A.

Definition

The term climate benchmark is employed to denote a high absolute accuracy SI traceable measurement of the state of the climate. Such a benchmark can be used as a reference point against which to evaluate change and assess the performance of climate models. The fidelity of the measurement must stand independent of the instrument characteristics, the sampling used or the time period sampled, thus as well as traceable absolute accuracy it needs to be well sampled in space and time such that uncertainties in sampling and natural variability, which all affect the accuracy of the benchmark, are minimised.

To be useful the benchmark needs, as a minimum, to be sensitive to the most significant and uncertain climate change signals and capture their climate impact, and this sensitivity must be maximised to enable earliest possible detection. In addition, in order to provide a significant constraint, the benchmark needs to provide the means of distinguishing different climate signals. For example, in the case of cloud, forcing varies with cloud fraction, cloud optical depth, cloud phase and particle size. So whilst it is necessary that a model can produce the correct decadal change in, say, shortwave cloud radiative forcing it is not a sufficient constraint, it must be clear that it is achieved by the same type of cloud. The ideal benchmark would provide the means to understand the change observed: allowing attribution of any observed change to specific causes and providing the information necessary to evaluate the deficiencies of climate models in terms of their ability to link change to the responsible forcing and feedback mechanisms.

Thus a well-designed benchmark can serve several inter-related scientific goals:

B.

Climate benchmarking product requirements

The choice of observations made and how they are obtained needs to be carefully considered. Data with sufficient information to resolve the inadequacies of models and probe the complexity of climate signals is required. Additionally, sampling and the treatment of the data must limit the aliasing of sampling bias error and the impact of natural variability on our observations, as these constitute noise on our measurements outside the controls of instrument design. In this section we address the product requirements of an Earth reflected solar climate benchmark. The requirements for incoming Total Solar Irradiance (TSI) are much simpler and outlined in [9].

We define three levels of benchmark products, each subsequent level adding to the requirements of the previous and providing additional scientific value. The consequential observational requirements are summarised in Table 1, but first we provide an overview of their scientific justification.

Table 1.

The measurement requirements summary, defined for the Benchmark Level 1-3.

The solar reflected benchmark requires an SI traceable measurement of the spectrally resolved Earth reflected radiance and incoming solar radiation with sufficient absolute accuracy to meet the product requirements stated in terms of reflectance.

The inescapable properties of climate change signals are that they are relatively small, broad scale and only emerge over long timescales; thus a benchmark doesn’t require high spatial or temporal resolution, but rather only needs to provide a relatively low resolution global picture for an average year or season. Each level of benchmark is a large scale average, and the primary requirements on the accuracy are on these ‘average products’. However, with each level there is an increasing complexity in the discrimination required by the instantaneous observations, increasing the demands on the instantaneous measurement and in particular its signal to noise ratio (SNR), which, together with an increasing degree of separation of the averages into different categories (and thus a reduction in the number of data points contributing to the average) will also increase the demands on the instantaneous measurement characteristics. Finally it must be remembered that the accuracy requirement on the benchmark products is relative to the true average, thus sampling error, including the sampling of year to year variability, as well as instrument absolute accuracy and noise or uncertainty in the categorisation of the averages all contribute to the overall uncertainty term.

Level 1: Baseline reflected solar benchmark: provides a benchmark measurement of the climate state which is adequate as a sensitive and robust measure of climate change and facilitates testing of predictions of climate forecast models. For the shortwave this comprises a high accuracy measurement of the spectrally resolved annual average reflected solar radiance. It comprises of 10-15 degree zonal (spectrally integrated and resolved) annual averages, with some resolution in longitude, for example 30 or 60 degree. Spectral resolution of 8 – 25 nm bandwidth and overall accuracies of 0.3 to 1 % (k=2) over the range 340 to 2350 nm.

Level 2: Climate feedback benchmark: In addition to the above this would provide a benchmark for key feedbacks: Cloud and surface Albedo. Additionally requires separation of measurements to clear-sky, all-sky and cloudy and higher spatial resolution to <~500 m. This means that in addition to uncertainties in the averages, instantaneous noise must also be <0.5% in certain spectral regions, for example 600 to 700 nm, to ensure accurate cloud detection.

Level 3: Attribution and evaluation benchmark: There are two aspects to this, the first (a) is based on direct use of the benchmark and the second (b) on the climate information that can be provided if the benchmark instrument is used to improve the calibration of key climate instruments to meet decadal climate signal accuracy requirements. In addition to the benefits of level 1 and 2 this would enable attribution of climate change and evaluate and improve the physics of climate models and processes. In practise this means improving the spatial resolution and maintaining high accuracy (and SNR) at the Instantaneous Field of View (IFOV). In this way it can provide detail on cloud characteristics and surface properties for a wide range of climate and operational applications. It also provides the means to upgrade the performance of sensors such as CERES, VIIRS, Landsat 8 and the new Copernicus sensors on Sentinel 2 and 3, binning the spatial and spectral measurements of TRUTHS to match the footprint and signatures of the other sensors.

C.

On-orbit reference sensor product requirements

A significant component of the level 3 benchmarking product is the ability for TRUTHS to act as an on-orbit reference calibration sensor to bestow SI-traceable calibration onto other sensors. There should be no confusion that climate benchmarking is the primary scientific driver for the TRUTHS mission and its demanding performance specification. However, the scientific value this secondary objective brings significantly outweighs the slight increase in complexity of the mission requirements as discussed below.

Reference calibration of other space assets from orbit requires that any convolution to match the footprint (spectral or spatial) does not contribute a significant uncertainty. Studies by the CLARREO team summarised in [2] using high resolution Schiamachy data have shown that 8 nm bandwidth and 4 nm sampling is the optimum needed to match sensors other than spectrometers. In practise this resolution is only needed for the spectral bands of other sensors which are relatively well defined and largely common e.g. Ocean colour bands.

Reference calibration for the full range of assets increases the demands on spatial resolution requirements, (ideally to ~ <50 m) whilst maintaining an IFOV SNR, for a typical target radiance level of >100 to 300 (the latter for the visible spectral range). Optimal targets have relatively high reflectance levels e.g. deserts, snow fields and the moon and so these SNR levels are not too onerous. Studies by the CLARREO team [3] have also shown that the ideal minimum swath is between 40 and 100 km to ensure sufficient overlap. The spectral, spatial resolution and SNR requirements, together with that for swath and IFOV exist within a trade-off space as defined by the spectrometer design, and specifically the limitation of available detectors. Studies recently performed by the TRUTHS science team, using the SSTL CHRIS-2 instrument as a baseline concluded these combined requirements could be achieved with low risk evolution of heritage instrumentation and are similar to the specifications of EnMAP for example. Work is in progress to design an optimal hyperspectral imager for TRUTHS and assess its likely performance.

The 609 km orbit altitude (table 1) is largely driven by the needs of reference calibration. Many EO sensors operate in 700+ km orbits and a large height gap between the relative orbits allows a longer orbital dwell time to maximise overlap times for simultaneous cross-calibrations, however, too low an orbit increases drag and impacts the fuel requirement.

Additional requirements are placed on orbital configuration and pointing ability. To minimise sampling noise to the sub 1% level necessary to achieve the required accuracy, 2000 samples (for a nominal SNR of <50) matched to within 1° in viewing angles and 5 minutes in time for each of the data configurations (scene spectral properties, polarisation, view angles etc.) is required. A pointing capability to ~ 1 km and a knowledge to <3x the IFOV is required to ensure a representative spatial matching can be made with the sensor under test.

D.

Solar irradiance product requirements

The measurement requirements for Total Solar Irradiance (TSI) are relatively easily defined and with the exception of the absolute accuracy needed to meet the climate benchmark criteria are based on best practise from existing heritage instruments e.g. VIRGO on SOHO. Similarly for solar spectral irradiance other than spectral resolution (where <1 nm is desired in the UV) the requirements are similar to those for Earth reflectance and lead us to the prospect of using the same hyperspectral imager for both applications. In this case, we use the native spectral resolution of the spectrometer together with long staring solar views to improve limitations posed by SNR.

A.

Current state-of-the-art

One immediate question would be whether the climate benchmarking product requirements could be achieved with current sensor capabilities i.e. is the novel TRUTHS SI-traceable method necessary. The majority of past, current and planned future missions are not designed for climate benchmarking, but rather to investigate specific processes or obtain specific retrieval products. Could these sensor concepts be adapted in a low risk way to climate benchmarking? As already discussed, the hyperspectral imager instrument requirements in terms of swath, coverage, spectral & spatial resolution and instantaneous SNR do not require a significant advancement in technology. The demanding aspect of a climate benchmark mission concept is in the instrument calibration accuracy, its traceability to SI and the maintenance of these strict accuracy requirements throughout the mission lifetime. For any sensor data to be used as part of a climate record, the sensor measurements should be stable and traceable with an estimate of the uncertainties which must be below the detection threshold for that record.

Current optical imaging sensors fall into two categories; filter radiometers and spectrometers. Filter radiometer-based sensors are unable to fulfil the climate benchmark role as they do not give the full required spectral coverage requiring significant modelling to interpolate nor do they have the required accuracy with at best uncertainties at beginning of life being no better than a few %.

Spectrometers can deliver the required spectral range (250-2500 nm) at high spectral resolution but again are limited in terms of SI-traceable accuracy to typically 2 to 5% dominated by radiometry. Reducing calibration uncertainties in the top-of-atmosphere radiometric measurements is clearly the technologically disruptive innovation provided by TRUTHS which distinguishes it from its peers allowing climate quality data.

The limitations in current calibration methods are not in our understanding of the calibration equation (the relationship between observed radiance and digital signal recorded and processed by the sensor), which while invariably complex can be well characterised pre-launch. The limitation is in the simplified linear relationship, typically resulting in a gain and offset term and the derivation and monitoring of these values on-orbit. The transition from air to vacuum, pre-flight storage and the harsh launch environment all lead to changes in the calibration parameters that means the pre-flight values cannot be guaranteed in flight, and with much evidence to show that they can change significantly.

Current on-board calibration methods are usually referenced to the reflectance factor of a white lambertian diffuser, with pre-flight calibrated accuracies typically of 1-2% (k=2) (although uncertainties a factor of ten lower than this are achievable). Calibration to radiance is reliant on a reference solar irradiance e.g. [4], with an associated accuracy of ~2%; clearly, in need of significant improvement to attain the 0.3% required for climate benchmarking. However, the largest concern with this method is lack of reliability of measurements of degradation of the diffuser panel. This is often coupled with the application of a degradation model e.g. [5] that assumes an exponential form; however the coefficients in this model vary from sensor to sensor and mission to mission and need to be determined on-orbit, ideally through an SI-traceable method.

Strategies, such as a monitor detector, the use of multiple diffusers (as employed on MERIS) or comparison against a stable reference site (such as a vicarious calibration site on the Earth or the moon) allow the monitoring of diffuser degradation to some extent, but these methods are neither SI traceable or can be performed to an accuracy level necessary for climate benchmarking studies.

With on-board calibration sources it should be possible to achieve uncertainties <3% (k=1). Maintaining and achieving uncertainties below 3% is only possible if the source is actively monitored. In-flight monitoring techniques have been deployed with some degree of success (e.g. MERIS, MODIS-A) to provide an on-board calibration stability of <1%. However, even these techniques rely on assumptions about their stability and cannot claim to be fully traceable to SI. Achieving VIS-SWIR spectral radiances at the 0.3% uncertainty level clearly demand advances in on-board monitoring methods.

B.

Sensor vicarious cross-calibration

Even with on-board calibration systems it is generally necessary to verify the radiometric accuracy is achieved on orbit by independent means. There are a number of techniques available to provide a vicarious calibration of a satellite sensor. Although techniques have been developed to monitor the calibration using stable reference sites, these are based on an assumption of the long-term stability of the sites being <±1% per decade which has not yet been independently verified. A recent review paper [6] provides an overview of many of the techniques available and their current status. The paper [6] shows that the accuracy of such activities is limited to approximately 1% in the stability of an individual sensor and 2-3% in cross-calibration depending on spectral band and similarity of the spectral response function of the spectral bands of the compared sensors.

Limitations to vicarious calibration activities include:

A study by Lukashin [7] has developed the uncertainty attainable in sensor inter-calibration using an ideal reference sensor and simultaneous nadir observations concluded that in favourable observation conditions and with adequate polarisation distribution models a radiometric uncertainty of 0.3% (1σ) could be achieved in the target sensor. By limiting the cross-calibration measurements over well characterised vicarious sites, the combination of this technique with those discussed in [6] could potentially reduce the number of required samples from that described in the benchmark level 3 requirements.

In conclusion, the main limitation for climate benchmarking is the radiometric calibration. Current diffuser monitoring and vicarious techniques are inadequate for benchmarking and <1% is unattainable without a change in traceability methods.

A.

Cryogenic solar absolute radiometer (CSAR)

The Cryogenic solar absolute radiometer (CSAR) is the heart of the TRUTHS mission [8], providing the means to establish SI traceability in orbit. CSAR (and laboratory cryogenic radiometers) are based on the simple principle of ‘electrical substitution’, where optical power incident on a black (spectrally neutral) absorbing surface will cause a rise in temperature of the disk, which is sensed by a thermometer. With the optical power shut off, the same temperature rise is then established by passing a current through an electrical heater attached to the disk. The electrical power used to create that temperature rise can be equated to the optical power, and since electrical power can be measured with relative ease, the optical power can be determined through the substitution of electrical power.

Although similar instruments are already used in space to measure TSI, this mission will be the first to fly such an instrument, cooled to cryogenic (~20 K) temperatures, using an Airbus cooler. Cooling such instruments in terrestrial laboratories ~30 years ago led to a step change reduction in radiometric uncertainties of nearly a factor of 100 and are now the primary standard of choice at most of the World’s National Metrology Institutes. CSAR will be the primary standard providing SI traceability for the calibration of all onboard optical instrumentation. This standard will provide traceability to SI to an absolute accuracy of 0.1 % (k=2). In addition to serving as a primary standard for radiance and irradiance, CSAR will also provide measurements in its own right:-Total Solar Irradiance (TSI) at an uncertainty of 0.02 % (k=2), a factor of ten better than the currently operational ambient temperature radiometers.