Earth’s Skin Temperature Reveals Climate Change Signals Across Six Satellite Datasets

Deep beneath every weather forecast and climate model lies a deceptively simple measurement: the temperature of the planet’s outermost skin. Scientists call it Tskin, the radiometric temperature of the very top layer of land, ocean, and ice, and it responds instantly to sunlight, clouds, wind, and shifting land cover. While near-surface air temperature has long been the workhorse of climate monitoring, it is only an indirect proxy for the energy exchanged between the surface and the atmosphere. Skin temperature, by contrast, is a direct readout of that exchange, which is why it has been designated an Essential Climate Variable. Yet despite its physical importance, Tskin has remained surprisingly underexploited in climate research, largely because the satellite records that measure it come from different instruments with different strengths, weaknesses, and blind spots.

That gap is what motivated a team led by Sarah Safieddine of LATMOS/IPSL in Paris to attempt something no one had done before: a global, day-and-night intercomparison of six skin temperature datasets spanning 2008 to 2022. The study, published in the journal Earth Observation, brings together the Moderate Resolution Imaging Spectroradiometer (MODIS) on NASA’s Terra satellite, the European Space Agency’s Land Surface Temperature Climate Change Initiative product (LST CCI v3.00), the ESA CCI and Copernicus Climate Change Service sea surface and sea ice temperature analysis, the ERA5 reanalysis produced by the European Centre for Medium-Range Weather Forecasts, and two distinct products derived from the Infrared Atmospheric Sounding Interferometer (IASI) flying on the Metop satellites: the EUMETSAT all-sky Climate Data Record and a newly developed clear-sky neural-network retrieval. The verdict after fifteen years of data is strikingly coherent: the fingerprint of climate change is unmistakable across nearly all of them, with a handful of fascinating regional exceptions.

The technical challenge of comparing these products is considerable, because each measures something subtly different. Thermal infrared instruments such as MODIS and IASI retrieve skin temperature from radiances in the atmospheric window between 8 and 13 micrometers, where most natural surfaces emit close to a perfect blackbody. By inverting Planck’s law, accounting for surface emissivity, the instruments convert measured radiance into temperature. But infrared retrievals only work under cloud-free skies, and atmospheric water vapour always leaves its mark. Microwave radiometry can peer through clouds, but its surface emissivity is far more variable and its spatial resolution much coarser. Reanalysis products like ERA5 offer complete, all-sky fields, yet their skin temperature is a model variable constrained by data assimilation rather than a direct observation, making it sensitive to how the model parameterises soil, vegetation, and snow. No single dataset can claim to be the universal truth, which is precisely why systematic intercomparison matters.

The team harmonised everything onto a common one-degree grid, restricted the analysis to comparable local solar times between roughly 09:30 and 10:30 in the morning and 21:30 and 22:30 in the evening, and applied dataset-appropriate cloud filtering. Over land, the results were largely reassuring. Daytime global means agreed within about 2 kelvin across the datasets, and the deseasonalised anomalies, which strip away the annual cycle to reveal underlying variability, tracked each other closely. The major modes of interannual variability, including the strong 2015 to 2016 El Niño warming and the subsequent 2020 to 2022 La Niña cooling, appeared consistently in every record, a sign that the fundamental climate signal is robust regardless of which instrument is doing the measuring.

Two systematic quirks did emerge, however, and both carry lessons for anyone using these products. The ESA LST CCI dataset ran consistently warmer than the others, but the team traced this to viewing geometry rather than calibration error: LST CCI restricts retrievals to view zenith angles within 22 degrees of nadir, while MODIS averages across its full 60-degree swath, where longer atmospheric paths and angular emissivity effects systematically depress retrieved temperatures. More concerning were step-like discontinuities in the LST CCI anomalies at sensor transitions, when the record switched from AATSR to MODIS Terra in 2012 and to Sentinel-3B’s SLSTR-B in 2018, indicating residual inhomogeneities that users of this version must treat with caution. At night, MODIS revealed a second surprise: a pervasive cold bias relative to every other product, larger than can be explained by its slightly later overpass time alone. Previous validation studies have found average nighttime biases of around minus 1.6 kelvin against ground stations, with cloud-screening issues and surface heterogeneity in snowy, icy, and arid regions likely contributing.

Over the ocean, the picture was cleaner still. Because the sea surface is far more homogeneous than land, inter-dataset biases between IASI, ERA5, and the ESA CCI/C3S analysis generally stayed below 1 kelvin, with root-mean-square errors near 1 kelvin over the open ocean. The largest discrepancies clustered at high latitudes, where cloud detection, sea-ice contamination, and weak thermal contrast degrade infrared retrievals, and where the very definition of skin temperature diverges across products. One subtle caveat the authors highlight concerns clear-sky sampling itself: during the 2020 to 2022 La Niña, increased tropical cloudiness reduced the availability of clear-sky IASI retrievals, pulling the IASI global mean cooler in a way that reflects sampling changes rather than genuine surface cooling. It is a reminder that even carefully constructed satellite climate records can be modulated by where and when clouds happen to allow a glimpse of the surface.

The heart of the study lies in its trend analysis. Using the robust Theil-Sen estimator on yearly deseasonalised averages, with significance assessed by the Mann-Kendall test, the team mapped warming and cooling across the globe for morning and evening overpasses separately. The dominant signal is unambiguous: skin temperatures have risen since 2008, more strongly over land than over the ocean, and most intensely at high northern latitudes. Alaska and Siberia light up in every dataset, a spatial fingerprint of Arctic amplification, the well-documented phenomenon by which the Arctic warms nearly four times faster than the global average. Daytime land trends ranged from +0.04 to +0.12 kelvin per year depending on the dataset, while nighttime land trends spanned +0.05 to +0.11 kelvin per year, with the all-sky ESA CCI/C3S product showing a modest oceanic warming of about +0.03 kelvin per year consistent with the clear-sky records.

But the most eye-catching findings are the exceptions, places where the planet’s skin is cooling even as the globe warms. India shows significant daytime cooling across all six datasets, a signal the authors link to rising aerosol emissions, which scatter sunlight and offset part of the greenhouse gas warming, compounded by cloud effects that vary seasonally and by time of day, and by land-use change and greening. Parts of central and eastern Africa cool as well, plausibly because increased cloud cover and latent heat release from more intense rainfall suppress surface temperatures. And in the southeastern Pacific off South America, a coherent zone of cooling sits squarely over the Humboldt Current upwelling system, where strengthening alongshore winds drive Ekman divergence and draw cold subsurface water to the surface, a wind-driven mechanism that can produce local cooling even amid global ocean warming, further modulated by Pacific decadal variability on these short timescales.

The study also showcases the growing sophistication of retrieval science. The new IASI neural-network product uses 87 carefully selected window-region channels, discards those sensitive to carbon dioxide to improve long-term stability, and incorporates monthly, time-varying emissivities from the CAMEL database, an important refinement given that neglecting emissivity trends can overestimate global skin temperature trends. The EUMETSAT Climate Data Record, retrained on more than 120 million IASI observations, produces smoother time series that pass significance tests more often, while the neural-network product offers greater independence from ERA5. With the IASI constellation guaranteed through at least 2030 and the IASI-New Generation series extending coverage toward 2045, Earth’s skin temperature is poised to become a cornerstone of climate monitoring, offering a direct, diurnally resolved view of how our planet’s surface responds to a changing atmosphere, one morning and one evening at a time.

Subject of Research: Multi-satellite intercomparison of land, sea, and ice skin temperature trends from 2008 to 2022

Article Title: Assessing Earth's skin temperature trends: consistent signals from IASI, MODIS, ESA CCI and ERA5

Article References: Safieddine, S., Sinnathamby, S., Hadji-Lazaro, J., Doutriaux-Boucher, M., Ghent, D., Whitburn, S., Clarisse, L., & Clerbaux, C. (2026). Assessing Earth's skin temperature trends: consistent signals from IASI, MODIS, ESA CCI and ERA5. Earth Observation, 1(1), 59-75. https://doi.org/10.5194/eo-1-59-2026

Image Credits: AI Generated

DOI: 10.5194/eo-1-59-2026

Keywords: skin temperature, land surface temperature, sea surface temperature, IASI, MODIS, ESA CCI, ERA5, climate change, Arctic amplification, aerosol cooling, upwelling, satellite remote sensing