Contributors: E. Carboni (UKRI-STFC RAL Space), G.E. Thomas (UKRI-STFC RAL Space)

Issued by: STFC RAL Space (UKRI-STFC) / Elisa Carboni

Date: 31/05/2023

Ref: C3S2_D312a_Lot1.2.1.5-v4.0_202305_PQAR_CCICloudProperties_v1.2

Official reference number service contract: 2021/C3S2_312a_Lot1_DWD/SC1

History of modifications

List of datasets covered by this document

Anchor
relateddocuments relateddocuments

Related documents

Acronyms

List of Figures

List of Tables

General definitions

The "CCI product family" Climate Data Record (CDR) consists of two parts. The ATSR2-AATSR Cloud Properties CDR is formed by a TCDR brokered from the ESA Cloud_cci project and an ICDR derived from the SLSTR on board of Sentinel-3. ICDR uses the same processing and infrastructure as the TCDR. Both TCDR and ICDR data have been produced by STFC RAL Space.

These Cloud Properties datasets from polar orbiting satellites consist of these variables: Cloud Fractional Cover (CFC), Cloud Phase (water/ice), Cloud Optical Thickness (COT), Cloud particle Effective Radius (CER), Liquid/Ice Water Path (LWP/IWP), and Cloud Top Pressure (CTP), Height (CTH) and Temperature (CTT).

Bias (accuracy): Mean difference between TCDR/ICDR and reference data

b=\frac{\sum_{i=1}^N (p_i - r_i)}{N} \ \ (Eq. 1)

Where: p_i is the CDR product, b is the mean bias and r_i is the equivalent value from the reference dataset. N is the number of observations.

bc-RMSE (precision): Bias corrected root mean squared error to express the precision of TCDR/ICDR compared to a reference data record

bc- RMSE=\sqrt{\frac{\sum_{i=1}^N ((p-b)-r)^2}{N}} \ \ (Eq. 2)

Where: p_i is the CDR product, b is the mean bias and r_i is the equivalent value from the reference dataset. N is the number of observations.

Stability: The variation of the bias over a multi-annual time period

Table 1: Summary of variables and definitions

Table 2: Definition of processing levels

Table 3: Definition of various technical terms used in the document

Scope of the document

This document provides a description of the product validation results for the Essential Climate Variable (ECV) Cloud Properties. This CDR comprises inputs from two sources: (i) brokered products from the Cloud Climate Change Initiative (ESA's Cloud_cci), namely those coming from processing of the Advanced Along-Track Scanning Radiometer (A)ATSR) data and (ii) those produced under this contract for the Climate Data Store, specifically those coming from processing of the Sea and Land Surface Temperature Radiometers (SLSTR).

The Thematic Climate Data Record (TCDR) is the product brokered from the European Space Agency Cloud Climate Change Initiative (ESA Cloud_cci) ATSR2-AATSR version 3.0 (Level-3C) dataset. This is produced by RAL from the second Along-Track Scanning Radiometer (ATSR-2) on board the second European Remote Sensing Satellite (ERS-2) spanning the period 1995-2003 and the Advanced ATSR (AATSR) on board ENVISAT, spanning the period 2002-2012.

In addition, the Interim Climate Data Record (ICDR) is the product derived from the Sea and Land Surface Temperature Radiometer (SLSTR) on board of Sentinel-3 and spans the period from January 2017 to present. Validation for this SLSTR derived product for the period from January 2017 to June 2022 is described in this document. In addition to the validation of the individual products from Sentinel-3A and -3B, the merged product (Sentinel-3A+3B) is also validated spanning the period from October 2018 to present (Sentinel-3B provides data from October 2018 on).

Executive summary

This document provides a description of the product validation results for some of the Essential Climate Variable (ECV) Cloud Properties. These specific products are brokered to (in case of (A)ATSR) or produced for the Climate Data Store (in the case of SLSTR) by the Copernicus Climate Change Service (C3S).

The TCDR is a brokered version of ESA's Cloud_cci ATSR2-AATSR version 3.0 (Level-3C) dataset, produced by RAL from the second Along-Track Scanning Radiometer (ATSR-2) on board the second European Remote Sensing Satellite (ERS-2) which operated in the period 1995-2003 and the Advanced ATSR (AATSR) on board ENVISAT which operated in the period 2002-2012. In addition, the Sea and Land Surface Temperature Radiometer (SLSTR) on board of Sentinel-3 has been operating from 2017 to present and provides the input to the ICDR. The validation of the ICDR is over the period from January 2017 to June 2022 with not just the individual products from Sentinel-3A/B but also the merged product spanning the period from October 2018 to June 2022. The retrieval algorithm is presented in [D2] and the validation methodology refers to the Cloud_cci Product Validation and Intercomparison Report [D1]. The same methodology is applied to the SLSTR dataset.

Poulsen et al. (2019) [D3] is the paper describing the dataset that includes cloud properties as well as Surface Radiation Budget and Earth Radiation Budget products. This document will mainly refer to the Cloud_cci Product Validation and Intercomparison Report [D1].

Table 2-1 provides a summary of the estimated accuracies of the TCDR together with the GCOS requirements. The validation results are provided in [D1] section 7-2, with a recommendation for use.

For the ICDR, the intercomparison with MODIS (using only the first 5 years of SLSTR data) show biases consistent with values found for the TCDR vs MODIS comparison (D1, section 4.1) for Cloud Fractional Cover (CFC), Cloud Top Pressure (CTP) and Liquid Water Path (LWP), but higher value for Ice Water Path (IWP) SLSTR-B, with a global averaged bias of -25 g/m³ for SLSTR-A and -40 g/m³ for SLSTR-B (TCDR IWP bias with MODIS was -29 g/m³).

This document is divided in different sections: 

• the first section presents a brief description of validation methodology together with a series of references for further information;
• the second section present the results of the validation and comparison of TCDR and ICRD data.
• the third section presents the compliance with user requirements and include recommendation on the usage and know limitations

1. Product validation methodology

The validation methodology is described in section 2.4 of [D1]. In addition, the validation methodology is also described in the corresponding Product Quality Assurance Document (PQAD) [D4]. In summary, the methodology uses the bias between the Cloud_cci product and the reference data to estimate the accuracy of the dataset. The bias corrected root mean squared error (bc-RMSE) is used to express the precision of CDR compared to a reference data record, which is also known as the standard deviation about the mean. Stability is calculated as the variation of the bias over a multi-annual time period.

Anchor
validation validation

2. Validation results

2.1 Validation results for the TCDR

The validation results for TCDR products are provided in [D1], section 3 and 4. The evaluation is divided in validation against high quality and satellite-based reference observations Please find information on the sensors in [D1]: CALIOP (Annex A.1), AMSR-E (A.2), DARDAR (A.3) and MODIS-C6.1 (A.10) (CALIOP, AMSR-E and DARDAR) and an intercomparison to well-established, satellite-based cloud datasets of similar kind (e.g. MODIS Collection 6.1). Table 2-1 provides a summary of the resulting TCDR accuracies.

1 Value obtained from the difference between CDR LWP trend (0.99g/m²/decade) and MODIS LWP trend (8.06g/m²/decade) divided by mean MODIS C6.1 LWP (123g/m²).
2 Value obtained from the difference between CDR IWP trend (-2.27g/m²/decade) and MODIS IWP trend (7.82g/m²/decade) divided by mean MODIS C6.1 IWP (208g/m²).

2.1.1 Cloud Fractional Cover (CFC)

Cloud Fractional Cover (CFC) is validated against CALIOP and reported in [D1], section 3.1.1 and compared with MODIS and reported in [D1], section 4.1.1. A slight underestimation of cloud occurrences is found in the Cloud_cci data compared to CALIOP, which is primarily due to a lack of sensitivity of passive imager data with respect to optically very thin clouds.

2.1.2 Cloud Top Height (CTH)

Cloud Top Height is validated against CALIOP and presented in [D1] section 3.1.3.

For liquid clouds only very small biases (0.12km) and bc-RMSE (0.97km) are found.

For ice clouds, the strong underestimation of cloud top height is evident and it is a common feature for all three Cloud_cci datasets. It is mainly caused by high-level, optically thin clouds. Biases are around -3.5 km and bc-RMSE around 2.3 km. Removing the optically very thin cloud layers at the top of the CALIOP profiles, improves the agreement between Cloud_cci and CALIOP substantially.

2.1.3 Cloud Top Pressure (CTP), Cloud Optical Thickness (COT) and Cloud Effective Radius (CER)

These parameters are compared against MODIS. As the quantities are also derived from satellite observations, it is not valid to calculate measures of bias and RMSE, as the MODIS values cannot be necessarily considered to be better. Nonetheless a comparison is useful. This comparison is documented in the following sections of [D1]: Cloud Top Pressure (CTP) is compared in section 4.1.2. Cloud Effective Radius (CER) is compared in section 4.1.5 and 4.1.6. Cloud Optical Thickness (COT) is compared in section 4.1.3 and 4.1.4 for liquid and ice cloud.

2.1.4 Liquid Water Path (LWP) and Ice Water Path (IWP)

Validation of Liquid Water Path (LWP) is presented in [D1] section 3.1.4. It is carried out against AMSR-E products and compared with MODIS and reported in [D1] section 4.1.7.

Ice Water Path (IWP) is validated against the DARDAR IWP product and reported in [D1] section 3.1.5 and compared with MODIS which is in section 4.1.8.

Validating liquid water path over ocean against AMSR-E gives very convincing results for the CDR dataset, with bc-RMSE values of 25 g/m², only small biases (-1.44 g/m²) and high correlations (0.76). Validating ice water path against the combined CALIPSO-CloudSat product DARDAR shows good agreement with correlations of 0.45. There is a general underestimation of IWP by Cloud_cci which in terms of relative bias partly exceeds 50%.

2.2 ICDR comparison with MODIS

The first 5.5 years and 3.5 years (of SLSTR-A and SLSTR-B respectively) of products have been compared against MODIS (Collection 6.1 Terra) following the same methodology described in [D1] section 4.1. We estimate the bias, i.e. mean differences, and the monthly mean global average of C3S and the MODIS data. To compute the monthly mean global average of both datasets we considered only the valid data between -60° and +60° latitude.

Table 2-2 shows the bias results from this comparison for both TCDR data (from D1) and for the ICDR. Except for the IWP, all the properties show better results (lower bias) for the ICDR in comparison to TCDR.

Figure 2-1 and Figure 2-2 show an example of ICDR monthly products for March 2017 and the equivalent monthly product from MODIS. These figures are for illustrative purposes so the user knows what to expected. Nonetheless, note that for this month, the ICDR and MODIS datasets are spatially similar across all four properties. However, there are some small differences observed. For example CTP seems much higher in the tropics in the MODIS product compared to the ICDR. For a more detailed analysis please go to the [D1].

3. Application(s) specific assessments

In addition to the extensive product validation (see chapter 2 for results and chapter 2/3 in [D4] for validation methodology) a second assessment is introduced to evaluate the Interim Climate Data Record (ICDR) against the Thematic Climate Data Record (TCDR) in terms of consistency. Since frequent ICDR deliveries make detailed validation not feasible, a consistency check against the deeply validated TCDR is used as an indication of quality. This is done by a comparison of the following two evaluations:

• TCDR against a stable, long-term and independent reference dataset
• ICDR against the same stable, long-term and independent reference dataset

The evaluation method is generated to detect differences in the ICDR performance in a quantitative, binary way with so called Key Performance Indicators. The general method is outlined in [D6] chapter 3. The same difference between TCDR/ICDR and the reference dataset would lead to the conclusion that TCDR and ICDR have the same quality (key performance is "good"). Variations or trends in the differences (TCDR/ICDR against reference) would require a further investigation to analyze the reasons. The key performance would be marked as "bad". The binary decision whether the key performance is good or bad is made in a statistical way by a hypotheses test (binomial test). Based on the TCDR/reference comparison (global means, monthly or daily means) a range is defined with 95% of the differences are within. This range (2.5 and 97.5 percentile) is used for the ICDR/reference comparison to check whether the values are in or out of the range. The results could be the following:

• All or a sufficient high number of ICDR/reference differences lies within the range defined by the TCDR/reference comparison: Key performance of the ICDR is "good"
• A smaller number of ICDR/reference differences is within the pre-defined range: Key performance of the ICDR is "bad"

3.1 Results

The results of the KPI test are summarized in Table 3-1.

Percentiles were calculated based on the comparison of the TCDR using the Advanced Along Track Scanning Radiometer (AATSR) instrument against MODIS Collection 6.1 (Terra satellite only) as reference dataset for the variables Cloud Fractional Cover (CFC), Cloud Top Pressure (CTP), Ice Water Path (IWP) and Liquid Water Path (LWP). Percentiles were based on the time from 2002-2012 with monthly means and applied to the ICDR from 01/2017 (10/2018) to 06/2022 for Sentinel-3A (Sentinel-3B and merged product Sentinel-3A+B) based on measurements of the Sea and Land Surface Temperature Radiometer (SLSTR).

Most of the ICDR months are outside the TCDR-based KPI limits and leading to “bad” KPI tests. Therefore, the ICDR is not stable in relation to the TCDR. This is due to multiple reasons starting with the fact of a five year gap (2012-2016) between TCDR and ICDR. In addition, TCDR and ICDR are based on different instruments with SLSTR on Sentinel-3 and (A)ATSR/ATSR-2 on Envisat/ERS-2, respectively. Differences occur due to a lower bias between ICDR and reference dataset and a subtraction of the monthly means (based on the TCDR) to remove the annual cycle leads to values outside of the KPI range (see method in [D6], chapter 3.2.2). On the other hand, IWP has a higher bias compared with MODIS which can be caused by different cloud classification due to differences in the visible channels between (A)ATSR and SLSTR. Please note that significant changes between 01/2017 - 12/2020 and 01/2017 - 12/2021 are due to bugfixes.

4. Compliance with user requirements

The validation results for the TCDR products are presented and described in detail in [D1], section 7. In this section, a summary highlighting recommendations on usage is presented. More detailed information about the user requirements is provided in the Target Requirements and Gap Analysis Document (TRGAD) [D5].

Table 2-1 (section 2) provides an overview of the GCOS requirements for the Cloud Properties and the values achieved by TCDR. It should be noted that GCOS requirements are targets and are often not attainable using existing or historical observing systems. The Cloud_cci doesn't meet the frequency requirement (3h) due to the nature of the satellite observations, but exceeds the spatial resolution (50 km GCOS target). ICDR accuracies (estimated with the first 5 years of data) are consistent with TCDR accuracies apart from IWP where we find slightly higher bias in comparison with MODIS (section 2.2).

For nearly all validations for which a trusted reference data source is available, the compliance to GCOS requirements could be shown, e.g. cloud fractional cover stability, cloud top height accuracy and stability, liquid water path accuracy and stability and ice water path stability. Cloud fractional cover accuracy is close to the GCOS target requirements. For effective radius and optical thickness no reliable reference data is available for accuracy compliance analysis.

A general problem is the assessment of the stability. Stability assessments are based on comparisons to MODIS, which in turn however, is not an entirely reliable source itself as it is sometimes characterized by significant trends which may or may not be true.

The following are recommendations on the usage and known limitations (from [D1] table 7.1):

Cloud Fractional Cover (CFC)

• Discrimination of heavy aerosol and cloud is not optimal, thus aerosol is sometimes flagged as clouds in such conditions. It is advised to be careful in the interpretation of cloudiness in periods with dust / volcanic ash outbreaks. Cloudiness is overestimated in these conditions.
• Cloud detection during polar night over snow and ice is generally difficult.
• Cloud detection in twilight conditions is of rather poor quality due to the small number of channels used.
• The ATSR2-AATSR cloud detection consistency between ATSR2 and AATSR is not optimal.
• Due to a limitation of passive imagers, the cloud fraction is usually biased towards lower values, compared to CALIOP, for example.

Cloud Top Pressure (CTP)

• In semi-transparent (ice) cloud conditions, the cloud top will be assigned too low.
• Multi-layer clouds are not modelled hence the CTH for cases of an upper layer of thin cirrus will effectively retrieve a radiative height (approx. 1 optical depth into the cloud).

Cloud optical thickness (COT)

• COT is a daytime product only.
• In cases the wrong phase (liquid/ice) is assigned, the optical thickness is likely to have significant errors.
• In the case of incorrectly assigned surface BRDF the optical depth is likely to be biased. With high BRDF the COT will be biased low. With low BRDF the COT will be biased high.
• In case of sub-pixel clouds or cloud borders the COT is likely to have significant errors.
• In case of optically thin clouds above (especially poorly) a defined highly reflecting surface, the COT retrieval might be problematic.
• For very optically thick clouds, the measurements go into saturation and thus the sensitivity of the measurement to the COT is small. Those values should be accompanied by large uncertainty values.

Cloud effective radius (CER)

• CER is a daytime product only
• In cases where the wrong phase (liquid/ice) is assigned, the effective radius is likely to have significant errors.
• In case of sub-pixel clouds or cloud borders, the effective radius is likely to have significant errors.

Liquid water content (LWP)

• Monthly mean LWP is computed using only daytime measurements.
• Since LWP is computed from retrieved COT and CER, the same limitations as for COT and CER apply for LWP.
• The method used assumes vertically homogeneous clouds, which might deviate from the true situation. In case of vertically inhomogeneous cloud layers, e.g. multi-layered clouds, the LWP retrieval is likely to show large errors, since the CER is retrieved from the uppermost cloud layers and may not be representative for the entire vertical column.
• In cases of a wrongly assigned cloud phase, i.e. ice cloud is treated as liquid cloud, the retrieved LWP will show large errors.

Ice water content (IWP)

• Monthly mean IWP is computed using only daytime measurements.
• Similar limitations as mentioned for Cloud liquid water path (LWP)

Related articles