treat 1806–1835 as experimental. |
| ERA-20C, CERA-20C | 1900 – 2010, monthly | Surface pressure and marine winds (CERA-20C is coupled to an ocean) | A second opinion on early-20th-century patterns. |
| Observed analyses | Varies | Station, ship, gauge or satellite data gridded without a weather model | Checking reanalysis temperature, precipitation or SST against observation-based products. |
The dataset menus show each record's first and last year and grey out datasets that lack the variable you've chosen. If you pick a variable first and then change dataset, the switchboard keeps your variable if it can and otherwise moves to the first one the new dataset carries.
A reanalysis runs one model throughout, but the observations feeding it change enormously, and those changes can show up as steps or trends that the atmosphere never made. Keep these boundaries in mind whenever a time series crosses them.
| Era | What changes | What to watch for |
|---|---|---|
| Before ~1900 | Scattered land barometers and ship logs | Only surface-pressure reanalyses exist. Large-scale pressure patterns are meaningful; precipitation, humidity and upper-air details are model-shaped. |
| 1900 – 1940s | Denser station networks, more shipping | Wartime gaps in ship data can affect oceans. Still surface-only upstream of 1940. |
| 1940 – 1957 | Upper-air observations begin in earnest | ERA5's early decade and R1's first years rest on a thin radiosonde network, especially away from North America and Europe. |
| 1958 | International Geophysical Year; radiosonde network expands | Many analyses treat 1958 onward as the reliable upper-air period. JRA-55 starts here. |
| 1979 | Global satellite soundings arrive | The most common source of spurious jumps. Southern Hemisphere and oceanic fields improve abruptly. Many reanalyses start here. |
| ~1998 – 2006 | New satellite sounders, then GPS radio occultation | Smaller shifts, mostly in the stratosphere and in humidity. |
The surface-pressure-only reanalyses and CORe exist partly to avoid these jumps. When a trend matters, confirm it in one of them and in a full-input reanalysis before believing it.
WRIT is monthly: every WRIT map and series is built from monthly means, so the shortest thing it can show is one month. For individual days you need the event workflow, which uses PSL's daily and 6-hourly composite tools. Those exist here for R1 and 20CRv3 only; ERA5 and JRA-3Q aren't offered at daily resolution by these PSL tools.
Daily means are UTC days, midnight to midnight Greenwich time. That is not a local calendar day and not a 12Z-to-12Z observer day. For an eastern-US event that ends in the evening, part of it lands on the next UTC date; for comparisons with cooperative-observer or 12Z-aligned gauge totals, list both UTC dates or use 6-hourly times.
The 6-hourly tools accept only 00, 06, 12 and 18 UTC. 20CRv3 is produced every 3 hours, but PSL's plotting tool exposes the 6-hourly synoptic times, so a request for 03 UTC is refused rather than silently rounded.
A mean is the plain average for your dates. A climatology is the long-term average for those calendar months or days, ignoring the years you entered. An anomaly is mean minus climatology. The base period that defines “climatology” matters: a 1981–2010 base makes recent years look warmer than a 1991–2020 base does.
In the monthly map tool and the daily tools, PSL fixes the base period; it's not a form field, and the switchboard says so in the slip under “Decided by NOAA, not by you”. The result page states it; note it before comparing with other work. The time-series workflows let you set the base period yourself and apply it identically to both datasets when comparing.
Standardized anomalies divide each anomaly by that calendar month's standard deviation over the base period, so months with different variability become comparable.
A composite averages several dates or years to bring out what they have in common. It's a strong tool with three habitual traps. First, small samples: eight El Niño winters give a noisy average, and none of these PSL tools test significance, so a colourful pattern can be sampling noise. Second, selection: if the years were chosen because of the outcome you're mapping (the wettest summers, say), the composite will of course show wetness; it's the other fields that carry information. Third, averaging different kinds of events blurs them, and a composite pattern may never have occurred on any single date.
A continuous period in the event workflow is a composite too: every day or 6-hourly time from start to end, equally weighted.
Difference maps subtract dataset B from A after PSL places both on a common grid. Differences in raw means of surface fields, especially 2 m temperature, sea-level pressure and precipitation, often reflect how each model defines its surface and orography rather than any disagreement about the weather, so the slip suggests anomaly differences for those. For time series, the switchboard sends the same box, months, land/sea mask and base period for both datasets, so the only thing that differs is the dataset.
Only compare over years both datasets cover. The slip checks this and names the dataset that runs short.
A box the size of Pennsylvania contains only about one by three grid cells in R1 (2.5°) and a few dozen in 20CRv3 (1°). The slip warns when a box holds 16 or fewer cells of a dataset whose grid is known. Such a series describes the regional-scale atmosphere over and around the box, not the state's own climate, and it won't reproduce station or PRISM statistics.
The land-only and ocean-only options use PSL's land–sea mask for each dataset. For a mostly-land box, “land only” keeps coastal ocean cells from diluting the signal.
The correlation is Pearson's r over the years your series and the index have in common. PSL doesn't detrend and doesn't adjust significance for autocorrelation, so two quantities that both trend will correlate even with no physical link. For a first look that's fine; for a claim, download both series (the CSV link on NOAA's page) and test properly.
Index composites let PSL select years for you. “The N most extreme years” takes the top or bottom N; “percentile” takes every year beyond that percentile; “raw value” and “standardized value” use a threshold in the index's units or in standard deviations. The index is averaged over the same months you map. PSL lists the selected years on the result page, and that list is the actual sample, so keep it with the figure.
Sent to names the PSL tool. Why this tool gives the reason in one or two sentences; for events it also says whether the source was chosen automatically. Checks lists problems first: ✕ must be fixed before the button unlocks, ! is a caution you can proceed through, ✓ confirms something was verified. Decided by NOAA, not by you lists every methodological choice the PSL tool makes that you can't set, so that nothing is silently assumed. Request URL is the exact request, and it's the most reproducible record of what you did.
NOAA's page shows the image and, below or beside it, download links. WRIT pages usually offer the plotted data as NetCDF and time-series pages offer CSV and TXT. These files are temporary on PSL's server; download what you need promptly. The page also ends with PSL's requested citation line for publications.
If NOAA's page reports an error despite a clean slip, the most common reasons are a variable that exists for a dataset but not at every level or date, or a server-side change since the switchboard's field list was captured. The Original NOAA form link at the bottom of the slip opens the native form so you can compare.
Three ways to keep or share a configuration. The request URL reproduces the plot at NOAA with no dependence on this page. The share link reopens the switchboard with every setting restored; when this page is viewed inside another app's frame, the share link may point at the frame rather than the page, in which case use the recipe. The recipe is a small JSON record with the title, the request URL, every setting, and every caution and NOAA-made choice at the time; paste it into the box on the History tab to reload it. Opened plots are also kept in the History tab on this device.
Each button loads the settings into the Build tab. Look at the slip before opening the plot; each one was chosen to show a particular check.
Hurricane Agnes, June 21–23 1972. R1 daily precipitable-water anomaly over the Mid-Atlantic and Northeast. Shows automatic source choice and the precipitation-field caution you'd get if you switched variable.
The Great Blizzard, 12 March 1888, 12 UTC. 20CRv3 6-hourly sea-level pressure. Shows the pre-1900 caution and the synoptic-hour rule.
St. Patrick's Day flood, March 16–19 1936. 20CRv3 daily 500 hPa height anomaly as a continuous period.
Pennsylvania summer rainfall, ERA5 vs JRA-3Q, 1958–2025. JJA precipitation anomaly averaged over land in the Pennsylvania box with a 5-point running mean. Shows the small-region warning.
El Niño winters. ERA5 DJF 500 hPa height anomaly for the 8 highest Niño 3.4 winters, 1950–2024. Shows the year-labelling note for seasons across the new year.
NAO and Pennsylvania winter temperature. Scatterplot of the DJF 2 m temperature anomaly against the CRU NAO index.
How far does 20CRv3 sit from ERA5? DJF sea-level pressure, 20CRv3 minus ERA5, 1950–1979. Shows the raw-mean-difference caution.
Dates outside a dataset's record: NOAA would error, or in some tools quietly substitute a nearby date. Variables a dataset doesn't carry. More than 20 listed years or dates, or more than 50 years in a range, because the PSL forms cap them. 6-hourly times other than 00/06/12/18 UTC. A continuous period that crosses a calendar year, which the daily tools can't express. Event dates after 17 March 2026, since R1 ended and no other daily source is wired in. Event dates spanning both before 1948 and after 2015, since no single source covers them. Climatology in the 6-hourly tools, which don't offer it. Each refusal names the rule and how to get around it.
The button stays grey. Look for ✕ in the slip; each one says what to change. A dataset is greyed out. It doesn't carry the variable you chose; pick the variable first. NOAA shows an error page. Open the original NOAA form from the slip and try the same settings there; if the form fails too, the dataset may be temporarily unavailable at PSL. The plot looks blocky. R1 is on a 2.5° grid and 20CRv3 on 1°; that's the data, not the drawing. Try “Grid cells” under More controls to show it honestly. History is empty on another device. History lives in this browser only.
Reanalysis: a model-based reconstruction of past weather that assimilates observations. WRIT: PSL's Web-based Reanalyses Intercomparison Tools. 20CR: the 20th Century Reanalysis; V3 is the current version. R1: the NCEP/NCAR Reanalysis 1. Composite: an average over chosen dates or years. Anomaly: departure from a climatological mean. Lag: a shift applied to every date, to look before or after an event. Synoptic hours: 00, 06, 12 and 18 UTC. UTC day: 00Z to 00Z. Land–sea mask: the grid of land and ocean cells a dataset uses.
These PSL tools cover jobs the switchboard doesn't build requests for yet. Each link opens NOAA's own form.
Plots you've opened, most recent first. Stored only in this browser.
Paste a recipe (the JSON from “Copy recipe”) or a share link.