Investigating sea surface temperature diurnal

Remote Sensing of Environment 183 (2016) 1–12

Contents lists available at ScienceDirect

Remote Sensing of Environment journal homepage: www.elsevier.com/locate/rse

Investigating sea surface temperature diurnal variation over the Tropical Warm Pool using MTSAT-1R data Haifeng Zhang a,b,d,⁎, Helen Beggs c, Leon Majewski c, Xiao Hua Wang a,b, Andrew Kiss a,b,d a

The Sino-Australian Research Centre for Coastal Management, The University of New South Wales, Canberra, Australia School of Physical, Environmental and Mathematical Sciences, The University of New South Wales, Canberra, Australia Bureau of Meteorology, Docklands, Melbourne, Australia d ARC Centre of Excellence for Climate System Science, Australia b c

a r t i c l e

i n f o

Article history: Received 16 September 2015 Received in revised form 14 April 2016 Accepted 14 May 2016 Available online xxxx Keywords: SST Tropical Warm Pool MTSAT-1R Validation Diurnal variation

a b s t r a c t Diurnal variation (DV) of sea surface temperature (SST) plays an important role in air–sea interaction. We have validated four months (January to April 2010) of the version 3 Australian Bureau of Meteorology reprocessed Multifunction Transport SATellite-1R (v3 MTSAT-1R) SST data over the Tropical Warm Pool (TWP) region (90°E to 170°E, 25°S to 15°N) against both drifting buoy and Advanced Along-Track Scanning Radiometer (AATSR) SST data. Validation against collocated point measurements from drifting buoys, under conditions where the surface ocean is well-mixed, shows that overall the v3 MTSAT-1R SSTs perform well with an average bias of 0.00 °C and a 0.73 °C standard deviation (STD). The average daytime and night-time mean bias is −0.06 °C and 0.08 °C, respectively. For all hours of the diurnal cycle, the mean biases are within ±0.25 °C, indicating the consistency between day and night v3 MTSAT-1R data. However, on average, the v3 MTSAT-1R SSTs are overestimated at cold SSTs and underestimated at warm SSTs. Similar results are obtained from validation against the AATSR satellite SSTs but with smaller STD (0.48 °C) and smaller average daytime and night-time mean biases (−0.04 °C and 0.06 °C, respectively). These results indicate that the v3 MTSAT-1R data set is suitable for SST DV investigations and validation of DV models. Using the validated v3 MSTAT-1R data, together with surface wind speed and solar shortwave insolation (SSI) outputs from the Australian Community Climate and Earth-System Simulator – Regional (ACCESS-R) numerical prediction model, we investigate SST DV events over the TWP region. Good correlation is found between DV events and low wind and high SSI conditions. The dominant role of wind speed in SST DV events over the SSI is also revealed. © 2016 Elsevier Inc. All rights reserved.

1. Introduction Sea surface temperature (SST) is recognized as one of the most important variables in climate and weather studies. It plays a key role in constraining the exchange of moisture and heat in air–sea interaction and is able to exert influences on both short and long term climate dynamics. SSTs are also controlled by atmospheric conditions. For example, the sea surface wind causes turbulent mixing in the upper ocean, thus changing the ocean's near surface temperature profile. Warm SST also increases near-surface air temperature and reduces the stability of the atmospheric boundary layer. The role of SST is of particular significance in the Tropical Warm Pool (TWP) region, defined as the western equatorial Pacific Ocean and eastern Indian Ocean, which exhibits some of the highest average SSTs over a large expanse of the earth's surface (Fig. 1). Previous investigations have shown that the ⁎ Corresponding author at: The Sino-Australian Research Centre for Coastal Management, The University of New South Wales, Canberra, Australia. E-mail address: [email protected] (H. Zhang).

http://dx.doi.org/10.1016/j.rse.2016.05.002 0034-4257/© 2016 Elsevier Inc. All rights reserved.

seasonal, inter-annual and long-term natural climate fluctuations in the equatorial and mid-latitude regions are particularly sensitive to the SST distribution in the warm tropical oceans (e.g. An, Kim, Im, Kim & Park, 2012; Park, Yeh & Kug, 2012; Vecchi, Clement & Soden, 2008). Diurnal variation (DV) of SST, or diurnal warming, normally refers to the daily temperature fluctuation in the upper few meters of the ocean. The thermal stratification of the ocean surface layer is the result of the interaction of several factors, including the heat exchange between the atmosphere and the ocean, turbulent mixing, and the absorption of solar insolation. Under most conditions, there is net loss of heat from the ocean with the skin, or interfacial, layer being cooler than the water below (Fairall et al., 1996). However, during the day, especially under calm winds and clear skies, a warm layer can develop in the uppermost 5–10 m of the ocean (Gentemann, Minnett, Le Borgne & Merchant, 2008). Diurnal warm layers have typical temperature differences relative to the body of water below on the order of 0.5–3 ° C, but in some cases can reach values up to 5–8 °C (e.g. Gentemann et al., 2008; Karagali, Høyer & Hasager, 2012; Merchant et al., 2008). Previous studies have shown that a better understanding of DV events

2

H. Zhang et al. / Remote Sensing of Environment 183 (2016) 1–12

Fig. 1. Annual average SST in 2010, obtained from Global Australian Multi-Sensor SST Analysis data (Zhong & Beggs, 2008). The box shows the TWP study domain.

is essential to better represent the air–sea interaction in weather and climate models (e.g. Clayson & Bogdanoff, 2013; Fairall et al., 1996). Ignorance of the SST DV could lead to errors in the surface flux estimates. For instance, Clayson and Bogdanoff (2013) have found that significant portions of the tropical oceans experience total heat flux differences as high as 10 Wm− 2 on a yearly average when the SST DV effects are taken into consideration. SST DV events have been studied for decades. Before the 1980s, insitu SST measurements served as the primary data source (e.g. Halpern & Reed, 1976; Stommel, Saunders, Simmons & Cooper, 1969). Since remotely sensed data from Advanced Very High Resolution Radiometer (AVHRR) sensors on-board the National Oceanic and Atmospheric Administration (NOAA) series satellites became widely available in the 1980s, satellite data soon greatly facilitated the research in terms of temporal and spatial data coverage (e.g. Böhm, Marullo & Santoleri, 1991; Hawkins, Clancy & Price, 1993; Stuart-Menteth, Robinson & Challenor, 2003). Remotely sensed observations of SST may be obtained from infrared or microwave sensors on-board either geostationary or low earth orbit satellites. In order to take advantage of the high quality in-situ data, most studies have been conducted on regional scales, such as over the Mediterranean Sea (Böhm et al., 1991; Castro, Wick & Buck, 2014), the Mutsu Bay (Kawai, Otsuka & Kawamura, 2006), the North and Baltic Seas (Karagali & Høyer, 2013; Karagali et al., 2012), the western Pacific Ocean (Kawai & Kawamura, 2005), and even the Arctic Ocean (Eastwood, Le Borgne, Péré & Poulter, 2011). Global studies are fewer mostly due to the requirement of continuous high quality data coverage both spatially and temporally (e.g. Clayson & Bogdanoff, 2013; Stuart-Menteth et al., 2003). The TWP region is an ideal area for DV studies due to the frequent occurrence of large DV events associated with the region's relatively calm winds, strong insolation, and sensitivity to changes in air–sea fluxes. Webster, Clayson, and Curry (1996) addressed the relationship between clouds, radiation, and the diurnal cycle of SST in the TWP region. They concluded that the amplitude of the diurnal cycle is largest for the greatest insolation and lowest wind speed, and that the influence of surface wind speed on the diurnal cycle amplitude of SST is nonlinear. In Soloviev and Lukas (1997), large diurnal warming events were observed from very high resolution measurements of near-surface thermohaline and turbulence structures made using bow-mounted probes and a free-rising profiler. In Kawamura, Qin, and Ando (2008), the amplitude and temporal characteristics of SST DV were examined as well using a combination of in-situ and AVHRR data. Tanahashi, Kawamura, Takahashi, and Yusa (2003) also studied the SST DV events using the Geostationary Meteorological Satellite (GMS) data over the open ocean. In 2005, the Multi-functional Transport SATellite-1R (MTSAT-1R) was launched by the Japan Aerospace Exploration Agency (JAXA) on behalf of the Japan Civil Aviation Bureau and the Japan Meteorological Agency as a successor to the GMS series satellites. MTSAT-1R was in a geostationary orbit above 140°E and carried the Japanese Advanced Meteorological Imager (JAMI) on board that captured full-disc imagery on an hourly basis during the period 2005–2010 using five spectral

channels (wavelengths of 0.6–12.0 μm). MTSAT-1R provided a further step forward in observing technology for SST DV. So far, published works focus more on SST retrieval algorithms and algorithm validation than on the application in analysing the SST DV events (e.g. Kawamura, Qin, Sakaida & Qiu, 2010a; Kawamura, Qin, Sakaida & Setiawan, 2010b). With the version 3 (v3) MTSAT-1R data specially processed for the Tropical Warm Pool Diurnal Variability project (TWP + project, see details in Section 2) by the Australian Bureau of Meteorology (Bureau), we are offered a valuable opportunity to investigate SST DV events in detail within the TWP region. The focus of this paper is to validate the v3 MTSAT-1R data set and to characterise the derived SST DV events. Section 2 introduces the data sets and the methods used in this work, followed by Section 3 which illustrates the validation results. The SST DV events are investigated in Section 4. Discussion and conclusions regarding the results are presented in Section 5. 2. Data and methods 2.1. Data 2.1.1. MTSAT-1R data The v3 MTSAT-1R data set provides SST at approximately 10 μm depth (SSTskin), calculated from radiance observations of spectral channels centred at 3.7, 10.8 and 12.0 μm. These full-disc radiances measured at approximately 10 μm depth were regressed against drifting buoy SST observations at 20 to 30 cm depth during the period June 2006 to June 2010 to produce a sub-skin SST (for detailed SST definition please refer to Donlon et al., 2007). The sub-skin is converted to a skin SST by subtracting a constant 0.17 °C (Beggs et al., 2013) to account for the average cool-skin effect, following the method of Beggs, Kippo, and Underwood (2012b). In order to reduce temporal and spatial biases relative to the version 2 (v2) MTSAT-1R SSTs which were originally reprocessed by the Bureau for the Integrated Marine Observing System (IMOS) (Beggs et al., 2012a), the v3 processing system applied correction factors to account for a number of geometric and temporal properties, including pixel and line position, observation hour, solar declination and earth-sun distance (Beggs et al., 2013; Majewski, Griffin & Beggs, 2013). The v3 MTSAT-1R SST data used in this study were produced to fulfil the requirements of the TWP+ project and are available on request ([email protected]). The TWP+ project is conducted as a collaboration between the Bureau, Group for High Resolution SST (GHRSST), IMOS, Météo-France, University of Edinburgh (UoE) and REMote Sensing Systems (REMSS) for the study of DV over the TWP region (https://www.ghrsst.org/ ghrsst/tags-and-wgs/dv-wg/twp/). The TWP+ dataset is a comprehensive data collection compiled by assembling different data sources (insitu data, satellite data, and models) that provide a variety of parameters, including SST, wind speed, and solar insolation, over the TWP study region (90°E–170°E, 25°S–15°N) and time period (1 January to 30 April 2010). The v3 MTSAT-1R data, produced in December 2012, are the key component of the TWP + data set for validation of DV


models. This data set has been processed to minimize day-night and regional temperature biases so that it is suitable for the comprehensive study of the frequency, amplitude, and spatial extent of the SST DV events over the TWP region and its associated forcing factors (Beggs et al., 2013). The spatial resolution of the v3 MTSAT-1R data is 0.05° × 0.05°. Each file contains 24 h of hourly SSTskin data along with other useful ancillary information for each pixel. Ancillary information used in this study include the Sensor Specific Error Statistics (SSES) of bias with respect to drifting buoy SSTs (https://www.ghrsst.org/ ghrsst/tags-and-wgs/stval-wg/sses-common-principles/), proximity to cloud flag (“proximity confidence”), land mask, and distance of that pixel to land. It is worth noting that the Bureau has four years (June 2006 to June 2010) of the earlier v2 MTSAT-1R data (Beggs et al., 2012a; available on request). As this v2 is updated to v3 (currently ongoing), it could improve the temporal coverage for future DV studies.

2.1.2. In-situ data Drifting buoy data within the same period, also obtained from the TWP + data set, are used to validate the v3 MTSAT-1R data over the TWP + domain. Typically, the depth at which the buoys measure SST is approximately 20–30 cm. There are also other in-situ SST measurements in the TWP + data set such as those from moored buoys and ships. However, because there are much fewer of these measurements and the depths at which they measure SST are not relatively constant, ranging from 1 m to 10 m or even deeper, they are not adopted in this work in order to keep the comparison consistent.

3

2.2. Method 2.2.1. Quality control A series of relatively strict Quality Control (QC) steps are performed on the MTSAT-1R data prior to being used in the study. First, data b15 km offshore are rejected in order to filter out possible land contamination. Second, data of the highest Quality Level (QL) of 5 are selected. A quality level (or “proximity_confidence”) flag of 5 means that the SST observation is at least 5 pixels (~ 20 km at nadir) from an identified cloud. Spatial and temporal filters are also applied. Spatially, the value is retained only when the four adjacent neighbours (north, south, west, and east directions) are all of QL 5. Temporally, the value is rejected also if the two data values of the prior and post hour are not of QL 5. The last QC step is to apply the SSES bias (SSES_Bias) correction. This bias is subtracted from the SST at each pixel. QC is also carried out on the AATSR data. In each of the files, there is an uncertainty variable that estimates the quality of the values. Each file contains an uncertainty variable that is calculated by the combination of three components: 1) radiometric noise, which reflects the sensitivity of the retrieval coefficients to instrument noise; 2) retrieval error, which is effectively the fitting error in the linear regression to create the retrieval coefficients; and 3) sampling error, which represents the fact that not every pixel is observed in the grid cell and is a weighted standard deviation of the SST observations in the grid cell. Embury (2011, personal communication) suggested that we discard ARC v1.1 SST values with uncertainty over 0.3 °C in order to exclude SST observations along cloud edges. It was also mentioned that how well the uncertainty estimate works was still being investigated. However, empirical evidence indicates that a threshold of 0.3 °C is quite strict.

2.1.3. AATSR data Skin SSTs derived from the Advanced Along-Track Scanning Radiometer (AATSR) are also utilised to cross-validate the MTSAT-1R data. The AATSR instrument operated from 2002 to 2012 on the European Space Agency (ESA) polar-orbiting Environmental Satellite (EnviSat) with local equator crossing times of 10 am and 10 pm daily with a deviation of up to ± 5 min (https://earth.esa.int/web/guest/missions/esaoperational-eo-missions/envisat/operations). Taking advantage of the dual-view scanning geometry as well as the high quality instrumental calibration, the AATSR SSTs are retrieved through radiative transfer simulations, meaning that they are not tuned to in-situ measurements like those of MTSAT-1R and AVHRR (Merchant & Le Borgne, 2004). The exact version used in this study is the ARC (ATSR Reprocessing for Climate) v1.1 (Embury & Merchant, 2012; Embury, Merchant & Corlett, 2012a; Embury, Merchant & Filipiak, 2012b). In Embury et al. (2012a), the validation of ARC v1.1 against tropical moored buoys, which are considered to be more accurate than drifting buoys, showed b0.03 °C bias and b 0.18 °C robust standard deviation. According to the results of comparison between several data sets in Merchant et al. (2014, their Table 2), this data set is the “most accurate and stable SST product available” due to its combined features, including independence from in-situ data, and homogeneity and stability throughout the time-series. The spatial resolution of the ARC v1.1 gridded data set is 0.1° × 0.1° and the temporal resolution is twice daily, compared to 0.05° × 0.05° and hourly resolution for the MTSAT-1R data set.

2.2.2. Collocation steps After the QC steps, the collocations between MTSAT-1R and drifting buoy data are then constructed following a set of criteria. Spatially, the positions of the in-situ value and MTSAT-1R pixel should be within 0.025° both longitudinally and latitudinally. Temporally, the difference between the two values should not exceed half an hour. Next, for the consideration of sufficient turbulent mixing for satellite SST validation, only conditions with wind speed N6 ms− 1 during the day and N2 ms−1 during the night are selected (based on empirical results reported in Donlon et al., 2002). Before a pair of collocations are finalised, from each in-situ measurement, 0.17 °C is subtracted to produce an effective cool skin equivalent to the in-situ measurement which is taken at 20–30 cm below the cool skin (Beggs et al., 2012b). The 0.17 °C cool-skin constant correction is consistent with the correction applied to the MTSAT-1R SSTs to convert from drifting buoy to skin depths (Section 2.1.1). For collocation between MTSAT-1R and AATSR data, the temporal window is also set to be within half an hour. As both instruments measure skin SST, no wind speed filtering is required in this case. Spatially, for each AATSR value, there are four adjacent MTSAT-1R values lying in the four directions (north, south, west, and east) due to the grid resolution difference. If at least three of the four adjacent MTSAT-1R values are valid, their mean value then forms a collocated pair with the centred AATSR value.

2.1.4. Meteorological data The wind speed at 10 m height above sea level and solar shortwave insolation (SSI) data used in the investigation of DV events are from the Bureau's hourly, 0.375° resolution, Australian Community Climate and Earth System Simulator-Regional (ACCESS–R) 24-hour forecasts. This regional numerical weather prediction model was jointly developed by the Bureau and the Commonwealth Scientific and Industrial Research Organisation (CSIRO) and is able to predict a variety of surface parameters such as shortwave and longwave flux, friction velocity, sensible and latent heat flux, and wind stress (Puri et al., 2013).

2.2.3. Diurnal variation calculation The daily SST variation (dSST) at a given time within a day is calculated as the difference between that SST value and the foundation SST (SSTfnd). The maximum dSST at one location within one day is referred as dSSTmax. SSTfnd is a concept defined by GHRSST which refers to the temperature free of diurnal temperature variability or, more specifically, the temperature at the first time of the day when the heat gain from the solar radiation absorption exceeds the heat loss at the sea surface (https://www.ghrsst.org/science-and-applications/sst-definitions/; Donlon et al., 2007). Without an accurate DV model, it is difficult to obtain a reliable value of SSTfnd from satellite observations. Foundation

4


SSTs are therefore normally estimated from the pre-dawn SSTs or, to increase the data amount, by the average values of the night-time SSTs. The start (e.g., 22:00 or 24:00 local solar time (LST)) and end (e.g., 4:00, 5:00, or 6:00 LST) times of the night-time window need to be determined. Several published papers have systematically compared the differences between all these night-time windows, and whether to adopt only the same day or a few more days before and after has been also discussed (e.g. Karagali & Høyer, 2014; Karagali et al., 2012). Karagali and Høyer (2014) concluded that DV estimates obtained from these different SSTfnd data are not significantly affected in terms of their amount, or seasonal and spatial distribution. Therefore, in this work, the night-time window for SSTfnd is from 0:30 to 5:30 LST of the same day.

defined by

SI ¼

rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi h i ffi b 2 N a a − SSTb −SST 1 ∑ SST − SST i¼1 i i N SSTb

ð1Þ

where N is the number of collocations and the overbar denotes an average over the N collocations; SSTa represents the MTSAT-1R data and SSTb is either the in-situ buoy data or the AATSR data, all in °C. MAD is robust to outliers and gives a suggestive confidence estimation. To obtain a MAD, we first calculate the median as the estimate of the mean. Then the differences from this median are calculated. Finally, we take the median of the absolute values of those differences as a MAD. 3.1. In-situ validation

3. Validation of MTSAT-1R data Before the new v3 MTSAT-1R can be used with confidence, validation is necessary. Mean bias, standard deviation (STD) of the bias, scatter index (SI), correlation coefficient (R), and the Median Absolute Difference (MAD) are the statistical parameters used in both validations. SI is

In-situ data used in this work are the drifting buoy data (Section 2.1.2) over the TWP+ domain. After all the necessary QC and collocation steps (Section 2.2), 2126 collocated pairs are yielded. The density plots of local daytime collocations, night-time collocations and all collocations with statistical results are shown in Fig. 2. To highlight the biases, the y

Fig. 2. In-situ validation results: (a) daytime (7:00–19:00 LST) collocations density plot within each 0.2 °C × 0.2 °C bin. The y axis is the MTSAT-1R data minus in-situ data and the x axis is the in-situ data; (b) same as (a) but for night-time (19:00–7:00 LST) collocations; (c) same as (a) but for all collocations; (d) collocation numbers, bias and STDs (bars) over the in-situ SST range, in 1 °C bins; (e) same as (d) but over each local hour. No black dot at 17–18 LST in (e) means there is no collocation, while no error-bar means there is only one collocation. (f) quality-controlled MTSAT-1R data amount distribution over each local hour.


axis is changed to the difference between MTSAT-1R and in-situ data. For the daytime (7:00–19:00 LST) validation (Fig. 2a), a small negative bias (− 0.06 °C) is observed. The STD of the individual point differences is 0.71 °C. In night-time validation, the bias is a positive 0.08 °C and the STD is slightly higher (0.74 °C). The standard error in the mean is STD/ √N = 0.02 °C in both cases, so these biases are statistically significant, with a confidence exceeding 95%. The MAD values for the daytime and night-time collocations are the same, both being 0.41 °C (Table 1). These results indicate that the algorithm to minimize the day-night bias in the reprocessing of the v3 MTSAT-1R data has been effective. For all collocations combined (Fig. 2c), the 0.00 °C bias means that overall the collocated values are very close. The overall STD is 0.73 °C. These values are very similar to the results in Majewski et al. (2013). Their results show that over the entire MTSAT-1R footprint and from June 2006 to June 2010, the bias between MTSAT-1R SSTs and in-situ data is b 0.1 °C with a standard deviation of ~0.7 °C. To explore MTSAT-1R's performance over each in-situ SST 1 °C range, the number of collocations, average bias, and STD value are calculated as displayed in Fig. 2d. Based on this trend line, clearly an overestimation with amplitude of 0.3 °C–1.2 °C is observed in the MTSAT-1R data over the relatively cold SST situations (in-situ SSTs b 27 °C). The 27–28 °C range, which has not only the most collocations but also the smallest bias and STD values, marks MTSAT-1R's best performance over all the ranges. For the warmer SST situations (28–30 °C), the biases and STDs fluctuate but are on the order of ±0.2 °C. For the limited number of in-situ SSTs N31 °C, MTSAT-1R underestimates the SST by as much as ~ 3 °C. Over the whole range, the trend of overestimation at cold SSTs and underestimation at warm SSTs, including the bias amplitudes, is highly similar to the in-situ validation in Kim et al. (2011, their Fig. 6), although they used two-year (March 2006 to February 2008) MTSAT-1R data retrieved over the north-western Pacific region. Considering the very small number of collocations, further investigation was continued in the cross-validation study (Section 3.2). To investigate the performance of the MTSAT-1R data over different times of the day, the collocations within each local hour are analysed and shown in Fig. 2e. The number of collocations for local times differ remarkably from one another, with the largest number falling between 7:00–10:00 LST and the smallest between 5:00–6:00 LST (around local sunrise) and 17:00–19:00 LST (around local sunset). Further investigation shows that the numbers of collocations are basically determined by the numbers of MTSAT-1R measurements rather than the in-situ data. As shown in Fig. 2f, there exist two evident troughs at 5:00–6:00 LST and 17:00–19:00 LST for all four months of MTSAT-1R data. Notwithstanding the lack of collocations around sunrise and sunset, Fig. 2e indicates that the biases between MTSAT-1R and in-situ SSTs are relatively constant over the 24 h period, with mean deviations no more than ± 0.25 °C. This is also in accordance with Beggs et al. (2013) and Majewski et al. (2013) which found that the hour-to-hour differences of the v3 MTSAT-1R and drifting buoy SSTs are b0.1 °C with the exception of day-night transitions (b0.2 °C). One likely reason why MTSAT-1R has far fewer quality controlled data with more questionable quality at sunrise and sunset periods is due to the cloud screening method (Merchant, Harris, Maturi & Maccallum, 2005). The cloud mask depends on the MTSAT-1R visible

5

and infrared channels during the daytime and on the 3.7 μm and 11 μm infrared channels at night. Dusk and dawn are times when the cloud clearing process cannot use the visible channels because the daylight is weak, yet the 3.7 μm channel is still contaminated by the visible component. Therefore, during these periods, there is greater uncertainty in the cloud detection. Most SST observations are flagged as QL b 5, therefore not included in this analysis. The local solar time of dusk and dawn varies by latitude and time of the year so we see a reduction either side of local dusk and dawn. 3.2. Cross validation For cross validation of MTSAT-1R, the ARC v1.1 AATSR skin SST data set (Section 2.1.3) is selected to serve as reference satellite data due to its relatively small uncertainties (Embury et al., 2012a; Merchant et al., 2014). The temporal resolution of this data set is twice daily at ~ 10:00 LST and ~ 22:00 LST. Therefore the validation process is also divided into daytime validation and night-time validation. The results are displayed in Fig. 3. As expected, two much larger collocation sets are obtained in comparison with in-situ validation. Shown in Fig. 3a and b are the density plots of the day and night collocations together with their statistics. To highlight the biases, the y axis is again changed to the difference between MTSAT-1R and AATSR data. The average day and night mean biases are −0.04 °C and 0.06 °C, respectively, slightly better than the − 0.06 °C and 0.08 °C biases from in-situ validation. Both day and night STDs are 0.48 °C, which are much smaller than their in-situ validation counterparts (0.71 °C and 0.74 °C), indicating a better consistency between the two satellites. This should partly be attributed to both providing skin SST measurements and both sampling over broad areas (~ 5 km2 for MTSAT-1R and ~10 km2 for AATSR) rather than point measurements. Both SI values decrease from 0.026 in in-situ validation to 0.017. The R values also increase slightly. Combining all the collocations together, the bias reduces to nearly zero but other validation parameters change very slightly (Table 1). The number of collocations, average bias, and STDs over each 1 °C SST range (based on AATSR data) for day and night validation are shown in Fig. 3c and d, respectively. For the daytime validation, the largest number of collocations falls within the 27 °C–28 °C range, while for the night-time validation they fall within the 28 °C–30 °C range. For most of the AATSR SSTs b 27 °C, in both plots, an overestimation can be found in MTSAT-1R data with most bias values above zero (0 °C–0.4 °C). While generally consistent performance is observed for the 27 °C–30 °C ranges, MTSAT-1R data tend to underestimate SSTs N31 °C with an amplitude of up to 2 °C. The overall characteristics are in accordance with that of the in-situ validation (Section 3.1) but with smaller bias values. Spatial distributions of bias values, STDs, and collocation numbers for each 2° × 2° bin are also investigated (Fig. 4). Over most of the study domain, the bias values are within −0.2 °C–0.2 °C and the STDs are within 0 °C–0.4 °C for both day and night validations. Large biases and STDs are mainly found in the eastern Indian Ocean. In Fig. 4e and f, it is shown that most of the collocations are clustered around the north-west coast of Australia, the west coast of the Philippines, and the west coast of Thailand. This distribution is basically determined by

Table 1 Statistics of both in-situ validation and cross-validation. Num represents the number of collocations, STD the standard deviation of the bias, SI the scatter index, R the correlation coefficient and MAD the Median Absolute Difference (see definitions in Section 3). In the in-situ validation, daytime is defined as from 7:00 to 19:00 local solar time (LST) and night-time from 19:00 to 7:00 LST, while in the cross-validation day/night times refer to 10:00/22:00 LST.

INSITU

AATSR

Day Night All Day Night All

Num

Bias (°C)

STD (°C)

SI

R

MAD (°C)

1138 988 2126 292,489 236,639 529,128

−0.06 0.08 0.00 −0.05 0.06 0.00

0.71 0.74 0.73 0.48 0.48 0.48

0.026 0.026 0.026 0.017 0.017 0.017

0.845 0.902 0.883 0.944 0.934 0.940

0.41 0.41 0.41 0.29 0.30 0.29

6


Fig. 3. Results of MTSAT-1R vs AATSR cross-validation: (a) daytime collocations density plot within each 0.2 °C × 0.2 °C bin. The y axis is the MTSAT-1R data minus AATSR data and the x axis is the AATSR data; (c) collocation numbers, bias and STDs (bars) over different SST 1 °C ranges (based on AATSR SSTs) of the daytime collocations; (b) and (d) are the same as (a) and (c) respectively but for the local night-time collocations. Note the local equator crossing times for AATSR on EnviSat were approximately 10:00 LST and 22:00 LST.

Fig. 4. The spatial distribution of (a) bias, (c) STD, (e) collocation count for daytime collocations of MTSAT-1R SSTs with AATSR SSTs in 2° × 2° bins; (b), (d), and (f) are the same as (a), (c), (e) respectively, but for night-time collocations.


7

4.1. Whole domain study

Fig. 5. Spatial distribution of the data density of quality controlled satellite measurements: (a) for AATSR data; (b) for MTSAT-1R data. If all data pass QC steps, the maximum count should be 120 (days) × 2 = 240 for AATSR, and 120 (days) × 24 (hours) = 2880 for MTSAT-1R.

the spatial distributions of the amount of quality controlled satellite measurements (Fig. 5). Since both satellites carry infrared sensors and cannot penetrate clouds to observe SST, this distribution also reflects the cloud pattern over this period. 4. Diurnal variation events The SST DV events over the study domain are investigated using the validated four months of MTSAT-1R data. This section consists of two subsections: Subsection 4.1 illustrates the spatial distribution characteristics and statistics of the DV events over the whole TWP + research domain, while Subsection 4.2 focuses on the relationship between SST DV, wind speed and SSI over a selected region off the north-western Australian coast where the densest data are found.

For each of the four months, monthly mean dSSTmax is calculated over each pixel and the spatial distribution is studied. The occurrence, i.e. days in a month, of the dSSTmax N 1 °C is also calculated. As shown in Fig. 6a, January has large blank areas without any DV events mostly due to the lack of available quality controlled measurements. Only for the regions to the east of Papua New Guinea and to the west of Australia, scattered DV events with amplitudes as large as 1 °C–1.5 °C are found. However, the occurrence of dSSTmax N 1 °C over those regions is rarely N2 days (Fig. 6b). The DV distributions of February (Fig. 6e) and April (Fig. 6m) are similar, with monthly mean dSSTmax being the highest along the coastal seas of New Guinea to the extended eastern regions (up to ~2 °C) and over the northern coast of Australia (~1 °C). Most of the DV events with dSSTmax N 1 °C occur b5 days in a month (Fig. 6f and n). In March, as seen in Fig. 6i, the large region lying to the west of the north-western coast of Australia experiences high amplitude DV events. In this area, dSSTmax N 1 °C is observed over at least 5 days (Fig. 6j). DV is also found along the north-eastern coast of Australia and along the east coast of Vietnam. It is worth noting that the DV distributions of the four months are to a large extent determined by the availability of data, as shown in Fig. 5b. For large regions of the western tropical Pacific and eastern Indian Ocean, generally referred to as the centre of the TWP region, the lack of data due to heavy clouds prevents further detailed studies. The monthly mean wind speed and SSI are also plotted to investigate their relationship with DV events. As expected, for all four months, the weakest winds are found over the equatorial oceans (10°S to 10°N across the whole domain). In January (Fig. 6c) and February (Fig. 6g), the lowest average winds, at about 3–4 ms− 1, are found mainly off the western coast of Indonesia and the eastern coast of Papua New Guinea. This distribution is in excellent agreement with that of the large DV events. In March, a significant low wind tongue is observed near the north-west coast of Australia, corresponding very well to the region with frequent, large amplitude DV events. SSI shows a seasonal pattern with the highest SSI found in the most southern region in January and in the most northern zone in April. In between, March has a transforming pattern with high SSI distributed over a large area of the northern domain and the north-west coast of Australia. The northwest coast of Australia is the same region where low wind speed and

Fig. 6. Spatial distribution of all four months in 2010: (a) monthly mean dSSTmax for January; (b) frequency (days in a month) of dSSTmax N 1 °C for January; (c) monthly mean daytime (7:00–19:00 LST) wind speed for January; (d) monthly mean daytime (7:00–19:00 LST) SSI for January. The second, third, and fourth rows are the same as the first but for February, March, and April, respectively.

8


frequent DV events occur. Over all four months, it is shown that large DV events are generally consistent with a combination of low wind and high SSI. The spatial distribution maps of the DV events, wind speed, and the SSI are all very similar with other studies (e.g. Clayson & Weitlich, 2006; Kawai & Kawamura, 2005), even though these authors used different satellite or modelled data over different time periods. For example, Fig. 9a in Clayson and Weitlich (2006) shows basically the same areas with large DV events within the TWP region, although the DV amplitudes in their work are averaged over five years (1996– 2000) of modelled data. In order to better understand the distribution of the DV amplitudes and the features of the SST diurnal cycles, some statistical analyses are conducted. In Fig. 7a, it is shown that most DV events (22.2%) over the TWP+ domain and period have amplitudes of 0.2 °C–0.4 °C. However, in 2.82% of the cases (289,493 out of 10.3 million pixels) and 0.0023% of the cases (231 out of 10.3 million pixels), amplitudes N 2 °C and N5 °C are found. The highest percentage of all these dSSTmax values, 22.58%, are measured at 15:00–16:00 LST (Fig. 7b), while abnormally few (0.68%) are observed during the 17:00–20:00 LST period. This is because the 17:00–20:00 LST period corresponds to around sunset when far fewer MTSAT-1R observations are available (Section 3.1). Choosing only those pixels with dSSTmax N 2 °C, the monthly shape of the diurnal cycle is plotted against local hours in Fig. 7c and the STDs of the calculated dSST values in Fig. 7d. The dSSTs in Fig. 7c and d are the average dSST values over the specified month for each hour. The lowest dSSTs and STDs are found at 5:00–6:00 LST, i.e. pre-dawn, for all four months as expected. The highest dSSTs are found at slightly different local solar times for the four months, with 12:30 LST for January and 14:30 LST

for the other three months. Apart from these features, two abnormal jumps of both dSSTs and STDs are observed in some months at around 6:00–7:00 LST and 17:00–19:00 LST. This is to a large extent due to larger uncertainties in the cloud screening method around sunrise and sunset which cause unstable data quality (Section 3.1). Finally, we observe that the dSSTs and STDs at 23:30 are markedly larger than the 00:30 values. There are two factors that contribute to this difference. The first is that dSST is calculated as the difference from SSTfnd, itself the average of SST from 0:30 to 5:30 LST. Thus early morning dSST values approach 0 °C. The second is that for DV events N 2 °C, the cooling before midnight cannot totally cancel the heating during the day unless there are sudden strong wind bursts causing significant turbulent mixing. Fig. 8 shows the monthly diurnal cycles for all QC'd pixels under different daytime average wind speed conditions. Generally, these shapes share similar features as in Fig. 7c in terms of the trends. For the calm situation (wind speed b3 ms−1), for all four months, the amplitudes of monthly mean DV can reach N 1 °C. All the dSSTs reach their maximum at 14:00–15:00 LST except for January at 12:00–13:00 LST. When the wind speed is N3 ms−1 but b6 ms−1, the amplitudes of all months reduce to 0.3 °C–0.65 °C. If the wind speed is N 6 ms− 1, the diurnal cycle becomes very weak and hardly recognizable (Fig. 8c). For any wind speed condition, it is interesting to observe that even though almost all the dSSTs at 23:30 LST are higher than the 00:30 LST values, the differences become smaller as wind speeds increase. For wind speeds b3 ms− 1, the difference is around 0.15 °C–0.25 °C, which reduces to 0.05 °C–0.15 °C when the wind speed is N 3 ms−1 or N6 ms− 1. Finally, we observe abnormal jumps or falls at all sunrise

Fig. 7. Over the whole study region for four months: (a) pixel number distribution over different dSSTmax values; (b) distribution of pixels with dSSTmax over local hours; (c) monthly shape of the diurnal cycle; (d) monthly shape of the STDs of the calculated dSST values. Note that in (c) and (d), to highlight the curves, only pixels with dSSTmax N 2 °C are selected, and all the 24 dSST values at that pixel of the day are included in the calculation. Then the average dSST of all the dSSTs at each hour is plotted in (c), and the STD of all the dSSTs at each hour is plotted in (d).


9

Fig. 8. Monthly shape of the diurnal cycles under different wind speed conditions: (a) wind speed b3 ms−1; (b) 3 ms−1 b wind speed b6 ms−1; (c) wind speed N6 ms−1. Wind speeds are the mean daytime (7:00–19:00 LST) values. The dSSTs at each hour are the average of all measurements calculated.

and sunset times. The foundation SST construction method, the extraordinarily small data amount and low data quality all contribute to this. 4.2. Regional study Given that the densest data are clustered over the north-west coast of Australia (Fig. 5b), this region (110°E–130°E, 25°S–10°S) is selected to conduct regional studies, mainly to investigate the relationship between dSSTmax, wind speed and SSI.

Fig. 9 shows a case study of the significant DV events that occurred during a consecutive four day period from 5th to 8th March 2010 LST. High DV amplitudes of up to 4 °C over areas of the order of ~40,000 to ~100,000 km2 can be observed. These locations with the highest DV amplitudes and the regions with the lowest wind speeds are well matched. The positive correlation between dSSTmax and SSI is also illustrated by the locations where there are both high dSSTmax and SSI. However, SSI only plays a secondary role in comparison with wind speed, since high SSI does not necessarily lead to high dSSTmax. It is interesting to note

Fig. 9. Relationship between dSSTmax (left column) and daily mean wind speed (middle column) and SSI (right column) for four consecutive days (from 5–8th March 2010) over the northwestern Australian coast. Note that daily mean wind speed and SSI are the average values of 7:00–19:00 LST.

10


that the low SSI regions are found to match cloud distributions causing missing SST data in this particular case study, giving us some confidence in the ACCESS-R model's 24-hour forecasts of SSI for the TWP+ period and region. Following this case study, the relationship between the three parameters is investigated. The plot of the average DV amplitudes for each 10 Wm−2 × 0.4 ms−1 bin is displayed in Fig. 10. It is found that for wind speeds b6 ms−1, DV events are observed when SSI ≥ 400 Wm−2. The strongest DV events are observed when wind speed ≤ 1.2 ms−1 and SSI ≥ 500 Wm−2. At high wind speed (N10 ms−1), the few DV events that are observed have SSI ≥ 650 Wm−2 and minor amplitude (~0.5 °C). No DV events are observed when wind speed N 6.4 ms−1 and SSI b 600 Wm−2. These findings are also consistent with previous studies (e.g. Soloviev & Lukas, 1997; Webster et al., 1996). Further study to quantify the dependence of dSSTmax on the wind speed and SSI is considered necessary. The correlation coefficients between the three variables (referred to as RWS and RSSI hereafter) are calculated over the four months and the spatial distributions are shown in Fig. 11c for RWS and Fig. 11d for RSSI. Note that the inputs of wind speed and SSI are the daytime mean values. Also, since the spatial resolutions are different for the SST data (0.05° × 0.05°) and wind speed and SSI data (0.375° × 0.375°), centred around each daily mean wind speed and SSI pixel value, all adjacent (within 0.2° × 0.2°) dSSTmax values of the same day are averaged to form a pair. We observe that for the locations in the middle of the region with frequent DV events (Fig. 11a), most of the RWS values are between − 0.5 to − 0.7 (Fig. 11c). That shows a strong negative correlation between wind speed and DV events. These locations are also consistent with the densest data locations (Fig. 11b). However, over most of the region, the positive RSSI values rarely reach over 0.5 (Fig. 11d). This indirectly demonstrates that the wind speed plays a dominant role in DV events.

5. Discussion and conclusions Regarding the Bureau's four months (1st January to 30th April 2010) v3 MTSAT-1R data over the TWP+ Project domain (90°E to 170°E, 25°S to 15°N), this work focuses on: 1) the validation of the data set including in-situ and cross-validation; and 2) investigation of the derived SST DV events in combination with the wind speed and SSI data obtained from the ACCESS-R model.

Fig. 10. Relationship between dSSTmax and the daily mean SSI and wind speed is investigated by pixels in the selected region and time period (four days from 5-8th March). Shown in the figure are the average dSSTmax values over each 10 Wm−2 × 0.4 ms−1 bin. Note that the SSI starts at 400 Wm−2 as there are few pixels falling below this value which are therefore omitted. Also the daily mean wind speed and SSI are the average values of 7:00 to 19:00 LST.

From the in-situ validation, the v3 MTSAT-1R data overall prove to be of acceptable quality with a 0.00 °C bias and a 0.73 °C STD. The minimized day-night bias in this new version data is demonstrated by the relatively constant values (within the order of ±0.25 °C) over the 24hour period of the diurnal cycle, including the day-night transitions (Fig. 2e). On average, the daytime and night-time biases compared with collocated drifting buoy observations (adjusted for the “cool-skin effect” and selected for well-mixed surface ocean conditions) are − 0.06 °C and 0.08 °C, respectively. There is obvious overestimation found in MTSAT-1R data when the in-situ SSTs are b27 °C (cold SST condition) and underestimation when in-situ SSTs are N32 °C (warm SST condition), with magnitudes of b1.5 °C and b2.5 °C respectively. For the middle SST ranges where the densest collocations are yielded, MTSAT-1R's performance is in excellent agreement with that of in-situ data. However, the numbers of the in-situ collocations under the cold and especially the warm SST conditions are very limited. Therefore, the cross-validation is followed. In the cross-validation with ARC v1.1 AATSR satellite data, the daytime and night-time biases are − 0.04 °C and 0.06 °C, respectively. The underestimation and overestimation are also observed in both day and night validations although with smaller magnitudes (b 0.5 °C and b 2.0 °C). To ensure that the AATSR data are reliable, they are also validated against drifting buoy data over the same TWP+ domain and period. The results, not shown in this paper, suggest a 0.06 °C bias and 0.42 °C STD from 454 collocations. Although the collocated pairs are not large in number, they perform consistently well over all SST ranges with no evident underestimation (overestimation) over cold (warm) SSTs. Thus far we can draw the conclusion that the v3 MTSAT-1R data perform well overall with small day-night biases, but tend to overestimate (underestimate) the temperature under cold (warm) SST conditions. These biases at the temperature extremes may be due to the method of generating coefficients for the regression algorithm between buoy and satellite SST observations. As there are relatively few matchups at extreme conditions, the extremes are under-represented in the training data. However, since the SST DV events over the TWP + domain and period have been demonstrated to have amplitudes up to several degrees, the frequency, amplitudes and spatial patterns derived from the MTSAT-1R data should be robust. The influences of sunrise and sunset effects on MTSAT-1R cloud identification are revealed in both data amounts and qualities and are not negligible. In addition to the largely missing data at these times, the data qualities near sunrise and sunset times are also relatively more questionable, indicated by the large STDs at 6:00–8:00 LST in Fig. 2e. The average dSSTs at each local hour under different wind speed conditions are also investigated. The observed average dSSTmax is 1 ° C–1.2 °C for wind speeds b3 ms−1, 0.4 °C–0.6 °C for wind speeds of 3– 6 ms− 1, and b0.2 °C for wind speeds N 6 ms− 1. This can potentially offer some useful references in choosing the wind speed thresholds for accounting or not for the effects of SST DV events in future numerical weather prediction models. Investigation into the DV events derived from MTSAT-1R data is followed together with analysis of wind speed and SSI characteristics. Due to the warm SSTs and strong evaporation and air–sea interaction, there are persistent clouds over the TWP region which often prevents the infrared sensor from measuring the SST. This results in large areas of data gaps. Despite missing SST data, the spatial distribution of large monthly mean dSSTmax has shown its high correlation with low wind speed and high SSI. The case study over the northwest coast of Australia has shown very good spatial consistency between the large DV events (dSSTmax as high as 4 °C over areas as large as ~40,000 to ~100,000 km2) and low winds. The correlation coefficients between the dSSTmax and wind speed, and dSSTmax and SSI, are shown for the TWP + region and period. The negative correlation between dSSTmax and wind speed has values up to −0.7, while the values of the positive correlation between dSSTmax and SSI do not exceed 0.5, demonstrating the dominant role of wind speed in DV events.


11

Fig. 11. Spatial distribution over four months of: (a) the hours with dSST values N1 °C; (b) number of quality controlled data; (c) correlation coefficients between dSSTmax and the wind speed; and (d) correlation coefficients between dSSTmax and the SSI.

The results of this SST DV study may have potential applications in short-term or long-term climate models, whose air–sea interaction accuracy can be enhanced if DV effects are considered. Although the temporal coverage is relatively small in this study, the v3 MTSAT-1R data have been demonstrated to be useful for DV studies over the TWP region. The data set is of sufficient accuracy to quantify the amplitude of SST DV events. The v3 MTSAT-1R data are therefore likely to play an important role in the inter-comparison of several DV models over this region in future studies as part of the TWP+ Project. Reprocessing the longer record of v2 MTSAT-1R radiances (June 2006 to June 2010) archived at the Bureau to v3 SST records could be of great use in the future. The MTSAT-2 and Himawari-8 geostationary satellites have also been launched and are designed to be operational from 2010 to 2015and from 2015 to 2022, respectively. Sea surface temperature observations from these two satellites are currently being processed by the Bureau. Using these new geostationary satellite SST data sets together with the newly reprocessed 23-year (1992–2015) IMOS 0.02° resolution AVHRR SST data set (http://imos.org.au/sstproducts.html), will allow DV events over the TWP region to be more thoroughly examined in future studies. Acknowledgements This is a Sino-Australian Research Centre for Coastal Management Publication Number 31. The authors would like to thank Sandra Castro, Gary Wick, Chris Merchant, Jon Mittaz and Andy Harris for helpful discussions relating to the enhancement and application of the MTSAT-1R and ACCESS-R data sets for TWP+ DV studies and Matthew Wheeler for reviewing the manuscript. The v3 MTSAT-1R SST data set was produced as a collaboration between the Bureau of Meteorology, Integrated Marine Observing System (IMOS) Project and the NOAA Centre for Satellite Applications and Research (STAR), with significant contributions from Eileen Maturi, Jon Mittaz and Andy Harris. If one would like to acquire the data, please contact Dr. Helen Beggs ([email protected]). The ARC v1.1 data set was kindly provided by the ATSR Reprocessing for Climate Project. These data are hosted and made available by the UK Natural Environment Research Council Earth Observation Data Centre (http://www.neodc.rl.ac.uk). The authors

would also like to thank the two anonymous reviewers whose detailed comments have greatly improved the clarity and quality of this paper. The research was funded, in part, by the ARC Centre of Excellence for Climate System Science (grant CE110001028). References An, S. -I., Kim, J. -W., Im, S. -H., Kim, B. -M., & Park, J. -H. (2012). Recent and future sea surface temperature trends in tropical pacific warm pool and cold tongue regions. Climate Dynamics, 39, 1373–1383. http://dx.doi.org/10.1007/s00382-011-1129-7. Beggs, H., Majewski, L., Kruger, G., Verein, R., Oke, P., Sakov, P., ... Griffin, C. (2012a). Report to GHRSST13 from Australia — Bluelink and IMOS. Proceedings of the GHRSST XIII Science Team Meeting (pp. 149–163) Tokyo, 4–8 June 2012 https://www.ghrsst.org/ files/download.php?m=documents&f=121207144549GHRSSTXIIIProceedingsIssue1Rev0.pdf Beggs, H., Kippo, H., & Underwood, M. (2012b). Enhancing ship of opportunity sea surface temperature observations in the Australian region. Journal of Operational Oceanography, 5(1), 59–73. http://dx.doi.org/10.1080/1755876X.2012.11020132. Beggs, H., Majewski, L., Griffin, C., Verein, R., Sakov, P., Huang, X., ... Tingwell, C. (2013). Report to GHRSST14 from Australia — Bluelink and IMOS. Proceedings of the GHRSST XIV Science Team Meeting (pp. 104–121). USA: Woods Hole 17–21 June 2013 https:// www.ghrsst.org/documents/q/category/ghrsst-science-team-meetings/ghrsst-xivwoods-hole-ma-usa/ Böhm, E., Marullo, S., & Santoleri, R. (1991). AVHRR visible-IR detection of diurnal warming events in the western Mediterranean Sea. International Journal of Remote Sensing, 12(4), 695–701. http://dx.doi.org/10.1080/01431169108929686. Castro, S. L., Wick, G. A., & Buck, J. J. H. (2014). Comparison of diurnal warming estimates from unpumped Argo data and SEVIRI satellite observations. Remote Sensing of Environment, 140, 789–799. http://dx.doi.org/10.1016/j.rse.2013.08.042. Clayson, C. A., & Bogdanoff, A. S. (2013). The effect of diurnal sea surface temperature warming on climatological air–sea fluxes. Journal of Climate, 26(8), 2546–2556. http://dx.doi.org/10.1175/jcli-d-12-00062.1. Clayson, C. A., & Weitlich, D. (2006). Variability of tropical diurnal sea surface temperature. Journal of Climate, 20, 334–352. http://dx.doi.org/10.1175/JCLI3999.1. Donlon, C. J., Minnett, P. J., Gentemann, C., Nightingale, T. J., Barton, I. J., Ward, B., & Murray, M. J. (2002). Toward improved validation of satellite sea surface skin temperature measurements for climate research. Journal of Climate, 15(4), 353–369. http://dx.doi.org/10.1175/1520-0442(2002)015b0353:TIVOSSN2.0.CO;2. Donlon, C., Robinson, I., Casey, K. S., Vazquez-Cuervo, J., Armstrong, E., Arino, O., ... Rayner, N. (2007). The global ocean data assimilation experiment high-resolution sea surface temperature pilot project. Bulletin of the American Meteorological Society, 88(8), 1197–1213. http://dx.doi.org/10.1175/BAMS-88-8-1197. Eastwood, S., Le Borgne, P., Péré, S., & Poulter, D. (2011). Diurnal variability in sea surface temperature in the Arctic. Remote Sensing of Environment, 115(10), 2594–2602. http://dx.doi.org/10.1016/j.rse.2011.05.015. Embury, O., & Merchant, C. J. (2012). A reprocessing for climate of sea surface temperature from the along-track scanning radiometers: a new retrieval scheme. Remote Sensing of Environment, 116, 47–61. http://dx.doi.org/10.1016/j.rse.2010.11.020.

12


Embury, O., Merchant, C. J., & Corlett, G. K. (2012a). A reprocessing for climate of sea surface temperature from the along-track scanning radiometers: initial validation, accounting for skin and diurnal variability effects. Remote Sensing of Environment, 116, 62–78. http://dx.doi.org/10.1016/j.rse.2011.02.028. Embury, O., Merchant, C. J., & Filipiak, M. J. (2012b). A reprocessing for climate of sea surface temperature from the along-track scanning radiometers: basis in radiative transfer. Remote Sensing of Environment, 116, 32–46. http://dx.doi.org/10.1016/j.rse. 2010.10.016. Fairall, C. W., Bradley, E. F., Godfrey, J. S., Wick, G. A., Edson, J. B., & Young, G. S. (1996). Cool-skin and warm-layer effects on sea surface temperature. Journal of Geophysical Research, 101(C1), 1295. http://dx.doi.org/10.1029/95jc03190. Gentemann, C. L., Minnett, P. J., Le Borgne, P., & Merchant, C. J. (2008). Multi-satellite measurements of large diurnal warming events. Geophysical Research Letters, 35(22). http://dx.doi.org/10.1029/2008gl035730. Halpern, D., & Reed, R. K. (1976). Heat budget of the upper ocean under light winds. Journal of Physical Oceanography, 6(6), 972–975. http://dx.doi.org/10.1175/15200485(1976)006b0972:HBOTUON2.0.CO;2. Hawkins, J. D., Clancy, R. M., & Price, J. F. (1993). Use of AVHRR data to verify a system for forecasting diurnal sea surface temperature variability. International Journal of Remote Sensing, 14(7), 1347–1357. http://dx.doi.org/10.1080/01431169308953961. Karagali, I., & Høyer, J. L. (2013). Observations and modeling of the diurnal SST cycle in the north and Baltic Seas. Journal of Geophysical Research, Oceans, 118(9), 4488–4503. http://dx.doi.org/10.1002/jgrc.20320. Karagali, I., & Høyer, J. L. (2014). Characterisation and quantification of regional diurnal SST cycles from SEVIRI. Ocean Science, 10, 745–758. http://dx.doi.org/10.5194/os10-745-2014. Karagali, I., Høyer, J., & Hasager, C. (2012). SST diurnal variability in the North Sea and the Baltic Sea. Remote Sensing of Environment, 121, 159–170. http://dx.doi.org/10.1016/j. rse.2012.01.016. Kawai, Y., & Kawamura, H. (2005). Spatial and temporal variations of model-derived diurnal amplitude of sea surface temperature in the western Pacific Ocean. Journal of Geophysical Research, 110, C08012. http://dx.doi.org/10.1029/2004JC002652. Kawai, Y., Otsuka, K., & Kawamura, H. (2006). Study on diurnal sea surface warming and a local atmospheric circulation over Mutsu Bay. Journal of the Meteorological Society of Japan, 84(4), 725–744. http://dx.doi.org/10.2151/jmsj.84.725. Kawamura, H., Qin, H., & Ando, K. (2008). In-situ diurnal sea surface temperature variations and near-surface thermal structure in the tropical hot event of the Indo-Pacific Warm Pool. Journal of Oceanography, 64, 847–857. http://dx.doi.org/10.1007/s10872008-0070-9. Kawamura, H., Qin, H., Sakaida, F., & Qiu, C. (2010a). Advanced sea surface temperature retrieval using the Japanese geostationary satellite, Himawari-6. Journal of Oceanography, 66, 855–864. http://dx.doi.org/10.1007/s10872-010-0069-x. Kawamura, H., Qin, H., Sakaida, F., & Setiawan, R. Y. (2010b). Hourly sea surface temperature retrieval using the Japanese geostationary satellite, multi-functional transport satellite (MTSAT). Journal of Oceanography, 66, 61–70. http://dx.doi.org/10.1007/ s10872-010-0005-0. Kim, M. -J., Ou, M. -L., Sohn, E. -H., & Kim, Y. (2011). Characteristics of sea surface temperature retrieved from MTSAT-1R and in-situ data. Asia-Pacific Journal of Atmospheric Sciences, 47(5), 421–427. http://dx.doi.org/10.1007/s13143-011-0027-6.

Majewski, L., Griffin, C., & Beggs, H. (2013). Spatial and temporal biases in MTSAT-1R SST. Presented as a poster at the 4th Asia Oceania Meteorological Satellite User's Conference. http://dx.doi.org/10.13140/RG.2.1.3861.6805 Melbourne, Australia, 9–11 October 2013. Merchant, C. J., & Le Borgne, P. (2004). Retrieval of sea surface temperature from space, based on modeling of infrared radiative transfer: capabilities and limitations. Journal of Atmospheric and Oceanic Technology, 21(11), 1734–1746. http://dx.doi. org/10.1175/JTECH1667.1. Merchant, C. J., Harris, A. R., Maturi, E., & Maccallum, S. (2005). Probabilistic physically based cloud screening of satellite infrared imagery for operational sea surface temperature retrieval. Quarterly Journal of the Royal Meteorological Society, 131, 2735–2755. http://dx.doi.org/10.1256/qj.05.15. Merchant, C. J., Filipiak, M. J., Borgne, P. L., Roquet, H., Autret, E., Piolle, J. -F., & Lavender, S. (2008). Diurnal warm-layer events in the western Mediterranean and European shelf seas. Geophysical Research Letters, 35, L04601. http://dx.doi.org/10.1029/ 2007GL033071. Merchant, C. J., Embury, O., Roberts-Jones, J., Fiedler, E., Bulgin, C. E., Corlett, G. K., ... Donlon, C. (2014). Sea surface temperature datasets for climate applications from phase 1 of the European Space Agency Climate Change Initiative (SST CCI). Geoscience Data Journal, 1(2), 179–191. http://dx.doi.org/10.1002/gdj3.20. Park, J. -Y., Yeh, S. -W., & Kug, J. -S. (2012). Revisited relationship between tropical and North Pacific sea surface temperature variations. Geophysical Research Letters, 39, L02703. http://dx.doi.org/10.1029/2011GL050005. Puri, K., et al. (2013). Implementation of the initial ACCESS numerical weather prediction system. Australian Meteorological and Oceanographic Journal, 63, 265–284. Soloviev, A., & Lukas, R. (1997). Observation of large diurnal warming events in the nearsurface layer of the western equatorial Pacific warm pool. Deep Sea Research Part I, 44(6), 1055–1076. http://dx.doi.org/10.1016/S0967-0637(96)00124-0. Stommel, H., Saunders, K., Simmons, W., & Cooper, J. (1969). Observations of the diurnal thermocline. Deep Sea Research, 16, 269–284. Stuart-Menteth, A. C., Robinson, I. S., & Challenor, P. G. (2003). A global study of diurnal warming using satellite-derived sea surface temperature. Journal of Geophysical Research, 108(C5). http://dx.doi.org/10.1029/2002jc001534. Tanahashi, S., Kawamura, H., Takahashi, T., & Yusa, H. (2003). Diurnal variations of sea surface temperature over the wide-ranging ocean using VISSR on board GMS. Journal of Geophysical Research, 108(C7). http://dx.doi.org/10.1029/2002JC001313. Vecchi, G. A., Clement, A., & Soden, B. J. (2008). Examining the tropical Pacific's response to global warming. Eos, Transactions American Geophysical Union, 89(9), 81–83. http:// dx.doi.org/10.1029/2008EO090002. Webster, P. J., Clayson, C. A., & Curry, J. A. (1996). Clouds, radiation, and the diurnal cycle of sea surface temperature in the tropical Western Pacific. Journal of Climate, 9, 1712–1730. http://dx.doi.org/10.1175/1520-0442(1996)009b1712:CRATDCN2.0.CO;2. Zhong, A., & Beggs, H. (2008). Operational implementation of global Australian multisensor sea surface temperature analysis. Analysis and prediction operations bulletin no. 77http://cawcr.gov.au/projects/SST/GAMSSA_BoM_Operational_Bulletin_77.pdf