Climate Science Reanalysis

Real monthly climate reanalysis data from the NCEP/NCAR Reanalysis 1 project (Kalnay et al., 1996), downloaded directly from NOAA’s Physical Sciences Laboratory: surface air temperature and sea level pressure, area-weighted over the Niño 3.4 region (5°S-5°N, 170°W-120°W, the standard ENSO monitoring box), January 1948 through February 2026 (938 months), ingested through linaldb’s own NetCDF/CF-convention connector, with zero netCDF4, xarray, or h5py in the query path.

Native CF-convention ingestion

USE DATASET FROM "nino34_air_slp.nc" AS nino34
SELECT * FROM nino34 LIMIT 5

The connector applies each variable’s real scale_factor/add_offset packing and _FillValue/missing_value masking automatically, and surfaces units/ standard_name/long_name as inspectable field metadata, the same real semantics a xarray.open_dataset() call would apply, done natively in Rust.

A real bug was found and fixed building this exact dataset: NCEP/NCAR’s own files declare scale_factor/add_offset/missing_value as 1-element arrays rather than true 0-d scalars, which the connector was silently ignoring, disabling CF unpacking and fill-value masking against real-world reanalysis files with no warning at all. Fixed upstream and re-verified against this exact data before this page was written.

A real warming trend, via LSTSQ

LET fit = LSTSQ design_matrix air_nino34   -- design_matrix columns: [decimal_year, 1]

Recovers a real trend of +0.093 °C/decade for surface air temperature over the tropical Pacific Niño 3.4 region, cross-checked against numpy.linalg.lstsq on the same 938-month series. Smaller than global land-average warming trends, honestly, since this is an open-ocean-adjacent tropical region, not a land station.

The real annual cycle, via FFT

LET spectrum = FFT air_anomaly
LET mag = MAGNITUDE spectrum

The magnitude spectrum’s peak lands at bin 78 of a 938-point real FFT: a 12.03-month period, correctly recovering the real seasonal cycle from real reanalysis data. Cross-checked against numpy.fft.rfft on the same series.

The real ENSO signature, via COVARIANCE

LET cov = COVARIANCE air_nino34 WITH slp_nino34

-0.159, negative, which is real, meaningful climate physics: this is the Walker circulation / ENSO teleconnection, where a warmer tropical Pacific correlates with lower sea level pressure in the same region. Cross-checked against numpy.cov.

Try it yourself

• View the full notebook
• Try it in the Playground