Skip to content

Observing the climate

How a global mean temperature is arrived at, what reanalyses add, and why observation is a work of reconstruction.

Updated on 19 August 2026 · Climate observation, Modelling and projections

A "global mean temperature" exists nowhere as a direct measurement. It is a reconstructed quantity, and understanding how it is reconstructed disposes of most of the objections raised against it.

The problem with stations

The observing network has never been regular. Stations cluster in inhabited, wealthy regions; they have moved, changed instruments, and watched their surroundings become urban. Crudely averaging their readings would produce a figure governed by that geography rather than by the climate.

The solution is a shift of object: we do not average temperatures, we average anomalies — each station's departure from its own normal. Anomalies behave coherently across wide regions where absolute temperatures do not, which makes it possible to interpolate between stations without inventing anything.

Global mean temperature anomalyDeparture from the 1951-1980 reference period, combining land and ocean surfaces.

NASA Goddard Institute for Space Studies — GISS Surface Temperature Analysis (GISTEMP v4) · accessed 15 November 2026

Reanalyses: filling the gaps with physics

Stations do not cover the oceans, the deserts or the poles evenly. A reanalysis answers this by running a weather model over the past, constraining it continuously with every available observation — stations, buoys, ships, radiosondes, satellites.

The result is neither a pure observation nor a free-running simulation: it is the state of the atmosphere most consistent with all the measurements and with physical law. These products have become the reference for studying extremes, because they provide complete, regular fields Muñoz‐Sabater et al., 2021.

Measuring directly what can be measured

Not every quantity requires that reconstruction. Atmospheric CO₂ concentration is measured directly, at sites chosen far from local sources, and its record is among the most reliable in climate science. That is what makes it the anchor point of the radiative balance and of the carbon budget Friedlingstein et al., 2023.

Observation and modelling are not opposed

Observations serve to build models, to evaluate them, and to constrain their parameters — foremost among them climate sensitivity. Intercomparison protocols systematically confront simulations with observations over the historical period before granting any credit to their projections Eyring et al., 2016.

Sources

  1. GISS Surface Temperature Analysis (GISTEMP v4) — NASA Goddard Institute for Space Studies
  2. Climate indicators — Temperature — Copernicus Climate Change Service
  3. Trends in Atmospheric Carbon Dioxide — NOAA Global Monitoring Laboratory

Studies cited in this brief

Pierre Friedlingstein et al. · Earth system science data · 2023

  • Carbon cycle and greenhouse gases
  • Climate observation