1. Introduction

Sea surface temperature (SST), identified as one of the essential climate variables (Bojinski et al. 2014) that critically contributes to the characterization of Earth’s climate by the World Meteorological Organization Global Climate Observing System (GCOS) (WMO 2022), is monitored globally from satellites. To anchor them to the International System of Units (SI), validation of these measurements is performed by field-deployed infrared radiometers (or technically “radiation thermometers”) that are shipborne to gather data collocated with the satellite measurements. Both the satellite and the shipborne infrared radiometer derive the sea surface skin temperature (SST_skin) (Donlon et al. 2007). To gain confidence in the satellite measurements it is therefore essential to ensure accuracy and reliability of these field-deployed radiometers through comparisons. In Ohring et al. (2005), the target accuracy for satellite-derived SST is given as 0.1 K, and the shipborne infrared radiometers therefore currently aim to have similar accuracies so as to serve to underpin the current satellite missions that require less than 0.3 K (k = 1) accuracy (Donlon et al. 2012).

The calibration and validation community within the Committee on Earth Observation Satellites (CEOS) has previously held comparison exercises to verify the accuracy of these radiometers (Rice et al. 2004; Barton et al. 2004; Theocharous et al. 2010, 2017, 2019; Theocharous and Fox 2010; Barker-Snook et al. 2017a,b). However, 6 years had passed since the last comparison, and it was considered timely to repeat/update the process, and so a renewed CEOS International Thermal Infrared Radiometer Intercomparison (CRIC) was conducted in 2022.

The comparison took place during two weeks in June of 2022. The first week involved the laboratory-based comparisons of blackbody (BB) reference standards and radiometers at NPL, which is described in an accompanying paper (Yamada et al. 2024). The second week was devoted to the field-based comparison, at the end of Boscombe Pier in Bournemouth, United Kingdom, evaluating the differences in radiometer response when viewing sea surface targets, in particular, the effects of external environmental conditions such as sky brightness temperature. This was different from the previous comparison in 2016, at which time the comparison took place at an inland water reservoir. Unlike the previous comparison, land surface temperature measurement was not a part of the 2022 comparison.

This paper covers the results of the field comparison of the radiometers of the participants. Details of the comparison results can be found in a full report (Yamada et al. 2023c).

2. Overview of the comparison

c. Comparison venue

The field comparison was conducted at the Boscombe Pier in Bournemouth, on the south coast of the United Kingdom. The pier is located in the center of a few kilometers’ stretch of sandy beach and extends southward to the Poole Bay, which is part of the English Channel. At the end of the pier, a corner was fenced off so that the radiometers, data acquisition systems, and other additional instruments could be placed. A generator was placed on site to supply the necessary electricity. A photograph of the pier viewed from the land is shown in Fig. 1a, and a map with its geolocation is in Fig. 1b.

View Full Size

View and location of the comparison venue: (a) View of Boscombe Pier. The neighboring Bournemouth Pier is also visible in the distance. (b) Location of Boscombe Pier (source: Google Maps 2023).

Citation: Journal of Atmospheric and Oceanic Technology 41, 3; 10.1175/JTECH-D-23-0060.1

• Download Figure
• Download figure as PowerPoint slide

3. Participants’ radiometers and their installation at the comparison site

b. Installation of the radiometers

The field comparison exercise was conducted by having the participants’ radiometers simultaneously retrieve the SST_skin. The radiometers were mounted on the platform outside the railing at the end of the pier, extending to the sea, viewing approximately the same area of the sea surface side by side. Care was taken by each participant to avoid the antifall wires, attached to the pier, from obstructing the radiometer field of view.

The six radiometers after installation are seen in Fig. 2. RAL’s large rectangular SISTeR is seen in the foreground. UoV’s two cylindrical-shaped CIMEL radiometers are placed next (one down below viewing the sea at 25° from nadir, the other inside the railing viewing the sky at the complementary angle (25° from zenith) to correct the specular reflection of the downwelling sky radiance on the sea (Niclòs et al. 2004, 2009), then KIT’s two Heitronics radiometers with their distinctive rectangular long hoods are seen attached to a tall pole. Last, three large drum-shaped ISAR radiometers belonging to DMI, CSIRO, and UoS, respectively, are visible toward the far end. The UoV radiometers were removed during rain and each evening to avoid possible contamination.

View Full Size

Radiometers installed side by side retrieving the SST_skin at Boscombe Pier.

Citation: Journal of Atmospheric and Oceanic Technology 41, 3; 10.1175/JTECH-D-23-0060.1

• Download Figure
• Download figure as PowerPoint slide

Table 1 lists the participants’ radiometers. The SST_skin is derived by each participant applying correction for emissivity and for sky brightness reflection. The sea surface emissivity values assumed for these corrections are also listed, which were individually determined according to each measurement condition (i.e., wavelength and view angle from nadir).

Table 1.

Participants’ radiometers and their installation.

View Table

4. Comparison results

Figure 3 shows the comparison results of the SST_skin reported by all participants, where Fig. 3a is the reported SST_skin values, and Fig. 3b shows the SST_skin averaged over 20-min intervals for each participant, with the standard uncertainty claimed by each participant shown in error bars. The horizontal axis in these graphs and in all graphs that follow is the date and time in coordinated universal time (UTC). The reference value is also shown overlayed, evaluated as the simple mean of all participants’ results averaged over the 20-min interval. It was necessary to evaluate and compare averaged values because each participant reported data with differing intervals and timing. The averaging interval was chosen to ensure that sufficient data were averaged for all participants. Since UoV did not make measurements during the night, UoV data are included only in the comparison reference value during daytime hours. The KIT measurements showed an abrupt shift in the sea surface brightness temperature just before midnight of 21 June 2022, which resulted in a shift of approximately −0.4°C in the derived SST_skin, so KIT data on or after 22 June 2022 were also excluded from evaluating the reference value. Although RAL reported a drift in the SISTeR internal BB and therefore claims the results may require adjustment, RAL data have been included in the evaluation of the reference value since the effect of the reported 20 mK drift is insignificantly small in comparison with the scatter among participants. The difference from the reference value was evaluated for each participant and the result is plotted in Fig. 3c, with the error bars denoting the expanded uncertainty for k = 2. The plots show two types of error bars: the one for the temperature difference is the uncertainty of the participant measurement, the other black bar around zero (“ref”) is the uncertainty of the reference value evaluated as the standard error of the mean of the participant reported SST_skin.

View Full Size

Comparison results for SST_skin. The reference value is the arithmetic mean of all participants, excluding KIT on or after 22 Jun 2022. (a) Participant-reported values. (b) SST_skin averaged over 20 min. The error bar denotes standard uncertainty. (c) Difference of SST_skin averaged over 20 min from the reference value. Error bars are the expanded uncertainties (k = 2) of the participant measurements and of the reference value, respectively.

Citation: Journal of Atmospheric and Oceanic Technology 41, 3; 10.1175/JTECH-D-23-0060.1

• Download Figure
• Download figure as PowerPoint slide

Figure 4 shows the reported brightness temperatures of the sea surface and the sky from which the SST_skin was derived: Fig. 4a is the brightness temperature of the sea, and Fig. 4b is the brightness temperature of the sky.

View Full Size

(a) Sea surface brightness temperature and (b) sky brightness temperatures (RAL data not reported).

Citation: Journal of Atmospheric and Oceanic Technology 41, 3; 10.1175/JTECH-D-23-0060.1

• Download Figure
• Download figure as PowerPoint slide

Figure 5a shows the scatter among participants evaluated as the standard deviation of the SST_skin for each 20-min interval. KIT data on or after 22 June 2022 were not included for the same reason as for the evaluation of the reference value. Possible influencing factors are also plotted in Fig. 5: predicted tide time for high tide in Fig. 5a, air temperature and humidity in Fig. 5b, temperature difference between the average temperature for the internal ambient temperature BBs of the three ISARs and the air in Fig. 5c, and absolute temperature difference between the same average BB temperatures and the comparison reference value in Fig. 5d. Air temperature and humidity were measured on site close to the radiometers with a HygroVUE5 Digital Temperature and Relative Humidity Sensor from Campbell Scientific owned by KIT. The sensor had a radiation shield housing and was mounted about 30 cm above Boscombe Pier’s floor. Measurements were taken every 15 s and then averaged over 1 min. The manufacturer states an accuracy of ±0.3°C in air temperature (range: 20°–60°C) and ±3% in relative humidity (range: 80%–100%). In Fig. 5d the absolute temperature difference is evaluated, which falls to a minimum each day in the morning, noting that this corresponds to the timing of low standard deviation in Fig. 5a.

View Full Size

Scatter of measurement and influencing factors: (a) Standard deviation of the SST_skin reported by participants excluding KIT after 22 Jun 2022. Tide times for high tide are indicated by green arrows (source: Tide Times 2023). (b) Air temperature and relative humidity (measurement provided by KIT). (c) Temperature difference between the ISAR internal ambient temperature BBs and the air temperature. (d) Absolute temperature difference between the ISAR internal ambient temperature BBs and the comparison reference value (SST_skin).

Citation: Journal of Atmospheric and Oceanic Technology 41, 3; 10.1175/JTECH-D-23-0060.1

• Download Figure
• Download figure as PowerPoint slide

5. Discussion

The SST_skin reported by the participants, shown in Fig. 3a, all show the daily rise and fall of the temperature, with the highest temperature varying by about 2°C depending on the day. The reference temperature during the 5-day period was in the range from 16.8° to 19.8°C, so the comparison was able to cover this 3.0°C temperature range, which is approximately 2 times as large as the previous comparison at a water reservoir (Barker-Snook et al. 2017b).

The agreement among all participants is relatively good initially on 20 and 21 June. However, as mentioned in the previous section, an abrupt shift in the KIT value occurred just before midnight of 21 June, after which the offset of approximately −0.4°C remained. The pilot later notified KIT of this irregularity, but nothing could be found by KIT upon reviewing the measurement data that can explain the abrupt change in behavior, and so no change was made to the data. The shift is seen in the sea brightness temperature (Fig. 4a) but not in the sky brightness temperature (Fig. 4b). Two separate radiometers are used for the two measurements by KIT, so it appears that the radiometer measuring the sea was affected.

It is also noticeable in Fig. 3a that the scatter of the SST_skin reported by CSIRO (plotted in yellow) was large until the afternoon of 20 June, when it suddenly decreased and stayed small thereafter. A closer inspection reveals that this is seen in both the sea surface brightness temperature and the sky brightness temperature, which is expected because for the ISAR the same radiometer measures both temperatures. Upon notification from the pilot after submission of data, CSIRO investigated the cause but could not find any clue to what might have caused this and thus the result stays as it is. A possible explanation is some noise through the power line was present in the beginning but somehow disappeared afterward. The scatter does not influence the measurement in a systematic way and is taken into account in the slightly larger measurement uncertainty for this period as shown in Fig. 3c.

In the sky brightness temperature data of Fig. 4b, each day just around 1200 UTC except for 23 June, a spike is seen for measurements by DMI, RAL, CSIRO, and UoS. KIT data do not show this phenomenon. UoV submitted data after eliminating this part of data for two days (21 and 22 June). The spike is thought to be caused by the light from the sun coming into the field of view of the radiometer when measuring the sky brightness temperature: the pier extends toward the south with a slight tilt to the east, and the radiometers were all aligned straight out to this direction, and, considering the geolocation of Bournemouth, which is to the west of Greenwich, all seem to indicate the cause to be the sun coming in view just before noon. This did not happen on 23 June, because the sun was behind the clouds. KIT had a different orientation from the others for its sky viewing radiometer so was not affected. The sea brightness temperature measurement is not affected and there is no evidence of sun glitter, and therefore the effect of the spike in the sky brightness temperature measurement results in a slight dip in the SST_skin corrected for the reflection, but this is short and hardly noticeable except for the spikes seen in Figs. 3b and 3c for UoV (who took care to eliminate the spikes in the sky brightness temperature and therefore were affected by spikes in the average) around noon for two days, and has insignificant effect on the overall result of the comparison.

The scatter among the participants’ reported SST_skin over time was evaluated in Fig. 5a. As seen from the plot, the scatter becomes largest around or before 1200 UTC each day. To investigate the cause, the ambient temperature and humidity are plotted in Fig. 5b. The peak in the scatter (Fig. 5a) shows some resemblance to the rise of the ambient temperature in Fig. 5b. However, the fall of the ambient temperature does not correspond to the valley of the scatter. In Fig. 5c, difference of the mean of the ISAR internal ambient BB temperatures from the air temperature is shown. Individual ISAR BB temperatures follow the mean temperature within 0.5°C for most of the time. The increase in the temperature difference during daytime of 2°–5°C is considered the effect of insolation. When one compares Fig. 5a with Fig. 5c, we notice that there is a similarity, indicating the possibility of the instrument being affected by this. However, for this case also, the rise and fall of the scatter before and around 1200 UTC in Fig. 5a is not reflected entirely in Fig. 5c at the same timing. In Fig. 5d the absolute difference of the mean of the ISAR internal BB temperatures from the comparison reference value (i.e., mean retrieved SST_skin) is shown. When one compares Fig. 5a with Fig. 5d, we notice that there is a remarkable similarity. Change in SST_skin is relatively small in comparison with the instrument internal BB temperature, and the difference between them becomes zero when they cross over. This occurs twice daily, in the morning, and in the afternoon or during night, which are seen as dips in Fig. 5d. Increases in the absolute temperature difference around the dips are well reproduced in the peaks in the standard deviation plotted in Fig. 5a. Since the ISAR and the SISTeR both utilize the internal BB, it is understandable that they achieve the best accuracy when the SST_skin and the internal BB temperature are about the same. This is because the infrared detector only needs to detect the small difference in the incoming fluxes between the sea view and the internal BB view to give the small correction to the precise reading of the contact thermometer monitoring the ambient blackbody to determine the sea brightness temperature. When these temperatures deviate the uncertainty of measurement will increase, which corresponds to the observed increase in the scatter. UoV’s CIMEL radiometer measures the internal cavity radiance to account for the target-ambient temperature difference in a similar way (Legrand et al. 2000) so can be expected to follow the same trend.

Figure 4b shows the sky brightness temperature, which is representative of the cloud condition: it is low when it is sunny, and higher when there is cloud. If we compare Fig. 4b with Fig. 5a, we do not see any relation between the two, indicating that the correction for the reflected sky radiance is working well. It should be noted that this is when the sea surface emissivity values adopted varied among the participants in the range from 0.986 to 0.992. A variation of approximately 0.005 in emissivity is equivalent to approximately 0.23°C variation in SST_skin on a clear day and approximately 0.07°C with overcast cloud, which indicates that the participants adopted emissivity values that represent each one’s measurement conditions appropriately.

Variation in relative humidity from around 35% to over 90% is observed in Fig. 5b. However, from visual inspection no correlation is seen between this and Fig. 5a. No occurrence of dew was observed. On 23 June, precipitation was encountered, but the ISAR and SISTeR instruments, equipped with rain detectors, were shielded automatically, CIMEL was removed out of the rain, and no comparison measurements took place. Wind condition was mostly stable with wind speed ranging from 0 to 5 m s⁻¹, averaging around 2 m s⁻¹ for all days, according to Windfinder (2022). No breaking of waves was observed, and no surfers were noticed in the area. Tide time shows no apparent correlation with scatter in Fig. 5a.

The error bars in Fig. 3b, for the participants’ measurement and for the reference value, represent the expanded uncertainty for k = 2. Overlap of the two error bars means agreement of the measurement with the reference. All participants show good agreement throughout the 5-day comparison period, the only exception being a slight deviation by KIT for a short period after the unexplained abrupt shift on 21 June just before midnight.

The difference of the mean participant values from the reference value, shown in Fig. 3c, were (with the exception of KIT on or after 22 June) all within ±0.07°C. This is a more than 2-times improvement when compared with the previous comparison in 2016 for the same six participants (Barker-Snook et al. 2017b). The improvement is striking if one takes into account that the temperature range of this current comparison was 2 times as wide as in the previous comparison. The reason for the improvement is not clear, although it can most likely be attributed to the good measurement conditions and careful undertaking by the participants. Also, as compared with the last comparison at an inland water reservoir, in this comparison the participants were all able to view the sea facing south so as to avoid shadows. Favorable weather conditions with fewer clouds and less precipitation could also have contributed. Good mixing of the water by tide and waves may have played a role. The observed sea surface was clear of any obstacles at all times.

The uncertainty of the comparison reference value is around 0.1°C (k = 2). The participant measurement uncertainty ranged from roughly 0.1° to 0.75°C (k = 2). Therefore, the reference value uncertainty is small enough and the comparison accurate enough to verify the agreement between the participant radiometers. It should be noted that the agreement evidenced here is with respect to the mean of the participant measurements. This means the comparison only supports agreement among the participants when retrieving SST_skin (even though the uncertainties reported by participants include calibration uncertainty). There is a possibility of a systematic offset in the reference value due to systematic offsets in the measurement and calibration of the radiometers involved, which is the subject of investigation in the two laboratory-based comparisons of the CRIC, where direct comparisons are made against NPL reference standards (Yamada et al. 2024).

6. Conclusions

The SST_skin retrieving capabilities of six participating institutes were evaluated through a comparison of radiometers at Boscombe Pier, Bournemouth, on the south coast of England as a part of the CRIC. During the comparison, which took place during five days in June 2022, the six radiometers viewed the sea surface from the end of the pier and the measured temperatures were compared after correcting for sea surface emissivity and reflected radiation from the sky.

All participants’ reported values agreed with the reference value within the uncertainties. Here, the reference value was evaluated as the simple mean of the participant reported SST_skin. The mean of the difference from the reference value taken over the whole comparison period was evaluated for each participant, and all were found to be within 0.07°C, which is one-half of what was reported in the previous comparison in 2016 (Barker-Snook et al. 2017b).

Although the derived SST_skin range was about 3°C, which is 2 times as wide as the previous comparison, it is still limited when considering the actual SST_skin that one needs to retrieve in the ocean. It would be of interest to conduct a similar comparison at a different location or during a different season to cover a wider or different temperature range.

An issue that became apparent was an abrupt shift in one of the radiometers: the reading shifted by −0.4°C from the middle of the comparison period for an undetermined reason, and data after this shift were excluded from the comparison. The shift is quite obvious when comparison is made with other radiometers, but if no data from other radiometers or thermometers were available, it would be extremely difficult to detect. When the radiometer is deployed on board a ship, even if one detects there was a shift through a recalibration after return from the trip, it will be impossible to identify whether the shift was abrupt or gradual, or when it had happened. Some of the radiometers in the current comparison have an internal cavity source or two high-precision internal blackbodies to track and correct for any drifts. The result of the current comparison confirms the importance of such systems.

The field comparison results verify the agreement of the participant radiometers among themselves in practical conditions. Agreement with the SI has been substantiated in the laboratory part of the comparison, the results of which are presented in an accompanying paper (Yamada et al. 2024), and details in full reports (Yamada et al. 2023a,b). The combined comparison results for the two parts support the reliability of the SST_skin retrieval by the radiometers.

Data availability statement.

Datasets for this research are included in Yamada et al. (2023c; https://ships4sst.org/sites/shipborne-radiometer/files/documents/FRM4SST-CRICR-NPL-003_ISSUE-1.pdf).