To achieve the BSRN target uncertainty requirements of 2% or 5 W m^-2 (Ohmura et al., 1998; McArthur, 2005), measurements of downwelling SW irradiance on land-based platforms incorporate redundant measurements. An unshaded and ventilated secondary-standard pyranometer is deployed to measure the downwelling SW radiation (we will refer to downwelling SW irradiance measurements from a single pyranometer as global SW radiation going forward). Redundant measurements are also made from summing the measured direct normal component with a pyrheliometer and the diffuse irradiance measured with a ventilated shaded pyranometer. We will refer to this way of measuring the downwelling total SW irradiance as the component sum SW going forward. This implementation requires a motorized sun tracker that keeps the pyrheliometer sensor normal to the incoming radiation and the diffuse pyranometer fully shaded.

Since the unshaded pyranometer is affected by the cosine response error, thermal offset and tilting, downwelling SW irradiance derived by the component sum method described above is generally considered more accurate than the global irradiance measured by a single unshaded pyranometer, although recent secondary-standard instruments have excellent performance with respect to the cosine response. On a moving platform, maintaining a sun tracker is generally not feasible, due to platform motion. Even fixed towers tilt a few degrees in response to wave motion (e.g., di Sarra et al., 2019). Upkeep of moving parts in the corrosive sea salt environment and power limitations further limit feasibility of maintaining a sun tracker on ocean platforms.

Pyranometers measuring global SW have improved significantly over time, and though they do not meet the accuracy of a tracker system, some models come close. Broadly, the specifications for a research-grade marine pyranometer should meet the highest standard for calibration uncertainty; have minimal directional dependency; and a fast response time.

The ISO 9060:2018 standards define a spectrally-flat Class A pyranometer that multiple manufacturers can currently meet, providing a starting point to evaluate the current state of the art for accuracy of SW irradiance sensors. We consider a number of the criteria in the current standard and how they are important for measurements from ocean-based platforms. For the purpose of measuring the radiation energy budget accurately, global SW measurements should be made with a thermopile-based sensor rather than one using a silicon (Si) photodiode sensor, or what is considered a spectrally-flat pyranometer in ISO 9060:2018 standards. Si-based sensors have a non-linear spectral response for SW wavelengths between 0.3 and 1.1 μm, and do not measure the portion of the SW spectrum between 1.1 and 4 μm (e.g., Vignola et al., 2016, 2017), which introduces a poorly-constrained bias dependent on sky conditions (e.g., atmospheric water vapor content). However, we note that Si-based sensors have a very high sampling rate which might be useful for experimental studies on the impact of tilts and motion on measurements.

Pyranometer measurements should prioritize minimization of cosine response errors, the directional dependence due to non-linearities in the thermopile response to the incident angle of incoming radiation, ideally through choice of sensors or possibly by corrections (described more in Section 8.3). This is particularly important for high-latitude measurements where the solar elevation angle is always relatively low. This is one of the largest error sources in previous generations of pyranometers (e.g., Michalsky et al., 1995), and an area where manufacturers have significantly improved technologies in recent years (e.g., Vuilleumier et al., 2014). Figure 2 shows the sensitivity of pyranometer calibration factors (responsivities) to solar zenith angle for different pyranometer models. As is typical for older models of pyranometers, the Eppley Precision Spectral Pyranometers (PSPs) vary by about 8% over the course of the day, while newer models from Kipp and Zonen (CM21, CM22) vary by 1-2%. Pyranometers are typically calibrated at a solar zenith angle of 45°, in clear-sky conditions, as shown by the blue vertical line in Figure 2 . Thus, a pyranometer is generally most accurate at solar zenith angles of 45° and during overcast conditions when diffuse radiation is dominant, with positive and negative deviations from the cosine response partially balancing each other over the course of diurnal and seasonal cycles; the degree to which this cancellation occurs depends on deployment latitude.

Differences between downwelling SW measured by an unshaded pyranometer and the component sum can be easily found at land-based BSRN sites (Gueymard and Myers, 2009). A preliminary analysis comparing measurements from unshaded pyranometers and the component sum from BSRN data archived over the period 2000-2018, showed a median monthly RMSD of ~7 W m^-2 with first and third quartiles of ~4 and ~12 W m^-2, respectively, over all stations and months in the timeseries. The comparison was calculated for 1-minute measurements for each month and station in the archive. The average bias was nearly null, with first and third quartiles of -2 and +2 W m^-2, respectively. A range of instrumentation is used at BSRN stations, including pyranometers that meet ISO 9060:2018 Class A standards and older instrumentation with larger cosine errors.

Figure 3A shows NASA’s CERES SYN1deg data product (Rutan et al., 2015) calculated (hourly) SW irradiance down at the surface minus US DOE Atmospheric Radiation Measurement (ARM) station (SGP E13) observed values from component sum (black) and unshaded pyranometer (red) during clear skies. Results in 3a) show an overall positive bias (<1%) that linearly increases with respect to total SW as the cosine of the solar zenith angle [cos(SZA)] increases. However the calibration of the Eppley pyranometer to 45° solar zenith angle introduces non-linearity into the comparison resulting in larger negative biases for cos(SZA) > 0.8. Buoys historically, and in the foreseeable future, will likewise have unshaded pyranometers. Figure 3B shows calculated SW irradiance compared to buoy observations for the same clear sky conditions though collected from 46 different buoys. Overall bias is over 2% with significantly higher random error (RMS >10%), most likely associated with satellite cloud retrievals over tropical oceans using geostationary imagers and buoy motion. Again, the polynomial fit is significantly concave. These results are for hourly, ‘clear sky’ comparisons and suggest care should be used at high time resolutions as unshaded pyranometers give more meaningful values for daily averages and beyond. Likewise, the climate regime of a buoy may affect temporal averages if the buoy sits in a predominantly clear sky location. Indeed, many buoys in Figure 3B are located in tropical subsidence zones. Modern pyranometers have greatly reduced this solar zenith angle dependence though this is not the case for the historical record.

IR loss should also be considered as a potential systematic bias in measurements. While ventilated radiometers may have a reduced IR offset compared to non-ventilated ones, the thermal offset varies among different models of pyranometers and with environmental conditions and should be taken into account and properly corrected (Dutton et al., 2001; Gueymard and Myers, 2009; Wang et al., 2018). It is recommended that the thermal offset be mitigated if possible through use of radiometers with minimal offsets and ventilation when appropriate. However, since ventilation is difficult to maintain for pyranometers installed on many ocean platforms due to limited power availability, and funding for new instrumentation that mitigates thermal offsets may not be available, the thermal offset may represent a non-negligible bias in SW irradiance measurements. In these cases, correction of thermal offset in a SW radiometer should be applied, which requires simultaneous measurements of raw signals from a LW radiometer, reinforcing the need to include both sensors in the instrument suite. Methods for correcting for thermal offset are described in Section 8.2.

Response time is an important consideration on moving platforms, particularly when motion correction is applied to a measurement. Using an instrument with a faster response time than the 10-s minimum response time (time for 95% response) in the ISO 9060:2018 Class A standard is possible and desirable. Current thermopile-based pyranometers are available with a 3-s response time and could be created with less than 1-s response time (William Beuttell, personal communication). The response time of the pyranometer should be sufficiently fast to capture the sampling requirements of the scientific use of the measurements, which will be discussed more in Sections 3.2 and 7.4.

Finally, in the previous generation of instrumentation, sensor degradation was linked to degradation of the black coatings used on the thermopiles (Wilcox et al., 2001; Riihimaki and Vignola, 2008) causing potentially rapid drift from the initial calibration. It is recommended not to recoat sensors in response to sensitivity decline, as this decline happens most rapidly in the first year so recoating the sensors leads to a greater instability in the measurements. The sensitivity decline should be handled by further calibration when possible, otherwise it is recommended to retire the instrument from operations. Newer instrumentation with non-organic coatings do not experience the same rate of decline in sensitivity. Instrument manufacturers consulted here reported that sensor degradation is not likely to be significant after a potential sensitivity decline of ~0.5% in the first year of use. This stability is a specification listed in the ISO 9060:2018 standard which manufacturers need to meet. Current BSRN standards recommend regular calibrations (every 1-2 years). For older sensors this is necessary to maintain accurate calibrations. While newer sensors may not require such frequent calibrations for stability, regular checks also help identify any other instrument changes or repairs needed. Thus, currently regular calibrations (every 1-2 years) against a reference standard traceable to the World Radiometric Reference (WRR) are recommended following BSRN recommendations, as discussed in Section 9. We also recommend further conversations between the ocean community and BSRN to determine optimal calibration procedures for accuracy and practicality .

When a pyranometer is not level it can change the apparent zenith angle of the direct solar radiation beam and thus alter the measurement significantly, particularly under clear skies and low sun angles when the aspect of the tilt is aligned with the direction of the sun. This introduces a challenge for measuring downwelling SW irradiance on moving platforms and care should be taken to mitigate the impact on measurements.

Platforms at sea can have both persistent tilts and varying tilts due to the impact of waves and swell on the platform. Changing cargo loads, fuel, and ballast tank levels, and trimming of the vessel can be the source of persistent tilts on ships that vary during a cruise. Surface buoys with wind loading and or drag forces on the hull and mooring line may have persistent tilts as well as pitch and roll. Systems on sea ice may be impacted by persistent, abrupt, and/or slowly-varying tilts from ice freeze/melt and dynamic activity.

We discuss here the three most common ways to deal with platform tilt and motion: averaging, stabilized platforms, and tilt correction post-processing. Which one is most appropriate for a given application depends both on deployment limitations and scientific priorities.

The wave motion that influences ocean platforms during calm sea conditions is fairly regular with a period on the order of 1-20 seconds (e.g., Toffoli and Bitner-Gregersen, 2017). If the desired scientific need of the measurement is for longer-term averages, such as daily or monthly average data from fixed towers or moored buoys for comparisons with model or satellite data, much of the variability caused by platform motion can be treated as a random fluctuation that will approximately cancel out with averaging (e.g., Colbo and Weller, 2009; di Sarra et al., 2019). A mean tilt to the sensor for certain wind and wave conditions, or due to the imbalance of the platform, however, will not average out and instead remains a systematic error or bias in the measurements. MacWhorter and Weller (1991) showed that a mean tilt of 10° can cause an error of 40% in the SW irradiance. If averaging is used as a means to mitigate platform motion, then a method is needed to also characterize any impact from mean sensor tilt relative to the direction of incoming solar radiation. The other two methods for tilt mitigation require additional instrumentation or equipment and may be cost-prohibitive or otherwise technically infeasible on some platforms. The unquantified systematic and random errors caused by waves and mean tilt may well make measurements unusable for some data users, such as the satellite community that needs measurements of known uncertainty for validation of derived products at the surface. Thus, the community recommends that best practices for characterizing mean tilt be developed through future intercomparison experiments .

Additionally, for averaging to be effective, the sampling must be of high enough resolution to subsample the period of motion and not selectively oversample one direction of motion over others. Both the instrument response time and the sampling frequency need to be considered when averaging. This is discussed further in Section 7.4.

Averaging may not provide sufficient accuracy for some scientific purposes or platforms, e.g. the need for higher temporal resolution data, such as 1-min average data needed for studying cloud effects, and platforms such as aircraft or USVs where substantial platform motion or mean tilts are too great for averaging to provide sufficient correction. Other mechanical or advanced post-processing methodologies are needed in these situations.

On platforms with sufficient power and weight capacity, such as a ship or larger aircraft, a mechanically stabilized platform can be used (Wendisch et al., 2001; Bucholtz et al., 2008). These platforms can keep an instrument level, compensating the ship motion, though active motion compensation may be more challenging in rough seas. Stabilized platforms can also be costly, and limited in the motion they can correct for so may not be practical in all deployments or wave conditions.

Another possibility is to use simultaneous measurements of sensor orientation and direct SW irradiance to calculate a correction to the downwelling SW. Because the orientation of the sensor primarily changes the measurement of the direct beam and not the diffuse irradiance, correcting downwelling SW irradiance requires knowledge of the separation of the irradiance components. A calculation described by Long et al. (2010) assumes that diffuse irradiance is isotropic and unchanged by tilt which they showed was a valid assumption when the instrument was within 10° tilt from horizontal. Equation 1, shows how to calculate total downwelling horizontal SW (G) from tilted measurements of downwelling SW irradiance (G_T), the cosine of solar zenith angle ( μ o ), cosine of the tilt angle ( μ T ), and the ratio of diffuse to direct normal irradiance (K).

The tilt angle is defined as the angle between a normal to the tilted surface and the incoming direct beam, as illustrated in Figure 4 . Details of this derivation are found in Long et al. (2010). This correction has been successfully applied to the downwelling SW measurements from USV saildrones (Zhang et al., 2019), where the SW irradiance is measured at 5 Hz by an SPN1 Sunshine Pyranometer and the three-axis motion of the platform is measured at 20 Hz by a dual GPS-aided IMU VN-300.

In order to use the post-processing method of SW irradiance measurements, measurements of the partitioning between diffuse and direct SW radiation is necessary. Additionally, measurements of diffuse and total downwelling SW irradiance together can be used to estimate clear sky irradiances (Long and Ackerman, 2000), and cloud fraction (Long et al., 2006), and thus give more information for studies of cloud radiative effects and forcing. Also, the quantification of the diffuse and direct contributions helps separate the impact of clouds and aerosols on surface irradiance (e.g., Qiu, 2001; Riihimaki et al., 2009).

An example of an instrument capable of measuring the diffuse downwelling SW component on moving platforms including aircraft, ships, and an autonomous vehicle (Long et al., 2010; Zhang et al., 2019; Gentemann et al., 2020) is the SPN1 radiometer (Wood, 1999). The SPN1 instrument uses a fixed shading device that always shades at least one of seven sensors and keeps at least one sensor unshaded (see photograph inset in Figure 5 ), and thus has no moving parts. Another strength of the SPN1 instrument for moving platforms is that the small thermopile sensors have a fast response time of 200 ms (Delta-T Devices Ltd, 2019), which is sufficient to sample the impact of wave motion on the instrument’s effective zenith angle. The instrument design also minimizes IR loss, keeping offset errors to 3 Wm^-2 or less (Delta-T Devices Ltd, 2019).

However, due to the configuration and spectral range of the SPN1, step functions on the order of 10-20 W m^-2 can occur over the course of the day as different sensors become shaded and unshaded ( Figure 5 ). Badosa et al. (2014), also found a 5-10% low bias in the diffuse SW radiation measurements depending on the conditions (e.g. Figure 5C ). Thus, for the highest quality measurements, it is recommended that when a lower accuracy instrument is used to measure diffuse irradiance, a Class A pyranometer be fielded alongside as described in Long et al. (2010).

In addition to the SPN1, several deployments have used a fast-rotating shadowband on broadband and filter radiometers on ships, without the need for a stabilized platform (e.g., Reynolds et al., 2001; Witthuhn et al., 2017), though none were commercially available at the time of publication.

To summarize the recommendations for the selection of SW instrumentation, we have created a decision tree ( Figure 6 ). Class A pyranometers refer to spectrally-flat Class A pyranometers according to the ISO 9060:2018 standard. There are a number of pyranometers that meet this standard, with different specifications and sizes as well as power requirements. A list of some commonly used pyranometers in the BSRN network are listed in Table 2 to give readers an idea of the potential range of specifications for measurements meeting this standard, though this list should not be considered exhaustive.

For the purposes of calculating air-sea fluxes, Bradley and Fairall (2007), recommended calculating upwelling SW irradiance using a constant broadband albedo of 0.055. The date, time, and latitude-based albedo parameterization of Payne (1972) gives a better representation of albedo as a function of solar zenith angle and wind speed than a simple fixed value, and is now implemented in the forthcoming COARE version 4.0 bulk flux algorithm (James Edson, private communication).

Jin et al. (2004) developed a parameterized model for ocean surface albedo from observations taken during the CLAMS experiment (Smith et al., 2005) and a coupled ocean/atmospheric/sea ice radiative transfer model (Jin et al., 2006, 2023). The parameterization can be used to derive broadband ocean surface albedo as a function of the cosine of the solar zenith angle, wind speed, aerosol/cloud optical depth, and chlorophyll concentration. Hogikyan et al. (2020) compared albedo calculations from the CERES and ISCCP satellite products with a constant 0.055 albedo at equatorial, sub-tropical, and sub-polar buoy locations in the Pacific Ocean. The CERES dataset (based on the Jin et al., 2004 model) and the ISCCP model both include solar zenith angle, but use different inputs and models to describe sea surface conditions (e.g., foam, spray, glint, waves) as well as clouds and aerosol. Both models show reasonable agreement with a constant 0.055 albedo value in the tropics, but large spatial and temporal variability in surface albedo exists beyond simple zenith angle dependence (Hogikyan et al., 2020). A similar model to the Jin parameterization was recently developed by Wei et al. (2021). The model compared well with observations from the Chesapeake Light Tower and CERES ocean surface albedos and is optimized for the Rapid Radiation Transfer Model (RRTM) (Clough et al., 2005) which is the radiative transfer model used in a number of current GCMs. Nonetheless because of the complexity of the ocean surface there is a need for further work to test and improve albedo parameterizations for use in these calculations ( Cronin et al., 2019 ).

Rather than measuring upwelling LW, a more realistic approach for many platforms is to estimate upwelling LW from measurements of downwelling LW plus either the measurement of surface skin temperature or its estimate from a subsurface measurement used as input in an algorithm that accounts for cool skin and warm layer effects (Fairall et al., 1996, 2003). Bradley and Fairall (2007) recommend using an emissivity ( ε ) of 0.97 and calculating upwelling LW using Equation 2:

where σ is the Stefan-Boltzmann constant, T s is the ocean surface skin temperature (in K), and L W u p and L W d n are upwelling and downwelling LW irradiance, respectively.

This calculation requires accurate skin temperature estimation or measurements (Donlon et al., 2014). Ideally, the skin temperature is measured directly with downward looking radiometers, or IR thermometers that are corrected for reflected radiation by a separate upward looking device or the same device that is occasionally rotated to look upwards (Donlon et al., 2014). More typically, a thermistor is used to measure the temperature at some depth. Thermistors that can be towed very close to the sea surface (e.g., a sea-snake) require an adjustment for cool skin (on the order of 0.25°C). Thermistors at depth (e.g., from surface moorings or drifters) require correction for diurnal warming and then adjustment for cool skin (Fairall et al., 1996; Marullo et al., 2016). A vertical array of temperature sensors may constrain estimates of the warm layer but not the cool skin since it is very shallow (<= 1 mm).

The location of instrument installation is critical and should be chosen to reduce or eliminate shading and thermal emission of platform structures and minimize radio frequency interference from antennas and other electronics. Sensors should be positioned on the highest point possible to avoid shadows and IR heating. On ships, it is also recommended that radiometers be placed forward of the stack, as stack gas can be sufficiently warm to produce IR radiation in a measurable range. However, the highest level on a platform is also often desirable for other meteorological instruments that are particularly susceptible to flow distortion, such as anemometers and rain gauges. If space constraints make it impossible to avoid having objects in the field of view of the radiometer, it is recommended to take into account the cosine response of the sensor (i.e., have the object as low in the radiometer’s field of view as possible) and consider the reflectivity/emissivity of the object. For instance, one can calculate the impact of structures on LW measurements using the formula in (Bradley and Fairall 2007, Appendix C). They show diagrams of potential radiometer placement on board ships that include considerations that minimize shading/heating, but also keep instrumentation accessible to technicians who clean and maintain them. If it is not possible to eliminate the influence of platform structures, another way to mitigate this problem is by installing a second radiometer that will be shaded by platform structures at different times and combining the two datasets. If shading cannot be avoided, compromised SW data should be removed from the dataset when possible.

Additionally, as sensor leveling is critical, care should be taken to align the sensor with the waterline to remove mean tilt biases. When calculating a tilt correction, as described in Section 3.2, the sensors should be aligned with respect to what the tilt sensors call level.

When considering locations for radiometer, particularly on ships or large fixed ocean platforms, it is possible to conduct Electromagnetic Interference (EMI) testing to determine the “cloud” of EMI influence by other systems on the platform. Such testing can determine the level of EMI between nearby satellite, radio, or other EM emitting systems at the site proposed for the radiometers. If EMI is moderate to severe, enough to break up or mask the radiometer signal, then an alternate location for the radiometers should be chosen.

The BSRN standards dictate using a fan to ventilate pyranometers and pyrgeometers to minimize IR loss biases and reduce dust, frost, and condensation on domes. Using ventilators is impractical on many ocean platforms due to power and space limitations. However, the use of ventilators on platforms where space and power are not limiting factors, such as ships, should be further explored, especially ships that frequent high-latitude locations where heating and ventilation may prove necessary in freezing conditions. If ventilators are not used, some field experience suggested that removing the sun shields may be advisable to reduce uneven heating of radiometer bodies at low sun angles. Tests are recommended to determine the relative impact of these three configurations: ventilated, unventilated with a sun shield, and unventilated without a shield to make further recommendations about when these configurations are advisable .

In high-latitude locations, such as over sea ice, where freezing conditions are common, ice-mitigation is needed to avoid instantaneous errors of +/- 100-200 W m^-2 (SW) and +60 W m^-2 (LW) (Cox et al., 2021). Mitigation is typically achieved with a combination of heating and ventilation, but if carefully-designed, ventilation alone is sufficient, which reduces power consumption (Cox et al., 2021).

Some groups use off-the-shelf (i.e., commercial) instrumentation and data loggers to acquire output directly from instrumentation, while others use modified electronics and instruments for better performance in ocean environments. This is particularly true for power-limited, unattended platforms like buoys and USV.

Thermopile output voltages from both SW and LW radiometers are on the order of 10 µV per W/m² of incoming radiation or less, thus these small voltage signals need to be amplified prior to digitization. Due to the smaller signal out of a thermopile for LW measurements, there is a greater need to quantify and control the stability and accuracy of amplifiers and other signal conditioning electronics than with pyranometers. Radio frequency interference can also introduce noise to the signal. When building electronics to amplify and digitize thermopile output for datalogging, it is recommended to keep the amplifier and digitization near the transducer . Inline amplifiers in the signal cable should be avoided.

Some analog LW radiometers used internal circuitry to combine case and dome thermistor readings with thermopile output to provide a voltage as a measure of the downwelling LW. Users found that recording the thermistor readings and the thermopile output for use in computing downwelling LW yielded more accurate results.

Marine use exposes radiometers to salt spray and visits from sea birds which leave guano behind. Some users replace the stock housings (holding the thermopile and dome) with more corrosion resistant material (for example, replacing aluminum with stainless steel) and/or apply coatings to exposed metal surfaces. As these changes may influence the solar heating and heat held by the housings, it is important to quantify any change in radiometer performance by comparison with unmodified standards. To reduce birds roosting on radiometers, vertical rods have been added to a ring fitted around the radiometer case (Figures 7B, C); if this is done the impact of shadowing and of infrared emission in the near field of the radiometer should be assessed.

Modifications for ventilation to reduce thermal gradients across the radiometer and for mitigation of dew/frost on the dome have been made and should be assessed for impact on performance and calibration.

Additionally, vendors are now marketing digital pyranometers and pyrgeometers and we think it would be valuable to test these for ocean use. A major recommendation of the surface radiation community working group at the 2020 OBPS workshop was to work with manufacturers to standardize instrument modifications for widespread marine application (Cronin et al., 2021).

More discussion in the community about best practices for instrument modification related to developing electronics for low-power environments and adequately sealing instrumentation and electronics for robustness in ocean environments is recommended, including the ability to pass along these needs to instrument manufacturers, who may be able to provide custom-built instrumentation for marine environments, making more accurate measurements more accessible and reproducible .

BSRN standards for land-based irradiance measurements require 1-minute averages (minimum, maximum and standard-deviation) of 1-second samples, and recommend that operators record the raw 1-second data for reprocessing purposes, if possible. Sampling rates on ocean-based platforms are not standardized, and it is recommended that sampling-rate standards be developed for better consistency across platforms .

Two fundamental scientific frequencies need to be taken into account in order to determine appropriate sampling rates: the typical period of wave motion (on the order of a few seconds), and the typical timescales of growth and dissipation of convective clouds (on the order of minutes).

If a tilt correction is going to be applied, the sampling rate needs to be at least twice the frequency of platform motion, and coincident with measurements of pitch, roll, and heading. As periodic wave motion in calm seas typically has a period of 1-20 s (e.g., Toffoli and Bitner-Gregersen, 2017), the sampling rate will generally need to be greater than 1 Hz for tilt correction. As discussed in Section 3, it is important to use a SW sensor with sufficiently fast time response to capture this motion. The corrected high-temporal resolution data can then be averaged to a more practical resolution like 1-minute data for most purposes, which also should capture changes in radiation due to cloud evolution overhead.

If the primary scientific scope of the measurements is long-term averages, the sampling rate needs to be sufficiently high to not alias the measurement by selectively sampling a given orientation relative to the wave slopes. In this case, an instrument with a longer temporal response may naturally perform some of that averaging if its temporal response is slower than the platform’s motion that houses the instrument. Sensitivity tests of the sampling rate with instruments of different temporal response could be done to help determine requirements for sampling rate .

Three recommendations for maintenance in marine environments were deemed particularly important by the working group. First, instruments need to be packed with care so that the domes are not broken during transport as damage to domes will compromise the measurements (e.g. Figure 7A shows the results of improper packing).

Second, to take good measurements, moisture within pyranometers and pyrgeometers needs to be kept to minimal limits. In attended instrumentation, such as on ships, this can be accomplished by changing desiccant regularly. On unattended platforms, instruments must be robust enough to seal out all moisture for extended deployments. This may require instrument modifications if the desired instrument does not meet this criterion.

Finally, cleaning of sensors from dust, aerosols, salt deposits, bird droppings, and biofouling is important to maintain good measurements. On ships and other attended platforms, daily cleaning is recommended, with weekly cleaning at a minimum. Cleaning can be done with a soft cloth or delicate wipes, using distilled water in warm temperatures, or with alcohol when temperatures are cold or for soiling due to organic or oil-based substances. More details are included in the Supplementary Materials , under ship-based recommendations.

On unattended platforms, such as buoys, cleaning may only be able to be performed during deployment of instruments, or otherwise naturally when rain occurs. Since rain occurs at different intensities and frequencies across the globe and throughout time (annually, seasonally, subseasonally, diurnally), cleaning by rain cannot be counted on all the time or at all locations. Redundant measurements can help assess some quality control issues such as biofouling or a bird sitting on an instrument that typically do not impact both instruments simultaneously. Additionally, some automated cleaning systems are being designed for fixed-platforms, and it is recommended to test these systems for their ability to keep domes clean and give more insight into the magnitude of this cleaning error on systems that can’t be cleaned .

Power limitations may make it impractical to deploy automated cleaning systems on many unattended platforms. In these cases, it is important to be able to quantify the possible error associated with dome soiling. One practice that can help quantify this error is to calibrate instruments before and after cleaning at the end of a deployment (Colbo and Weller, 2009). This may not be possible for drifters which are typically not recovered post-deployment. Waliser et al. (1999) investigated the effect of dome cleaning on SW irradiance measurements in the North Atlantic Ocean. They recalibrated the SW radiometers after substitution on the buoys following 8-month unattended deployments before and after cleaning, and found differences smaller than 2%. This can be done with side-by-side tests using instrumentation on a ship when buoys are being serviced, or done at a calibration facility after radiometers have been recovered from the field.

These tests alone are not likely to fully quantify the errors as sensors are subject to changing conditions such as rain and contaminants don’t necessarily build up at a measurable or predictable rate over time. Observation sites which are not far from stations where reference measurements are carried out on land, such as for the Lampedusa observatory (e.g., di Sarra et al., 2019), give the opportunity to investigate systematic differences. Di Sarra et al. (2019) compared SW irradiance measurements made at the Lampedusa observatory with those made on the island of Lampedusa, at 15 km distance, where regularly cleaned instruments are operational. They found that the average effect of dome cleaning on ocean observations at Lampedusa is negligible. Some land-based tests also showed fairly negligible (less than 1%) changes in SW irradiance measurements from pyranometers when assessing 2 years of data from 25 sites that were cleaned every 2 weeks (Myers et al., 2001). Geuder and Quaschning (2006) tested the impact of pyranometer soiling in dry conditions with more mineral dust deposition and infrequent rain for periods of time up to over 100 days and found an average of about 1% error due to lack of cleaning, though with individual cases with errors of up to 5% from heavier soiling, and one case with a 17% error due to bird droppings on the pyranometer dome. Foltz et al. (2013a) studied the impact of dust deposition on measurements made on a moored buoy west of tropical Africa, where intense desert dust transport and deposition occurs. They found a negative bias in SW irradiance with respect to satellite analyses and model calculations, and estimated a 14% decrease in SW irradiance produced by dust deposition by comparing a freshly calibrated sensor with a dirty one removed from the buoy. Foltz et al. (2013a) also used precipitation measurements on buoys as an indicator of when domes would have been cleaned by rain. This might be a potential way to evaluate measurement uncertainties in locations with high soiling.

For the case of instrumentation that is unattended in the field for long periods of time, such as on buoys, care should be taken to check the impact of environmental degradation on the measurements. This information can help better quantify achievable measurement uncertainty in the field.

When possible, comparisons against freshly calibrated instruments on the same or nearby platforms should be performed before unattended instruments are retrieved for maintenance and calibration. This gives a field comparison against a well-maintained instrument. This can be done with overlapping deployments on ships or new surface moorings alongside those that have been in the water for an extended period.

Additionally, calibrating instruments retrieved from the field before any cleaning or maintenance is done can give an estimate of the impact of soiling. The condition of the instruments should be documented and preserved, including material deposited on the dome.

Broad checks can be made with data in the field to ensure values are reasonable. Measurements should be taken 24 hours a day when feasible and nighttime offsets examined to look for reasonable values. Because of the thermal offset experienced by some pyranometers, values will often be negative at night, particularly under clear skies when the temperature difference between a clear sky and the thermopile is greatest. The magnitude of these negative offsets depends on the atmospheric conditions and the instrument in use, but generally is no more than a few W m^-2 in newer pyranometers that meet ISO 9060:2018, Class A standards. For older generation instruments like the Eppley PSP, these offsets may be larger (up to 10 Wm^-2 or even larger).

Measurement values should be checked using expected ranges of values for each variable. While these limits can be set more specifically for a given location and climatology in post-processing, as described in Section 8.4, in general the downwelling SW irradiance values should be 0 at nighttime, average to no more than 1200 W m^-2 at solar noon. On clear days, SW irradiance measurements should exhibit a smooth diurnal curve. On cloudy days, SW values should generally decrease compared to clear skies, though cloud effects on partially cloudy days may cause downwelling SW measurements to exceed 1500 W m^-2 for short durations up to 10-20 minutes.

Downwelling LW irradiance values can range from 40 to 700 W m^-2 (Long and Shi, 2008), although smaller realistic ranges can be defined at specific latitudes, with higher values in warmer, humid locations like the tropics and lower values in dryer, colder locations like polar regions. Downwelling LW values should increase when the sky is cloudy (particularly in the presence of low, optically-thick clouds).

It is technically possible to install LW and SW radiometers for which the sign of the recorded voltage is opposite the intended measurement but the instrument is working properly. It is good practice to reconcile the expected actual signs of the voltages with expected radiation values at radiometer installation. As discussed in Sections 7 and 8.4, it is advantageous to deploy duplicate SW and LW sensors on the same platform or in overlapping deployments of new and existing to-be-recovered moorings so that outputs can be checked and corrected if needed.

Finally, it is recommended that the community create climatologies or climatology tools to show typical values by location, time, and relationships between variables to make it easier to quality control data consistently .

Negative offsets in pyranometer measurements caused by IR loss by the thermal exchange between an all-black detector and the atmosphere can be reduced through instrument engineering or removed through post-processing adjustments to the data. IR loss is the cause of negative pyranometer output at night, though it can also lead to a negative bias during the day. When simultaneous measurements from a pyrgeometer are available, this bias can be corrected. Several calculations are available for this correction. In the previous version of the BSRN manual (McArthur, 2005), no consensus was reached about which correction method should be used, though it was recommended that regardless of the methodology used, adjusted measurements should be thoroughly checked at each installation. We recommend working with the BSRN community to determine when and what IR loss corrections should be applied for the next manual .

One widely used estimate of thermal offset uses measured pyrgeometer detector flux, case, and dome temperatures (Dutton et al., 2001). This correction determines regression coefficients ( b 0 ,   b 1 ,   b 2 ) empirically by fitting pyranometer and pyrgeometer measurements at night, when solar zenith angles are larger than 108° (Equation 3).

Where o s is the offset in W m^-2, N e t I R is the flux measured by the pyrgeometer detector (the irradiance signal that comes from the pyrgeometer alone without correction terms from case or dome thermistors), σ is the Stefan-Boltzmann constant, and T d o m e and T c a s e are the dome and case temperatures of the pyrgeometer in K. For some models of pyrgeometer only the case temperature is measured and the dome temperature is assumed to equal the case temperature so the last term in the equation is effectively zero.

Other simpler forms of the offset corrections correlate the pyranometer offset with the pyrgeometer detector flux only (e.g., the detector only correction in Younkin and Long, 2003), or with the pyrgeometer IR signal (e.g., Wardle et al., 1996).

Much of the research on IR loss correction was done using the previous generation of pyranometers. A recent study by Wang et al. (2018) examined the impact of applying a thermal loss correction on newer and older generation pyranometers, ventilated and unventilated. For newer pyranometers, the average thermal offset was quite small, generally<2 W m^-2. The study examined the magnitude of the deviation of global pyranometer measurements from downwelling SW irradiance measured by the component sum method. Instrument-specific recommendations were then given for the best thermal offset correction method, including that described in Dutton et al. (2001), a detector-flux only correction, or no correction.

The cosine response of a SW pyranometer can be significantly non-linear and yield significant errors when using a single calibration value, due to the dependence of the measurement on the direction of illumination. This is particularly true of older generation pyranometers. Newer pyranometers have more fully minimized and quantified this error so that, for example, pyranometers meeting Class A specifications given in ISO 9060:2018 should have cosine response errors<10 W m^-2 (as discussed in Section 3.1).

One possible approach to mitigate this error in downwelling SW measurements is to calibrate pyranometers with respect to solar zenith angle. However, the instrument cosine response is dependent on sky conditions so it isn’t straightforward to determine how to apply that calibration. When a single calibration coefficient is used, set at 45° incidence, the pyranometer will overestimate solar irradiance during clear sky conditions when the solar elevation is low (i.e., near sunrise and sunset) and underestimate irradiance when the solar elevation is high. This can be corrected to some extent if it is characterized as illustrated in Figure 2 . However, this correction is only strictly valid for clear sky conditions when the majority of the signal comes from the direct normal irradiance. Diffuse radiation generally does not have a directional component, and a calibration at 45° is more accurate for these times (see argument in Vignola et al., (2017), Chapter 6). Therefore, for an accurate correction, it must be determined whether there is a direct component to the measurement. Most land-based measurements do not use a zenith angle dependent correction because pyranometer-measured downwelling irradiance is considered a secondary measurement and more accurate measurements can be made with the sum of the direct and diffuse measurements.

More work should be done to test the impact of a solar zenith angle dependent calibration for ocean platforms, to fully understand under which conditions it would be recommended .

Estimates of clear sky irradiance at a given time and location can help with both quality control and scientific interpretation of the data. A number of methodologies exist to calculate clear sky irradiance, particularly for SW; some discussion of their accuracies is given here to help direct the community towards the most useful models.

The first factor that is needed in most clear sky SW irradiance methods is an accurate calculation of the solar position for a given place and time (e.g., ephemeris calculations of solar zenith angle, elevation angle, and azimuth angle). As a convenience to direct the community to useful calculations, we suggest a few models that are of known accuracy, though this is not meant to be exhaustive. Michalsky (1988) gives a model with stated azimuth and zenith angle accuracy of 0.01°, and includes printed fortran code in the appendix. An implementation of a solar position algorithm in C (https://midcdmz.nrel.gov/spa/), with an accuracy of 0.0003°, is described by Reda and Andreas (2004). This algorithm is widely used, and has been implemented in commonly used tools like pvlib available for Matlab and Python (Holmgren et al., 2018; https://pvpmc.sandia.gov/applications/pv_lib-toolbox/) and the Python library pysolar (Stafford et al., 2021; https://pysolar.readthedocs.io/en/latest/).

With an accurate calculation of solar position, SW clear sky irradiance can be estimated using an empirical approach such as that of Long and Ackerman (2000), which fits a curve to points that are identified as clear, or by modeling the clear sky. An advantage of the empirical approach is that estimates of other atmospheric parameters like aerosol loading and water vapor concentrations are not needed. Empirical fits to the data will also remove some measurement errors when creating ratios or differences between measured and calculated clear sky irradiance. For example, if a pyranometer has a calibration bias, the clear sky curve will have that same bias so it will be reduced in calculations of the transmissivity. However, the Long and Ackerman (2000) algorithm requires measurements of diffuse and total irradiance to identify clear skies, and the impact of sensor motion must be calculated, so it is not applicable to all measurements.

Care must be taken when applying a clear-sky fit function found with data from a pyranometer to measurements made by another model of pyranometer, because the fit may be affected by instrument characteristics, such as the cosine response. Furthermore, the clear sky fit found in a certain surface condition may not be suitable if the surface albedo changes, as could be the case for measurements taken in polar regions. The Long and Ackerman (2000) method handles these challenges by fitting to as many clear sky days as possible within the time series and interpolating the fit parameters between those clear days. This can give more inaccurate results when a site has frequent cloudy skies as there will be fewer clear sky fits.

The SW irradiance can be modeled using physical radiative transfer calculations or with a combination of radiative transfer and parameterized elements. All models require some characterization of inputs like aerosol, water vapor, and surface albedo. Sun et al. (2019) performed a robust comparison of 75 different clear sky models against 75 high quality ground-based land stations representing five climate regimes (i.e., equatorial, arid, temperate, cold, polar) using input data from the MERRA-2 reanalysis data. The MAC2 (Davies and McKay, 1982) and REST2v5 (Gueymard, 2008) models stood out as the top performing models globally, though some models performed better in particular climate regimes. Most of these models were coded using the R computer language by the authors, and are available online (Bright and Sun, 2018).

Clear-sky, surface LW estimates are based on the emissivity of the atmosphere. If temperature and humidity profiles of the atmosphere are available for a given location and time, clear-sky estimates can be calculated via radiative transfer models. There are also a number of parameterized models that use measurements of the ambient air temperature and humidity to approximate the emission of the atmospheric profile. These parameterizations generally take the form of Equation 5, where an effective clear-sky broadband emissivity ( ε c ) is estimated assuming that the atmosphere can be treated as a single slab:

where L W c   is the clear sky downwelling surface LW, ε c is an effective clear-sky broadband emissivity, σ is the Stephan-Boltzman constant, and T a is the ambient air temperature in K.

The challenging piece of this equation is estimating the effective clear-sky emissivity, and a number of models exist to make that calculation. These models are based on surface parameters (like air temperature, water vapor partial pressure) or bulk quantities, such as integrated water vapor, or a combination thereof. Flerchinger et al. (2009) compare 13 algorithms at 21 sites and find that the Dilley and O’Brien (1998), Prata (1996), and Ångström (1915) algorithms calculate clear sky the most accurately. Guo et al. (2019) compare 7 algorithms at 71 different global sites and claim that Carmona et al. (2014) performs the best, but found that different methods worked better in different climate regions. Shakespeare and Roderick (2021) published a model that includes the ability to also constrain CO₂ concentrations and lapse rate to better tailor a clear sky estimate to a particular time and place.

Clear sky downwelling LW can also be estimated empirically using fits to the data. For example, Long and Turner (2008) use empirically identified clear sky periods and surface temperature and humidity measurements to estimate LW clear sky. As they discuss, because ~90% of the surface LW signal comes from the lowest 1 km of the atmosphere, an accurate surface clear sky LW calculation depends primarily on how well the near surface temperature and humidity profile is known. They base their estimates on the Brutsaert (1975) formula of where the effective clear-sky emissivity in Equation 5 is:

In Equation 6, e is the vapor pressure in hPa, and C is the effective temperature/humidity lapse rate coefficient which Brutsaert (1975) calculated to be 1.24 for a standard atmosphere. Long and Turner (2008) scale this estimate to identified clear periods. This identification of clear sky intervals is one of the challenging pieces: during the day this problem can be solved by exploiting the SW irradiance data, while at night the method proposed by Marty and Philipona (2000), based on LW irradiance, air temperature and humidity measurements, may be applied.

Another empirically derived method is given in the appendix of Fairall et al. (2008), based on a number of ship cruises in the Pacific Ocean, with over half the measurements taken near the equator. A fit based on latitude and surface humidity is given in Equation 7, though an additional version that includes the integrated water vapor as a dependent variable is also given in the reference. This equation can also be adjusted to fit data from a particular field campaign.

where q_a is the specific humidity in g/kg, and l a t is latitude.

In future work, we recommend specific testing for the best clear sky models for ocean measurements at different latitudes, in order to find site-specific clear sky models based on the available atmospheric parameters .