Cropper, T. E., Berry, D. I., Cornes, R. C., & Kent, E. C. (2023). Quantifying Daytime Heating Biases in Marine Air Temperature Observations from Ships, Journal of Atmospheric and Oceanic Technology https://doi.org/10.1175/JTECH-D-22-0080.1

Instrumental records of global mean temperature currently go back to 1850. They are combination of air temperature over land and sea-surface temperature over the ocean. Sea surface temperature is used over the ocean because there’s a problem with the air temperatures measured by ships: during the day, the ship warms up and biases the air temperatures measured on its decks. This means that air temperatures can only be used if they were taken during the night, when the ship cools down again, thus cutting the number of available measurements in half. Latterly, we also have grillions of SST observations from satellites and buoys for which there is sadly no air temperature equivalent. Moored buoys, like those in the TAO array, make air temperature measurements, but they also suffer from biases associated with the thermometer screen. Amazingly, the combined effect of all these biases means that we don’t really know what the diurnal cycle of air temperature is over the ocean. In contrast, hourly observations of SST from drifting buoys, Argo floats and moored buoys, as well as more frequent observations from satellites, give us a very good understanding of the diurnal cycle of the sea surface itself.

Using sea-surface temperature is not without its own problems. There are vanishingly few SST measurement before 1850, which is why global temperature records start then (or later). SSTs are also afflicted by systematic errors that changed over the past 170+ years as the instruments used to measure them were modernised*. SST measurements are also noisier than marine air temperature measurements. Furthermore, recent results (at least on a centennial time scale) suggest that using SST as a proxy for air temperature underestimates long-term global temperature change.

If the daytime heating bias could be fixed, then air temperatures could be used in place of sea-surface temperatures and it might be possible to extend the global temperature record back even unto the eighteenth century. This would be great for all sorts of reasons and comes with the added bonus that we’d have a proper air temperature dataset for the first time††.

This paper produced by the GloSAT project** is a step towards achieving that. They use a physically-motivated model (based on earlier work) to statistically estimate the daytime heating bias for different ships. The model depends on solar radiation (cloud cover), wind speed as well as the time of day and location. Applying this to ships is not straightforward as cloud cover and wind speed aren’t always available. Going further back in time, night time measurements of air temperature aren’t always available either and these are needed to perform the fit. Consequently, parameters have to be estimated on a group of ships that have the necessary information and applied to ships that don’t. To keep the task manageable, this is done for a subset of ships and the results grouped to give a set of stock coefficient, the further selected to give a subset that match the available data in the earliest periods.

The paper plots out a way towards generating a data set of marine air temperatures that are corrected for the daytime heating bias as well as quantifying the uncertainties. The precise accounting is going to be complex as different components of the corrections will have errors that correlate at different levels from the individual measurement (uncorrelated errors from small-scale variability), to the particular day (synoptically correlated), for all measurements from the same ship (correlated for one ship) or same set of coefficients (correlated across all ships using them), uncertainty in the climatologies used (large scale correlated in space and time possibly), and finally structural uncertainty (possibly quite complex correlations). This keeping track of all the uncertainties is a hard problem and causes downstream difficulties for anyone trying to use the data – whatever methods they use need to be ready for complex error models. The paper lays a good ground work for all of this.

The paper underscores the importance of digitisation – the more data we have, the better – and recovery of all variables from log books, not just key ones like air and sea temperature (digitisation is a part of the GloSAT project). It’s also a reminder that the current, modern network of marine air temperature measurement is inadequate and has been deteriorating since the late 1980s and that we don’t know something as simple as the average diurnal cycle of air temperature over ~70% of the Earth’s surface.


Actually, night marine air temperatures are used to bias adjust sea-surface temperatures, because the physics of a cooling bucket depends on the ambient air temperature (amongst other things). This paper suggests that early SST might therefore be over adjusted because warm-biased marine air temperatures were used to correct them in the very earliest years.

* I say “modernised” and not “improved” because the changes weren’t always made with accuracy aforethought. Early changes in sampling buckets from wooden to canvas were driven more by ease of deployment than scientific accuracy. This was remedied mid-century with specialised insulated buckets of varying design. However, around the same time, it was noted that it was safer and easier*** to sample the water sucked into the engine room for various purposes. While this can be done with good accuracy, it rarely is. Engine room inlet measurements are often biased and of low quality****.

†† OK… There’s reanalyses, but they assume part of what we want to test and, if they ingest air temperatures, are ingesting biased data so aren’t free from this problem.

** I’m a member of the advisory board. I wasn’t involved in this work so I may have got things wrong.

*** I don’t know which factor was more important, but bucket measurements continue to this day, though they are now quite rare*****.

**** Some instructions suggest opening a tap and running water from the inlet over the bulb of the thermometer. Other installations had an in situ thermometer, but were prone to fouling and incrustation or had widely-spaced gradations; 5 degree spacings are not unheard of and many data are given to the nearest whole degree.

***** Not, one hopes, through attrition.