Zhijie Li, Zhaoyi Wang, Yun Li , Yu Zhang, Jingjing Zheng, Shan Gao

National Marine Environmental Forecasting Center China

Keywords:CGOFS reanalysis climatology SST SSS SLA

A B S T R A C T The global high-resolution marine reanalysis products that were independently developed by the National Marine Environmental Forecasting Center based on the Chinese Global Oceanography Forecasting System (CGOFS),are evaluated by comparing their climatologies with internationally recognized data from WOA (Word Ocean Atlas), SODA (Simple Ocean Data Assimilation), AVISO (Archiving, Validation, and Interpretation of Satellite Oceanographic Data), and C-GLORS (Global Ocean Reanalysis System). The results show that the SST RMSEs of CGOFS and SODA against WOA are 0.51 °C and 0.43 °C respectively; and in the North Pacific, the SST of CGOGS is closer to that of WOA than SODA. The SSS RMSEs of CGOFS and SODA compared with WOA are 0.48 PSU and 0.40 PSU, respectively. CGOFS can reproduce the main large-scale ocean circulation globally, and obtain a similar vertical structure of the Equatorial Undercurrent as SODA. The RMSE of the CGOFS global sea-level anomaly against AVISO is 0.018 m. The monthly averaged sea-ice extents are between those of SODA and C-GLORS in each month; the growth and ablation characteristics of the ice volume are consistent with SODA and C-GLORS;but the ice volume of CGOFS is greater than that of SODA and C-GLORS. In general, the climatology of the CGOFS global high-resolution reanalysis products are basically consistent with similar international products, and can thus provide reliable data for the improvement of marine science and technology in China.

1. Introductio n

With the continuous development of marine scientific research,ocean reanalysis products generated by integrating numerical models forced by atmospheric fluxes and observations through a data assimilation system ( Balmaseda et al., 2015 ) are increasingly in demand. To date, dozens of global ocean reanalysis products with different time scales, different elements, and different spatial resolutions,have been released ( Carton and Giese 2008 ; Balmaseda et al., 2008 ;Ferry et al., 2010 ; Forget et al., 2015 ). They provide a strong guarantee for marine scientific research. Numerous studies have investigated the ocean circulation and climate variability on the basis of these reanalysis products, including air—sea heat exchange ( Yeo and Nam 2020 ), maximum wave height assessment ( Barbariol et al., 2019 ), ocean current examination ( Artana et al., 2019 ), ENSO forecasting and exploration( Kumar and Hu 2012 ; Tang et al., 2003 ; Heyer et al., 2017 ).

The quality of reanalysis products is affected by errors caused by the numerical model, assimilation scheme, and observation data themselves. It is highly necessary to evaluate the reappearance effect of reanalysis products on different-scale marine phenomena and dynamic processes, which provides guidance to improve the quality of reanalysis products, as well as assistance with improving the reliability of research results based on marine reanalysis products.

Intercomparison of different ocean reanalysis products is a simple but effective method to evaluate the quality of a new reanalysis system, which has been carried out as a precedent by GODAE Ocean View Intercomparison and Validation Task Team ( Ryan et al., 2015 ;Divakaran et al., 2015) and ORA-IP (Ocean Reanalyses Intercomparison Project) ( Balmaseda et al., 2015 ). In this paper, the global highresolution ice—sea coupling reanalysis products of the Chinese Global Oceanography Forecasting System (CGOFS) are evaluated by comparing them with internationally recognized global analysis and reanalysis products from WOA (Word Ocean Atlas), SODA (Simple Ocean Data Assimilation), AVISO (Archiving, Validation, and Interpretation of Satellite Oceanographic Data), and C-GLORS (CMCC Global Ocean Physical Reanalysis System). The characteristics, including sea surface temperature, sea surface salinity, near-surface velocity, and sea-ice extent, are compared in the global and typical oceans.

The paper is structured as follows: The model configuration and data are described in section 2 . The results are presented in section 3 . Conclusions are given in Section 4.

2. Model configuration and data

2.1. Model configuration

CGOFS, developed by the National Marine Environmental Forecasting Center, is based on the NEMO (Nucleus for European Modelling of the Ocean) ( Madec et al., 2008 ) oceanic general circulation model and the LIM3 (Louvain-la-Neuve Sea Ice Model) ( Rousset et al., 2015 )dynamic—thermodynamic sea-ice model. The global horizontal resolution is 1/12°. The bathymetry adopts NRL DBDB2 (Naval Research Laboratory Digital Bathymetry Data Base 2 min resolution) terrain. The

Z

-coordinate is adopted in the vertical direction. The model’s depth range is 0—5728 m, which is vertically divided into 75 layers. The model employs NCEP GFS global weather forecast fields as atmospheric forcing fields. The assimilation system of CGOFS is based on the ensemble LESTKF (local error subspace Kalman filter) ( Evensen 2003 ;Nerger et al., 2005 ) and PDAF (parallel data assimilation framework) Nerger and Hiller (2013) . Thirty ensemble samples are used for each assimilation. The second-order exact ensemble square root filter T?dter and Ahrens (2015) is used to generate the ensemble. The observation operator employs the inverse distance weight interpolation method. The assimilation observation data include satellite observation fusion OSTIA (Operational Sea Surface Temperature and Sea Ice Analysis) SSTs ( Donlon et al., 2012 ), AVISO sea-level anomalies (SLAs)( Pujol et al., 2016 ), and Argo and Tao observations (about 100 observations assimilated daily). Climatologies are obtained by averaging the reanalysis from 1981 to 2010.

2.2. Data

Four kinds of internationally recognized products are employed to carry out the intercomparison to evaluate the quality of the reanalysis products of CGOFS. These products are SODA, WOA, C-GLORS, and AVISO.

In the present study, the monthly averaged data, such as temperature, salinity, current velocity, ice thickness, and ice fraction of version 3.4.2 of SODA Carton and Giese (2008) are used. The spatial range covers the whole world. The horizontal resolution is 0.5°and the vertical direction is divided into 50 layers with unequal spacing.

Monthly averaged temperature and salinity data from WOA18(World Ocean Atlas 2018) ( Garcia et al., 2019 ) are employed to evaluate the surface and profile of global ocean temperature and salinity. The WOA18 dataset was computed by objective analysis of historical observations that were scientifically quality-controlled in the World Ocean Database 2018. Its horizontal resolution is 1°and the vertical direction is divided into 102 layers.

The sea surface current, sea-ice concentration, and sea-ice thickness data from version 5 of the CMCC Global Ocean Physical Reanalysis System (C-GLORSv5) ( Storto et al., 2016 ; Storto and Masina 2016 ) are used to assess the sea-surface current, sea-ice extent, and volume. C-GLORSv5 is the newly upgraded ocean reanalysis produced at CMCC, based on NEMO and coupled to the LIM2 sea-ice model, which is configured with a horizontal resolution of 1/4°.

The satellite altimeter data of AVISO are employed as the reference to evaluate the SLA of CGOFS. AVISO fuses a variety of satellite observations (TOPEX/Poseidon, Jason-1, ERS-1/ERS-2, Envisat, GEOSAT). The spatial resolution is 1/3°×1/3°on the Mercator grid.

3. Model evaluation results

3.1. Sea surface temperature

The global climatology sea surface temperature (SST) maps of CGOFS, SODA, and WOA are presented in Fig. 1 (a-c), which show that the SST distribution of CGOFS and SODA are similar to the reference data of WOA. The highest SSTs are located in the equatorial area, and gradually decrease from the equator to the poles, where the SST drops to about 0C.

Fig. 1. Climatology maps of global SST from (a) CGOFS, (b) SODA, and (c) WOA. Climatology maps of global SSS from (d) CGOFS, (e) SODA and (f) WOA. Climatology maps of global sea-surface current speed from (g) CGOFS, (h) SODA, and (i) C-GLORS. Zonal current in the equatorial Pacific and Atlantic from (j) CGOFS and (k)SODA. (l) Ni?o3.4 index of CGOFS and OISST.

Fig. 2. Bias of (a) CGOFS SST against WOA and (b) SODA SST against WOA. Bias is plotted as CGOFS minus WOA, and SODA minus WOA, respectively, with positive bias meaning the climatological SST of CGOFS is warmer than its counterpart, and vice versa. Bias of (c) CGOFS SSS against WOA and (d) SODA SSS against WOA.RMSE of CGOFS sea-surface current speed against (e) SODA and (f) C-GLORS. (g) RMSE of CGOFS SLA against AVISO.

Fig. 3. RMSE statistics in different ocean regions for CGOFS (a) SST, (b) SSS, (c) sea-surface current, and (d) sea-level anomaly, and the error in the vertical distributions of (e) the global sea temperature profile and (f) global sea salinity profile.

Fig. 4. Monthly average sea-ice extent of the (a) Arctic and (b) Antarctic, and the sea ice-volume of the (c) Arctic and (d) Antarctic for 30 years from 1981 to 2010.

Fig. 2 (a, b) shows the SST biases of CGOFS and SODA minus WOA,respectively. In most of the ocean areas, the error is at a low level, with a global overall RMSE of 0.51 °C and 0.43C for CGOFS and SODA,respectively, compared with WOA ( Fig. 3 (a)), whereas higher errors are apparent near the Gulf Stream and Antarctic Circumpolar Current.

A global view from Fig. 3 (a) shows that in most of the regions, the SST errors of CGOFS and SODA against WOA are at the same level, and the SODA SST fits that of WOA slightly better than CGOFS. Notably, the CGOFS SST is closer to that of WOA than SODA in the North Pacific.

The Ni?o3.4 index values of the CGOFS reanalysis data and OISST remote sensing observation data are shown in Fig. 1 (l). It can be seen that the Ni?o3.4 index given by the CGOFS reanalysis data is highly consistent with the observation results, with a correlation coefficient of 97.4%.

3.2. Sea surface salinity

The climatology of the global sea surface salinity (SSS) of CGOFS,SODA, and WOA are shown in Fig. 1 (d-f), from which we can see that they have a similar pattern on the whole. The SSS in the equatorial area is low, and reaches its highest values in the subtropical ocean area.From the subtropical zone to the poles, the salinity gradually decreases to below 34 psu at the poles.

The SSS biases were calculated to show the distribution of their discrepancy. From Fig. 2 (c, d), it can be seen that the biases in all the oceans except the Arctic are at a low level, which can be further proved from Fig. 3 (b). The global SSS RMSEs of CGOFS and SODA against WOA are 0.48 PSU and 0.40 PSU, respectively. On the whole, CGOFS and SODA both agree well with WOA in all regions except the Arctic Ocean.

3.3. Temperature and salinity profile

In order to understand the vertical coincidence of CGOFS and SODA with WOA, the bias and RMSE for temperature and salinity profiles were calculated to show how the errors change with depth.

Fig. 3 (e) shows that the bias of the global climatology of the sea temperature profile of CGOFS against WOA is negative (except for the depth range of 380—1300 m), and that between SODA and WOA is basically positive. In the depth range of 0—2500 m, the bias of CGOFS and SODA against WOA is small, and the absolute value is within 0.16 °C. In the depth range from 2500 m to the bottom, the absolute values of the bias of CGOFS and SODA against WOA increase with depth, and reach a maximum value of about 0.4 °C at the bottom. As for the RMSE, the profiles of the two RMSEs are similar in the whole depth range, but the RMSE of SODA against WOA is slightly smaller.

It can be seen from Fig. 3 (f) that the maximum absolute biases of CGOFS and SODA against WOA are 0.15 PSU and ? 0.08 PSU. Within the depth range of 0—80 m, the absolute bias decreases with the increase in depth. Below 80 m in depth, the biases of CGOFS and SODA against WOA fluctuate slightly around zero, and the absolute bias between SODA and WOA is slightly smaller. The RMSE between SODA and WOA is smaller at the depth of 0—800 m, and the RMSE between CGOFS and WOA is smaller within the depth of 800 m to the bottom.

3.4. Surface current

The climatology of the global surface current of CGOFS is evaluated by comparing it with the SODA and C-GLORS reanalysis products. Generally, CGOFS, SODA, and C-GLORS can reproduce the main large-scale ocean circulation globally, including the Equatorial Current, Kuroshio,Gulf Current, Agulhas Current, Antarctic Circumfluence, and so on, as shown in Fig. 1 (g-i). In order to better understand the vertical structure of the typical ocean circulation, we take the Equatorial Undercurrent(EUC) in the equational Pacific and Atlantic as an example and draw its vertical distribution map. Its flow pattern in CGOFS and SODA are compared, but C-GLORS is excluded because of a lack of data. From Fig. 1 (j)and k it can be seen that the shape and strength of the EUC of CGOFS and SODA are basically the same.

The global map of the sea surface current RMSE presents inconsistency between CGOFS and SODA in the tropical Pacific, tropical Atlantic, and Indian Ocean, as shown in Fig. 2 (e) and f. By contrast, the RMSE between CGOFS and C-GLORS is smaller in the corresponding areas, which can also be reflected in the regional statistics of the RMSE shown in Fig. 3 (c). The global RMSE of the CGOFS surface current speed against SODA and C-GLORS is 0.05 m sand 0.04 m s, respectively.The biggest RMSE is in the tropical Pacific, which reaches 0.052 m sand 0.071 m scompared with C-GLORS and SODA, respectively.

3.5. Sea-level anomaly

The satellite observations of AVISO are employed as a reference to evaluate the SLA of CGOFS. It can be seen from the global map of the SLA RMSE of CGOFS against AVISO in Fig. 2 (g) that higher RMSEs are located along the path of the Agulhas Current, Gulf Current Extension,Kuroshio Extension, and Antarctic Circumpolar Current. In general, the SLA RMSEs of CGOFS against AVISO are all at a low level in different ocean regions, with a global RMSE of 0.018 m, as shown in Fig. 3 (d).

3.6. Sea ice

The sea-ice extent and ice volume were chosen as an index to evaluate CGOFS. The isoline with a sea-ice concentration of 0.15( Heinrichs et al., 2006 ) is used as the sea-ice edge to calculate the seaice extent. Based on this, the sea-ice volume is obtained by multiplying the area of each grid within the edge line of the sea ice with the corresponding sea-ice thickness.

The sea-ice extent and sea-ice volume have regular growth and ablation characteristics. From Fig. 4 (a, b) it can be seen that the maximum sea-ice extent of the Arctic appears in March, and the minimum appears in September. In the Antarctic, the maximum sea-ice extent appears in September and the minimum appears in February. The monthly average sea-ice extent of CGOFS is between those of SODA and C-GLORS, both in the Arctic and Antarctic. The average absolute values of the relative errors of the CGOFS sea-ice extent against C-GLORS and SODA are 6% and 15% in the Arctic, and approximately 19% and 21% in the Antarctic.

Monthly average sea-ice volume comparisons are shown in Fig. 4 (c, d), from which it can be seen that in the Arctic, the sea-ice volume reaches a maximum in April and then decreases to its minimum value in September. In the Antarctic, the greatest sea-ice volume is in September, and the lowest value is in February. Compared with C-GLORS and SODA, CGOFS produces a consistently overestimated ice volume in both the Arctic and Antarctic. The average absolute values of the relative errors of the CGOFS sea-ice volume against C-GLORS and SODA are 13% and 30% in the Arctic, and approximately 29% and 40%in the Antarctic.

4. Conclusions

In this paper, the reanalysis products of CGOFS are evaluated through intercomparison with WOA, SODA, AVISO, and C-GLORS. Results show that the global climatology of the SST and SSS maps of CGOFS and SODA have the same pattern as in WOA. The global SST RMSEs of CGOFS and SODA against WOA are 0.51 °C and 0.43 °C. The global SSS RMSEs of CGOFS and SODA against WOA are 0.48 and 0.40 PSU,which indicate that compared with WOA, CGOFS, and SODA have a similar error level. The sea temperature and salinity profiles of CGOFS have also been proven to be essentially consistent with WOA. CGOFS can reproduce the main large-scale ocean circulation globally, and obtain a similar vertical structure of EUC with SODA. The global RMSE of the CGOFS SLA against AVISO is 0.018 m. The sea-ice extent of CGOFS is between those of C-GLORS and SODA, while the sea-ice volume of CGOFS is overestimated compared with SODA and C-GLORS.

CGOFS is an eddy-permitting ocean model system, whereas the data of SODA, WOA, AVISO and C-GLORS used for comparison are derived from monthly averaged data with low horizontal resolution. This means the assessments are insufficient for mesoscale eddies, which will be further explored in future research. In general, the reanalysis products of CGOFS are basically accurate, as proven preliminarily by the consistency with SODA, WOA, AVISO, and C-GLORS reported in this paper,albeit with some discrepancies.

Funding

This study was supported by the National Key R&D Program of China[grant number 2016YFC1401802 and 2016YFB0201105].