ABSTRACT
The paper focuses on the comparison between the National Oceanic and Atmospheric Administration/National Aeronautics and Space Administration Advanced Very High Resolution Radiometer Oceans Pathfinder sea surface temperature (SST) dataset and SST as derived from the Along-Track Scanning Radiometer (ATSR) onboard the European Remote Sensing Satellite (ERS-1) (ASST). These two datasets provide a unique opportunity for comparing, on global scales, two independent satellite-derived SST retrievals. The comparison was done for data between 1992 and June of 1996. In a preliminary step, mean values and standard deviations of the residuals as defined by the differences between the Modified Pathfinder SST (MPFSST) algorithm and the collocated in situ Pathfinder matchup database were calculated. Globally, as defined by the mean difference, the MPFS ST was colder than the in situ data by -0.01 deg C with a standard deviation of 0.54 deg C. However, these results were found to vary between ocean basins. The Caribbean showed the largest difference, with a warm mean difference of 0.24 deg C and a standard deviation of 0.56'C.
Mean differences and standard deviations of the residuals as defined by MPFSST - ASST were calculated. The loss of the 3.7-pm channel onboard the ATSR-1 instrument had a larger effect on the nighttime differences and, thus, application of the model to remove residual cloud cover only had a significant impact on the nighttime statistics. A mean difference of 1.40 deg C, with MPFSST warmer than ASST, and a standard deviation of 0.57 were calculated after the application of the cloud removal model to the ASST. To confirm that part of the differences between the MPFSST and the ASST was due to residual cloud cover, a set of empirical orthogonal functions (EOFs) was extracted from the MPFSST - ASST difference maps, before and after applying the cloud removal model to the ASST. A significant drop from 36% to 14% in the percent variance explained by the first mode indicates that applying the cloud removal algorithm has removed a significant signal from the difference maps. The mean bias for the summation of the first two EOFs is reduced from 0.590 to 0.34'C and the standard deviation from 0.19 to 0.16'C. Thus, a minimum 0.25'C of the signal in the difference maps is due to residual cloud cover in the ASST data. It is concluded that, with improved cloud detection and atmospheric corrections being applied to the ASST, along with improvements to the MPFSST, achieving a O'C mean difference and a standard deviation of < 0.3C for global climate studies is possible.
1. Introduction
Recent results (Merchant and Harris 1999) indicate that sea surface temperature (SST) datasets must still be cross-validated before conclusions can be reached about climate changes due to changes in global SST. Two of these datasets, SSTs as derived from the Along-Track Scanning Radiometer (ATSR) on board the European Remote Sensing Satellite (ERS-1) (ASST), and SSTs from the National Oceanic and Atmospheric Association/National Aeronautics and Space Administration Advanced Very High Resolution Radiometer (AVHRR) Oceans Pathfinder Project (PFSST) (Kilpatrick et al. 2001, hereafter KPE) provide a unique opportunity to compare two independent satellite-derived SST datasets. The independence of the datasets comes from the physically different (design, specs, etc.) instruments. The Pathfinder algorithm uses a best fit to in situ data and thus the SST is tuned toward a bulk temperature, although the AVHRR instrument still senses a skin temperature (KPE). The exact nature of this tuning is still a subject of research. The ASST uses radiative transfer theory to determine SSTs and so does not rely on in situ data. In this approach, radiosonde data, in conjunction with a radiative transfer model, are used to determine the coefficients for given atmospheric conditions. If the goal of reaching 0.1 IC (Merchant and Harris 1999) accuracy for determining climate change is to be reached, a thorough understanding of the differences and errors between these datasets is crucial. Hurrell and Trenberth (1999) show that significant differences still exist between SST datasets and that critical evaluation of these differences is necessary. Unlike comparisons between SST datasets that use common algorithms and in situ data, the ASST and PFSST provide a unique opportunity for comparing SST retrievals that use different algorithms, one physically based and one empirically based.
Biases in the derivation of SSTs from satellites may come from different sources, including atmospheric aerosols and water vapor (KPE; Merchant and Harris 1999). Other biases may occur due to cloud contamination (Jones and Saunders 1996; Jones et al. 1996). For both the PFSST and ASST, the correction for aerosols is difficult to define due to the differences in the sizes, compositions, concentrations, and temporal variability of the aerosols (May et al. 1992). Water vapor acts to attenuate the signal and has been incorporated into models of sea surface temperature (Emery et al. 1994). In addition, a bias due to skin-bulk temperature differences should exist between the PFSST and the ASST. This arises from the tuning of the PFSST toward a bulk temperature, while the ASST measures a skin temperature with no tuning toward an in situ dataset. Thus, part of the differences between these two algorithms should be attributable to the skin-bulk temperature differences. However, this difference is difficult to quantify because of the tuning of the Modified Pathfinder SST (MPFSST) algorithm. Additionally, these differences also vary due to winds and mixing processes (Schluessel et al. 1990). In ideal conditions, light to moderate winds, the skin layer should be cooler than temperatures at, say, 1 -in depth. This, along with the removal of the diurnal variability, justifies the separation, when possible, of all comparisons into daytime and nighttime statistics.
To better understand the differences between these two datasets, a brief review of the algorithms will be given. A more detailed and historical description of the Pathfinder algorithm may be found in KPE.
The purpose of this paper is to statistically compare the SSTs as derived from these two instruments and to determine the space-time characteristics of their differences. It is not to conclude which dataset is better. At the present time the algorithms for both the empirically derived MPFSST and the dynamically derived ASST are being modified and improved; thus, conclusions about which dataset is better are premature and perhaps not necessary. Additionally, results vary regionally, indicating that the different algorithms may perform differently under different oceanic and atmospheric conditions. It is the intent of this paper to indicate where improvements might occur. To achieve this goal, sections 2 and 3 are devoted to a description of the resolutions of the datasets; section 4 describes an algorithm for removing residual cloud contamination in the ASST data; section 5 validates the MPFSST against the matchup in situ database; section 6 compares the MPFSST with the operational satellite multichannel sea surface temperature (MCSST) product, and the ASST with the MCSST; section 7 contains the results of the comparisons between the MPFSST and the ASST data; section 8 examines the empirical orthogonal functions (EOFs) of the difference maps; and section 9 concludes and summarizes the results. The summary and conclusions section will discuss the implications of the results on future research and goals.
2. NOAA/NASA Pathfinder
The MPFSSTs exist in different temporal (daily, 8 day, and monthly) and spatial (9, 18, and 54 km) resolutions and are available through the Physical Oceanography Distributed Active Archive Center (PO.DAAC) at the Jet Propulsion Laboratory, California Institute of Technology. To interpolate the data to a 10 grid a simple Gaussian weighting was applied to the 54-km daily files. The weighting was such that spatial and temporal e-folding scales of 30 and 10 days were applied. Since the interest was to examine statistical differences between the datasets on basin to global scales, the application of these weighting factors provided smoothed global maps of SST, at a 10, 7-day temporal and spatial resolution, respectively, with few gaps due to cloud cover.
To overlap with the time period of ASST availability, the Gaussian weighting was applied to the daily, 54-km MPFSSTs from 1992 to 1996. The smoothing of the MPFSST filled in data gaps, maximizing the amount of collocated and smoothed estimates used in calculating the covariance matrix and not compromising the global statistics. This dataset was then compared with the ASST.
3. Averaged sea surface temperature from the ERS 1 alongtrack scanning radiometer
5. MPFSST validation against the matchup database
To determine how well version 4.1 of the Pathfinder algorithm was performing, collocated in situ SST values from the matchup database (MATCHUP) were compared with the satellite-derived SST values. Table 2 shows the mean, standard deviation, and number of points for the residuals as defined by MPFSST - MATCHUP in the different ocean basins.
The matchup in situ database consists of buoy and ship data that have been used in the regression to determine the coefficients in Eq. (1). Thus, the comparison does not use two independent datasets but nonetheless provides an indication of how well the algorithm is performing over a given ocean basin. For a more detailed description of the matchup database and additional comparisons see KPE. Residuals were calculated as MPFSST - MATCHUP.
Globally, the mean value for the residuals is -0.01 deg C (MPFSST colder than MATCHUP) with a standard deviation of 0.55 deg C. The main result for all the basins is that a standard deviation of approximately 0.5C is identified. These results are consistent with those of KPE, who report an accuracy for the Pathfinder data of 0.02 0.53. The small discrepancy between these two numbers is most likely based on KPE using the entire matchup between 1985 and 1999, while only the version 4.1 matchup database (199496) was used in this analysis. This was done to derive a statistic that was based on the latest version of the algorithm. At the time of this research version 4.1 of the algorithm was available through PO.DAAC for 1994-96. Nonetheless, the results show very close agreement with those of KPE.
Because the matchup data are used directly in the calculation of the coefficients, a mean difference of approximately O'C is expected. However, when the same statistics are examined for different ocean basins, the performance of the algorithm is shown to vary regionally, with the Caribbean showing the largest difference. The North Atlantic has an approximate mean value for the residuals of 0. deg C with a standard deviation of 0.54 deg C. Such a small mean difference and standard deviation could be a function of the large number of matchups found in the North Atlantic. The North Pacific has a mean difference of -0.03 deg C (MPFSST colder than buoy) and a standard deviation of 0.55 deg C. Both the Indian Ocean and South Pacific also have mean differences reflecting a colder MPFSST. These mean values could be due to an inadequate compensation for water vapor in the tropical atmospheres (see KPE). In general, the mean differences are consistent globally except for the Caribbean. The Caribbean has a mean difference of 0.24 deg C (MPFSST warmer than MATCHUP), which could be indicative of an overcompensation due to the atmospheric corrections. All the mean differences are not, however, statistically different from zero, with standard deviations around 0.5C.
6. MPFSST versus MOST and MOST versus ASST comparisons
To further validate the MPFSST and ASST against other datasets, comparisons were done with the MCSST as well as the MPFSST versus the operational Reynolds optimally interpolated (01) weekly data product (RSST). The RSST uses in situ data as well as the National Center for Environmental Prediction (NCEP) operational SST product. The NCEP operational satellite SST dataset uses the same algorithm as the MCSST. Thus, the RSST is a combination of both in situ and AVHRR-derived SST. For more details on the processing of this data see Reynolds and Smith (1994). Figure 1 shows the latitudinal averages between the daytime MPFSST and the RSST. The RSST data are not separated into daytime and nighttime fields, so the comparisons are done only versus the weekly RSST analysis. Figure 1 indicates a bias of -1.0C (MPFSST - RSST) during the Mt. Pinatubo eruption. Reynolds and Smith (1994) found that this bias appears in the RSST when compared with in situ data. The cold bias appearing in all the comparisons indicates that the effect that aerosols have on SST retrieval can be significant and needs to be corrected before conclusions can be reached about global climate change. The mean global difference between the RSST and the nighttime MPFSST is O'C with a standard deviation of 0.49 deg C. However, if latitudes greater than 500 are excluded, the standard deviation is reduced to 0.37C. The standard deviation is not surprising considering the RSST is not separated into daytime and nighttime differences. The daytime-nighttime mean difference of the MPFSST is 0.30 deg C, which indicates that the diurnal warming and cooling could explain a significant amount of the variability found between the MPFSST and the RSST. It is important to note that these numbers should not be treated as an absolute accuracy for the MPFSST because the datasets are not independent, and the operational nature of the NCEP Reynolds OI analysis precludes having used the improved calibration and cloud detection techniques of the MPFSST (KPE). A better comparison is done using the daytime and nighttime MCSST data from 1985 to 1999.
7. MPFSST versus ASST comparisons
The ASST provides a unique independent dataset for comparing directly with the MPFSSTs. Because the in situ data from the pathfinder matchup database is used in the MPFSST algorithm, they unfortunately do not provide an independent assessment of the performance of the algorithm. In addition, neither does the MCSST nor the RSST provide unique fields for assessing the independent accuracy of the MPFSST. This section will deal with a direct comparison between the MPFSSTs and the ASST for the given overlap period between 1992 and 1996. In addition, the statistics will be calculated before and after the application of the cloud removal algorithm to the ASST data in the hopes of identifying possible explanations for the differences between the two satellite-derived SSTs.
Figures 4a and 4b show the latitudinal averages of the differences as defined by MPFSST - ASST for 1992 through the middle of 1996 for both the daytime and nighttime passes before the application of cloud removal. Clearly evident are larger values in the nighttime differences. In addition, a large annual cycle in the nighttime differences is evident in both the Northern and Southern Hemispheres. The annual cycle is not as large in the daytime passes, but both the daytime and nighttime passes show an average mean difference of approximately 10, with the MPFSST colder than the ASST. A negative difference is also evident in the nighttime data at the end of 1992 and the beginning of 1993 between approximately 15 deg S and 20 deg N, consistent with the eruption of Mt. Pinatubo in the Philippines.
Figures 5a and 5b are the same as Figures 4a and 4b, except that the cloud removal algorithm has been applied. The amplitude of the annual cycle in the nighttime data has been reduced significantly, especially in the Southern Hemisphere. However, the negative difference in the nighttime data is most likely due to the loss of the 3.7-,um channel on the ATSR instrument. The presence of the mean difference is most likely due to the inability to use the visible channels in the AVHRR for nighttime atmospheric corrections and/or the loss of the 3.7-pm channel on board the ATSR-1 instrument (Merchant and Harris 1999; Reynolds and Smith 1994; Reynolds and Marisco 1993). Table 4 summarizes the results from the latitudinal averages of the MPFSST - ASST differences.
Before the application of the cloud removal technique, the MPFSST - ASST mean differences (see Table 4) are less than those reported for MCSST - ASST (see Table 3). The reduced mean differences between the MPFSST/ASST and the MCSST/ASST comparisons could be an indication that the MPFSST is performing, on global scales, better than the MCSST. However, such results must be interpreted with caution until independent in situ validation of these datasets, including the ASST, is performed. Although these comparisons were done between the MPFSST with the Gaussian weighting applied and the ASST, results were similar using the noninterpolated MPFSST. For example, mean differences were 1.380 and 1.57 deg C for the daytime and nighttime fields, respectively. Thus, the application of the Gaussian smoothing had minimal effect on the calculation of the global statistics.
Once the cloud removal technique is applied to the MPFSST/ASST comparisons the differences between the nighttime passes is reduced with a mean of 1.40 deg C (MPFSST warmer than ASST) and a standard deviation of 0.6C. These results are consistent with the mean of 1.39 deg C and standard deviation of 0.59'C reported by Merchant and Harris (1999) for the dual-2 Z95 nonaerosol robust coefficients. An attempt was made to differentiate the statistics based on May 1992, when the 3.7-jim channel was lost on board the ATSR instrument. However, no significant differences were noticed when the statistics were separated before and after May 1992. In addition, a second iteration of applying the cloud removal algorithm reduced the mean of the difference by only 0.01 PC, from 1.40 to 1.39'C. It was concluded that, using the annual and semiannual harmonic fit as the algorithm for cloud removal, one iteration effectively removed as much of the cloud contamination as possible. Additionally, the cloud removal algorithm was also applied to the MPFSST data. The mean difference for the nighttime data was reduced from 1.390 to 1.36'C. The mean difference for the daytime data, as seen from Table 4, essentially remained unchanged, with the standard deviation actually increasing. The results indicate that 80% of the reduction in the mean difference in the nighttime fields due to the application of the cloud removal algorithm occurs in the ASST data. These results are consistent with those reported by Merchant and Harris (1999), which were based on a direct comparison with the Tropical Atmosphere-Ocean (TAO) data array in the Tropical Pacific. Results here indicate that the means found in Merchant and Harris (1999) in the Tropical Pacific might be representative of global means and standard deviations. Because Merchant and Harris (1999) compared the ASST directly with the TAO data, no attempt is made in this paper to compare the ASST with buoy or in situ data. The focus here has been on a direct comparison with the MPFSST and global statistics. The point should be made that, because comparisons are made between the daytime and nighttime fields only, the temporal window for matching a MPFSST and ASST value is approximately 12 h. Thus, the time constraint is not as tight as with the MPFSST matchup database, where the window is 30 min and the spatial constraint is _0.1C (KPE). The mean difference between the daytime and nighttime fields of the MPFSST is 0.30C 0.1 PC and the ASST is 0.26 deg C + 0.2'C, indicating that mean differences on the order of PC cannot be explained by the diurnal variability. Thus, the 12-h uncertainty in collocating the MPFSST and ASST cannot cause the observed mean differences.
Figures 6a and bb show the mean difference between the daytime MPFSST and ASST for the entire globe before and after applying the cloud removal. Clearly the cloud removal does not significantly change the differences between the MPFSST and the ASST for the daytime data. Figures 7a and 7b show the same differences for the nighttime data, Several maxima appear off the North American, South American, and African Coasts. None of these maxima appear in the daytime differences, leading one to suspect that they are not due to changes in SST. The maxima, however, are significantly reduced by application of the cloud removal algorithm to the ASST data. To further determine whether the differences might be due to cloud contamination, the flag information in the MPFSST dataset was used to create a map of percent cloud cover based on the tree algorithm used in version 4.1 of the algorithm (KPE). Figures 8a and 8b show the percent cloud cover and number of cloud free pixels defined by version 4.1 of the MPFSST algorithm. Highs in the percent cloud cover are clearly identified with regions associated with maxima in the differences. These results appear to indicate that areas of large differences between the MPFSST and the ASST also result from cloud contamination in either one or both of the SST datasets.
8. EOFs of difference maps based on MPFSST - ASST
To further determine the space-time scales of the differences between MPFSST and ASST, a set of EOFs was calculated based on the maps created from MPFSST - ASST. The EOFs should further confirm whether the differences and standard deviations are due primarily to cloud contamination and identify the space-time scales. Only the EOFs for the nighttime differences are shown because they were also representative of the daytime variability, except that the magnitude of the differences was greater in the nighttime data.
Figure 9 shows the percent variability explained by the first 10 EOFs of the difference maps of MPFSST - ASST before and after applying the cloud removal of Jones et al. (1996).
The triangles indicate the percent variability explained by the first 10 EOFs of the difference maps with no cloud algorithm applied. The first EOF explains approximately 36% of the variability, with the second EOF dropping to 2% of the variance. The diamonds indicate the percent variability explained by the first 10 EOFs of the difference maps with the cloud algorithm applied. The first EOF now explains only 14% of the variability. Clearly, the application of the cloud removal algorithm has significantly reduced the percent variability explained by the first mode. This indicates that after application of the cloud removal algorithm a greater percentage of the signal in the difference maps is now due to noise (Overland and Preisendorfer 1982). Thus, the application of the cloud removal algorithm has been successful in removing a statistically significant signal in the difference maps.
Figures 10a and 10b show the temporal amplitudes for the first two EOFs of the difference maps before cloud removal. Both modes are dominated by an annual cycle in the data. The spatial part of these EOFs (Figs. 11 a,b) indicates that the variability is concentrated in areas in the Pacific off of North and South America, and in the Atlantic off of Africa and along 45 deg S. The second EOF has maxima in areas of the western boundary currents and, once again, off the North American coast.
Figures 12 and 13 show the temporal and spatial parts of the EOFs of the difference maps of MPFSST-ASST with the cloud removal algorithm applied. Figures 11a and 1 lb show the temporal part for the first two EOFs. Both modes once again are dominated by an annual signal.
However, the spatial part of the modes (Figs. 13a,b) indicates that the magnitude of the amplitudes of the EOFs has been reduced from the case with no cloud removal applied. These results are summarized in Table 5.
The first column of Table 5 indicates that the first two EOFs were summed and whether the cloud removal algorithm was applied. Because the application of the cloud removal algorithm did not change the statistics of the daytime comparisons, only the nighttime comparisons are shown. The second column is the mean bias for the summation of the first two EOFs, and the third column is the standard deviation. With no cloud removal applied, a mean bias of 0.59 deg C is accounted for by the first two EOFs. This is consistent with the first two EOFs, explaining 38% of the variability. Table 3 shows that with no cloud removal applied the mean difference, as defined by MPFSST - ASST for the nighttime passes, was 1.56 deg C. With the cloud removal algorithm applied the bias is reduced to 0.34 deg C with a standard deviation of 0.16'C. Although application of the cloud removal algorithm has reduced the mean difference from 0.590 to 0.34'C for the first two EOFs, a significant portion remains to be explained. Barton et al. (1995) did detect a mean difference of 0.40 between retrievals with and without the 3.7-/ann channel. Therefore, it is possible that the remaining mean difference between the MPFSST and the ASST might be due to the loss of the 3.7-/an channel. Other possibilities for explaining the remaining differences after application of the cloud removal is the 12-h collocation uncertainty between the MPFSST and the ASST and/or the skin-bulk temperature differences. Mean daytime-nighttime differences in both the MPFSST and the ASST of 0.3C, along with skin-bulk temperature differences (Schluessel et al. 1990) and the loss of the 3.7-arm channel on board the ATSR- 1 instrument, could account for the remaining variability.
9. Conclusions and summary
To get an independent comparison, SSTs as derived using the MPFSST were compared directly with SSTs from the ATSR instrument on board the European Remote Sensing Satellite ERS-1 (ASST). Such a comparison, with a residual cloud removal applied, yields a global mean for the residuals, as defined by MPFSST - ASST, of approximately 1.40 deg C (ASST colder than MPFSST) and a standard deviation of 0.56 deg C. Diurnal differences on the order of 0.3C cannot explain the remaining mean differences, especially since the comparisons were separated into daytime and nighttime fields. Comparisons were also done between the multichannel sea surface temperature (MCSST) operational product and the ASST. Mean differences for the MCSST/ASST comparisons for both the daytime and nighttime retrievals were greater than for similar MPFSST/ASST comparisons, indicating a possible improvement of the MPFSST over the MCSST. These results must be interpreted with caution until comparisons are done with independent in situ data. The MPFSST-ASST comparisons are consistent with those reported by Merchant and Harris (1999) in the Tropical Pacific using the dual-2 Rutherford Appleton Laboratory prelaunch coefficients, which were not aerosol robust. After application of the residual cloud removal algorithm it appears that there is a mean difference (MPFSST - ASST) around 1.4C with a standard deviation of around 0.6C. Differences on the order of PC were also observed by Perez-- Marerro (1998) in an area off the Canary Islands. To determine the space-time scales of the differences between the MPFSST and ASST, a set of EOFs was extracted from the difference maps before and after cloud removal. Results indicate that up to 36% of the variability could be associated with contamination of clouds in coastal areas off of North America, South America, and Africa. The remaining bias and standard deviation could be explained by the use of the dual-2 algorithm and the loss of the 3.7-jim channel on the ATSR instrument and/or collocation uncertainty/skin-- bulk temperature differences. For purposes of validation, the MPFSST was directly compared with the Pathfinder matchup database, which consists of in situ buoy and ship data.
Globally, version 4.1 of the Pathfinder algorithm (MPFSST), with respect to collocated in situ data from the matchup database, has a mean difference of -0.01C and a standard deviation for the residuals of 0.540C. In this case, the in situ and satellite-derived SSTs are collocated if they lie within a 30 min and 0. iC window. However, these results vary depending on the ocean basin with the North Atlantic having a mean difference of approximately O'C and the Caribbean a mean difference of 0.24'C and a standard deviation of 0.56 deg C. Positive means indicate that the MPFSST is warmer than the collocated in situ data while negative means indicate MPFSST is colder. The global mean and standard deviation of the residuals are encouraging for future uses of the dataset in global climate research. Because the matchup database is used to calculate the version 4.1 MPFSST coefficients, it does not provide an independent check of the algorithm, thus the importance of the comparison with an independent dataset such as the ASST. Merchant et al. (1999) and Brown et al. (1993) show that aerosol-- robust retrievals are possible using two-channel dual view (which is the situation for most of the ATSR- 1 mission). Using improved dual-3 aerosol-robust coefficients, the comparisons with the tropical Pacific data improved significantly. How much of the remaining 1.4C can be attributed to the loss of the 3.7-,um channel on board the ATSR-1 instrument, skin-bulk temperature differences, and/or uncertainty in collocating the two datasets still needs to be answered in future research.
Thus, future research should include global comparisons between improved MPFSST and ASST processed using the current dual-3 aerosol-robust coefficients. Indications are that these mean differences and standard deviations should be reduced significantly using improved coefficients for both the ASST and the MPFSST. A next step in this research needs to compare SSTs from the ATSR-2 mission with the MPFSST to examine if mean differences and standard deviations are reduced. This would answer the question of how much of the remaining differences is due to the loss of the 3.7-,um channel. Work also needs to be done in improving the collocation of the MPFSST and the ASST datasets. Indications are that a maximum of 0.3C of the mean difference could be attributed to uncertainty in collocating the data within a 12-h window. Nonetheless, it has been shown that better cloud detection needs to be a component of future MPFSST and ASST products.
The results are encouraging for future research in the comparison and use of these two datasets in global climate research. Comparisons need to be done using the improved coefficients and cloud detection techniques being applied to the ASST-2 data. It is feasible that, after application of new coefficients and improved aerosol detection in both the MPFSST and the ASST algorithms, a global mean difference of O'C and standard deviation of less than 0.3C can be achieved. This could then lead toward jointly using these datasets in global climate research.
Acknowledgments. The authors would like to thank Richard Reynolds at the National Oceanic and Atmospheric Administration, Robert Evans at the Rosenstiel School of Marine and Atmospheric Sciences, University of Miami, Andy Harris at the U.K. Meteorological Office, and Elizabeth Smith at Old Dominion University for many useful discussions concerning the manuscript. Christopher Mutlow at the Rutherford Appleton Laboratory is kindly thanked for providing the ATSR data. The research was carried out at the Jet Propulsion Laboratory, California Institute of Technology, under contract with the National Aeronautics and Space Administration.
[Reference]
References
[Reference]
Barton, I. J., A. J. Prata, and R. P. Cechet, 1995: Validation of the ATSR in Australian waters. J. Atmos. Oceanic TechnoL, 12, 290-300.
Brown, J., 0. B. Brown, and R. Evans, 1993: Calibration of Advanced Very High Resolution Radiometer infrared channels: A new approach to nonlinear correction. J. Geophys. Res., 98, 18 157-18 269.
[Reference]
Emery, W. J., Y. Yu, and G. A. Wick, 1994: Correcting infrared satellite estimates of sea surface temperature for atmospheric water vapor attenuation. J. Geophys. Res., 99 (C3), 5219-5236.
Evans, R., and G. Podesta, 1996: AVHRR pathfinder SST approach and results. EOS Trans. Amer. Geophys. Union, 77 (46), F-354.
Hurrell, J. W., and K. E. Trenberth, 1999: Global sea surface temperature analysis: Multiple problems and their implications for climate analysis, modeling, and reanalysis. Bull. Amer. Meteor. Soc., 80, 2661-2678.
[Reference]
Jones, M. S., and M. A. Saunders, 1996: Global remnant cloud contamination in the along-track scanning radiometer data: Source and removal. J. Geophys. Res., 101, 12 141-12 147.
-, and T. H. Guymer, 1996: Reducing cloud contamination in ATSR averaged sea surface temperature data. J. Atmos. Oceanic Technol., 13 (2), 492-506.
[Reference]
Kilpatrick, K. A., G. P. Podesta, and R Evans, 2001: Overview of the NOAA/NASA Pathfinder algorithm for sea surface temperature and associated matchup database. J Geophys. Res., in press.
May, D. A., L. L. Stowe, J. D. Hawkins, and E. P. McClain, 1992: A correction for Saharan dust effects on satellite sea surface temperature measurements. J. Geophys. Res., 97, 3611-3619.
McClain, E. P., W. G. Pichel, and C. C. Walton, 1985: Comparative performance of AVHRR-based multichannel sea surface temperatures. J. Geophys. Res., 90, 11 587-11 601.
Merchant, C. J., and A. R. Harris, 1999: Toward the elimination of bias in satellite retrievals of sea surface temperature. Part 2: Comparison with in situ measurements. J. Geophys. Res., 104 (CIO), 23 579-23 590.
M. J. Murray, and A. M. Zavody, 1999: Toward the elimination of bias in satellite retrievals of sea surface temperature. Part 1: Theory, modeling and interalgorithm comparison. J. Geophys. Res., 104 (CIO), 23 565-23 578.
Murray, M. J., M. R. Allen, C. T. Mutlow, and A. M. Zavody, 1998: Actual and potential information in dual-view radiometric
[Reference]
observations of sea surface temperature from ATSR. J. Geophys. Res., 103 (C4), 8153-8165.
Overland, J. E., and R. W. Preisendorfer, 1982: A significance test for principal components applied to a cyclone climatology. Mon. Wea. Rev., 110, 1-4.
Perez-Marrero, J., 1998: Aplicacion de la Radiometria infrarroja a las observaciones oceanograficas en el atlantico centro oriental. Ph.D. thesis, Departamiento de fisica, Universidad de Las Palmas de Gran Canariz, 198 pp.
[Reference]
Reynolds, R. W., and D. C. Marisco, 1993: An improved real-time global sea surface temperature analysis. J. Climate, 6, 114-119.
-, and T. M. Smith, 1994: Improved global sea surface temperature analysis using optimum interpolation. J. Climate, 7, 929-948.
Schluessel, P., E. Emery, H. Grassl, and T. Mammen, 1990: On the bulk-skin temperature difference and its impact on satellite remote sensing of sea surface temperature. J. Geophys. Res., 95, 13 341-13 356.
Walton, C. C., 1988: Nonlinear multichannel algorithm for estimating sea surface temperature with AVHRR satellite data. J. Appl. Meteor., 27, 115-124.
Zavody, A. M., C. T. Mutlow, and D. T. Llwellyn-Jones, 1995: A radiative transfer model for sea surface temperature retrieval for the along-track scanning radiometer. J. Geophys. Res., 100, 937-952.
[Author Affiliation]
Jorge Vazquez-Cuervo and Rosanna Sumagaysay
Jet Propulsion Laboratory, California Institute of Technology, Pasadena, California
[Author Affiliation]
Corresponding author address: Dr. Jorge Vazquez-Cuervo, JPL/ Caltech, M/S 300/323,4800 Oak Grove Dr., Pasadena, CA 91109.
E-mail: jv@pacific.jpl.nasa.gov
In final form 28 November 2000.
(c) 2001 American Meteorological Society
