Evaluation of NOAA/AVHRR Sea Surface Temperature at Full HRPT Resolution in the Northwest Pacific Ocean

CHEN Yan, QU Liqin, *, and GUAN Lei

Evaluation of NOAA/AVHRR Sea Surface Temperature at Full HRPT Resolution in the Northwest Pacific Ocean

CHEN Yan1), 2), QU Liqin1), 2), *, and GUAN Lei1), 2)

1),,,266100,2),,266237,

TheNational Oceanic and Atmospheric Administration (NOAA) Polar Orbiting Environmental Satellites (POES) High Resolution Picture Transmission (HRPT) data in the Northwest Pacific Ocean has been acquired through the SeaSpace ground station located at the Ocean University of China since October 2000, and these data have been processed by the TeraScan system. The sea surface temperature (SST) products in the Northwest Pacific Ocean derived from Advanced Very High Resolution Radiometer (AVHRR) are evaluated. We compared the SST products with the buoy SSTs during the stable operational period of each satellite. There are a total of 33715 and 71819 matchups acquired for daytime and nighttime, respectively, between the NOAA/AVHRR SSTs and buoy SSTs. For each satellite, the biases and standard deviations at daytime are smaller than those at nighttime. The monthly biases at daytime generally oscillate around 0℃, except for NOAA-15. By contrast, the monthly biases at nighttime mostly oscillate around −0.5℃. Both daytime and nighttime biases exhibit seasonal oscillations for all satellites. The seasonal biases of the SST difference at daytime between each satellite and buoy are mostly within±0.25℃, except for the negative bias of −0.58℃ in May for NOAA-18. The seasonal biases of the SST difference at nighttime are mostly around −0.5℃, and NOAA-16 has a lower bias,., −0.86℃, in April. These results indicate that the accuracy of the SST products is inconsistent for each satellite during different periods. It is suggested that the NOAA/AVHRR data should be reprocessed to provide highly accurate SST products.

NOAA/AVHRR HRPT data; sea surface temperature (SST); buoy data; validation; Northwest Pacific Ocean

1 Introduction

The Northwest Pacific Ocean is a crucial marginal sea inthe Pacific Ocean. It includes East China Sea, South China Sea, Yellow Sea, and Sea of Japan. Moreover, the Kuro- shio Current, the world’s second-warmest current, flows through it. The mean state and variation of sea surface tem- perature (SST) over the Northwest Pacific Ocean are the key to study regional air-sea interaction (Sakaida and Ka- wamura, 1992; Lee, 2005). SST can initially be collected from multiplemeasurements, such as ships, buoys and offshore platforms, and these measurements are usually accurate but are limited by time and space. During the past decades, satellites are widely used owing to their high spatial-temporal resolution. Furthermore, SSTs can be retrieved from thermal infrared and passive microwave satellites’ sensors, but they have their own advantages and shortcomings (Wentz, 2000; Emery, 2001). Infrared sensors have high spatial resolution, but the observations for SST are affected by cloud and aerosols (Guan and Kawamura, 2003). Microwave sensors can penetrate cloud and aerosols (Reynolds, 1993), but the observations for SST are affected by rain (Rapp, 2008)and side lobe contamination near land in coastal waters (Castro, 2012), also they have lower spatial resolution than infrared sensors.

The Advanced Very High Resolution Radiometer (AV- HRR) on theNational Oceanic and Atmospheric Administration (NOAA) Polar Orbiting Environmental Satellites (POES) have been providing SST products at a high spatial resolution of 1.1km since the 1980s. There have been a number of studies on the accuracy of SSTs from NOAA/ AVHRR, particularly on the development of SST validations, algorithms, and applications. A common method to validate satellite SST accuracy is to compare the collocated satellite SSTs with theSSTs. The multichannel SST (MCSST) and nonlinear SST (NLSST) algorithms havebeen proposed by using brightness temperature differences among thermal infrared channels to improve the accuracy of the SST data (McClain, 1985; Walton, 1988). TheNOAA/AVHRR and buoy matchups between 1989 and 1997 revealed that the differences between NOAA/AV- HRR retrieved SSTs and buoy SSTs ranged within 0.2–0.4℃ over the nine-year period (Walton, 1998). The SSTs derived from NOAA-12 and NOAA-14 have been validated withSSTs and their biases both have been found to be positive , but the biases of NOAA-14, which were less than 0.5℃, were lower than those of NOAA-12 (Li, 2001). AVHRR SSTs retrieved by using an op- timal estimation with a simple empirical bias correction model have been obtained with bias and standard deviations (SD) of −0.06℃ and 0.44℃ (Merchant, 2008). Validation of the AVHRR-derived SSTs withSSTs has revealed that the biases and SDs of AVHRR SSTs were −0.43±0.76℃ and −0.33±0.79℃for daytime and nighttime and the regional biaseswere large in the northern South China Sea (Qiu, 2009). AVHRR SSTs have been validated with drifter data over the seas around South Korea, and the root-mean-square errors (RMSEs) of MCSST and NLSST were evaluated to be less than 1℃ in most cases, and the dependencies of these biases on atmospheric and oceanic conditions were revealed (Park, 2011). Detailed validation results for all SST data in Climate Change Initiative (CCI) phase 1 products have reported that daytime SSTs in the SST CCI Along Track Scanning Radiometer (ATSR) and AVHRR products were generally noisier than nighttime SSTs and have larger robust standard deviation (RSD) relative to global drifting buoys (Corlett, 2014). AVHRR SSTs have been va- lidated in the East Japan Sea by using surface drifter SST as ground truths from 2005 to 2010, and the SST biases (satellite-drifter) have also demonstrated diurnal variationswith a relatively higher RMSE from 0.80℃ to 1.00℃ during nighttime and a smaller RMSE of approximately 0.50℃ during daytime (Park, 2015). NOAA has re- processed the AVHRR/3 global area coverage (4km) SST data from five NOAA and two Metop satellites from 2002 to 2015, by using variable regression SST coefficients in each satellite’s most stable operational periods, thus the stability of the SST time series has been further improved (Ignatov, 2016). The validation analyses have been conducted for all levels (L2P, L3U and L3C) of both the SST CCI ATSR and SST CCI AVHRR records within phase 2 of the ESA SST CCI project in order to offer sa- tellite-based time series of SSTs since 1981 for climate applications (Merchant, 2019). Therefore, the detail- ed validation of the SSTs derived from NOAA/AVHRR should be conducted to analyze the accuracy of the AVHRR SST products.

In this study, the AVHRR SSTs in the Northwest Pacific Ocean are evaluated by usingSST quality mo- nitor (iQuam) buoy SSTs asmeasurements. First, the AVHRR anddata are introduced in Section 2. Then, the comparison results between AVHRR andSSTs are discussed in Section 3. Finally, the conclusions are presented in Section 4.

2 Data and Methods

2.1 NOAA/AVHRR Data

In this study, the NOAA/AVHRR data have been acquired through the SeaSpace ground station located at Ocean University of China (OUC) since October 2000, and these data have been processed by the TeraScan system. The TeraScan system separates the AVHRR data from the HRPT dataset, and then the AVHRR SSTs have been retrieved through cloud detection, geometric correction and MCSST algorithms. The SST products are projected to a 0.01˚× 0.01˚ grid within 10˚–50˚N and 105˚–145˚E. Herein, we select the SST data from the AVHRRs on several satellites during their stable operational periods (Merchant, 2019), and the details are shown in Table 1. The coverage of the OUC ground station is shown in Fig.1.

Table 1 Summary of the characteristics of the satellite data used

Fig.1 The data coverage area received by the OUC ground station(inside the red circle).

2.2 iQuam in situ Data

TheSST data are from theSST quality monitor (iQuam) system developed by the NOAA NESDIS/STAR. The iQuam data files preserve much information, including SST, wind direction, wind speed and so on (Xu and Ignatov, 2013). The iQuam data include ship and buoy data, but we only select the buoy data with the highest data quality as thedataset. The monthly iQuam data files have been provided online in the NetCDF format, and can be downloaded from its website (ftp://ftp.star.ne sdis.noaa.gov/pub/sod/sst/iquam/v2.10/). Moreover, the i- Quam version 2.10 data have been used.

2.3 Generating the Matchups of Satellite and in situ SST

In this study, we obtained a matchup dataset between AVHRR and buoy data within a spatial window of 0.01˚ and a temporal window of 0.5h. Also we used the 5×5 block to detect the cloud data, and only when all SST data in the block around the center pixel of AVHRR SST are valid values, the pixel can be selected into the matchup dataset. The pixels wherein the solar zenith angles were below 85˚ were selected as the daytime dataset and the rest were put in the nighttime dataset (Ackerman, 2010; Wang, 2014). The overall flowchart of the matchup procedure is shown in Fig.2.

3 Comparisons Between the AVHRR and in situ SST Data

3.1 Statistical Analysis to the AVHRR and in situ SST Matchups

To evaluate the accuracies of AVHRR SST, we calculated the statistic parameters, including the matchup num- bers, bias, the minimum SST difference (min), the maximum SST difference (max), median, SD and RSD by using the matchup data. The RSD used herein was 1.48 times the median absolute deviation from the median (Embury, 2012). The data were eliminated beyond thrice of the RSD from the median as outliers (Bevington and Robinson, 2003; O’Carroll, 2008; Dash, 2012).By removing the outliers, the numbers of daytime and night- time matchups were reduced by 3.71% and 4.27%, respectively. Finally, 33715 and 71819 matchups for daytime andnighttime between the AVHRR SSTs and buoy SSTs were obtained. The number and statistic parameters of the ma- tchups for each satellite are summarized in Table 2. The number of daytime matchups for each satellite is less than that at nighttime, and the biases and SD values exhibit si- milar trends. The daytime biases range from −0.08(NOAA- 17) to −0.32℃ (NOAA-16), and the nighttimebiases range from −0.32℃ (NOAA-15) to −0.60℃ (NOAA-16).The daytime SDs range from 0.62℃(NOAA-17) to 0.75℃ (NOAA-15), and the nighttime SDs range from 0.76℃ (NOAA-17) to 0.82℃ (NOAA-16).

Fig.2 Flowchart for the generation of the matchup datasets between AVHRR and in situ data.

Table 2 Statistics of the AVHRR and in situ SST matchup datasets for daytime and nighttime

3.2 Characteristics of the AVHRR and in situ Matchups

Fig.3 shows the monthly distributions of the number of matchups between each satellite anddata. The mon- thlynumbers of daytime matchups for each satellite are smaller than those at nighttime.The numbers of matchups in May and October are greater than those of the other months for each year, and it reaches the maximum in October owing to the cloud cover variation in the work region. No collocated points exist for some months because the ground station did not acquire the data during those months. For a more precise look at this seasonal feature, Fig.4 illustrates the seasonal distributions of the total num- ber of matchups from 2000 to 2017 between each satellite anddata.The numbers of matchups in spring (Marchto May) and autumn (September to November) are greater than those of other seasons at both daytime and nighttime. This is consistent with the distribution shown in Fig.3.

The probability distribution of the SST differences between AVHRR andmatchups is shown in Fig.5. For all satellites, the differences between AVHRR andSSTs show a nearly symmetrical distribution at daytime, of which over 81.5% are within±1℃, and 54.1% are within±0.5℃. However, over 71.1% of the nighttime SST differences are within±1℃, and 41.6% of the nighttime SST differences are within±0.5℃. Moreover, the overall trend of the nighttime SST differences exhibits clear cold biases with a high frequency, indicating that most AVHRR SSTs are lower thanSSTs at nighttime.

Fig.6 presents the comparisons between NOAA satellite andSSTs at daytime and nighttime for the overall matchups. Most of the satellite SSTs seem to be well correlated with theSSTs by showing a linear relationship for both daytime and nighttime. The majority of the matchups lie in the temperature range from 20℃ to 30℃, and most of the SST differences are within±1.0℃. However, the satellite SSTs generally show a certain degree of underestimation at a relatively low temperature range from 5℃ to 15℃ at nighttime, particularly for NOAA-16 and NOAA-18. This requires a more detailed study on these satellites.

Fig.3 Monthly distributions of the number of matchups between AVHRR and in situ data: daytime (upper panel) and nighttime (lower panel).

Fig.4 Seasonal distributions of the number of matchups between AVHRR and in situ data: daytime (left panel) and nighttime (right panel).

Fig.5 Probability distribution histogram of SST differences between AVHRR and in situ matchups: daytime (left panels) and nighttime (right panels).

Fig.6 Scatterplots of SST matchups between AVHRR and in situ data: daytime (left panels) and nighttime (right panels).

To evaluate the stability of the satellite SSTs, we show the monthly variations of the bias and SD for AVHRR andSST differences in Fig.7. For NOAA-15, the monthlybiases generally oscillate between −1.06℃and 0.56℃ atdaytime and between −1.09℃ and 0.75℃ at nighttime. For NOAA-16, the monthly biases generally oscillate between −0.68℃and 0.03℃ at daytime and between −1.06℃and −0.01℃ at nighttime. For NOAA-17, the monthly biases generally oscillate between −0.64℃and 0.75℃ at daytime and between −1.04℃ and 0.09℃ at nighttime. For NOAA-18, the monthly biases generally oscillate be- tween −0.88℃and 0.21℃ at daytime and between −1.27℃ and −0.03℃ at nighttime. For NOAA-19, the monthly biases generally oscillate between −0.86℃and 0.20℃at daytime and between −0.94℃ and 0.17℃ at nighttime. Furthermore, the daytime and nighttime biases exhibit seasonal periodical oscillations for all satellites, and it can be seen that the biases during spring (March to May) are generally lower than those during other seasons in each year, particularly in 2008.

Fig.8 shows the seasonal SST difference biases and SDs between each satellite anddata. The seasonal daytime biases generally change within ±0.25℃ except forthe maximum negative bias of −0.58℃ in May for NOAA-18, and the nighttime biases generally change around −0.5℃ except for the maximum negative bias of −0.86℃ in April for NOAA-16. The biases in spring (March to May) are lower than those in other seasons during both daytime and nighttime, and this is consistent with the distribution shown in Fig.7.

4 Conclusions

In this study, we evaluated the accuracy of NOAA/ AVHRR SSTs retrieved from the OUC SeaSpace ground station using the high-quality buoy data. For each satellite, the biases and standard deviations at daytime are smaller than those at nighttime. The monthly biases at daytime generally oscillate around 0℃, except for NOAA-15. By contrast, the monthly biases at nighttime mostly oscillate around −0.5℃. Both daytime and nighttime biases exhibit seasonal oscillations for all satellites. The seasonal biases at daytime are mostly within ±0.25℃, except for the ne- gative bias of −0.58℃ in May for NOAA-18. The seasonal biases at nighttime are mostly around −0.5℃, and only NOAA-16 has a lower bias,., −0.86℃, in spring. Overall, NOAA-17 exhibits the best performance and NOAA-16 performs worst. The quality of the AVHRR SSTs in the Northwest Pacific Ocean are not as good as previous studies. For example, the AVHRR SSTs retrieved by using an optimal estimation with a simple empirical bias correction model have the bias and standard deviations (SD) of −0.06℃ and 0.44℃ (Merchant, 2008).

Fig.7 Time series of the monthly biases (dots) and SDs (bars) for AVHRR and in situ SST differences at daytime (red) and nighttime (black). (a)–(e) represent NOAA-15 to NOAA-19, respectively.

Fig.8 Seasonal distributions of the biases and SDs between AVHRR and in situ data: daytime (upper panel) and nighttime (lower panel).

Above all, the accuracy of the SST data is inconsistent for each satellite during different periods. The failures of cloud detection may cause large underestimation at nighttime. It is suggested that the NOAA/AVHRR data from the OUC SeaSpace ground station should be reprocessed to improve the cloud detection and SST retrieval accuracy.

Acknowledgement

This work has been supported by the National Key R& D Program of China (No. 2019YFA0607001).

Ackerman, S., Strabala, K., Menzel, P., Frey, R., Moeller, C., and Gumley, L., 2010. Discriminating clear-sky from cloud with MODIS algorithm theoretical basis document (MOD35). MODIS Cloud Mask Team, Cooperative Institute for Meteorological Satellite Studies, University of Wisconsin.

Bevington, P. R., and Robinson, D. K., 2003. Estimates of mean and errors. In:3rd edition. McGraw-Hill, New York, 1-55.

Castro, S. L., Wick, G. A., and Emery, W. J., 2012. Evaluation of the relative performance of sea surface temperature mea- surements from different types of drifting and moored buoys using satellite-derived reference products., 117: C02029.

Corlett, G., Atkinson, C., Rayner, N., Good, S., Fiedler, E., Mc- Laren, A.,., 2014. Product validation and intercomparison report. Project Document SST_CCIPVIR-UoL-001. http:// www.esa-sst-cci.org/PUG/documents.

Dash, P., Ignatov, A., Martin, M., Donlon, C., Brasnett, B., Rey- nolds, R. W.,., 2012. Group for high resolution sea surface temperature (GHRSST) analysis fields inter-comparisons- Part 2: Near real time web-based level 4 SST quality monitor (L4-SQUAM)., 77-80: 31-43.

Embury, O., Merchant, C. J., and Corlett, G. K., 2012. A repro- cessing for climate of sea surface temperature from the along- track scanning radiometers: Initial validation, accounting for skin and diurnal variability effects., 116 (4): 62-78.

Emery, W., Castro, S., Wick, G., Schluessel, P., and Donlon, C., 2001. Estimating sea surface temperature from infrared satellite andtemperature data., 82: 2773-2785.

Guan, L., and Kawamura, H., 2003. Study on the SST availabi- lities of satellite infrared and microwave measurements., 59 (2): 201-209.

Ignatov, A., Zhou, X. J., Petrenko, B., Liang, X. M., Kihai, Y., Dash P.,., 2016. AVHRR GAC SST reanalysis version 1 (RAN1)., 8 (4): 315, DOI: 10.3390/rs8040315.

Lee, M. A., Chang, Y., Sakaida, F., Kawamura, H., Cheng, C. H., Chan, J. W.,., 2005. Validation of satellite-derived seasurface temperatures for waters around Taiwan, Terrestrial., 16 (5): 1189-1204.

Li, X., Pichel, W., Clemente-Colon, P., Krasnopolsky, V., and Sapper, J., 2001. Validation of coastal sea and lake surface temperature measurements derived from NOAA/AVHRR data., 22 (7): 1285-1303.

McClain, E. P., Pichel, W. G., and Walton, C. C., 1985. Compa- rative performance of AVHRR-based multichannel sea surface temperatures., 90: 11587- 11601.

Merchant, C. J., Borgne, P. L., Marsouin, A., and Roquet, H., 2008.Optimal estimation of sea surface temperature from split-win- dow observations., 112 (5): 2469- 2484.

Merchant, C. J., Embury, O., Bulgin, C. E., Block, T., Corlett, G. K., Fiedler, E.,., 2019. Satellite-based time-series of sea surface temperature since 1981 for climate applications., 6: 223, https://doi.org/10.1038/s41597-019-0236-x.

O’Carroll, A. G., Eyre, J. R., and Saunders, R. W., 2008. Three- way error analysis between AATSR, AMSR-E, andsea surface temperature observations., 25 (7): 1197-1207.

Park, K. A., Lee, E. Y., Chung, S. R., and Sohn, E. H., 2011. Accuracy assessment of sea surface temperature from NOAA/ AVHRR data in the seas around Korea and error characteristics., 27 (6): 663-675.

Park, K. A., Lee, E. Y., Li, X. F., Chung, S. R., Sohn, E. H., and Hong, S., 2015. NOAA/AVHRR sea surface temperature accuracy in the East/Japan Sea., 8 (10): 784-804.

Qiu, C. H., Wang, D. X., Kawamura, H., Guan, L., and Qin, H. L., 2009. Validation of AVHRR and TMI-derived sea surface temperature in the northern South China Sea., 29: 2358-236.

Rapp, A. D., Kummerow, C., and Elsaesser, C., 2008. On the ef- fects of warm rain clouds in the tropics.. Fort Collins, Colorado, 0897.

Reynolds, R. W., 1993. Impact of Mount Pinatubo aerosols on satellite-derived sea surface temperatures., 6: 768-774.

Sakaida, F., and Kawamura, H., 1992. Estimation of sea surface temperatures around Japan using the advanced very high re- solution radiometer, (AVHRR)/NOAA-11., 48 (2): 179-192.

Walton, C. C., 1988. Nonlinear multichannel algorithm for estima- ting sea surface temperature with AVHRR satellite data., 27: 115-124.

Walton, C. C., Pichel, W. G., Sapper, J. F., and May, D. A., 1998.The development and operational application of non-linear algorithms for the measurement of sea surface temperatures with the NOAA polar-orbiting environmental satellites., 103 (C12): 27999-28012.

Wang, S., Cui, P., Zhang, P., Ran, M., Lu, F., and Wang, W., 2014. FY-3C/VIRR SST algorithm and cal/val activities at NSMC/CMA., 9261: 92610G-2.

Wentz, F., Gentemann, C., Smith, D., and Chelton, D., 2000. Sa- tellite measurements of sea surface temperature through clouds., 288: 847-850.

Xu, F., and Ignatov, A., 2013.SST quality monitor (iQuam)., 31 (1): 164- 180, DOI: 10.1175/JTECH-D-13-00121.1.

August 4, 2020;

October 9, 2020;

January 13, 2021

© Ocean University of China, Science Press and Springer-Verlag GmbH Germany 2021

Tel: 0086-532-66782907

E-mail: quliqin@ouc.edu.cn

(Edited by Chen Wenwen)