the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 11 Feb 2026
Assessing Earth’s Skin Temperature Trends: Consistent Signals from IASI, MODIS, CCI and ERA5
Abstract. Earth’s skin temperature (Tskin), i.e. land and sea surface temperature (LST and SST), directly reflects surface–atmosphere energy exchanges and is an Essential Climate Variable (ECV). Yet it remains less exploited than near-surface air temperature in climate monitoring. Here, we intercompare and assess the capability of several infrared sounders and Tskin products to monitor climate variability during morning and evening overpasses from a multi-sensor perspective over 2008–2022. Two Infrared Atmospheric Sounding Interferometer (IASI) satellite products are analysed: the EUMETSAT all-sky Climate Data Record (IASI-CDR) (all-sky) and a newly developed clear-sky neural-network product (IASI-NN). These IASI products are compared with the Moderate Resolution Imaging Spectroradiometer MODIS Terra Land Surface Temperature (LST) (v6.1), ESA LST CCI (v3.00), and ERA5 skin temperature.
Over land, daytime global means agree within ~2 K across datasets, but LST CCI is consistently higher, and deseasonalised anomalies are highly consistent, except for LST CCI, which exhibits sensor-transition discontinuities. At night, MODIS shows a prevalent cold bias relative to all other products. Over the ocean, inter-dataset biases between IASI and ERA5 generally remain below 1 K. Trend analyses reveal robust warming in Tskin since 2008 across the different datasets, while significant regional cooling is observed over India (daytime) and parts of central/eastern Africa, and in in the southeastern Pacific associated with the Humboldt upwelling system.
- Preprint
(2566 KB) - Metadata XML
-
Supplement
(418 KB) - BibTeX
- EndNote
Status: open (until 28 Mar 2026)
- RC1: 'Comment on egusphere-2026-400', Anonymous Referee #1, 26 Feb 2026 reply
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 94 | 51 | 11 | 156 | 19 | 22 | 13 |
- HTML: 94
- PDF: 51
- XML: 11
- Total: 156
- Supplement: 19
- BibTeX: 22
- EndNote: 13
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
This paper entitled "Assessing Earth's skin temperature trends: consistent signals from IASI, MODIS, CCI and ERA5" intends first to evaluate several datasets of skin temperature (over ocean and continents) and then to compare their respective trends.
Major comments:
- The datasets that are compared are not coherent. Some are clear-sky, other are all sky. Introducing cloudy skin temperatures in the analysis biases the dataset toward colder temperatures. Uncertainties are also higher below the clouds. Comparing datasets so different in nature is not reasonable and I suggest that you do the SAME cloud filtering for each of the datasets so that you can do an analysis on clear-sky only.
- The diagnostics that are used in the analysis are misleading in my opinion. You perform a lot of averaging on the data and then evaluate the biases on this averaged data. This is not correct. For instance, you can have a dataset with 50% of the samples with -5C of bias, and 50% of the samples with +5C biases, if you average these biases, you will come up with a 0 bias... This is what we can see for instance in Fig2 (column 1 and row 2) where you have very high values and low values, then compute the average bias in Table 3 (if I interpret correctly) at -0.63C. You cannot say that the overal bias between MODIS and ERA is -0.63C... This is highly misleading.
- Figure 1, first plot, the ESA CCI is 8°C higher than the other datasets. It means that their global average over land in the whole planet is 8°C higher than the other datasets, at 30°C... ESA CCI LST has been highly evaluated, there are papers and reports on its evaluation, so I believe there are an error in the way the data is represented. With such differences, you need to investigate what is happening, cannot just continue the analysis as if everything is normal.
- Figure 2, there is a longitudinal structure in the ERA5 data (the vertical bands). You mention this issue but do not solve it.
- You retrieval based on IASI information was trained on the IASU Eumetsat product that is based on IASI data and microwave observations, trained on ERA5 skin temperatures. First issue: training a IASI retrieval, on a IASI retrieval trained on ERA5 targets. I do not see the advantage compared to the retrieval based on ERA5 targets... Second, how can we inter-compare all these products that are not a all independent? And why are such large differences between the datasets?
Minor comments:
- You use in the title "CCI" for LST ESA CCI. CCI is used by NASA for many variables so this is ambiguous.
- Abstract: in general, no paragraphs in an abstract.
- Line 26: "in in"
- In your NN model, you have the pixel number (in the orbit). What is important in terms of radiative transfer (and then for the retrieval) is the incidence angle. There is a relationship between the pixel number and this incidence angle. I believe giving the angle would really be more adequate than this indicator, discrete, and non monotonic information.
- Figure 3 : I already commented on the fact that these numbers are misleading. You are averaging spatially the biases that are positive and negative, this can give you artificially low medium bias. We can see clearly in the maps of Figure 2 that biases are truly non neglectable. To obtain numbers that can synthetize the bias and RMS between the two datasets: 1) you need to do you analysis for each retrieval (don't average on 1° boxes, as this already smooth everything) and everything needs to be done at the native resolution; 2) you need to use the same filters (for clouds, etc.) so that you can compare couple of retrievals for same location and same time; 3) you can represent the maps of bias and RMS at native resolution; 4) if you want to obtain a single number, you can aberage the pixel RMS (as they are positive) but not the biases (I would recommend the average of the pixel biases).