Why Environmental Datasets Sometimes Disagree: A Case Study of Global Temperature Records

Current figure

The most recent complete-year global average surface temperature anomalies from two major datasets differ by approximately 0.03°C. For 2023, NASA’s GISTEMP v4 reported an anomaly of 0.89°C above the 1981–2010 baseline, while the UK Met Office’s HadCRUT5 reported 0.86°C above the same baseline. This small discrepancy is typical of the differences observed between leading global temperature records.

Measurement date

The figures refer to the calendar year 2023, covering January through December. Global temperature datasets are updated monthly as new station and satellite data become available. The annual values are typically finalised by mid-January of the following year, with minor revisions possible as additional data are incorporated.

Previous figure

For 2022, NASA GISTEMP reported an anomaly of 0.79°C, while HadCRUT5 reported 0.76°C — again a difference of 0.03°C. The year-on-year change from 2022 to 2023 was +0.10°C in both datasets, reflecting the transition from a La Niña to an El Niño phase. The absolute difference between the two datasets remained stable at 0.03°C.

Long-term trend

Since the late 19th century, all major global temperature datasets show a clear warming trend of approximately 0.08°C per decade, accelerating to about 0.18°C per decade since 1981. The small differences between datasets have been persistent but have not grown over time. The table below illustrates the anomalies and differences for selected years.

All values are anomalies relative to the 1981–2010 baseline. The long-term agreement confirms that the warming signal is robust and independent of methodological choices.

Data source

The two datasets compared here are:

• NASA GISTEMP v4 – produced by NASA’s Goddard Institute for Space Studies (GISS), using surface air temperature data from weather stations and sea surface temperature data from the Extended Reconstructed Sea Surface Temperature (ERSST) dataset.
• HadCRUT5 – produced by the UK Met Office Hadley Centre in collaboration with the Climatic Research Unit at the University of East Anglia, combining land station data (CRUTEM5) and sea surface temperature data (HadSST4).

Other widely used datasets include NOAA GlobalTemp, Berkeley Earth, and the JMA’s global temperature record. All are publicly available and updated regularly.

Methodology

Global temperature datasets are constructed by combining millions of individual temperature measurements from land stations, ships, buoys, and satellites. The key steps include:

• Data collection and quality control: Raw observations are screened for errors, inhomogeneities, and biases (e.g., station moves, instrument changes).
• Interpolation: Because weather stations are unevenly distributed, datasets use statistical methods to estimate temperatures in unobserved regions. NASA GISTEMP uses a reference station method that extrapolates anomalies up to 1,200 km, while HadCRUT5 uses a kriging-based approach that provides uncertainty estimates for each grid cell.
• Baseline selection: Anomalies are calculated relative to a fixed reference period (e.g., 1981–2010) to remove the influence of absolute temperature differences between locations.
• Combining land and ocean data: Land and ocean records are merged, often with different treatments of sea ice regions and coastal zones.

These methodological choices, while scientifically justified, lead to the small differences observed between datasets.

Why annual values fluctuate

Year-to-year variations in global temperature are dominated by natural climate phenomena, particularly the El Niño–Southern Oscillation (ENSO). El Niño events tend to warm the planet, while La Niña events cool it. Other factors include volcanic eruptions (which inject cooling aerosols into the stratosphere) and solar variability. The differences between datasets can also fluctuate slightly from year to year due to updates in input data, station coverage, and methodological refinements. For example, the transition from HadCRUT4 to HadCRUT5 in 2020 reduced the long-standing cool bias in the Arctic, narrowing the gap with NASA GISTEMP.

Regional variation

Disagreements between datasets are often larger at regional scales, especially in data-sparse areas such as the Arctic, Antarctica, and parts of Africa. The table below shows the 2023 annual anomalies for selected regions from NASA GISTEMP and HadCRUT5, highlighting where differences are most pronounced.

Larger discrepancies in polar regions reflect sparse observational networks and different interpolation techniques. In well-observed areas like North America and Europe, differences are typically less than 0.05°C.

Meaning and limitations

The small differences between global temperature datasets do not indicate uncertainty about the reality of long-term warming. Instead, they reflect the inherent challenges of constructing a global average from incomplete data. Each dataset is a best estimate, and the spread between them provides a lower bound on structural uncertainty. Users should consider the following limitations:

• All datasets are subject to revisions as new data become available and methods improve.
• Regional and short-term trends are less certain than global and long-term trends.
• Differences in baseline periods can cause confusion; always compare anomalies using the same reference period.
• Uncertainty ranges (typically ±0.05°C for annual global means) should be taken into account when interpreting small differences.

Despite these caveats, the close agreement among independent datasets strengthens confidence in the observed warming trend.

Next expected update

Both NASA GISTEMP and HadCRUT5 are updated monthly. The next annual figure for 2024 is expected to be released in mid-January 2025. Preliminary data for 2024 indicate it may be the warmest year on record, with anomalies likely exceeding 1.0°C above the 1981–2010 baseline in both datasets.

Downloadable chart or table

The table below provides annual global mean temperature anomalies from 1979 to 2023 for both datasets, along with the difference. The full underlying data can be downloaded from the respective sources: NASA GISTEMP and HadCRUT5.