1. Introduction

Sea surface temperature (SST) measurements obtained from infrared instruments deployed on earth-orbiting satellites have provided meteorologists and oceanographers with synoptic views of the dynamic thermal character of the ocean surface for over 20 yr (e.g., Robinson and Donlon 2003). Such SST measurements are fundamentally important to agencies and institutions tasked with the study of climate variability, operational weather and ocean forecasting, military operations, validation and forcing of ocean and atmospheric models, and ecosystem assessment and fisheries research, among others. Despite the importance of satellite SST data, it remains costly and difficult to produce timely and unequivocal evidence that satellite instruments deliver SST datasets to the accuracy defined in their design specifications. Validation data collection and analysis is an important (but often overlooked) component of every earth observation activity, being fundamental to the success of the whole satellite mission since independent error limits for the geophysical data products cannot be demonstrated without a rigorous validation program. New systems [such as the Global Ocean Data Assimilation Experiment (GODAE) High Resolution Sea Surface Temperature Pilot Project (GHRSST-PP; see Donlon et al. 2007)] are now developing high-resolution (<10 km, daily, accurate to <0.5 K) global SST analyses, for which it is extremely important to have accurate knowledge of uncertainty for all of the complementary SST observations used in the merging process. In the case of infrared satellite SST datasets now being obtained from the Environmental Satellite (Envisat) Advanced Along-Track Scanning Radiometer (AATSR; see Edwards et al. 1990) to an accuracy of better than 0.3 K (O’Carroll et al. 2004), the quality of in situ measurements for validation has become the limiting factor to establishing the absolute accuracy of the satellite SST product. To properly validate satellite SST observations, contemporaneous in situ measurements of SST must be obtained. Accurate and well-documented validation datasets provide the means to test satellite instrument performance, verify the atmospheric correction strategies and geophysical algorithms used to derive SST estimates from top of the atmosphere radiances, and quantify the accuracy (bias and standard deviation) of the derived SST data products.

Given the rapidly varying thermal character of the air–sea interface (time scales <10 s; Jessup et al. 1997), comprehensive SST validation datasets require that in situ measurements are matched with satellite observations to within narrow spatial and temporal limits. Ideally, in situ observations should be obtained at regular short (10 min or less) time intervals as block averages in order to properly sample the spatial and temporal variations of the SST field in a similar way to bulk aerodynamic flux estimates. They should also properly sample the various atmospheric conditions (including both horizontal and vertical structures) for which SST retrieval algorithms are expected to function. Most importantly, SST validation data should consider the significant differences that exist between the surface skin temperature (SSTskin) of the ocean when measured by an infrared radiometer and the subsurface temperature (SSTdepth) measured by conventional contact thermometers (typically, but incorrectly, reported without specification of the measurement depth). In this paper, SSTskin is defined as the radiometric skin temperature measured by an infrared radiometer operating in the 10–12-μm spectral wave band. As such, it represents the actual temperature of the water at a depth of approximately 10–20 μm. This definition is chosen for consistency with the majority of infrared satellite and ship-mounted radiometer measurements. SSTskin measurements are potentially subject to a diurnal cycle including cool skin layer effects (especially at night under clear skies and low–wind speed conditions) and warm layer effects in the daytime. In some cases, thermal stratification during high-insolation, low–wind speed conditions (e.g., Yokoyama and Tanba 1991; Donlon et al. 2002; Stuart-Menteth et al. 2005) effectively decouples the SSTskin from the SSTdepth, rendering subsurface thermometry seriously inadequate for constructing SST validation datasets in these conditions. In geographical areas characterized by persistent low wind speed and diurnal variability, the use of infrared radiometers to measure the SSTskin from ships is the only way to provide suitable in situ measurements for validating infrared satellite SST measurements (Donlon et al. 2002). Recent years have seen the development of several ship-based radiometer designs capable of target measurements with accuracy better than 0.075 K. Most notable are the scanning infrared sea surface temperature radiometer (SISTeR; see Donlon et al. 1999a), the marine atmospheric emitted radiance interferometer (M-AERI; Minnett et al. 2001), the DAR011 (Barton et al. 2004), and the Calibrated Infrared In situ Measurement System (CIRIMS) radiometer. CIRIMS was specifically developed to provide autonomous SSTskin observations (Jessup et al. 2002) and has been deployed for several years in an autonomous configuration on research vessels (see http://cirims.apl.washington.edu/CruiseMaps.html). However, the CIRIMS is both large and costly and has yet to be deployed from a voluntary observing ship (VOS) platform in a truly autonomous mode, as it requires frequent maintenance. Despite these developments, the regular collection of a globally comprehensive in situ SSTskin dataset has remained an elusive goal for the agencies and scientists tasked with the sustained validation of satellite SST data. The available number of SSTskin validation data is limited in number, spatial, and temporal extent (e.g., Thomas et al. 1995; Donlon et al. 1999a; Kent et al. 1996; Donlon and Robinson 1997; Noyes et al. 2006) and there is a clear need for in situ infrared radiometers capable of sustained long-term operation (i.e., without operator intervention) to provide adequate satellite SST validation measurements.

The lack of data is largely due to the fact that maintaining and deploying delicate optical instrumentation aboard ships is both difficult and costly. While all of the in situ radiometer systems noted above continue to provide high-quality SSTskin datasets, most are not capable of autonomous deployment because they have no means to automatically protect their optical components from the harsh marine environment. Instead, they rely on an instrument operator to manually close and protect them during poor weather conditions or heavy seas. The requirement for an operator incurs a considerable overhead cost and presents logistical problems for sustained operations. Furthermore, instrument operators may be exposed to danger when securing and protecting ship-mounted radiometers (often located in dangerous areas such as the ship foremast) in deteriorating weather and increasing sea state. An autonomous measurement system could dramatically improve the probability of successful satellite and in situ matchups through the use of VOS platforms, avoiding operator overhead cost and maximizing measurement times at sea.

This paper describes a fully autonomous infrared radiometer system that has been developed specifically for deployment aboard VOS to validate Envisat AATSR SSTskin data products in a sustained measurement program. The infrared SST autonomous radiometer (ISAR) is a self-calibrating instrument measuring in situ SSTskin accurate to 0.1 K rmse. ISAR is capable of continuous deployment for up to 3 months aboard VOS in all weather conditions while maintaining accurate calibration. Four ISAR instruments are now being used in operations based in the Caribbean and in the Bay of Biscay. Coupled with additional supporting sensors and its own ability to log data from external instruments and an interface to onboard instrument computing facilities, the ISAR provides a complete measurement package for a variety of seagoing experiments.

We present a complete review of the ISAR instrument system, first discussing the issues associated with measuring SSTskin from ships. A comprehensive description of the design and calibration of the ISAR instrument is then given. We use laboratory data to quantify the accuracy and performance of the ISAR instrument both before and after a 3-month deployment. We discuss an intercomparison with contemporary infrared ship-mounted radiometers at sea that provides evidence for the accuracy of ISAR SSTskin observations. We present an example dataset collected in European waters and discuss the new opportunities for validating satellite SSTskin retrievals that are now possible using the ISAR instrument. Finally, we outline our plans for further development of the ISAR instrument.

2. Measuring the radiometric temperature of the ocean surface

Figure 1 shows the geometry and spectral radiance components that must be considered when measuring SSTskin radiometrically. The infrared radiation emitted by the sea surface varies with the absolute temperature of the water surface. If the sea surface behaved as a perfect radiator, then the absolute temperature could be determined simply by measuring its spectral radiance L_λ over a finite spectral bandwidth, and inverting the Planck equation. Seawater has an emissivity slightly less than unity that varies with wavelength λ and the viewing angle from nadir θ. As a result, a small proportion of radiation originating from the (typically much cooler) atmosphere is reflected at the sea surface into the field of view (FOV) of the radiometer. If no allowances were made for reflected sky radiation, the resulting SSTskin retrieval would be too cool. To measure SSTskin accurately from a ship, radiometric measurements of both the sea surface radiance and the downwelling atmospheric radiance must be obtained and the value of seawater emissivity ɛ should be known accurately. In the 9–12-μm wave band ɛ has a maximum value (>0.98) when viewing a calm sea surface at θ < 40°. However, for θ > 40°, ɛ decreases significantly (Bertie and Lan 1996).

Consider a radiometer viewing a sea surface at temperature T_s and an incidence angle θ; the downwelling radiance incident on the sea surface can be found using

where τ_path is the transmittance along the atmospheric path and

which is the spectral radiance emitted by a blackbody (BB), and is the spectral radiance emitted at air temperature T_air, averaged over the atmospheric path. The upwelling radiance from the sea surface is given by

and the upwelling radiance arriving at the aperture of the radiometer is

As τ_path approaches unity, L_sea is given by

When the radiometer is less than 30 m above the sea surface and the relative humidity is below ∼95%, τ_path is very close to unity for measurements in the region 9.6–11.5 μm. This working assumption (τ_path = 1) in (5) introduces errors of less than 0.05 K in retrieved values of SSTskin and while this error is small, it is not insignificant when the goal is an uncertainty of 0.1 K.

The above analysis assumes that the ocean surface is flat and that the component of L_sky reflected at the sea surface into the radiometer FOV comes from a zenith angle θ. However, when the sea is rough, radiance from many parts of the sky can be reflected from suitably oriented surface facets into the radiometer field of view (Donlon and Nightingale 2000). This uncertainty is the subject of ongoing research into the variation of ɛ(λ, θ) with sea state (Watts et al. 1996; Wu and Smith 1997) and to proceed practically we assume the average reflected sky radiance to be that from the direction that reflects in a calm sea.

The radiometer combined detector and wave band filter (if used) spectral response function is defined as ζ(λ) in output units per unit radiance. The output signal S_sea of the radiometer when viewing the sea is then

where the limits of integration are chosen to span the bandwidth of the detector and filter as defined by ζ(λ). The output when viewing the sky S_sky is

Within the narrow wave band 9.5–11.5 μm, ɛ(λ, θ) and B(λ, T) vary only slowly with wavelength and so Eq. (7) can be separated, to a good approximation, into a combination of the band-averaged values ɛ_B(θ), L_sky, and B_B(T), giving

where

Equation (8) can be written

so that, finally

Measurements of the sea and sky radiance responses, S_sea and S_sky, which are ideally obtained almost simultaneously by looking downward at the incidence angle θ and upward at the zenith angle θ, are required to solve (11). We stress that the time difference between sea and sky measurements must be small to limit errors associated with rapidly changing atmospheric radiance conditions due to clouds of different species and height and therefore different radiative temperature (Donlon and Nightingale 2000).

As an engineering solution to this issue, we choose a rotating mirror to direct radiance onto a single detector while alternately viewing the sea surface and atmosphere at a variety of different angles. An alternative approach adopted by the CIRIMS radiometer (Jessup et al. 2002) utilizes two separate detector systems with carefully matched optical components: one to view the sea surface and the second to view the sky. Our approach has been proven by several existing radiometer systems including the SISTeR, M-AERI, and DAR011, and we believe this design is preferential to the use of two separate detector systems because

• a single optical pathway within the instrument guarantees an identical spectral measurement for all measurements;
• a common calibration is obtained for all measurements so that systematic measurement gains and offsets can be eliminated in a straightforward way;
• the system is cost-effective and robust, requiring only a single detector, one set of optical components, and one calibration system; and
• a compact instrument design is possible.

3. Description of the infrared SST autonomous radiometer

ISAR is a compact (570 mm × 220 mm cylinder) infrared temperature measuring system that employs two reference blackbody cavities to maintain the radiance calibration of a special Heitronics KT15.85D radiometer (hereafter simply called KT15) to an accuracy of ±0.1K. The ISAR instrument consists of the following subsystems:

• a fore-optics system to route target radiance to the detector,
• a detector and blackbody calibration subsystem,
• an internal control and data acquisition computer subsystem,
• an environmental protection subsystem incorporating a storm shutter, and
• an external RS485 interface to which additional atmosphere and ocean sensors (e.g., air temperature, solar radiation, and SSTdepth) may be connected, powered, and data collected.

Figure 2a shows ISAR in the laboratory with the associated optical rain gauge in front. Figure 2b shows them mounted on the bridge wing of the P&O vessel Pride of Bilbao.

e. Calibration of KT15 detector analog output

Calibration of the KT15 detector is required to compensate for the impact of the ZnSe window and gold mirror in the ISAR optical chain and to maintain the calibration of the KT15 itself. Each of the ISAR calibration blackbodies are positioned at the end of the ISAR optical path (Fig. 3) so that the radiometric impact of the ZnSe window and gold mirror are identical for all instrument views (i.e., both target views and BB views). During typical SSTskin measurement operations, the ISAR scan drum is programmed to measure four different radiances in a measurement scan sequence: L_scene(λ_o), the downward nadir angle view of the sea surface; L_atm,θ(λ_o), the upward azimuth view of the sky; L_bba, the radiance of the ambient temperature blackbody; and L_bbh, the radiance of the heated blackbody. Note that ISAR is capable of viewing many different user-defined target angles in a measurement sequence.

The voltage output of the radiometer is related to the incoming radiance integrated over the radiometer’s frequency pass band by a linear relationship. Radiometric calibration of the instrument consists of determining the linear relationship between detector output voltage, as measured by the 18-bit analog-to-digital converter, and the integrated incoming radiance. The ISAR calibration system is designed to minimize the impact of any nonlinearity in the radiometer characteristics by optimizing the onboard calibration system for the limited range of global ocean temperatures. Each of the blackbody calibration reference cavities is periodically viewed by the detector. One of these is allowed to “float” at the ambient instrument temperature, which is always close to the temperature of the sea surface, and the other cavity is heated to ∼12 K above the ambient temperature (i.e., warmer than the actual SST). With this arrangement the accuracy of ISAR calibration is best in the temperature range of the actual SST because linearity is only assumed over a small range of temperature; the effects of any nonlinearity in the system are minimized. Clearly the calibration is degraded at the much colder temperatures (∼180 K) recorded when viewing a clear dry atmosphere. As <2% sky radiance is typically reflected at the sea surface, the impact of calibration errors due to extrapolation on the overall accuracy of SSTskin determination is minimal even when considerable errors of >5 K are evident.

The signal measured by the KT15 when viewing the ISAR ambient blackbody (BB_amb) is given by

where T_amb is the thermometric temperature of BB_amb and the signal when viewing the hot blackbody (BB_hot) is given by

where T_hot is the thermometric temperature of BB_hot. The radiometric gain G of the system can then be derived from (12) and (13) using

where C denotes measured detector counts, B is the Plank function, T is the radiometric temperature of the ambient or hot blackbody cavities, and ζ is the ZnSe window filter function. By substitution, the calibration offset can be found using

This scheme assumes an emissivity of 1.0 for the blackbody cavities and that the millivolt output of the KT15 detector is proportional to the radiance. The radiance of each blackbody cavity is calculated using the temperatures measured by the embedded thermistors and KT15 instrument specific radiance-to-temperature and temperature-to-radiance functions based on spectral integration of the Plank function across the KT15 bandpass shown in Fig. 3.

In practice, the KT15 radiometer would require a view aperture of 100 mm to eliminate all stray radiance from the FoV despite the fact that the optical system uses a focused lens rather than the 10 mm used in the ISAR design. To compensate for stray radiance, we assume that all strays emanate from the ZnSe window and surrounding ISAR bulkhead material. The temperature of the main bulkhead in this area is measured by an accurate thermistor set close to the ZnSe window. The stray radiance is accounted for by adjusting the emissivity value of the calibration blackbody cavities and introducing a reflected radiance term derived from the main bulkhead temperature. Based on extensive laboratory calibrations using the Concerted Action to Study the Ocean Thermal Skin (CASOTS-II) reference cavity (C. J. Donlon and G. J. Fisher 2005, unpublished manuscript) and several separate ISAR instruments, we have determined that ɛ′ = 0.997. Then L_BBx becomes

where ɛ′ is the modified calibration BB emissivity and L_stray is the radiance derived from the ISAR window/bulkhead thermistor measurements.

Figure 8 shows the retrieved brightness temperatures when ISAR views the sky and sea surface together with the retrieved SSTskin during a typical deployment of ISAR in the Bay of Biscay. At the start of the record, clear skies and corresponding cool sky temperature values are found. Note the cool temperatures that are retrieved when viewing the sky. As clouds encroached, the sky brightness temperature increased from Julian day 78.8. from ∼190 to nearly 270 K. Under clear-sky midlatitude conditions, temperatures of 190 K are not uncommon with temperatures closer to the SSTskin for low cloud cover. There is often considerable variability in the sky temperature record as a consequence of small and broken cloud at different heights and therefore different temperatures. Because of this characteristic, we tend to measure the sky brightness temperature based on only a limited number of samples, but sample as often as possible. The bottom panel shows the direct impact of reflected sky brightness temperature at the sea surface. Under clear skies, a temperature deviation of nearly 0.5 K is apparent between the retrieved SSTskin and sea surface brightness temperature. This is reduced to ∼0.1 K as the sky temperature approaches 270 K.

4. Evaluation of the ISAR calibration system

Many factors contribute to uncertainty in the radiances measured by ISAR during deployment that influence the final value of SSTskin. These include the effect of high humidity variations, direct and rapid warming by solar radiation, electronic drift, specific deployment geometry, appropriate knowledge of seawater emissivity, and the algorithms used to derive geophysical parameters from engineering sensor outputs. Confidence in the instrument is improved by laboratory and field intercomparisons either with other instruments observing the same area of the ocean surface at the same time or calibration reference blackbody cavities (e.g., Rice et al. 2004).

Several ship-mounted radiometer intercomparison experiments have been conducted in the past (Donlon et al. 1999b; Kannenburg 1998) and, recently, the second Miami, Florida, radiometer intercomparison workshop brought together all of the major in situ infrared radiometer systems used in the world, including ISAR. A full account of the laboratory calibration experiments is reported by Rice et al. (2004). An at-sea deployment of all radiometers viewing the same area of the ocean surface during a 2-day deployment that traversed the Gulf Stream Current between Florida and Bimini Island is reported by Barton et al. (2004). This experiment concluded that the differences between radiometer measurements are at the ±0.1-K limit imposed by the measurement technique (Donlon and Nightingale 2000).

Although ISAR is a self-calibrating instrument, for the reasons stated above it is important to determine its accuracy with an independent reference blackbody over a representative temperature range before and after each instrument deployment to provide confidence limits for the data collected during deployment (e.g., Donlon et al. 1999b). Independent calibration using a reference blackbody source is essential to monitor any degradation of calibration due to the inevitable contamination of optical surfaces and the ISAR calibration blackbody cavities that are exposed to the marine environment. Ideally, the comparison should be conducted at several different ambient temperatures to simulate operation in different climates and atmospheric conditions. This can be accomplished using temperature controlled rooms, but difficulties are encountered at ambient temperatures below the local dewpoint due to condensation on the surface of the reference blackbody. A thin condensation layer will itself support a large skin temperature gradient and effectively decouple the measured radiance from the value derived by measuring the temperature of the cavity itself.

Figure 9 presents laboratory calibration of the ISAR instrument 03 (ISAR-03) just before and shortly after a 3-month autonomous deployment in the Bay of Biscay (reported in the following section). Calibration data were obtained using a CASOTS-II reference calibration blackbody (C. J. Donlon and G. J. Fisher 2005, unpublished manuscript). This blackbody reference cavity is similar to the design proposed by Fowler (1995) for the third-generation NIST reference blackbody and has an emissivity value of >0.999 using a 30% aperture reduction.

The CASOTS-II water bath temperature was measured with a Hart Scientific 1504 resistance bridge (serial: A1B256) with Thermometrics ES255 100-Ω platinum resistance thermometer (PRT; serial: 203). The PRT and resistance bridge were calibrated to NIST standards and have an accuracy of ±20 mK. The ISAR field of view was carefully aligned with the center of the CASOTS-II blackbody and the front face of the instrument was 6 cm from the CASOTS-II BB aperture using a specially designed jig. For each primary validation dataset, ISAR data are processed to provide temperature data using the same processing code that generates operational SSTskin data. In Figs. 9a,b, the temperature range of measurement is included and the difference between ISAR and the reference blackbody is plotted, respectively. Mean calibration statistics are included in Fig. 9b. Throughout the deployment, the ISAR scan mirror degraded slightly due to salt deposition (confirmed by visual inspection). Contamination of the mirror surface was estimated to be <3% of the mirror surface area. While there is a small increase in the system noise following deployment, there was no significant change in the ISAR-03 instrument calibration relative to the CASOTS-II blackbody. The mean difference between calibrations taken before and after the 3-month autonomous deployment is ∼60 mK. In fact, the mean difference between ISAR and the CASOTS-II blackbody is close to the calibration accuracy of the Thermometrics PRT.

The calibration data presented in Fig. 9 form an essential part of the quality control procedures that ensure consistency between different ISAR units and provide a measure of the absolute accuracy for each deployment. Based on 21 months of experience, it is recommended that each ISAR be serviced after 2–3 months of operation. Service should include a calibration followed by general cleaning and maintenance and a final independent calibration before redeployment. Calibration data from several deployments demonstrate a repeatable consistency in the calibration of the ISAR instrument even when deployments have included several days of extreme conditions and provide independent verification of the ISAR instrument accuracy according to the Global Climate Observing System (GCOS) climate monitoring principles (Global Climate Observing System 2004). In summary, ISAR is repeatedly able to measure the SSTskin to an accuracy of ±0.1 K during a 3-month fully autonomous deployment. In addition, such a consistency of calibration demonstrates the success of the ISAR instrument design, which provides an elegant solution for the autonomous measurement of SSTskin in an extremely challenging environment.

We have deployed an ISAR instrument alongside other radiometer systems on the U.S. research vessel Ronald H. Brown between 31 October and 23 November 2004 as part of a research cruise through the Gulf of Mexico from Miami to Panama, and then across the Pacific Ocean from Panama to southern Chile. Data were obtained near simultaneously from an ISAR instrument, M-AERI, CIRIMS, and a conventional SST2m measurement made using a pumped thermosalinograph (TSG) system. During this cruise a real-time satellite telemetry system was operated successfully and reported all data in real time. In this type of intercomparison slightly different viewing geometries, radiometer calibration configurations, duty cycles, and sky temperature retrieval schemes are used by each instrument. However, Table 2 shows a very small difference of the measured SST between the three instruments and confirms that the ISAR agrees with the other radiometric sensors within very narrow tolerances (less than 0.1 K) and the mean difference between ISAR and SST2m data is comparable to that in the literature (e.g., Donlon et al. 2002). This intercomparison provides a complete and independent end-to-end test of the ISAR system that confirms the performance and accuracy of the ISAR instrument against peer systems in the field.

At-sea comparisons such as the one described here are an essential component—possibly the only way to truly evaluate the uncertainty of any SSTskin measurement system. Even with a perfect instrument, major contributions to measurement uncertainty come from estimations of surface emissivity, which are functions of view angle, and the treatment of atmospheric emission reflections, and these effects are not part of a laboratory calibration.

5. Example ISAR observations

Since March 2004, two ISAR systems have been used to collect SSTskin data in a sustained manner aboard the P&O Ferries vessel M/V Pride of Bilbao (PoB), which operates in the English Channel and Bay of Biscay. The ISAR instrument functioned without intervention apart from sporadic visits to check the system and download data while the ship was in port. Measurements were collected as part of the core validation program for the Envisat AATSR. The PoB makes two return passages per week between Portsmouth (England) and Bilbao (Spain). An ISAR system was installed on the starboard bridge wing of the PoB with an uninterrupted view of the sea surface free of ship bow wave effects. The height of the instrument above the sea surface was 26 m. The ISAR was programmed to view the sea surface at an angle of 25° from nadir and the sky at an angle of 25° from vertical. For the sea view observation, 40 samples were obtained; for the sky view 10 samples were obtained and 30 samples were taken while looking at each calibration blackbody. The total measurement sequence takes 140 s. The sky view observations are made immediately after the sea view measurements to minimize differences in environmental conditions (mainly clouds). A value for seawater emissivity was taken from Bertie and Lan (1996) for an angle of 25°, appropriately weighted by the ISAR field of view, which has an elliptical footprint on the sea surface of ∼6.3 m × 7.0 m. During normal operations, the ship has a speed of ∼20 kt. A hemispherical longwave pyrgeometer and shortwave pyranometer were connected to the ISAR using the RS485 instrument port and all data from these instruments were logged by the ISAR system.

A typical composite of ship tracks, determined by the ISAR onboard GPS unit during the deployment of ISAR over the period of 20 June until 21 September 2005, is shown in Fig. 10. The ship track includes coastal, shelf seas, and open-ocean environments. The study region is characterized with relatively dynamic SST features in the western approaches, coastal upwelling off the northwest tip of France, and warm riverine freshwater in the coastal region off Santander. The region of Portsmouth and Southampton also has many dynamic features. Warmer SSTskin observations are recorded in the southern regions of the ship track in the southern Bay of Biscay and close to the Spanish coast. The PoB has been equipped with two SSTdepth measurement systems: a TSG unit taking an input supply at a depth of ∼5 m that is calibrated to an accuracy of ±0.001 K and a SeaBird SBE-48 hull-mounted thermistor also at a depth of ∼5 m calibrated to an accuracy of ±0.002 K.

Figure 11 shows ISAR SSTskin observations plotted as a latitude–time Hovmöller plot for the period 2004–06. Gaps in the data are due to bad weather including mist/fog rain and conditions when there are atmospheric aerosol loads high enough to trigger a signal on the optical rain gauge, ship refits, or datalogger problems. All data when the ship was in port have been removed from the figure. Large and regular temperature excursions due to the predominantly north–south orientation of the ship track are clearly visible. The SST in this region and time of year (mid- to end of summer season) has quite large variability with a peak season warming at Julian day ∼220, at which point the SST begins to cool. The record-breaking warm European summer temperatures of 2006 are clearly visible in this plot. The lack of data in the latter part of 2005 is due to the presence of fog and mist (not seen in 2004 or 2006), which prevented the ISAR from making SSTskin measurements. The record also has considerable finescale thermal structure that is mostly due to the natural variability of the SSTskin (order of ±0.3 K).

In general the data return is high and throughout the record the ISAR closes for many short periods of poor weather and continues to measure once environmental conditions have improved. These data have been collocated with the corresponding Envisat AATSR satellite SSTskin data within ±2 h of satellite overpass resulting in several hundred matchup data points available for validation studies. This is an important result and a major achievement, as few matchup data points from in situ radiometers are available in midlatitude areas that are dominated by highly varied weather conditions, limiting the opportunity for satellite infrared SSTskin retrievals due to cloud cover. It is easy to see that a research cruise planned several months in advance in this area could easily return very little data due to bad weather.

Figure 12 shows the computed ΔT_5m (ISAR − SST_5m) for 2005–06. SST5m was measured by a SeaBird SBE-48 hull-mounted thermistor unit corrected for a lagged response due to the ship’s hull of approximately 10 min. The time difference was computed automatically by first minimizing the differences between the ISAR SST records followed by computation of the mean time shift over a period of several days to account for a slow change in response due to seasonal SST variability. A mean ΔT_5m of −0.2 ± 0.2 K is found when considering the entire dataset, which is in reasonable agreement with the mean value of −0.17 K reported by Donlon et al. (2002). Diurnal warming events with ΔT_5m in excess of 1.5 K and cool skin at night in excess of −0.3 K (though never exceeding −0.5 K in this case) are visible. Low wind speed and high solar radiation conditions lead to these strong thermal stratification conditions that effectively decouple the SST_5m from the SSTskin observations. As the AATSR instrument retrieves an estimate of the SSTskin, these events highlight the importance of using SSTskin to validate the satellite retrieval. In addition, it is clear that diurnal warming signals must be accounted for (either by removing the effect from the data or by having a system that is capable of resolving diurnal SST signals) when merging satellite and in situ SST data together, or when assimilating satellite SST observations into numerical prediction models.

6. Conclusions

To meet the requirements of an infrastructure for the operational validation of satellite-derived sea surface temperature measurements, a new in situ ship-mounted radiometer system has been developed called the infrared sea surface temperature autonomous radiometer (ISAR). ISAR is a single channel self-calibrating infrared radiometer capable of determining SSTskin to an accuracy of zero bias ±0.14 K rms (when compared with two contemporaneous independent research-quality radiometer measurements obtained during at-sea intercomparisons and repeated laboratory calibrations using reference radiance targets) and can be deployed on volunteer ships of opportunity for periods of up to 3 months. The instrument is fully automatic and can be programmed to measure up to 10 user-defined views of the sea surface or atmosphere and does not require any operator during normal operation. It is capable of automatically protecting itself from rain and sea spray using a storm shutter system triggered by an optical rain gauge. Four ISAR systems are currently operational. We demonstrate the capabilities of the system using data obtained in the Bay of Biscay and English Channel collected as part of the Envisat Advanced Along-Track Scanning Radiometer (AATSR) satellite SSTskin validation program. Independent primary instrument validation data collected before and after deployment using a CASOTS-II reference blackbody demonstrate that the ISAR provides SSTskin data within the design specification of ±0.1 K. Comparisons of data obtained simultaneously with peer instruments during a cruise in the Gulf of Mexico and Pacific Ocean verify that ISAR is capable of making SSTskin measurements accurate to ±0.065-K rms, which is equivalent to accepted research-grade infrared spectroradiometer systems. In summary, ISAR has achieved its design goals and is repeatedly able to measure autonomously the SSTskin to an accuracy of ±0.1 K during 3-month fully autonomous deployments. In addition, a consistency of calibration demonstrates the success of the ISAR instrument design in the harsh marine environment. The ISAR provides a unique and elegant solution for the autonomous measurement of SSTskin and associated parameters derived from other instruments that could be logged by the ISAR system RS485 external port aboard volunteer observing ships.