Quality Evaluation of Satellite Sea Surface Salinity Products in the Pacific Ocean

Sea surface salinity (SSS) is one of the most important variables in ocean dynamics and atmospheric climate. The launch of three salinity satellites, Aquarius, SMAP and SMOS, has greatly expanded the global sea surface salinity data field. The latest ocean surface salinity (CCI+SSS) fusion project fully utilizes the satellite data from SMOS, supplemented by SMAP and Aquarius satellite data, to merge into the CCI fusion product. In this paper, the data accuracy of the four SSS products, Aquarius, SMAP, SMOS and CCI, is evaluated in terms of both average deviation and root mean square error (RMSE) by comparing with the in situ SSS monthly gridded EN4.2.2 dataset in the Pacific Ocean. The results show that compared with the in situ EN4.2.2 data, the average deviation and root mean square error of the four SSS products are relatively small in the Pacific subtropical region, while the errors are generally higher in the cold water and offshore regions. Overall, the CCI product has the smallest average deviation and RMSE in the Pacific Ocean, which is the best performance among the four products.


Introduction
Salinity is one of the three most fundamental elements in ocean dynamics and plays an important role in the global atmospheric and oceanic processes such as ocean circulation, mesoscale ocean phenomena, and air-sea interactions [1] .Changes in ocean salinity affect the generation of barrier layers, deep water masses, and the formation of thermohaline circulation.Meanwhile, sea surface salinity is also an important factor affecting the sea air interface, and its seasonal and interannual variability is closely related to the sea air interaction, which is a basic variable for understanding and forecasting climate change [2] .
In the past few decades, most of the measured salinity data have come from Argo drifting buoys, CTDs, moored buoys, and some cruise observations.However, the spatial and temporal resolution of the in situ salinity data cannot effectively represent the fine spatiotemporal changes of ocean salinity.In the past decade, the launch of three ocean salinity satellites (SMOS, Aquarius and SMAP) has solved the problem of the lack of sea surface salinity data.The high temporal and spatial resolution satellite salinity products provide a new horizon for monitoring and analyzing the changes in meso-small scale sea surface salinity, as well as the relationship with ocean thermodynamic processes, the water cycle and climate change.As the three satellites, SMAP, SMOS and Aquarius, have large differences in retrieval methods, spatiotemporal resolution, and error correction strategies, this will undoubtedly result in differences in the quality of different satellite products.Before the application of satellite salinity data, the calibration and validation are guarantees of the accuracy and reliability of the data.W. Tang et al.  (2014) evaluated the salinity products of Aquarius satellite in the global ocean and compared them with Argo and ship buoy data, showing good agreement [3] .Fournier.S et al. (2019) evaluated the salinity products from SMOS and Aquarius satellites in the Arctic Ocean and compared them with Argo buoy data.The results showed similar performance, but with biases in winter and spring [4] .D. Dukhovskoy et al. (2019) evaluated the accuracy of the satellite SSS products in the Arctic Ocean and found that surface salinity measurement accuracy was high, but there were some errors in the deep sea [5] .Castelao et al. (2022) investigated how SMAP and SMOS SSS products capture changes in the narrow coastal zone of western Greenland.The analysis showed that various satellite SSS products capture seasonal salinity changes in the western sea area of Greenland [6] .
Previous studies evaluated the quality of salinity data by comparing satellite SSS products with in situ data, but mainly focused on three satellite data products: SMOS, Aquarius and SMAP.Due to the late release of CCI fusion SSS products, there are few related studies.No study has yet analyzed the three satellite SSS products and the sea surface salinity fusion product (CCI+SSS) products for long time series quality comparison.Therefore, this paper focuses on evaluating the accuracy of different satellite sea surface salinity products and their fusion products by comparing with in situ SSS data.This helps to identify the regions where the retrieval algorithm needs to be improved, and it can provide a reference for the scientific application of the satellite SSS products at different spatial and temporal scales.

Data and Quality metrics
2.1.Data 2.1.1.Satellite Salinity Gridded Products.This article compares the grid data of three salinity satellite products, Aquarius, SMOS, SMAP, and a fused SSS product CCI in the Pacific Ocean with measured data to evaluate the quality of salinity data.To maintain the same spatiotemporal scale, interpolate and grid the salinity satellite product data field with the measured sea salinity data.
Aquarius V4.0 CAP L3 data (hereafter referred to as Aquarius CAP) is an "S-GRID" product developed by the National Center for Atmospheric Research (NCAR).It is a statistical model-based data interpolation method that converts irregularly distributed data points into a regular grid format and fills in missing data.The temporal resolution is averaged month by month and the spatial resolution is 1° x 1°.This product is derived from L2 data regenerated by applying the Combined Active-Passive Algorithm (CAP) [7] , which utilizes real-time data from radiometers and scatterometers, and inverts salinity and wind fields by minimizing the sum of the root mean square errors between the model and the observed data [8] .Meanwhile, Aquarius CAP version V4.0 has revised the salinity deviation of the product and optimized the radiation calculation method.Past versions used the National Centers for Environmental Prediction's wind speeds in roughness correction, and the updated roughness correction improved performance at high wind speeds [9] .
The SMOS gridded products are mainly based on the latest release of SMOS CATDS L3 level V7, covering the data range from January 2010 to November 2021, by the French Research Institute for the Development of the Sea (CATDS).SMOS satellite observation data is processed using a gridded algorithm that interpolates data from discrete points observed by the satellite onto a grid to generate continuous ocean salinity distributions on a grid basis [10] .Compared to the V5 version, in terms of algorithm the new GIBBS algorithm is used for L1 level data processing [11] .For L2 level data processing, a new dielectric constant model and specific radio frequency interference filtering are used to invert L2 SSS and correct the instantaneous effects of rainfall.The debiasing method is similar to V5, but there is estimation bias in different periods and regions.The SSS data has better stability, reducing latitudinal seasonal bias and radio frequency interference contamination.It has a spatial resolution of 0.25° × 0.25° and a temporal resolution of month-by-month averaging [12] .
The main changes in SMAP version V5.0 data compared to version V4.0 are as follows: uncertainty estimation has been added to salinity inversion products and corrected based on external sea ice concentration products [13] .The AMSR-2 TB measurement used in the SMAP V5.0 product retrieves salinity near the edge of the sea ice and helps detect large icebergs near the Arctic [14] .The standard product in this version is a smoothed product with a resolution of about 70 km, which is gridded on a month-by-month time scale and has a spatial resolution of 0.25° × 0.25°.This product is obtained by smoothing.In the open ocean areas, the 70km data product filters out most of the noise, resulting in better data quality.CCI is a monthly averaged SSS gridded L3 product released by the European Space Agency's (ESA) Climate Change Initiative (CCI) alliance for Ocean Surface Salinity.The CCI dataset is based on multiple salinity satellites' long term climate data records, providing an improved calibration of the global sea surface salinity field with a horizontal resolution of 25km.The SSS+CCI project fully utilizes ESA's SMOS mission, supplemented by the SMAP and Aquarius satellite missions [15] .It has successfully reproduced large-scale interannual variability in areas with very large salinity variations, such as the Bay of Bengal and Atlantic River Plume, confirming the strong practicality of the product [16]   .

IN situ Salinity
Data.The EN4.2.2 data is a new version of the EN series datasets from the UK Met Office Hadley Center.This dataset is based on ocean temperature and salinity profile data obtained from WOD09, GTSPP, Argo, and ASBO sets since 1900.The latest version of the EN4.2.2 data was first compared to identify and remove duplicates from all data.Then, after a series of quality controls, the salinity profile series recorded by the Argo buoys were compared to detect errors.The profile data were exported to monthly NetCDF files with a resolution of 1° × 1° [17] .

Quality metrics
In this paper we perform a quality assessment of three satellite gridded products and sea surface salinity fusion products with field measured datasets in the Pacific Ocean to analyze the accuracy of satellite SSS data in terms of two metrics, the mean deviation (Bias) and the root mean square error (RMSE).
Bias is the difference between the mean value of all satellite SSS datasets and the field measured SSS datasets for the corresponding time period.The larger the deviation, the worse the accuracy of the satellite data. =   −   (1) RMSE is the square root of the ratio of the square of the deviation between satellite SSS data and in situ SSS data to the number of observations n.It measures the deviation between satellite data and in situ data and is highly sensitive to outliers in the data.
(2) From the sampling differences between salinity satellites and in situ datasets, the resolution of SMAP, SMOS and CCI fusion SSS products is about 40km, and the resolution of Aquarius is about 120km, while the in situ data, such as EN4.2.2, are real-time measurements of salinity at a single point in the ocean.Therefore, to ensure the accuracy of the salinity product quality assessment, it is important for salinity assessment to keep the different SSS products and real-time measurement datasets matched on the same time and space scales as much as possible.

Results
To set the same spatial and temporal scales, we interpolated all satellite SSS products onto the EN4.2.2 resolution (1 o *1 o ) so that the salinity products have the same spatial resolution as the in situ data.Two different time periods for comparative analysis were determined based on the time scales of salinity measurements by Aquarius and SMAP satellites.The first time period is from August 2011 to June 2015 (totaling 47 months), which we refer to as the Aquarius data time period.The second time period is from April 2015 to September 2020 (totaling 66 months), which is referred to as the SMAP data time period.The latitudinal distribution of the monthly mean SSS bias and the mean bias of the four SSS products and the in situ dataset EN4.2.2 over different time periods are shown in Figure 2 and Figure 3.All four products have biases in the near coastal areas of China, Indonesia and the Gulf of Mexico, with Aquarius and SMAP being more severe.In addition, these four SSS products exhibit significant deviations in high latitudes regions of the Northern Hemisphere due to factors such as sea ice cover and low temperature environments [18] , with Aquarius having a large negative bias and SMAP showing a significant positive bias.While in the Southern Hemisphere (above 60°S), all three SSS products except CCI have negative deviations.It can be seen that the overall deviations of SMOS and CCI from EN4.2.2 data are relatively small in these two time periods.The zonal mean deviations of Aquarius and SMAP become progressively larger with the change of latitude from subtropical to bipolar.Overall, comparisons with in situ data indicate that SMOS and CCI have the smallest deviation from measured salinity data between 40°S and 40°N in the Pacific Ocean, stabilizing within 0.1 PSU, which is a better performance among these four SSS products.Figure 4 shows the distribution of the monthly average root mean square error (RMSE) of the four SSS products compared to the EN4.2.2 in situ data over two time periods.From the figure, it can be seen that the satellite salinity products in the tropical ocean region and non-coastal areas are in better agreement with the in situ data, and the root mean square error is not more than 0.2 PSU.However, larger RMSE values can be observed in the high latitude and in the coastal areas of the Pacific Ocean.Among them, the RMSE of SMOS in the northern Pacific (30°N-50°N, 160°E-160°W) exceeds 0.3 PSU, and the RMSE of Aquarius and SMAP can reach 0.5 PSU for a large area of eastern Asia (40°N-60°N, 100°E-160°E).Figure 5 shows the monthly root mean square error time series between the four SSS products and in situ data.From the figure, it can be seen that the CCI fusion SSS product has the lowest RMSE, with a monthly average RMSE peak of only 0.32 PSU, and the overall monthly RMSE is located in the range of 0.2-0.25 PSU.The RMSE of SMAP is relatively high, with an average value of 0.27 PSU and a peak value of 0.42 PSU.And in the first half of 2015, the SMAP satellites in the early stage of operation there are higher outliers.It is also worth noting that the RMSE of SMOS and CCI have seasonal peaks around August-October in some years, and it is presumed that the inversion effect of the satellites may be affected by certain seasonal phenomena.
Overall, the comparative analysis of root mean square error between various SSS products and in situ salinity data EN4.2.2 has to some extent proven that SSS products can provide reliable ocean salinity information, especially in marine areas far from land, where CCI performs better than the other three products.However, it should be noted that due to various factors affecting SSS products, there is still a certain degree of uncertainty in their accuracy, such as poor performance in coastal and high-latitude waters.In order to study the data quality of the salinity satellite products in different oceanic regions, several representative oceanic regions are selected in this paper.As shown in Figure 6, these regions include the Estuary of the Yangtze River (EYR), as it represents the sea area adjacent to the land and with strong river runoff (ranging from 20°N-35°N, 120°E-135°E); The West Coast of South America (WCSA), as it represents the sea area close to the land coast (ranging from 15°S-40°S，100°W-110°W); The North Pacific (NP), as it represents cold water areas located at high latitudes (ranging from 40°N-65°N, 170°E-140°W); The Eastern Pacific South Equatorial Current (EPSEC) represents the sea area affected by ocean currents (ranging from 0-15°S，100°W-130°W).The results of the comparative analysis of the root mean square error of each SSS product with the actual salinity measurement data EN4.2.2 in the selected regions are shown in Figure 7.In the Eastern Pacific South Equatorial Current and the West Coast of South America, the root-mean-square errors of the four SSS products are generally low, with error values between 0.1 and 0.3 PSU.In the North Pacific, the RMSE of all four SSS products is higher, which may be due to the excessive surface roughness under the strong wind field at high latitudes, and the L-band radiometer becomes less sensitive to salinity and the accuracy of the dielectric model decreases in cold water, as well as the effect of sea-ice contamination [19] .In Aquarius, the RMSE began to decrease in June, and the error remained below 0.3 PSU from July to September, and then began to increase after September, which may be due to the fact that Aquarius can use radar data to make roughness corrections at high wind speeds and the temperature picks up in this time period.In the Estuary of the Yangtze River, the root mean square errors of SMOS, SMAP and CCI are distributed between 0.3 and 0.7 PSU, and all of them show the phenomenon of increasing and then decreasing, reaching the peak of errors in August.The Aquarius SSS already had errors of more than 1 PSU in January-March and December, and the change in error was opposite to the other three SSS products, reaching a low value in July.
In the nearshore waters, there are many factors that can lead to significant salinity deviations.This includes insufficient nearshore observation data, large radiation interference errors, and near ground pollution.Radio frequency interference pollution can lead to lower SSS values, and the large deviation and root mean square error of SMAP near the Yangtze River estuary are mainly due to radio frequency interference pollution.On the one hand, the erosion of the Yangtze River diluted the shallow sea area, resulting in salt stratification in the shallow sea area.Due to the difference in measurement depth between remote sensing data and observation data, there is a significant error between remote sensing data and observation data.At the same time, the error of three SSS products, except for Aquarius, is increasing when the Yangtze River enters the flood season from May to September.On the other hand, it is because the sampling of observation data is not sufficient, and due to the limited observation data in China's offshore areas, it may lead to significant errors in the monthly average grid products of EN4.2.2 in situ data [20] .Therefore, using EN4.2.2 in situ data to verify satellite SSS product data in these areas will result in significant errors.In addition, in the eastern waters of the Southern Equatorial Current, the root mean square error of the four SSS products in March and November is relatively large.This may be due to the presence of rainfall zones in the flow area of the Southern Equatorial Current, especially due to the significant decrease in salinity during the rainy season in March.In winter, the flow area of the Southern Equatorial Current is usually relatively humid, with high air humidity and relatively weak exposure to sunlight, At this point, the evaporation of water in the seawater is suppressed, leading to drastic changes in salinity in the seawater [21] .

Summary
This article conducts precision research and quality evaluation on different satellite sea surface salinity products and their fusion products based on measured ocean salinity data.It has certain application value and scientific significance for relevant scholars to choose suitable and accurate satellite salinity products.
This paper first obtains three salinity satellite products and fusion products from 2011 to 2020, as well as measured sea surface salinity products, and performs spatiotemporal interpolation matching between satellite product sea surface salinity values and on-site measurement values.By comparing the salinity fusion product and three salinity satellite products with the grid measured sea surface salinity product EN4.2.2 dataset, the average deviation and root mean square error of the Pacific region are calculated, and the inversion error size and possible causes of different latitudes and regions are analyzed.
According to the comparative analysis of monthly average scale salinity products and EN4.2.2, the CCI fused SSS products have the smallest deviation and root mean square error across the Pacific Ocean.In addition, both satellite salinity products and fused salinity products have good on-site data consistency in the tropical Pacific Ocean region and the high seas region, while their quality is poor in nearshore river water diluted areas and high latitude sea areas.
With the increasingly widespread application of salinity satellite products, quality evaluation has become increasingly important.In the future, the development of salt satellite product quality assessment can be envisioned from the following aspects: (a).With the continuous progress of technology and the improvement of data processing algorithms, the accuracy of salinity satellite products will continue to improve.It is necessary to strengthen research on data processing methods, correction algorithms, and other aspects.
(b).Establishing a sound quality control system is the key to ensuring the stable and reliable quality of salinity satellite products.In the future, it is necessary to strengthen the construction and practice of standardized operating procedures, on-site monitoring systems, and other aspects.
(c).Due to the complexity and variability of the marine environment, there are still some problems with the current salinity satellite inversion model.In the future, it is necessary to continuously optimize models and develop new sensors to improve data accuracy and coverage.

Figure 2 .Figure 3 .
Figure 2. Spatial distribution of (a) Aquarius, (b) SMOS, (c) CCI satellite deviation from EN. 4.2.2 during the Aquarius data time period, and (d) Latitudinal distribution of different satellite data deviations (in PSU)

Figure 4 .
Figure 4. Spatial distribution of root mean square error between SSS products and measured ocean salinity data EN4.2.2 (the first column is for the Aquarius time period, and the second column is for the SMAP time period)

Figure 5 .
Figure 5. Changes in Monthly Mean Square Root Error of Different SSS Products and Measured Marine Salinity Data EN4.2.2 from August 2011 to September 2020

Figure 6 .
Figure 6.Extent of the location of the four representative areas selected (Yangtze River Estuarine Region (EYR) waters, West Coast of South America (WCSA) waters, North Pacific (NP) waters, and Eastern South Equatorial Current (EPSEC) waters).

Figure 7 .
Figure 7. Changes in Monthly Mean Square Root Error of Different SSS Products and Field Measurements of Marine Salinity in Four Representative Regions EN4.2.2 (a) Estuary of the Yangtze River (EYR), (b) West Coast of South America (WCSA), (c) North Pacific (NP), and (d) Eastern Pacific South Equatorial Current (EPSEC)