Climate Data Science: How Scientists Know What They Know About Climate Change

Climate science rests on an extraordinary infrastructure of measurement, analysis, and modelling — a global observing system monitoring Earth's climate in near-real-time from the ocean floor to the stratosphere, combined with climate models of increasing complexity that simulate past, present, and future climate states. The temperature records documenting 1.2°C of warming since pre-industrial times are not a single dataset but a synthesis of hundreds of millions of measurements from thousands of weather stations, ocean buoys, and ships, subjected to rigorous quality control and bias correction by multiple independent research teams — who consistently reach the same conclusion despite using different methods, different datasets, and different analytical approaches. The convergence of evidence from this independent analysis is one of the strongest foundations of scientific confidence in modern climate science.

Ice Cores — Windows into 800,000 Years of Climate

Ice cores drilled from Antarctic and Greenland ice sheets provide the most detailed record of past climate available — extending back 800,000 years in the deepest Antarctic cores. As snow falls and compresses to ice, it traps tiny bubbles of air that preserve the atmospheric composition of that time — allowing scientists to measure past CO₂ and methane concentrations directly. The isotopic composition of the ice itself records past temperature. Dust and aerosol particles record past volcanic eruptions and wind patterns. The result is an 800,000-year record showing eight cycles of ice ages and interglacials — and confirming unambiguously that current CO₂ concentrations (422 ppm) are dramatically higher than any level in the ice core record (maximum pre-industrial: approximately 300 ppm). The correlation between CO₂ and temperature across 800,000 years of ice core data is one of the most powerful demonstrations of the greenhouse effect in Earth's history.

Climate Models — The Science's Crystal Balls

Climate models — mathematical representations of the Earth's climate system encoded as computer programs — are the primary tools for understanding how the climate responds to forcing and for projecting future states. Modern Earth System Models divide the atmosphere, ocean, land surface, and ice into three-dimensional grids and simulate physical, chemical, and biological processes at timesteps of minutes to hours. Their outputs are coordinated through the Coupled Model Intercomparison Project (CMIP6), whose most recent phase provides the scientific foundation for the IPCC Sixth Assessment Report. Despite their complexity, these models must be validated against the observed climate record — and their success in reproducing observed temperature trends, seasonal cycles, and the climate responses to volcanic eruptions and solar variability provides confidence that their projections of future climate change under different emissions scenarios are physically meaningful, even as specific regional projections carry significant uncertainty.

Satellite Earth Observation — The Climate's Eyes in Space

The revolution in Earth observation from space has provided climate science with measurement capabilities that were unimaginable a generation ago. Over 100 Earth observation satellites are currently operational, collecting data on surface temperature, sea surface height, ice extent, forest cover, aerosol loading, atmospheric composition, ocean colour, soil moisture, and dozens of other climate-relevant variables at global coverage and increasing spatial and temporal resolution. The GRACE satellite mission — measuring minute changes in Earth's gravitational field caused by shifts in the distribution of water mass — has produced the most comprehensive record of changes in land ice mass, groundwater storage, and ocean mass ever compiled, revealing patterns of ice loss from Greenland and Antarctica that are directly relevant to sea level projections. The OCO-2 and OCO-3 satellites measure atmospheric CO₂ at column-average resolution fine enough to identify individual power plant plumes and city-scale emissions — transforming the ability to independently verify national emissions inventories and detect emissions from unexpected sources.

Reanalysis Products — Reconstructing the Past Atmosphere

Climate reanalysis — the process of using numerical weather prediction models, constrained by all available historical observations, to reconstruct consistent global datasets of atmospheric conditions extending back decades or centuries — has become one of the most important tools in climate science. ERA5 (the fifth generation ECMWF Reanalysis), produced by the European Centre for Medium-Range Weather Forecasts, provides hourly estimates of global atmospheric, land, and ocean variables from 1940 to the present at 31-kilometre horizontal resolution — making it possible to analyse the temperature, wind, precipitation, and humidity at any point on Earth for any time in the past 80 years with a consistent methodology. NOAA's 20th Century Reanalysis extends this record back to 1836 using only surface pressure observations, providing a longer but lower-resolution reconstruction of large-scale atmospheric dynamics. These reanalysis products are fundamental infrastructure for climate attribution studies, extreme weather analysis, climate model validation, and the assessment of trends in climate variables over decadal timescales.

The detection and attribution of climate trends requires careful statistical analysis of observational datasets that contain multiple sources of uncertainty: instrument calibration errors, changes in observation methods over time, inhomogeneous station networks that have expanded and contracted, and natural variability that can obscure long-term trends in short records. The convergence of independent temperature records produced by NASA GISS, NOAA NCEI, the UK Hadley Centre, Berkeley Earth, and the Japanese Meteorological Agency — using different raw datasets, different quality control methods, and different gridding algorithms — on the same warming trend of approximately 1.2°C since the pre-industrial period is one of the strongest indicators of the robustness of the observed warming signal. The probability that five independent research teams using different methods would all arrive at the same erroneous conclusion by coincidence is essentially zero.