1. Introduction
Sea surface temperature (SST) is one of the key essential climate variables that can be used for climate change detection, air–sea flux calculations, and assimilation into ocean and atmospheric models (Donlon et al. 2010; Hollmann et al. 2013). In the last decades, many satellites have been able to observe the SST with good accuracy, and several satellite-based global SST products have been produced (Reynolds et al. 2002; Roberts-Jones et al. 2012; Merchant et al. 2012). These products have been used to monitor and determine global-scale climatic trends and variability of SST (IPCC 2013). On a regional scale, however, the SST variability and error characteristics may be significantly different from the global patterns (Marsouin et al. 2015; Merchant et al. 2006; Høyer et al. 2012). The Baltic Sea and the North Sea are regions that require special attention regarding the production of satellite SST records and the assessment of climatic variability (Høyer and She 2007; Karagali and Høyer 2014). Both the North Sea and the Baltic Sea are semienclosed basins, highly variable and influenced by large-scale atmospheric processes and by the vicinity of the land. In addition, the Baltic Sea is one of the largest brackish seas in the world. When analyzing regional-scale climate variability, all these effects have to be considered, which requires dedicated regional and validated SST products. Previously, sparse in situ observations and shorter records of nonvalidated satellite observations have been used to analyze the climatic SST signals in the North Sea and Baltic Sea (BACC Author Team 2008; BACC II Author Team 2015; Siegel et al. 2008; Lehmann et al. 2011). In this paper, we present a regional multisensor SST climate data record (CDR) developed with consideration of the regional conditions that apply to the North Sea and Baltic Sea. A daily gap-filled level-4 SST analysis is constructed from 1982 to 2012. A thorough validation against independent in situ observations has been performed to assess the performance of the SST CDR, with emphasis on the temporal stability of the record on annual and interannual time scales.
The satellite and in situ observational records used in this analysis are described in section 2. Section 3 presents the combination of these observations to construct the level-4 data record. Validation against independent data is presented in section 4, and results from the initial analysis of the climate data record are described in section 5. A summary and conclusions are found in section 6.
2. Data
a. In situ observations
The in situ observations used in this study are from the International Comprehensive Ocean–Atmosphere Data Set (ICOADS), version 2.5 (Woodruff et al. 2011, 1987), and have been quality controlled as described in Rayner et al. (2006). ICOADS consists of SST observations from drifting buoys, moored buoys, and ships. Only observations that have passed all quality control procedures are used in this study. The typical reference datasets used for climate assessment and satellite SST validation are either drifting buoy observations, Argo floats, or the global tropical moored buoy array (Hollmann et al. 2013; Merchant et al. 2012). However, because of the shallow water depths and the proximity to the coast, Argo floats and drifting buoys do not survive long in the North Sea and Baltic Sea, and other data types have to be used for the assessment, such as moored buoys and ship observations. From the positions of the drifting buoy observations, it was evident that drifters in, for example, the Baltic Sea were reporting temperature observations when ashore. In addition, ship observations from harbors had also passed the quality control procedures. To increase the quality of the in situ observations and avoid spurious values, all ship and drifting buoy observations within 2 km of the coast were discarded using a high-resolution land mask. Not all the available observations from moored buoys were included in ICOADS, and the in situ data were thus supplemented with quality controlled moored buoy observations from the Marine Environmental Monitoring Network in the North Sea and Baltic Sea (MARNET). Only the observations from the shallowest sensor were used for each buoy, provided that the sensor depth was not more than 4 m. It was shown in Karagali et al. (2012) that MARNET observations at depth were in general slightly colder than the near-surface observations. However for depths shallower than 4 m, this effect is <0.05°C. Figure 1 shows the number of available in situ observations for bins of 1° latitude and 2° longitude. The large number of observations in the German bight and Danish waters is due to the location of moored buoys from MARNET with long time series.
Total number of in situ SST observations in 1° latitude by 2° longitude bins, including drifting buoys, moored buoys, and ship observations.
Citation: Journal of Climate 29, 7; 10.1175/JCLI-D-15-0663.1
b. Satellite observations
The main satellite datasets used here are the Advanced Very High Resolution Radiometer (AVHRR) Pathfinder dataset, version 5.2, (Casey et al. 2010) and the Along-Track Scanning Radiometer (ATSR) Reprocessing for Climate (ARC) dataset (Merchant et al. 2012; Embury and Merchant 2012), version 1.1.1. Both satellites series are from instruments on polar orbiting platforms. The Pathfinder dataset spans from September 1981 to December 2012 and includes observations from the polar orbiting satellites NOAA-7, -9, -11, -14, -16, -17, -18, and -19. The spatial resolution of the Pathfinder observations is 4 km globally, with separate day and night products. The SST observations are thermal infrared observations from the AVHRR instrument and are therefore limited by cloud cover (Kilpatrick et al. 2001). The Pathfinder SST retrieval algorithm is based on the nonlinear split window algorithm (Walton et al. 1998), which has been used in several SST retrieval algorithms (Merchant et al. 2009; Pichel et al. 2001). The coefficients in the retrieval algorithm are determined through regression toward in situ observations, and the dataset thus represents the subskin temperature of the oceans (see, e.g., Donlon et al. 2010). Subskin observations are subject to diurnal warming effects, which can be significant in the Baltic Sea (Karagali et al. 2012; Karagali and Høyer 2014); therefore, only nighttime passes have been included in the analysis to minimize this effect. Only observations with quality levels from 4 to 7 have been included in the analysis. Validation studies have shown the Pathfinder AVHRR data to be accurate within 0.5°C (Marullo et al. 2007) in comparisons with in situ observations in the Mediterranean Sea. However, it is well known that AVHRR SST observations retrieved from a nonlinear split window algorithm can display significant seasonal bias variations in the middle and high latitudes due to atmospheric water vapor variability (Høyer et al. 2012; Kumar et al. 2003; Marsouin et al. 2015; Kearns et al. 2000).
The ARC dataset spans from the beginning of the ATSR-1 observations in August 1991 through the end of the Advanced ATSR (AATSR) instrument in April 2012 and has a spatial resolution of 0.1° on a regular latitude and longitude grid. The (A)ATSR instruments also observe in the infrared with a cloud limitation. The instruments are very well calibrated and with a high accuracy and have a dual view capability where all locations are observed twice, with different pathlengths through the atmosphere (Smith et al. 2012; Zavody et al. 1994). The ARC SST retrieval coefficients are derived using physical-based radiative transfer model simulations, and the algorithm is thus independent of in situ observations (Merchant et al. 2012).
The swath width of the (A)ATSR instruments is 500 km, compared to 2700 km for the AVHRRs, and the data return of the (A)ATSR instruments is about 10% of the Pathfinder data return in the region considered here. Several SST estimates are available in the ARC dataset. In this study, we decided to use the SST estimated at 1 m, where skin and diurnal variability effects have been accounted for. To increase the number of observations from the ARC dataset, it was decided to use both day and night data. The ARC dataset has been shown to be very accurate and stable, with a standard deviation of the differences between 0.2° and 0.3°C when compared against global drifting buoy observations (Embury et al. 2012). The ARC and Pathfinder datasets are thus complementary with respect to data coverage and accuracy.
c. Ice
Ice information is typically included in the global level-4 analysis through the use of global sea ice products based on microwave satellite observations. However, most of the global operational sea ice fields are derived using microwave observations, which are severely contaminated 50–100 km from the coast (Andersen et al. 2006; Comiso et al. 2003). In this study, we use a sea ice concentration dataset with a high spatial resolution of 5 km. It is produced by the Swedish Meteorological and Hydrological Institute (SMHI) and originates from digitized ice charts. The temporal resolution during the ice season is twice a week in the beginning of the period and almost daily in the last years. In case of missing data for a given day, the ice field that was closest in time was used to construct a daily sea ice record for all days from 1982 to 2012. The average number of days in a year with an ice concentration >15% in the SST product is shown in Fig. 2. It demonstrates that sea ice occurs most frequently in the Bay of Bothnia, with up to 100 ice covered days per year. However, as evident from Fig. 2, sea ice can occur in all parts of the Baltic Sea and Danish straits, demonstrating the need for careful treatment of sea ice in the SST analysis.
Average number of days in a year with ice concentration >15% from digitized ice charts.
Citation: Journal of Climate 29, 7; 10.1175/JCLI-D-15-0663.1
3. Construction of the satellite climate data record
The ARC dataset has been shown to be accurate, but with a limited data return, whereas the Pathfinder dataset has a better data return but also higher uncertainties. The complementarity between these two datasets and the data gaps because of clouds in both sets are good reasons to construct a level-4 product, which combines the two datasets and interpolates to produce a gap-free product. The SST observations have been combined with the ice mask in an optimum interpolation method to construct daily SST fields for the North Sea and Baltic Sea with a spatial resolution of 0.03° on a regular latitude and longitude grid. The level-4 grid has a higher resolution than the satellite datasets to be able to resolve the complex coastline in the Danish straits. The central time for the analysis is 0000 UTC (midnight), and observations within two days from the central time are included.
Several global reanalyses of sea surface temperature are available (Reynolds et al. 2002, 2007; Roberts-Jones et al. 2012). However, it has been demonstrated that estimating accurate regional and high-latitude satellite SSTs is more challenging than for the global applications (Høyer and She 2004, 2007; Høyer et al. 2012). The method used here was developed with special attention to the regional characteristics in middle and high latitudes (Høyer and She 2007) with a dynamical high-latitude bias correction method that uses accurate observations to adjust and improve less accurate satellite records (Høyer et al. 2014).
Based upon validation results (Embury et al. 2012; Lean and Saunders 2013), it was decided to use the ARC observations from August 1991 as the reference data against which the Pathfinder observations were adjusted dynamically with a temporal window of 7 days, as described in Høyer et al. (2014). Before the availability of ATSR-1 data in August 1991, the Pathfinder satellite observations were adjusted against in situ observations. To ensure a robust bias calculation, all types of ICOADS in situ observations were used for the reference, supplemented with three selected MARNET buoys (Fehmarn, EMS, and Deutsche Bucht) not included in the ICOADS. In addition, the bias correction window was increased from 7 to 15 days. The increased temporal window is slightly higher than the typical temporal scale of the SST errors as reported in Høyer et al. (2012) but was necessary to increase the number of observations available for reference. About 10% of the ICOADS data prior to 1991 (selected as the data from every 10th day) were withheld from the bias calculation to be used for the validation. After 1991, all ICOADS data were used for validation. In addition, the MARNET buoy records not used for the bias calculation were used for validation. The different observational products used for constructing the dataset are shown in Fig. 3, also outlining the observational product used as a reference against which the Pathfinder observations have been corrected. The mean bias corrections applied to Pathfinder observations for the years 2003–11 with respect to the AATSR observations are shown in Fig. 4. Figure 4 demonstrates a large annual variation in the Pathfinder data, with small biases in winter and significant negative biases during summer, with a peak in July. The shift from a reference based on in situ observations to the ATSR-1 satellite reference occurred in 14 August 1991. A consistent climate data record should be independent of the type of observations used as reference for calculating the bias corrections. The calculated Pathfinder biases during 1991 are included in Fig. 4 (dashed line) and show consistent bias values before and after the shift in the reference dataset at day 226. The spatial distribution of the mean monthly biases for July 2003–11 is shown in Fig. 5 and demonstrates that negative biases are seen throughout the full domain. However, a significant spatial variation in the Pathfinder corrections is also evident throughout the region. Pathfinder corrections from −0.5° to 0.8°C are seen in most of the Baltic Sea, whereas the North Sea typically displays negative biases from −0.2° to −0.4°C.
Overview of the availability of the datasets used to construct the SST analysis. The blue bars indicate the period for which the specific data type was used as a reference product for Pathfinder.
Citation: Journal of Climate 29, 7; 10.1175/JCLI-D-15-0663.1
Average Pathfinder biases (°C) for the years 2003–11 with respect to the AATSR observations (solid) and Pathfinder biases for the year 1991 (dashed) with respect to ICOADS observations before 14 Aug (day 226) and ATSR-1 afterward.
Citation: Journal of Climate 29, 7; 10.1175/JCLI-D-15-0663.1
Mean July bias for Pathfinder for years 2003–11 with respect to the AATSR data.
Citation: Journal of Climate 29, 7; 10.1175/JCLI-D-15-0663.1
Standard deviation of the first guess errors used for the optimum interpolation.
Citation: Journal of Climate 29, 7; 10.1175/JCLI-D-15-0663.1
4. Validation of the climate data record
In situ observations not included in the analysis have been used to perform an independent validation of the climate data record. The validation includes observations of all three data types available in ICOADS (drifting buoys, moored buoys, and ship observations), supplemented with the MARNET moored buoy observations not included in ICOADS. Both day and night observations have been used for the validation to ensure robust statistics for the spatial and temporal validation details. The difference in the overall validation was small, about 0.01°C between the day and night statistics, probably a result of the moored buoy and ship observations being at 1–5-m depths, where diurnal SST variability is reduced. In addition, the vast majority of drifting buoy observations are from the Norwegian Sea and the North Atlantic, where the diurnal variability is limited. Matching pairs of satellite analysis data and in situ observations were extracted throughout the record to perform the validation. Each in situ observation was paired with the satellite analysis SST value from the field that was closest in time (i.e., a maximum of 12 h apart). The value from the pixel containing the observation was extracted. Satellite analysis versus in situ pairs from regions covered with ice were discarded, and a filter was applied to the matchup pairs, removing matches that deviated more than 3 times the standard deviation from the mean differences. This quality control procedure was applied separately for each of the in situ data types to ensure a representative validation statistic. About 1%–1.5% of the matches were discarded in this quality control.
The overall validation statistics for the 30-yr daily fields are shown in Table 1 for the three types of observations included in the ICOADS data. The table shows a very good performance of the satellite analysis, concerning bias and standard deviation when compared against drifting buoy and moored buoy observations, whereas the ship observations display larger differences. The larger errors on the ship observations are in agreement with Kennedy et al. (2011) and are probably related to the way the observations are performed (i.e., by measuring the temperature of the engine intake water or the ship hull). The different observation types also include a sampling pattern. The vast majority of the drifting buoy observations are from the Atlantic Ocean and Norwegian Sea, with a few observations in the central Baltic Sea and almost none from the North Sea. The moored buoy observations are primarily from the coastal areas, with a few buoys in the central Baltic Sea. The ship observations cover the full domain, with highest density along the main shipping routes in the region. Table 1 also shows a difference in the performance of the satellite analysis for different regions, where the meridian of 9°W in the Skagerrak was chosen as the border between the North Sea and Baltic Sea. The North Sea validation statistics for bias and standard deviations are similar or better than the performance of global SST analysis (Roberts-Jones et al. 2012). This is in contrast to the validation statistics from the Baltic Sea, which show significantly larger standard deviations of the differences, compared to the North Sea, Norwegian Sea, and Atlantic Ocean. This applies for all in situ types and irrespectively of the sampling pattern of the in situ observation type. The elevated errors in the Baltic Sea are anticipated from the spatial distributions of the first guess errors (Fig. 6), the Pathfinder biases (Fig. 5), and the satellite validation results in Høyer and She (2004). The best agreement between satellite analysis and in situ observations is obtained with the moored buoy observations, which was not expected, considering that these observations are often placed in regional environments. However, the good performance indicates that the maintenance and quality control of the moored buoy observations in this region is very effective.
Overall validation statistics for the full climate data record for the full region (ALL), Baltic Sea (BS), and North Sea (NS), against drifting buoys, moored buoys, and ship observations not included in the analysis.
Yearly mean validation statistics have been calculated throughout the record to assess the temporal performance of the climate data record. Figure 7 shows the yearly mean biases and standard deviations for each of the in situ observation types and the number of observations used for the statistics. A minimum of 50 matches is required before a mean has been calculated. The large reduction in the number of matches prior to 1992 is because the majority of the data have been used for reference and therefore not included in the validation. The yearly validation results demonstrate that the yearly differences between the satellite analysis and in situ observations of drifting and moored buoys are, in general, <0.1°C, with insignificant temporal trends in the biases, taken over the full data record. Fitting a straight line to the differences between the CDR and in situ observations yields trends of 0.01°C decade−1 for both the drifting and moored buoys. The results from the ship comparisons stand out, as was the case for the overall statistics. The differences compared to ship observations are small from 1982 to 1995 but reach −0.3°C in the years before and after year 2000. These bias variations result in a linear trend of 0.09°C decade−1 for the ship comparisons. Figure 8 shows the performance of the product throughout the year, with 15 days average validation statistics for all in situ types. The drifting buoy comparison shows insignificant seasonal variability in the bias, whereas the moored buoys and the ship comparisons are smallest in winter and largest during summer. The largest variation is seen in the ship comparisons that reach −0.33°C in the beginning of March, whereas the moored buoy seasonal bias variation is less pronounced, with a minimum during February of −0.17°C.
Yearly validation statistics from drifting buoys (black), moored buoys (green), and ship observations (red). (top) Mean biases are solid lines with circles, and standard deviations are solid with crosses (°C). The dashed lines show the best linear fit to the biases. (bottom) The number of yearly satellite analysis vs in situ matchups throughout the record.
Citation: Journal of Climate 29, 7; 10.1175/JCLI-D-15-0663.1
The 15-day validation statistics during the year from drifting buoys (black), moored buoys (green), and ship observations (red). (top) Mean biases are solid lines with circles, and standard deviations are solid with crosses (°C). (bottom) The number of satellite analysis vs in situ matchups throughout the year.
Citation: Journal of Climate 29, 7; 10.1175/JCLI-D-15-0663.1
All data types show a seasonal variation in the standard deviations, with highest values during summer and lowest in the wintertime. This is in agreement with the results obtained in Høyer et al. (2012, 2014) and is probably attributed to mid- and high-latitude retrieval errors in the level-2 satellite products. In general, the validation of the CDR product demonstrates a very stable performance over the full record. If the comparisons against ship observations are disregarded, the performance of the SST CDR yearly validation meets the updated Global Climate Observing System (GCOS) requirements of a horizontal resolution of 10 km, an accuracy of 0.1°C, and a stability of 0.03°C decade−1 (GCOS 2006, 2010), thus making the product very suitable for detailed investigations of the SST variability with an interannual and decadal focus.
5. Analysis of the climate data record
It is outside the scope of this paper to examine the climate data record in detail. We will, however, present some results from the initial analysis of the interannual and decadal variability in the dataset and thus demonstrate the advantages of this dataset, compared to, for example, discrete in situ time series and coarse-resolution gridded datasets. To do this, mean monthly, annual SST, and anomaly SST time series have been produced from the daily reanalysis SST fields.
Monthly mean SST fields were calculated by averaging the daily SST fields of the month, and the yearly mean SST is the mean of all the mean monthly SST fields of the year. The monthly climatological SST was generated for each discrete month by averaging of all the available years. The monthly variability of SST in the region covered by the analysis is very large because of the continental influence, with annual amplitudes exceeding 12°C in the Baltic Sea (not shown). To reduce the seasonal impact, a time series was constructed of monthly anomalies, defined as the mean monthly SST minus the climatological SST of the same month. The original spatial resolution of 0.03° in latitude and longitude was maintained in the averaged products. To focus on interannual and decadal variability and reduce the seasonal variability, the monthly anomalies were further smoothed by applying a 13-month running mean, estimated as the average of the 6 months before and after a given month.
The climate data record is ideal for examining the spatial representativeness of the SST time series and defining the natural boundaries for a North Sea, a Danish strait, and a Baltic Sea region. The spatial correlations of the smoothed monthly anomalies were calculated for three positions representative of the central parts of the North Sea (56°N, 4.5°E), the Baltic Sea (56°N, 18°E) and the Danish straits (56.5°N, 11.5°E). The results are shown in Fig. 9. In the North Sea, correlations above 0.9 are limited to within the basin, whereas areas with correlation between 0.85 and 0.9 extend along the Norwegian coast and offshore of the United Kingdom. Correlations with the Danish straits and the Baltic Sea are lower than 0.8. When the Danish straits are considered, correlation values higher than 0.85 extend both toward the North Sea and the Baltic Sea; and, especially for the latter, correlation values reach up to 0.95. For the location in the central Baltic Sea, correlations higher than 0.85 extend as far away as the middle of the North Sea, the coastal parts of southern Norway, and the southern part of the Bay of Bothnia.
Spatial correlations of interannual SST anomalies for three selected locations in the (top) North Sea, (middle) Danish straits, and (bottom) Baltic Sea. The central positions are marked with the gray circle.
Citation: Journal of Climate 29, 7; 10.1175/JCLI-D-15-0663.1
The correlation regions were used to determine the boundaries of the three subregions. The North Sea domain was defined as the area constrained by an upper latitude of 60°N and the longitudes 3°W and 9°E. The Danish straits were defined as the region between 9° and 14°E, with an upper latitude limit of 60°N; and the central Baltic Sea extended eastward from 14°E and southward from a northern latitude of 59°N, to exclude the Gulf of Bothnia and Gulf of Finland, which have very strong seasonal cycles, as they freeze over most winters.
The mean monthly anomalies of all grid points within the defined domains were averaged to produce the mean monthly anomaly time series shown in Fig. 10 and their corresponding running means. Large variability of the monthly anomalies is identified for the Danish straits and the Baltic Sea, compared to the North Sea. This is justified by their proximity to land, enclosed nature, and shallower depths, giving rise to water temperatures more closely approximating land surface temperatures. In addition, variations in the frontal positions between the brackish waters in the Baltic Sea and the salty waters of the North Sea are likely to enhance the SST variability in the Danish straits. The smoothed SST anomaly curves for the three regions generally display correlation in the SST variations. The coldest period is seen in the beginning of the record, with 1987 being the coldest for all three regions. Two other cold periods are found in 1994 and 1996. All regions show that 2007 was by far the warmest year, with the smoothed anomalies being 0.5°–1°C warmer than any other year. The 2007 anomalies are highest for the Danish strait region and smallest for the North Sea. A general decrease in the anomalies is seen from 2007 to 2012 for all the regions, with negative anomalies during 2010 and 2011. In spite of these cold years, the overall linear trends from a least squares fit to the monthly anomalies are positive for all three regions, with slopes of 0.037° ± 0.008°C yr−1 for the North Sea, 0.039° ± 0.014°C yr−1 for the Danish straits, and 0.041° ± 0.013°C yr−1 for the Baltic Sea region. The confidence intervals mark the 95% limits for the slopes. These regional trends are a factor of 3–4 times higher than the reported global estimates of approximately 0.01°C yr−1 (IPCC 2013). The reason for this regional amplification is probably due to the variability in the larger-scale atmospheric circulation patterns. Decadal variability in the North Atlantic Oscillation index (Hurrell 1995) has been shown to be connected with the regional Baltic Sea climate (Moberg and Jones 2005), and the enhanced positive trend in the SST is in agreement with the Atlantic multidecadal oscillation (Schlesinger and Ramankutty 1994), which has been increasing during the period we are looking at here.
Average monthly SST anomalies (gray line) for the (top) North Sea, (middle) Danish straits, and (bottom) Baltic Sea. The solid black line is the 1-yr running mean, and the dashed line is the best linear fit.
Citation: Journal of Climate 29, 7; 10.1175/JCLI-D-15-0663.1
The spatially complete SST reanalysis fields make it possible to examine the spatial variations in the trends by calculating the linear trends for every grid point. This was done by fitting a straight line to the time series of monthly mean SST values from each grid point, for the entire 30-yr period of available data. Figure 11 shows the spatial distribution of trends, where values of 0.05°C yr−1 are identified in large parts of the domain. Locally, trends of up to 0.08°C yr−1 are found, both in coastal areas and in the open ocean, while inside the Gulf of Finland the value reaches up to 0.1°C yr−1. The smallest trends are found in the Nordic seas, north of the Faroe Islands, and in the Bay of Biscay.
Trends from a linear regression to the 30-yr monthly anomaly time series.
Citation: Journal of Climate 29, 7; 10.1175/JCLI-D-15-0663.1
The overall patterns in the SST trends in the Baltic Sea are in agreement with the trends reported in Lehmann et al. (2011) but with lower trend estimates, especially in the Danish straits and the central Baltic Sea. The difference is likely to be explained by the difference in the record length of 19 years (1990–2008), compared to the 30 years used for this analysis. A pattern with smaller trends along the southern part of the Swedish south coast is recognized in Fig. 11. This corresponds to an area characterized by frequent upwelling (Lehmann et al. 2012), and the smaller trends in this region might indicate an increase of the upwelling events over the last years. These findings are in agreement with Lehmann et al. (2011), who suggested that this was due to a change in the wind directions.
The calculation of annual averages in regions with seasonal ice cover is not a trivial task for the Baltic Sea, where large interannual variability and a general decrease in the sea ice has been observed from 1982 to today (Haapala et al. 2015). In this study, we have included ice covered regions, with a surface temperature of −1°C, to avoid a misrepresentation of the yearly average from missing data during the sea ice covered periods. The constant SST under ice of −1°C is an assumption that is not valid everywhere, because of the surface salinity gradients in the Baltic Sea ranging from almost fresh waters in the Bay of Bothnia to a salinity of up to 32 psu in the Danish straits and 35 psu in the northern part for the North Sea (Janssen et al. 1999). The corresponding freezing points for seawater are −0.24°C for a salinity of 3 psu and −2°C for a salinity of 35 psu.
To remove any effects from interannual and decadal changes in partially sea ice covered regions, the 30-yr trends have been calculated for the months of July–September. These three months were chosen as they have the warmest temperatures and no sea ice in any part of the domain. The spatial patterns of the linear trends fitted to every grid point are shown in Fig. 12. Note the different scaling on the color bar compared to Fig. 11. Figure 12 shows that the 30-yr trend in the summer is significantly larger than the trend for all the months. This is particularly the case in the Bothnian Sea and the northeastern part of the central Baltic Sea. The trends in these regions correspond to a 3°–5°C warming in July–September during the 30 years, compared to the 1°–2°C warming when all months during the year are taken into account.
Trends from a linear regression to the 30-yr monthly July–September anomaly time series.
Citation: Journal of Climate 29, 7; 10.1175/JCLI-D-15-0663.1
6. Summary and conclusions
The study has demonstrated that with a statistical merging and interpolation method designed for the mid- and high-latitude coastal and shelf seas, it is achievable to construct a regional SST climate data record that meets the GCOS requirements.
The collection and quality control of in situ data are crucial for regional applications. The usual SST in situ reference datasets, such as drifting buoys and profiling Argo floats, cannot be used alone to assess the quality of the dataset produced here because of the lack of observations in this region. Additional quality control on the drifting buoys was necessary to remove observations from the stranded buoys, and the database was supplemented with moored buoys and ship observations in order to obtain a spatially representative in situ dataset for an independent validation of the time series.
The value of a multisensor satellite product has been demonstrated in this study. The Pathfinder data have a much better coverage than the ARC data but also have larger uncertainties, with a seasonal variability of the bias that can reach −0.5°C during the summer months. Including both datasets in the analysis with a dynamical correction of the Pathfinder observations toward the ARC data utilizes the benefits of each dataset. Larger standard deviations during summer months were seen in the validation of the final level-4 product, but the seasonal bias variation was minimized. The use of in situ observations as a reference against which the Pathfinder observations can be corrected has been demonstrated to be effective, with a seamless transition from the in situ–based reference to the ARC-based reference. Elevated uncertainties in the in situ reference dataset compared to the ARC dataset and higher uncertainties on the earlier AVHRR instruments are probably responsible for the higher uncertainties in the beginning of the period.
Performing a proper validation of regional SST products in coastal and shelf seas can be challenging because of the lack of the traditional in situ reference observations, such as the drifting buoys and the profiling Argo floats. The SST observations obtained from ships have a better coverage in the North Sea and Baltic Sea, but they also have significantly larger uncertainties. In this region, the best comparisons between in situ observations and the satellite analysis were obtained using the moored buoys’ observations from, for example, the MARNET buoys. Small differences were seen despite the coastal locations of the buoy and sensor depths that can be up to 5 m. This indicates that the calibration and quality control from the data producers is efficient and that the moored buoys are important supplements to the in situ reference observations in this region. The importance of adding more data to the regional validation was evident from the validation results. All data types indicate that the Baltic Sea SSTs have higher uncertainties than the North Sea, Norwegian Sea, and North Atlantic. This conclusion would be hard to reach with the same confidence if only drifting buoys and Argo floats were used to assess the quality, which is often the case for global products. The overall validation results for the SST product with biases <0.1°C and standard deviations of 0.5°C are at the level of the global validation statistics (Reynolds et al. 2002; Roberts-Jones et al. 2012), which is very satisfactory for a regional product covering the coastal and shelf seas.
The annual signals in SST can be very large, with amplitudes up to 12°C in the Baltic Sea. To study interannual and decadal variability, it is therefore important to construct a monthly climatology and work on the anomalies with respect to the climatology. The spatial correlation of the interannual SST anomaly time series for central points in the North Sea, Danish straits, and central Baltic Sea showed that the North Sea interannual and decadal variability was separated from the Danish straits and the Baltic Sea region. Higher correlations were obtained between the southern part of the Baltic Sea and the Danish straits. The average SST curves also showed a similarity between the Baltic Sea and the Danish straits, as much larger monthly variability was identified in both regions compared to the North Sea. Despite these differences, the interannual and decadal variations and the linear trends of the SST were similar between the different subregions. The trends in the SST were highest during the spring and summer seasons and largest in the Baltic Sea.
It has been demonstrated that the SST CDR described in this paper meets the GCOS requirements for a satellite-based climate data record. The product has been developed within the EU 7th Framework Programme (FP7) project MyOcean and is included in the Copernicus Marine Environment Monitoring Services as part of the Ocean and Sea Ice Thematic Assembly Center. This implies that the full dataset is freely available for more in-depth analysis regarding the daily, annual, or decadal SST variability throughout the region. Regular updates to the CDR will be carried out, thus making it ideal for continued and sustained monitoring of the surface temperatures in the North Sea and Baltic Sea region.
Acknowledgments
Financial support was received from the MyOcean project, which has been funded by the European Commission under the FP7 space program and from the ESA STSE SSTDV:R.EX-IM.A.M. project. Digitized ice chart data were provided by Lars Axell from the Swedish Meteorological and Hydrological Institute (SMHI).
REFERENCES
Andersen, S., R. Tonboe, S. Kern, and H. Schyberg, 2006: Improved retrieval of sea ice total concentration from spaceborne passive microwave observations using numerical weather prediction model fields: An intercomparison of nine algorithms. Remote Sens. Environ., 104, 374–392, doi:10.1016/j.rse.2006.05.013.
BACC Author Team, 2008: Assessment of Climate Change for the Baltic Sea Basin. Springer Science & Business Media, 474 pp., doi:10.1007/978-3-540-72786-6.
BACC II Author Team, 2015: Second Assessment of Climate Change for the Baltic Sea Basin. Springer Science & Business Media, 501 pp., doi:10.1007/978-3-319-16006-1.
Casey, K. S., T. B. Brandon, P. Cornillon, and R. Evans, 2010: The past, present, and future of the AVHRR Pathfinder SST program. Oceanography from Space, V. Barale, J. F. R. Gower, and L. Alberotanza, Eds., Springer, 273–287, doi:10.1007/978-90-481-8681-5_16.
Comiso, J. C., D. J. Cavalieri, and T. Markus, 2003: Sea ice concentration, ice temperature, and snow depth using AMSR-E data. IEEE Trans. Geosci. Remote Sens., 41, 243–252, doi:10.1109/TGRS.2002.808317.
Donlon, C., and Coauthors, 2010: Successes and challenges for the modern sea surface temperature observing system. Proc. OceanObs09: Sustained Ocean Observations and Information for Society, Vol. 2, ESA Publication WPP-306, Venice, Italy, Intergovernmental Oceanographic Commission, doi:10.5270/OceanObs09.cwp.24.
Embury, O., and C. J. Merchant, 2012: A reprocessing for climate of sea surface temperature from the along-track scanning radiometers: A new retrieval scheme. Remote Sens. Environ., 116, 47–61, doi:10.1016/j.rse.2010.11.020.
Embury, O., C. J. Merchant, and G. K. Corlett, 2012: A reprocessing for climate of sea surface temperature from the along-track scanning radiometers: Initial validation, accounting for skin and diurnal variability effects. Remote Sens. Environ., 116, 62–78, doi:10.1016/j.rse.2011.02.028.
GCOS, 2006: Systematic observation requirements for satellite-based products for climate: Supplemental details to the satellite-based component of the implementation plan for the global observing system for climate in support of the UNFCCC. Global Observing System for Climate Tech. Doc. 107, WMO/TD-1338, 103 pp. [Available online at https://www.wmo.int/pages/prog/gcos/Publications/gcos-107.pdf.]
GCOS, 2010: Implementation plan for the Global Observing System for Climate in support of the UNFCCC (2010 update). Global Observing System for Climate Tech. Doc. 138, WMO/TD-1523, 186 pp. [Available online at http://www.wmo.int/pages/prog/gcos/Publications/gcos-138.pdf.]
Haapala, J. J., I. Ronkainen, N. Schmelzer, and M. Sztobryn, 2015: Recent change—Sea ice. Second Assessment of Climate Change for the Baltic Sea Basin, Springer, 145–153, doi:10.1007/978-3-319-16006-1.
Hollmann, R., and Coauthors, 2013: The ESA climate change initiative: Satellite data records for essential climate variables. Bull. Amer. Meteor. Soc., 94, 1541–1552, doi:10.1175/BAMS-D-11-00254.1.
Høyer, J. L., 2015: Product user manual for SST_BAL_SST_L4_REP_OBSERVATIONS_010_016 level 4 SST reprocessed products for the Baltic Sea. EU Copernicus Marine Service, version 1.1, 21 pp. [Available online at http://marine.copernicus.eu/documents/PUM/CMEMS-OSI-PUM-010-016.pdf.]
Høyer, J. L., and J. She, 2004: Validation of satellite SST products for the North Sea–Baltic Sea region. Danish Meteorological Institute Tech. Rep. 04-11, 31 pp. [Available online at http://www.dmi.dk/fileadmin/user_upload/Rapporter/TR/2004/tr04-11.pdf.]
Høyer, J. L., and J. She, 2007: Optimal interpolation of sea surface temperature for the North Sea and Baltic Sea. J. Mar. Syst., 65, 176–189, doi:10.1016/j.jmarsys.2005.03.008.
Høyer, J. L., I. Karagali, G. Dybkjær, and R. Tonboe, 2012: Multi sensor validation and error characteristics of Arctic satellite sea surface temperature observations. Remote Sens. Environ., 121, 335–346, doi:10.1016/j.rse.2012.01.013.
Høyer, J. L., P. Le Borgne, and S. Eastwood, 2014: A bias correction method for Arctic satellite sea surface temperature observations. Remote Sens. Environ., 146, 201–213, doi:10.1016/j.rse.2013.04.020.
Hurrell, J. W., 1995: Decadal trends in the North Atlantic Oscillation: Regional temperatures and precipitation. Science, 269, 676–679, doi:10.1126/science.269.5224.676.
IPCC, 2013: Climate Change 2013: The Physical Science Basis. Cambridge University Press, 1535 pp., doi:10.1017/CBO9781107415324.
Janssen, F., C. Schrum, and J. O. Backhaus, 1999: A climatological data set of temperature and salinity for the Baltic Sea and the North Sea. Dtsch. Hydrogr. Z., 51 (Suppl. 9), 5–245, doi:10.1007/BF02933676.
Karagali, I., and J. L. Høyer, 2014: Characterisation and quantification of regional diurnal SST cycles from SEVIRI. Ocean Sci., 11, 1093–1128, doi:10.5194/os-10-745-2014.
Karagali, I., J. L. Høyer, and C. Hasager, 2012: SST diurnal variability in the North Sea and the Baltic Sea. Remote Sens. Environ., 121, 159–170, doi:10.1016/j.rse.2012.01.016.
Kearns, E. J., J. A. Hanafin, R. H. Evans, P. J. Minnett, and O. B. Brown, 2000: An independent assessment of Pathfinder AVHRR sea surface temperature accuracy using the Marine Atmosphere Emitted Radiance Interferometer (MAERI). Bull. Amer. Meteor. Soc., 81, 1525–1536, doi:10.1175/1520-0477(2000)081<1525:AIAOPA>2.3.CO;2.
Kennedy, J., N. Rayner, R. Smith, D. Parker, and M. Saunby, 2011: Reassessing biases and other uncertainties in sea surface temperature observations measured in situ since 1850: 1. Measurement and sampling uncertainties. J. Geophys. Res., 116, D14103, doi:10.1029/2010JD015218.
Kilpatrick, K., G. Podesta, and R. Evans, 2001: Overview of the NOAA/NASA advanced very high resolution radiometer Pathfinder algorithm for sea surface temperature and associated matchup database. J. Geophys. Res., 106, 9179–9197, doi:10.1029/1999JC000065.
Kumar, A., P. J. Minnett, G. Podestá, and R. H. Evans, 2003: Error characteristics of the atmospheric correction algorithms used in retrieval of sea surface temperatures from infrared satellite measurements: Global and regional aspects. J. Atmos. Sci., 60, 575–585, doi:10.1175/1520-0469(2003)060<0575:ECOTAC>2.0.CO;2.
Lean, K., and R. W. Saunders, 2013: Validation of the ATSR Reprocessing for Climate (ARC) dataset using data from drifting buoys and a three-way error analysis. J. Climate, 26, 4758–4772, doi:10.1175/JCLI-D-12-00206.1.
Lehmann, A., K. Getzlaff, and J. Harlaß, 2011: Detailed assessment of climate variability of the Baltic Sea area for the period 1958–2009. Climate Res., 46, 185–196, doi:10.3354/cr00876.
Lehmann, A., K. Myrberg, and K. Höflich, 2012: A statistical approach to coastal upwelling in the Baltic Sea based on the analysis of satellite data for 1990–2009. Oceanologia, 54, 369–393, doi:10.5697/oc.54-3.369.
Marsouin, A., P. Le Borgne, G. Legendre, S. Péré, and H. Roquet, 2015: Six years of OSI-SAF METOP-A AVHRR sea surface temperature. Remote Sens. Environ., 159, 288–306, doi:10.1016/j.rse.2014.12.018.
Marullo, S., B. B. Nardelli, M. Guarracino, and R. Santoleri, 2007: Observing the Mediterranean Sea from space: 21 years of Pathfinder-AVHRR sea surface temperatures (1985 to 2005): Re-analysis and validation. Ocean Sci., 3, 299–310, doi:10.5194/os-3-299-2007.
Merchant, C. J., O. Embury, P. Le Borgne, and B. Bellec, 2006: Saharan dust in nighttime thermal imagery: Detection and reduction of related biases in retrieved sea surface temperature. Remote Sens. Environ., 104, 15–30, doi:10.1016/j.rse.2006.03.007.
Merchant, C. J., A. R. Harris, H. Roquet, and P. Le Borgne, 2009: Retrieval characteristics of non-linear sea surface temperature from the Advanced Very High Resolution Radiometer. Geophys. Res. Lett., 36, L17604, doi:10.1029/2009GL039843.
Merchant, C. J., and Coauthors, 2012: A 20 year independent record of sea surface temperature for climate from Along-Track Scanning Radiometers. J. Geophys. Res. Ocean, 117, C12013, doi:10.1029/2012JC008400.
Moberg, A., and P. D. Jones, 2005: Trends in indices for extremes in daily temperature and precipitation in central and western Europe, 1901–99. Int. J. Climatol., 25, 1149–1172, doi:10.1002/joc.1163.
Pichel, W., E. Maturi, P. Clemente-Colón, and J. Sapper, 2001: Deriving the operational nonlinear multichannel sea surface temperature algorithm coefficients for NOAA-15 AVHRR/3. Int. J. Remote Sens., 22, 699–704, doi:10.1080/01431160010013793.
Rayner, N., P. Brohan, D. Parker, C. Folland, J. Kennedy, M. Vanicek, T. Ansell, and S. Tett, 2006: Improved analyses of changes and uncertainties in sea surface temperature measured in situ since the mid-nineteenth century: The HadSST2 dataset. J. Climate, 19, 446–469, doi:10.1175/JCLI3637.1.
Reynolds, R. W., N. A. Rayner, T. M. Smith, D. C. Stokes, and W. Wang, 2002: An improved in situ and satellite SST analysis for climate. J. Climate, 15, 1609–1625, doi:10.1175/1520-0442(2002)015<1609:AIISAS>2.0.CO;2.
Reynolds, R. W., T. M. Smith, C. Liu, D. B. Chelton, K. S. Casey, and M. G. Schlax, 2007: Daily high-resolution-blended analyses for sea surface temperature. J. Climate, 20, 5473–5496, doi:10.1175/2007JCLI1824.1.
Roberts-Jones, J., E. K. Fiedler, and M. J. Martin, 2012: Daily, global, high-resolution SST and sea ice reanalysis for 1985–2007 using the OSTIA system. J. Climate, 25, 6215–6232, doi:10.1175/JCLI-D-11-00648.1.
Schlesinger, M. E., and N. Ramankutty, 1994: An oscillation in the global climate system of period 65–70 years. Nature, 367, 723–726, doi:10.1038/367723a0.
Siegel, H., M. Gerth, and G. Tschersich, 2008: Satellite-derived sea surface temperature for the period 1990–2005. State and Evolution of the Baltic Sea, 1952–2005: A Detailed 50-Year Survey of Meteorology and Climate, Physics, Chemistry, Biology, and Marine Environment, R. Feistel, G. Nausch, and N. Wasmund, Eds., John Wiley & Sons, 241–264, doi:10.1002/9780470283134.ch9.
Smith, D., C. Mutlow, J. Delderfield, B. Watkins, and G. Mason, 2012: ATSR infrared radiometric calibration and in-orbit performance. Remote Sens. Environ., 116, 4–16, doi:10.1016/j.rse.2011.01.027.
Walton, C., W. Pichel, J. Sapper, and D. May, 1998: The development and operational application of nonlinear algorithms for the measurement of sea surface temperatures with the NOAA polar-orbiting environmental satellites. J. Geophys. Res. Oceans, 103, 27 999–28 012, doi:10.1029/98JC02370.
Woodruff, S. D., R. J. Slutz, R. L. Jenne, and P. M. Steurer, 1987: A comprehensive ocean–atmosphere data set. Bull. Amer. Meteor. Soc., 68, 1239–1250, doi:10.1175/1520-0477(1987)068<1239:ACOADS>2.0.CO;2.
Woodruff, S. D., and Coauthors, 2011: ICOADS release 2.5: Extensions and enhancements to the surface marine meteorological archive. Int. J. Climatol., 31, 951–967, doi:10.1002/joc.2103.
Zavody, A., M. Gorman, D. Lee, D. Eccles, C. Mutlow, and D. Llewellyn-Jonesh, 1994: The ATSR data processing scheme developed for the EODC. Int. J. Remote Sens., 15, 827–843, doi:10.1080/01431169408954119.
