Measuring temperature onboard ships has a long history that stretches back over 250 years. Work is currently being undertaken to make use of these marine air temperature observations as a way of improving knowledge about current global temperature change.
The Origins of Ship-based Temperature Measurements
In September 1699 the ship HMS Paramour (Figure 1) set sail on its second voyage under the command of the astronomer Edmond Halley on an expedition to map the magnetic variation of the compass across the North and South Atlantic (Thrower, 1981). Halley took several scientific instruments onboard. Amongst them were two meteorological instruments that were still in their infancy, the barometer and thermometer, and the sequence of thermometer observations taken on the Paramour represent the first recorded ship-based measurements of marine air temperature (MAT) available in the global archive. Relatively few MAT measurements were recorded onboard sailing ships in the decades that followed. However, from the 1770s the number of records increase (Figure 2), and today ships still contribute millions of temperature observations annually, supplemented by observations made from buoys and other types of observing platforms.
The Evolving Marine Temperature Record
The track of the HMS Paramour on its 1699 voyage during which air temperature measurements were taken is shown in Figure 3. The record of observations provided by the ship is typical of many pre-nineteenth century observations in that they were largely recorded on expeditionary voyages. As a result of this, the global record of early MAT observations has sparse global coverage. However, the increasing ship traffic along global trade routes from the late eighteenth century provides an increase in data coverage. During this period the temperature data were principally recorded onboard ships traversing the trade routes of the British East India Company, between Europe and India or south-east Asia.
An increase in the quantity and quality of marine temperature observations occurs from the mid-nineteenth century. A particular catalyst for this were the agreements made during the international meteorological conference held in Brussels in 1853. During that conference many maritime nations agreed to share their meteorological observations in recognition of the importance of international cooperation for furthering understanding of global weather patterns. The conference had been orchestrated by the American naval officer and scientist Matthew Fontaine Maury, and following the conference many maritime nations began sending their oceanographic measurements to Maury at the US Naval Observatory in Washington, D.C. These data were compiled and re-distributed globally and the charts of global wind and ocean currents calculated from the data provided a vital tool for the optimization of global trade.
Despite a drop in coverage during the two World Wars, the global number of ship observations has increased markedly throughout the twentieth century. The latter two years shown in Figure 3 (1969 and 2021) share similarities in terms of the density and distribution of data. However, the coverage across the global ocean was greater in 1969. This reflects a shift from more ships taking fewer observations in 1969 to fewer ships taking more observations per ship in 2021.
Responding to the need for researchers to have open access to the archive of historical marine surface data, the Comprehensive Ocean-Atmosphere Dataset (COADS) was released in 1985 (Slutz et al., 1985). This database combined meteorological and surface ocean observations from several sources for the period 1854-1979; most of these data were ship-based. The COADS initiative was led by the US National Oceanic and Atmospheric Administration’s (NOAA) Environmental Research Laboratories (ERL) and provided an important source of data for analysing large-scale ocean-atmosphere processes, including global temperature change. The initiative has evolved over the last 40 years, to include many more data, including those from moored/drifting buoys, and in 2002 the name was changed to the International Comprehensive Ocean-Atmosphere Dataset (ICOADS), to reflect the importance of international cooperation to the success of the project (Freeman et al., 2019). ICOADS now provides the largest collection of surface marine observations to the research community (Freeman et al., 2017). In contrast to the handful of measurements made in the 18th century, there are millions of ship-based observations of air and sea surface temperature across the global ocean in 2021. Most of these ship observations are collated by the Voluntary Observing Ship (VOS) network (https://www.ocean-ops.org/reportcard2021/).
Monitoring global temperature change: Sea-Surface Temperature data
The record of surface temperature for terrestrial regions is constructed using observations of air temperature made at weather stations, typically in a screen at two metres in height. For the marine component there is a choice of using either temperatures from the atmosphere (MAT) or values recorded in the first few metres of the ocean; this variable is called sea-surface temperature (SST). Originally it was not thought to be important which variable was chosen and it was argued that on large space and time scales, anomalies of both values would be similar. SST was chosen because it is less variable than MAT, and therefore averages of temperature can be more readily constructed from sparse observations. SST is also better sampled particularly during the last 30 years, when observations recorded from drifting buoys have greatly enhanced global coverage (Kennedy et al., 2019).
SST observations were historically taken using samples of sea water from buckets that had been lowered to the sea surface from the ship. These buckets were then retrieved, and the temperature of the sample was taken using a thermometer. The material of the bucket, atmospheric conditions and the time taken to determine the temperature of the water in the bucket can all contribute to biases and increased uncertainty in the recorded value (Carella et al., 2017). In the mid-twentieth century, with the increasing size of these vessels, samples were more easily taken from the sea-water inlets that were used to cool the ship’s engines; in recent years, readings from hull-mounted sensors are the predominant source of SST data from ships (Kent et al., 2017).
Adjustments need to be applied to the SST data to account for these changing observation practices and to ensure homogeneity of the data over the course of the series. Random observation errors will average to zero if a sufficiently large sample is available. However, systematic errors are correlated across observing stations and need to be removed from the data. For example, it was shown by Chan et al. (2019) that certain SST data from the early twentieth century that were recorded in degrees Fahrenheit were truncated to whole degrees when the data were transferred to punch-cards. This caused a cool bias in the values from that source.
Over the last 40 years, observations of drifting buoys have grown to dominate the observational SST record and satellite observations have also become available (e.g. Merchant et al., 2014). These data have allowed globally complete, high-spatial resolution datasets to be constructed that provide unprecedented high-resolution depictions of SST and have allowed the quantification of complex features of the spatial and temporal variability of SST.
Marine Air Temperature Data
Although SST remains the principal variable for monitoring changes in temperature over the world’s oceans, MAT is also important variable. MAT is mainly used as a comparison to the SST data and if MAT datasets can be constructed with sufficient accuracy, they may be able to help resolve and identify errors in SST datasets, as long as the errors in MAT and SST are not correlated, i.e. the same systematic biases do not affect both variables.
Until the late twentieth century most MAT observations were recorded manually by the ship’s officers using mercury thermometers housed in marine screens to provide shelter from solar radiation (Figure 4). Since then, there has been a transition to the use of Automatic Weather Stations. The observation height of these data varies between ships and on average the height has increased due to the increasing length and height of the vessels (Kent et al., 2013). During the late nineteenth century, the observation height is very uncertain but likely to be around 6 m from the sea surface but on modern container ships the height can be over 40 m. The air temperature measurements are adjusted to a pre-determined reference height above sea-level (typically 10m) to ensure comparability of values between ships and to ensure that the homogeneity of the series is not affected by the increasing observation height over time. If this long-term adjustment was not applied to the data, a slight long-term cooling would occur in the data because of the increasing observation height. To adjust the temperature values to the reference height, knowledge about temperature gradients near the lower boundary of the atmosphere surrounding the ships is required in addition to the height of the thermometer. Hence metadata, i.e. where the observations were taken and the nature of the instrument, remain important sources of information for the homogenization of ship-based air temperature measurements (Moltmann et al., 2019)
An additional bias in the MAT data occur because of heating from the superstructure of the ship (Berry et al., 2004). As the ship warms during the day, the temperature measurements recorded onboard the vessels are artificially heated. To reduce the effects of this warming on the air temperature datasets, only the data recorded at night have been used to construct long-term climate datasets of MAT; these are referred to as Night Marine Air Temperature (NMAT) datasets. There are two NMAT datasets that are currently being produced: the CLASSnmat dataset, that is constructed by the National Oceanography Centre (Cornes et al., 2020) and UAHNMAT produced by the University of Alabama Huntsville (Junod & Christy, 2020). These datasets vary in the approach that is taken to produce gridded datasets from the individual ship observations. While this produces slight differences on the regional basis, the data are generally indistinguishable in terms of long-term global trends (Figure 5).
Comparing Apples to Oranges
On shorter timescales SST and MAT anomalies can vary significantly. However, on inter-annual and longer timescales – and across large regions – SST and MAT should be closely related (Kennedy et al., 2019). Indeed, this is the reason why SST and MAT anomalies are generally regarded as synonymous. However, in recent years a distinct difference has been observed in global average trends calculated from the SST datasets compared to the NMAT data. Over the last 30 years the observational record indicates that the global average NMAT has warmed at a slower rate compared to SST (Figure 5). However, it remains uncertain if this difference has occurred as a result of natural changes – perhaps a response to the rapid recent surface temperature increase – or as a result of unresolved biases in either the SST or NMAT data (Dunn et al., 2021). Work is ongoing to determine the cause of this difference.
The differential trends in the SST and NMAT observational datasets is contrary to the results from climate model simulations, which show trends of global average MAT rising faster than SST (Gulev et al., 2021). These simulations provide air temperature over all surfaces: ocean, atmosphere and ice as their standard output. Surface temperatures are rising fastest in polar regions, which are poorly sampled by most observations, and are also increasing faster over land than the ocean. Differences between MAT and SST, and between observations and models can therefore arise for several different reasons. To untangle the relative contributions of errors due to irregular sampling in the observations, measurement errors in SST and MAT, model errors in the climate simulations and real temperature differences is therefore challenging.
Ideally, we would make a like-for-like comparison of global temperatures between models and observations. This has been attempted by calculating an observation-like blend of air temperature anomalies over land with SST. This is not straightforward because many regions on the model grid are a mix of land and ocean, and temperature anomalies in sea-ice regions could be represented by either MAT or SST. One study concluded that it was likely that the uncertainty in extracting the required information from the model archive was as large as the signal itself (Jones, 2020).
Under the NERC-funded GloSAT project (https://www.glosat.org/), efforts are underway to construct a merged dataset that uses values of MAT adjusted to 2m elevation over the ocean along with the land-based 2m temperature values. This is in in contrast to the more usual combination of SST and terrestrial air temperature, and will provide a dataset that is more comparable to the model-derived 2m temperature estimates.
Extending the temperature record back in time: the GloSAT project
The Paris Agreement (https://unfccc.int/process-and-meetings/the-paris-agreement/the-paris-agreement) has a goal of limiting global warming to well below 2°C, and preferably to 1.5°C, compared to pre-industrial levels. The period 1850 – 1900 is used as the base-period to represent the pre-industrial climate. However, the industrial revolution was well underway by this time and the choice of base-period is therefore a pragmatic decision largely driven by the availability of sufficiently dense global data coverage and the relatively small global temperature changes due to human activities during the period.
The global temperature dataset that is being constructed under the GloSAT project will be extended back to the late eighteenth century to allow an estimate to be made regarding the amount of global warming that had already occurred by the mid-nineteenth century. This dataset will also provide the following advances in climate science:
- The extended data record will provide an extended comparison series for the evaluation of paleoclimate datasets and climate model simulations.
- The series will allow further analyses to be conducted regarding the impact on the climate system of several volcanic eruptions that occurred in the first half of the nineteenth century, most notably the Tambora eruption of 1815.
- NMAT datasets currently use only around half of the available MAT observations. Although adding daytime values has only a minor effect on increasing the effective global spatial coverage, the increase in the number of data using all-hours MAT data will reduce uncertainty in the gridded temperature values.
SST data were seldom recorded before the mid-nineteenth century and extending the global marine temperature dataset back before the 1850s therefore requires MAT observations. However, most of these air temperature values were recorded during the daytime and hence the heating biases that reside in these data need to be removed. The magnitude of this bias depends on the location of the ship, time of day and the physical properties of the ship, which means that different ships heat up by different amounts and rates (Figure 6). The night-time cooling rate also varies between ships.
This daytime-heating effect can be modelled using a heat-budget approach. For this model to be able to estimate the heating and cooling rate of ships, an estimate of the incoming solar radiation is required. This can be inferred from the amount of cloud cover. To determine the cooling rate, knowledge of the local airflow around the thermometer is needed, which can be determined by the ship’s relative wind speed (calculated from the ship’s speed and course, and the wind speed and direction). To calibrate the model on a ship-by-ship basis an estimate of the heating bias itself is required. This is difficult because the diurnal cycle of MAT recorded from the ship’s thermometer includes the true diurnal cycle, and the diurnal cycle of the heating bias, both of which vary with solar forcing and wind speed. As a compromise the GloSAT project will aim to adjust all MAT observations to the night-time mean. While this approach overestimates the effect of the heating bias, it has the advantage of ensuring that the adjusted data will all be relative to the night-time mean value, so any time-of-day bias in MAT observations will be minimal.
The continuing importance of ship-based measurements
The record of Marine Air Temperatures (MAT) remains an important source of information for monitoring global temperature change. As indicated in this article, the long record allows us to extend the temperature further back in time than was hitherto possible. However, the modern record has become a cause for concern due to the declining number of useable MAT observations. Observations of ship-based MAT peaked in the 1980s and has been declining since then due to fewer ships recording data. In the case of SST, the availability of readings made from drifting buoys has mitigated the decline in ship-based SST measurements (Kent et al., 2019). For MAT, however, ship-based measurements remain the main source of data for constructing global datasets. The decline in the number of ships recording meteorological observations needs to be stemmed if MAT is to remain a useful indicator of global temperature change.
 Deviations from a fixed 30-year average period.
 Paleoclimate series are constructed using environmental indicators such as coral reefs, tree rings or documentary sources from which air temperature values can be inferred.
Berry, D. I., Kent, E. C., & Taylor, P. K. (2004). An analytical model of heating errors in marine air temperatures from ships. In Journal of Atmospheric and Oceanic Technology (Vol. 21, Issue 8, pp. 1198–1215). https://doi.org/10.1175/1520-0426(2004)021%3C1198:AAMOHE%3E2.0.CO;2
Carella, G., Morris, A. K. R., Pascal, R. W., Yelland, M. J., Berry, D. I., Morak-Bozzo, S., Merchant, C. J., & Kent, E. C. (2017). Measurements and models of the temperature change of water samples in sea-surface temperature buckets: Temperature Change of Water Samples in SST Buckets. Quarterly Journal of the Royal Meteorological Society, 143(706), 2198–2209. https://doi.org/10.1002/qj.3078
Chan, D., Kent, E. C., Berry, D. I., & Huybers, P. (2019). Correcting datasets leads to more homogeneous early-twentieth-century sea surface warming. In Nature (Vol. 571, Issue 7765, pp. 393–397). https://doi.org/doi:10.1038/s41586-019-1349-2
Cornes, R. C., Kent, E. C., Berry, D. I., & Kennedy, J. J. (2020). CLASSnmat: A global night marine air temperature data set, 1880–2019. In Geoscience Data Journal (Vol. 7, Issue 2, pp. 170–184). https://doi.org/10.1002/gdj3.100
Dunn, R. J. H., Aldred, F., Gobron, N., Miller, J. B., Willett, K. M., Ades, M., Adler, R., Allan, R., P., Allan, R., Anderson, J., Argüez, A., Arosio, C., Augustine, J. A., Azorin-Molina, C., Barichivich, J., Beck, H. E., Becker, A., Bellouin, N., Benedetti, A., … Zotta, R. M. (2021). Global Climate. Bulletin of the American Meteorological Society, 102(8), S11–S142. https://doi.org/10.1175/BAMS-D-21-0098.1
Freeman, E., Kent, E. C., Brohan, P., Cram, T., Gates, L., Huang, B., Liu, C., Smith, S. R., Worley, S. J., & Zhang, H.-M. (2019). The International Comprehensive Ocean-Atmosphere Data Set – Meeting Users Needs and Future Priorities. Frontiers in Marine Science, 6, 435. https://doi.org/10.3389/fmars.2019.00435
Freeman, E., Woodruff, S. D., Worley, S. J., Lubker, S. J., Kent, E. C., Angel, W. E., Berry, D. I., Brohan, P., Eastman, R., Gates, L., Gloeden, W., Ji, Z., Lawrimore, J., Rayner, N. A., Rosenhagen, G., & Smith, S. R. (2017). ICOADS Release 3.0: A major update to the historical marine climate record. In International Journal of Climatology (Vol. 37, Issue 5, pp. 2211–2232). https://doi.org/10.1002/joc.4775
Gulev, S. K., Thorne, P. W., Ahn, J., Dentener, F. J., Domingues, C. M., Gerland, S., & Vose, R. S. (2021). Changing State of the Climate System. In V. Masson-Delmotte, P. Zhai, A. Pirani, S. L. Connors, C. Pean, S. Berger, N. Caud, Y. Chen, L. Goldfarb, M.I. Gomis, Haung, M., K. Leitzell, Lonnoy, E., J. B. R. Matthews, T. K. Maycock, T. Waterfield, O. Yelekçi, R. Yu, & B. Zhou, The physical science basis. Contribution of working group I to the sixth assessment report of the intergovernmental panel on climate change (pp. 287–422). Cambridge University Press. doi:10.1017/9781009157896.004
Jones, G. S. (2020). ‘Apples and Oranges’: On comparing simulated historic near‐surface temperature changes with observations. In Quarterly Journal of the Royal Meteorological Society (Vol. 146, Issue 733, pp. 3747–3771). https://doi.org/10.1002/qj.3871
Junod, R. A., & Christy, J. R. (2020). A new compilation of globally gridded night‐time marine air temperatures: The UAHNMATv1 dataset. In International Journal of Climatology (Vol. 40, Issue 5, pp. 2609–2623). https://doi.org/10.1002/joc.6354
Kennedy, J. J., Rayner, N. A., Atkinson, C. P., & Killick, R. E. (2019). An Ensemble Data Set of Sea Surface Temperature Change From 1850: The Met Office Hadley Centre HadSST.126.96.36.199 Data Set. In Journal of Geophysical Research: Atmospheres (Vol. 124, Issue 14, pp. 7719–7763). https://doi.org/10.1029/2018jd029867
Kent, E. C., Kennedy, J. J., Smith, T. M., Hirahara, S., Huang, B., Kaplan, A., Parker, D. E., Atkinson, C. P., Berry, D. I., Carella, G., Fukuda, Y., Ishii, M., Jones, P. D., Lindgren, F., Merchant, C. J., Morak-Bozzo, S., Rayner, N. A., Venema, V., Yasui, S., & Zhang, H.-M. (2017). A Call for New Approaches to Quantifying Biases in Observations of Sea Surface Temperature. Bulletin of the American Meteorological Society, 98(8), 1601–1616. https://doi.org/10.1175/BAMS-D-15-00251.1
Kent, E. C., Rayner, N. A., Berry, D. I., Eastman, R., Grigorieva, V. G., Huang, B., Kennedy, J. J., Smith, S. R., & Willett, K. M. (2019). Observing Requirements for Long-Term Climate Records at the Ocean Surface. In Frontiers in Marine Science (Vol. 6, p. 28). https://doi.org/10.3389/fmars.2019.00441
Kent, E. C., Rayner, N. A., Berry, D. I., Saunby, M., Moat, B. I., Kennedy, J. J., & Parker, D. E. (2013). Global analysis of night marine air temperature and its uncertainty since 1880: The HadNMAT2 data set. In Journal of Geophysical Research: Atmospheres (Vol. 118, Issue 3, pp. 1281–1298). https://doi.org/10.1002/jgrd.50152
Merchant, C. J., Embury, O., Roberts‐Jones, J., Fiedler, E., Bulgin, C. E., Corlett, G. K., Good, S., McLaren, A., Rayner, N., Morak‐Bozzo, S., & Donlon, C. (2014). Sea surface temperature datasets for climate applications from Phase 1 of the European Space Agency Climate Change Initiative ( SST CCI ). Geoscience Data Journal, 1(2), 179–191. https://doi.org/10.1002/gdj3.20
Moltmann, T., Turton, J., Zhang, H.-M., Nolan, G., Gouldman, C., Griesbauer, L., Willis, Z., Piniella, Á. M., Barrell, S., Andersson, E., Gallage, C., Charpentier, E., Belbeoch, M., Poli, P., Rea, A., Burger, E. F., Legler, D. M., Lumpkin, R., Meinig, C., … Zhang, Y. (2019). A Global Ocean Observing System (GOOS), Delivered Through Enhanced Collaboration Across Regions, Communities, and New Technologies. Frontiers in Marine Science, 6, 291. https://doi.org/10.3389/fmars.2019.00291
Slutz, R. J., Lubker, S. J., Hiscox, J. D., Woodruff, S. D., Jenne, R. L., Joseph, R. H., Steurer, P. M., & Elms, J. D. (1985). Comprehensive ocean-atmosphere dataset: Release 1. [Report]. NOAA Environmental Research Laboratories, Climate Research Program.
Thrower, N. J. W. (1981). The three voyages of Edmond Halley in the Paramore, 1698-1701. The Hakluyt Society.