This study’s comparison results between the SMAP and SMOS L3 daily product validated against the daily Argo profiles and moored-buoy data are presented in the following section. Tang et al. (2017) have previously validated SMAP L3 daily and monthly products obtained from JPL with in-situ salinity, and have confirmed that the SMAP SSS had the ability to accurately reflect the variabilities of the SSS on weekly and monthly timescales. In contrast with the method used by Tang et al. (2017), the daily in-situ SSS measurements were not averaged within eight-day periods, and only daily products were used in this study to detect the daily SSS errors of the satellite products.
Considering the fact that numbers of the daily Argo floats are lower in some regions, and they are inconsistent in the spatial resolution between the Argo floats and the satellites, two methods were taken in order to detect the differences between Argo floats and the satellites. First, the scatter plots were obtained by comparing satellite data with the nearest Argo floats within a distance of 0.5°, since the satellites represent the spatial average of the instantaneous measurements in a footprint of approximately 40 km. Second, the global distribution maps of the SSS errors were obtained by comparing the bias and RMSE between the satellite and the nearest Argo floats within a distance of 2.5°, since the Argo floats were sparsely distributed and the SSS has small changes in the distance of 2.5°.
Figure 3 details the scatter plots of the SMOS and SMAP daily SSS versus the Argo floats data from January 1, 2016 to December 31, 2016. The numbers of the collocated data had differed, the numbers of SMAP dataset (n=95 713) have a higher spatial resolution, and was therefore larger than that of the SMOS (n=42 799). As shown in Fig. 3, both the SMOS and SMAP daily SSS data have agreed with the Argo data. The results of this study’s comparison of the two scatter plots revealed that the SMAP had more high-salinity data between 38 and 40 than the SMOS. These differences were assumed to have probably been caused by the higher spatial resolution of the SMAP data, which were enabled the capture of a greater number of salinity signals. In addition, the SMOS SSS data displayed more deviations from the Argo data, with a larger RMSE (0.723) than that of the SMAP data (0.491). These results may due to the 8-day averaged process used in the SMAP daily product retrieval procedure.
Figure 3. Scatter plots of the SMOS (a) and SMAP (b) daily SSS data versus the Argo floats data SSS from January 1, 2016 to December 31, 2016.
As shown in Fig. 4, the distributions of the bias and RMSE for the SMOS and SMAP data versus the Argo data were observed to be inconsistent in the global ocean areas. It can be seen in the figure that the RMSE for the SMOS and SMAP products had similar distributions. Generally, both were lower in the subtropical regions and open ocean areas and were larger near the coastal areas and in the high-latitude ocean regions. Comparisons between the RMSE for the SMOS and SMAP products revealed that that the former was larger. However, the distribution of the bias of the two products was found to differ. For example, the SMAP had displayed a larger bias in the western Pacific Ocean warm pool, while the SMOS had a larger bias in the eastern Pacific cold tongue. These observed differences were assumed to have probably resulted from high precipitation and strong wind conditions, respectively. The bias of both products was found to be larger along the coastal and high-latitude regions, which were similar with the distribution patterns obtained from the monthly product data. The large coastal bias was assumed to have probably resulted from influences of RFI and land contamination.
Figure 4. Global distribution of the: bias of SSS (a, b); and RMSE of SSS (c, d) between the SMOS (a, c) and SMAP (b, d) versus the Argo floats data during the year 2016
In order to further evaluate the SSS error distributions, this study divided the ocean into five areas as follows: Indian Ocean (30°N–40°S, 40°–100°E); Pacific (60°N–40°S, 110°E–90°W); Atlantic (60°N–40°S, 70°W–20°E); west coast of the North Pacific (20°–50°N, 125°–145°E) and the open ocean areas of the Pacific (0°–40°N, 165°E–150°W). The last two areas were referred to in this study as the WCNP and OPPA, and were used to assess the errors between the coastal and open ocean regions. Table 1 lists the bias and RMSE for SMOS and SMAP data versus the Argo floats data within 0.5°, which were calculated for the five aforementioned areas. It was found that the bias of the SMOS was smaller than that of the SMAP. Meanwhile, the RMSE of the SMAP was observed to be smaller than that of the SMOS. The SMAP SSS have generally displayed a negative bias, while the sign of the SMOS SSS bias have been found to vary. Comparisons between the errors in the WCNP and OPPA indicated that the bias and RMSE in the WCNP were twice as large as those in the OPPA. The absolute bias and RMSE values for the WCNP had reached 0.3 and 0.8 , respectively. These findings suggested that quality control measures were required when using satellite SSS in the coastal regions. However, the satellite SSS was found to be of good quality in the open ocean areas, with the average daily bias and RMSE determined to be 0.039 and 0.372 , respectively. In this study, a comparison of the three largest oceans (Indian, Pacific, and Atlantic) showed that the errors between the Indian Ocean and Pacific were nearly the same, while the errors in the Atlantic had differed. The bias and RMSE of the Atlantic have nearly reached the values observed in the coastal areas, which may have been caused by the sparsity and locations of the Argo floats in the Atlantic (Fig. 2).
SMOS (daily) SMAP (daily) Bias RMSE Bias RMSE Global 0.007 0.723 −0.096 0.556 Indian Ocean −0.007 0.702 −0.065 0.432 Pacific 0.043 0.612 −0.050 0.443 Atlantic −0.217 0.886 −0.137 0.582 WCNP −0.206 0.864 −0.364 0.672 OPPA 0.039 0.454 0.039 0.372
Table 1. Bias and RMSE between the satellite and Argo SSS
For the purpose of further analyzing the large errors in the Atlantic and the error distributions of the three largest oceans, the number of matched Argo floats (n), and the bias and RMSE of the SMOS and SMAP versus the Argo data, were calculated in each latitude band (10°) for the Indian Ocean, Pacific, and Atlantic, respectively, as detailed in Fig. 5. Then, the monthly mean (based on the daily mean) precipitation, wind speed at 10 m, and SST were used in this study’s correlation analysis process in order to determine the contributions to the latitude-dependent error distributions of the SSS. The values for the n, bias, and RMSE were scaled in order to show within the same plot. To clarify, the values for the N and bias were scaled using different coefficients for each of the oceans, whereas the RMSE was scaled using one coefficient. Figure 5a and d show that the RMSE for each latitude band of the SMAP was less than that of the SMOS in the Indian Ocean. However, the bias showed the opposite trend. It was observed that the RMSE was lower in the tropical regions and increased with latitude. In regard to the SMOS, the minimum value occurred in the 30°S to 40°S band (0.382), while the minimum value of the SMAP occurred between 10°S and 30°S (0.328). The bias of the SMOS displayed small absolute values, which were lower than 0.04 and reached 0.000 5 in the 30°S to 40°S band. However, it was found that the bias of the SMAP, which displayed larger bias values, followed the same trend as the precipitation (correlation coefficient of 0.82). In addition, comparisons between the n, bias, and RMSE revealed that the number of matched Argo floats did not have any significant influences on the bias or the RMSE. The bias and RMSE were found to be mainly dependent on the locations of the Argo floats. As shown in Figs 5b and c, the Pacific had the highest number of matched Argos floats. In addition, the latitude-dependent distributions of the n and RMSE for the SMOS and SMAP were observed to be nearly the same. Both distributions displayed valleys (0.329 for the SMAP and 0.421 for the SMOS) in the 20°S to 30°S band, and peak values occurred at the high latitudes. A large RMSE also existed in the 10°N to 20°N band for the SMOS. Meanwhile, the largest RMSE for the SMAP occurred in the 0° to 10°N band. It was observed that unlike the SMOS, the bias of the SMAP also showed a precipitation trend (correlation coefficient of 0.81). Furthermore, the latitude-dependent RMSE of the SSS in the Pacific was found to be primarily influenced by the SST (correlation coefficient of –0.79). The Atlantic was confirmed to have the lowest N, with low values occurring particularly between 20°S and 20°N (Figs. 5c and f), which explained why it had displayed the largest errors. In the Atlantic, the SMAP had a lower RMSE than the SMOS, as was also the case for the Indian Ocean and Pacific. The RMSE values of both the SMOS and SMAP displayed valleys in the 20°S to 0° and 20°N to 30°N bands, and peak values occurred at the high latitudes. However, unlike the other two oceans, the bias of the SMAP in the Atlantic was found to be lower than that of the SMOS, while the RMSE of the SMAP remained lower than that of the SMOS. Correlation analysis indicated that the latitude-dependent RMSE of the SSS in the Atlantic was also primarily influenced by the SST (correlation coefficient of −0.66).
Figure 5. Distributions of the number of matched Argo floats (n; blue); bias (green); and RMSE (red) between the daily satellite and Argo floats SSS in each latitude band of the Indian Ocean (a, d), Pacific (b, e) and Atlantic (c, f).
In summary, comparisons between the daily SSS products of SMAP and SMOS versus the Argo floats data showed that the bias of the SMOS was lower than that of the SMAP. Meanwhile, the RMSE showed the opposite trend. The RMSE and bias were found to vary in the global oceans, with larger values along the coastal and high latitude regions, and lower values in the open ocean and tropical regions. These were found to be primarily caused by the influences of the SST and RFI on the SSS inversions. Moreover, the latitude-dependent RMSE of the SMAP was primarily influenced by the SST. It was observed that lower RMSE values consistently occurred with higher SST, lower wind speed, and weaker precipitation conditions. The bias of the SMAP was confirmed to be highly related to the precipitation conditions, which could be further used to correct the bias of the SMAP.
The daily measured SSS at a 1 m depth from the RAMA, TAO, and PIRATA buoy arrays during 2016 were used in this study. Since the RAMA, TAO, and PIRATA are moored-buoy arrays located in the tropical Indian Ocean, Pacific, and Atlantic, respectively, comparisons with the satellite SSS allowed for the performances of the satellite SSS products in the three largest oceans to be further explored. As shown in Fig. 6, despite the fact that they displayed different biases and RMSE values, the distribution patterns and high-density areas of the SMOS and SMAP products versus the buoy array data were obviously similar. It was observed that unlike the previous comparisons with the Argo data (Fig. 3), both the bias and RMSE of the SMAP (0.09 to 0.14 ; 0.26 to 0.37, respectively) were lower than those of the SMOS (−0.2 to 0.13 ; 0.56 to 0.57, respectively). Moreover, the comparison results of the scatter plots for the various oceans showed that the daily SMAP products had lower RMSE (0.255 ) in the Atlantic.
Figure 6. Scatter plots of the SMOS and SMAP daily SSS versus the moored buoy data, from January 1, 2016 to December 31, 2016. a. SMOS versus RAMA, b. SMOS versus TAO, c. SMOS versus PIRATA, d. SMAP versus RAMA, e. SMAP versus TAO, and f. SMAP versus PIRATA.
The time series of the SSS data at three buoys located in the Indian Ocean, Pacific, and Atlantic are shown in Fig. 7. It can be seen in the figure that the SMAP daily product have similar variabilities as those of the moored buoys in the three largest oceans, with correlation coefficients of approximately 0.8. In addition, the bias and RMSE values were calculated for the satellite SSS L3 daily products and the buoy SSS data. It was observed that at all three locations, the RMSE was always the largest for the SMOS daily data, whereas the SMAP daily data displayed a good agreement with the buoy data, with biases of approximately 0.1, and RMSE values lower than 0.3, as detailed in Table 2.
4.0°S, 80.5°E 0°, 156°E 0°, 10°W bias RMSE bias RMSE bias RMSE SMOS-daily –0.06 0.54 0.05 0.46 –0.36 0.68 SMAP-daily 0.03 0.28 0.17 0.30 0.14 0.24
Table 2. Bias and RMSE between the satellite and buoy SSS at three locations
Figure 7. Time series of the daily SMOS (blue) and daily SMAP (red) SSS products versus the moored buoys (black) at 4°S, 80.5°E (a), 0°, 156°E (b) and 0°, 10°W (c).
In conclusion, comparisons of the daily SMOS and SMAP SSS versus moored-buoy array data indicated that the RMSE of the daily SMAP product was lower than that of the daily SMOS product.
Validation and correction of sea surface salinity retrieval from SMAP
- Received Date: 2018-10-18
- Available Online: 2020-04-21
- Publish Date: 2020-03-01
- sea surface salinity (SSS) /
- soil moisture active passive (SMAP) /
- soil moisture and ocean salinity (SMOS) /
- validation /
Abstract: In this study, sea surface salinity (SSS) Level 3 (L3) daily product derived from soil moisture active passive (SMAP) during the year 2016, was validated and compared with SSS daily products derived from soil Moisture and ocean salinity (SMOS) and in-situ measurements. Generally, the root mean square error (RMSE) of the daily SSS products is larger along the coastal areas and at high latitudes and is smaller in the tropical regions and open oceans. Comparisons between the two types of daily satellite SSS product revealed that the RMSE was higher in the daily SMOS product than in the SMAP, whereas the bias of the daily SMOS was observed to be less than that of the SMAP when compared with Argo floats data. In addition, the latitude-dependent bias and RMSE of the SMAP SSS were found to be primarily influenced by the precipitation and the sea surface temperature (SST). Then, a regression analysis method which has adopted the precipitation and SST data was used to correct the larger bias of the daily SMAP product. It was confirmed that the corrected daily SMAP product could be used for assimilation in high-resolution forecast models, due to the fact that it was demonstrated to be unbiased and much closer to the in-situ measurements than the original uncorrected SMAP product.
|Citation:||Sisi Qin, Hui Wang, Jiang Zhu, Liying Wan, Yu Zhang, Haoyun Wang. Validation and correction of sea surface salinity retrieval from SMAP[J]. Acta Oceanologica Sinica, 2020, 39(3): 148-158. doi: 10.1007/s13131-020-1533-0|