1 INTRODUCTION

The upper ocean thermal structure is crucial to the development and intensification of tropical cyclones(TCs). Earlier studies have proven that sea surface temperature(SST)is an important factor that affects the genesis and development of TCs(DeMaria and Kaplan, 1994). It is well known that SST>26°C is needed to sustain TC genesis and development(Shay et al., 2000; Goni and Trinanes, 2003; Lin et al., 2005, 2008, 2009a, b). However, recent studies have found that the upper ocean heat content has greater influence on TC intensification(Emanuel, 1999; Bender and Ginis, 2000; Shay et al., 2000; Emanuel et al., 2004; Scharroo et al., 2005). In intensity forecasting, SST prior to cyclone development has been shown to be less critical than upper ocean heat content, particularly when forecasting rapid intensification(Schade and Emmanuel, 1999). In tropical regions, distributions of SST tend to be spatially homogeneous and differences in SST are generally not significant, whereas upper ocean heat content has distinct variation(Moon et al., 2009). Tropical cyclone heat potential(TCHP)is defined as the integrated heat from the sea surface to the depth of the 26°C isotherm(Leipper and Volgenau, 1972). TCHP is usually used to estimate the amount of heat available for cyclones to convert into latent heat energy(Wada and Usui, 2007; Shay and Brewster, 2010). It has been shown to be a much better predictor than SST alone in the statistical intensity prediction scheme(Law and Hobgood, 2007). Previous studies have shown that rapid intensification of cyclones requires them to obtain energy from the warm ocean with a deep mixed layer, which can be modulated by mesoscale ocean processes such as mesoscale eddies, rings, and fronts(Lin et al., 2005, 2008, 2009a, b; Wada and Usui, 2007; Shay and Brewster, 2010). TCs can cause a reduction of SST, which introduces a negative feedback loop in TC intensification process(Cione and Uhlhorn, 2003). TC self-induced SST cooling exceeding 2.5°C is regarded as an unsuitable condition for intensification, because such cooling would be sufficient to shut down the entire energy production of a TC(Gallacher et al., 1989; Emanuel, 1999). However, a warm and thick mixed layer can reduce TC-induced sea surface cooling, which can cause an increment in enthalpy flux from the ocean to the atmosphere and consequently favor the intensification of a TC(Lin et al., 2005, 2008, 2009a, b). Most major Category 4 or 5 TCs in various basins have been found to intensify rapidly over regions of high TCHP associated with warm eddies or a thick and warm mixed layers(Shay et al., 2000; Lin et al., 2005, 2008, 2009a, b). As the intensification of TCs depend on oceanic energy, underst and ing the interactions between upper ocean processes and TCs are critical for precise predictions of TC intensity.

Many earlier studies have focused on TC intensification in the Pacific and Atlantic oceans(DeMaria and Kaplan, 1994; Kaplan and DeMaria, 2003; Pun et al., 2007; Wada and Usui, 2007), but there have been fewer studies of TCs in the Indian Ocean. Previous studies in the Indian Ocean have reported cyclone intensification(dissipation)after traveling over anticyclonic(cyclonic)eddies in the North Indian Ocean(Aliet al., 2007). As well as the severe cyclone Sidr(2007)was intensified by an oceanic area with high TCHP values of 80 kJ/cm ² before it made l and fall(Lin et al., 2013) and Cyclone Nargis(2008), which caused enormous damage, encountered high TCHP values of nearly 100 kJ/cm ²(Lin et al., 2008; McPhaden et al., 2009). In situ observations in the Indian Ocean are not as widespread as in other ocean basins and therefore it is important to derive temperature/salinity(T/S)fields from satellite data to monitor the development and intensification of TCs. Satellite derived TCHP values are convenient for use over the large spatial and temporal scales of the North Indian Ocean(Nagamaniet al., 2012). A two layer model for estimating TCHP based on altimetry data has already been developed by Goniet al.(1996) and Shay et al.(2000), and further efforts have been directed towards a linear regression method that has greater vertical resolution(Guinehut et al., 2004, 2012). The method of this study is to reconstruct the upper ocean thermal structure by combining sea surface height(SSH) and SST using historical observed T/S profiles.

The remainder of this paper is organized as follows. Section 2 introduces the data and method used to derive the upper ocean thermal structure from satellite data. The T/S fields derived from the satellite remote sensing data are validated in Section 3. Section 4 verifies the satellite derived TCHP and Section 5 studies the intensification process of TC GONU using the satellite derived TCHP. Finally, a brief concluding summary is presented in Section 6.

2 DATA AND METHOD2.1 Data

The historical T/S profiles(Fig. 1)are based on a combination of data from the World Ocean Database 2009(WOD09)(Levitus, 2009), Global Temperature and Salinity Profile Project(GTSPP)(Sun, 2012), and Array for Real time Geostrophic Oceanography(ARGO). They were subjected to unified quality control, including checks on profile duplication, position/time, depth duplication, depth inversion, temperature and salinity range, excessive gradient, and stratification stability. In particular, drifting errors associated with ARGO salinity profiles were calibrated by employing the WJO calibration method developed by Wong et al.(2003). Following these checking procedures, the T/S dataset comprised 244 330 temperature profiles and 64 398 salinity profiles for the region 10°S–32°N, 25°–100°E from 1900–2010.

The satellite SST dataset was obtained from the Advanced Very High Resolution Radiometer, which had a spatial resolution of 0.25°×0.25° and daily temporal resolution(Reynolds et al., 2007). The merged Sea Surface Height Anomaly(SSHA)data was obtained from the Archiving, Validation and Interpretation of Satellite Oceanographic data, which have been configured since 1992 with the same spatial and temporal resolutions as the SST data(Ducet et al., 2000).

TC track data, which include the 6-hourly center position, central pressure, and maximum sustained wind speed, were obtained from the International Best Track Archive for Climate Stewardship(Knapp et al., 2010).

2.2 Method

Based on the historical T/S profiles, the relationships between subsurface temperature at each layer and SST and steric height were established using a multivariable regression method(Fox et al., 2002). The formulae for calculating the subsurface temperature from SST and steric height are given as:

where i is the index of the horizontal grid, j is the index of the observational location(i and j have two directions, including meridional and zonal directions, e.g., i _{lon, lat}, j _{lon, lat}), and k is the index of the vertical grid. i, is the weighted average temperature at grid point iand depth k, h is the satellite derived SSHA, and is the weighted mean of the steric height, which is calculated by integrating the historical T/S profiles(Eq.5). T_{j k} and are observational temperature and steric height at observational position j and depth k . is the weighted mean of , and a_ik, a_ik, and a_ik are regression coefficients. The weight b _{i, j} is a locally homogeneous correlation function:where x and y are longitude and latitude, and t is the time of year. L_x and L_y denote the zonal and meridional length scale, while L_t represents the time scale.

The steric heights were calculated using the method defined by Olbers et al.(1992):

where v(T, S, p)is the specific volume of seawater, v(0, 35, p)is the st and ard specific volume of seawater at 0°C and salinity 35, and H is the water depth.

The salinity profiles were derived using a weighted linear regression relationship between salinity and temperature by employing the observation profiles both having temperature and salinity; the functions are as follows(Fox et al., 2002):

where is the weighted mean of salinity at grid point i and depth k, S_{j k} is the observational salinity at observed position j and depth k, and a_ik is the regression coefficient.

Finally, the T/S dataset for 1993–2012 was derived from the SST and SSHA. The coverage of the dataset is 10°S–32°N and 25°–100°E with a daily temporal resolution and spatial resolution of 0.25°×0.25°. The vertical resolution is 26 levels at the st and ard layers(0, 5, 10, 15, 20, 25, 30, 35, 50, 75, 100, 125, 150, 175, 200, 250, 300, 350, 400, 450, 500, 600, 700, 800, 900, and 1 000 m).

3 T/S DATASET VERIFICATION

After obtaining the T/S dataset, it is necessary to validate it using independent observational data. Here, we used the independent GTSPP data from 2011–2012 for validation. After quality control, 439 76 observed profiles were employed to verify the satellite derived T/S fields. The monthly numbers of T/S profile observations are shown in Fig. 2f. The numbers of observations from January to December are 3 052, 3 613, 4 128, 3 789, 3 935, 3 731, 3 243, 3 324, 3 438, 3 965, 4 272, and 3 486. The biases between the satellite derived and observed temperatures are computed for the upper layers and are displayed in Fig. 2a–e. It can be seen that the bias of satellite derived temperature shows a pattern similar to a Gaussian distribution. The proportion of biases between -1°C and 1°C are 96.0%, 85.6%, 46.2%, 86.6%, and 91.3% at the sea surface, 50, 100, 200, and 400 m depth, respectively. Most biases of temperature over the vertical layers are small, except that the temperature at 100-m depth has a bias >3.5°C.

The monthly vertical distribution of bias, root mean square error(RMSE), and correlation coefficient between the satellite derived and observed temperatures are shown in Fig. 3. The RMSE in the upper mixed layer(upper ~50 m)is usually ＜1°C. The RMSE from 0–200-m depth is 1.14 on average with a maximum value of 2.06°C at 100-m depth. The correlation coefficients reveal the inverse pattern to the RMSE. The correlation coefficients are usually >0.9 in the mixed layer or below the thermocline, which indicate lower errors in the satellite derived temperature field. The correlation coefficient from 0–200-m depth is 0.81 on average with a minimum value of 0.69 at 150-m depth. The vertical RMSE and correlation coefficient averaged across all layers are 0.83 and 0.87, respectively. Moreover, the correlations at each level are statistically significant at the 99% confidence level based on the F-test. Regarding monthly variability, the RMSEs from March to August are small at each depth, and the correlation coefficients larger than for the other months.

The monthly vertical distribution of bias, RMSE, and correlation coefficient between the satellite derived and observed salinities are shown in Fig. 4. The bias and RMSE of salinity are larger from 0–200 m and gradually reduce with depth. The annual averages of RMSE and the correlation coefficient from 0–200 m are 0.27 and 0.91, respectively.

The RMSE of derived temperature was also compare with WOA result. Monthly averaged vertical distribution of derived RMSE and WOA RMSE in region of 5°–32°N, 25°–100°E were shown in Fig. 5. The RMSE of derived temperature was similar to WOA above 50m or below 300-m depth. At the thermocline, the derived RMSE was better than WOA, especially from September to November. And the RMSE of derived temperature was at least 0.25°C better than WOA at 100-m depth.

4 TCHP VERIFICATIONS

As demonstrated in previous researches(Hong et al., 2000; Cione et al., 2013), the intensification of TCs is caused mainly by high upper ocean heat content, which can be represented as TCHP. TCHP is the integrated heat content from the depth of 26°C isotherm to the sea surface, and recent studies have revealed that it can significantly affect the intensity of TCs, especially when values exceed 60–90 kJ/cm ²(Holliday and Thompson, 1979). TCHP can be calculated as follows(Leipper and Volgenau, 1972):

where D ₂₆ represents the depth of the 26°C isotherm calculated from derived temperature profiles, C _p is the heat capacity at constant pressure that has the value of 4 178 J/(kg·°C), ρ(z)is the density of water at depth z calculated using the seawater status equation defined by UNESCO(United Nations Educational, Scientific and Cultural Organization), and T(z)is the temperature at depth z .

To validate the satellite derived D26 and TCHP, the observed D26 and TCHP are calculated from the GTSPP T/S profiles from 2011–2012 and the satellite derived D26 and TCHP are interpolated to the same times and positions as the observational data. Monthly scatter diagrams of the observed and satellite derived D26 during 2011–2012 are shown in Fig. 6. The monthly coefficient of determination(R ₂)ranges from 0.47 to 0.80 and the maximum(minimum)value appears in December(February). With the D26 confirmed, the TCHP are calculated from the derived T/S profiles for comparison with the TCHP estimated from the observed data. Monthly scatter diagrams of the observed and derived TCHP are displayed in Fig. 7, which show that the maximum(minimum)R 2 is 0.82(0.58)in May(August).

The monthly bias, RMSE, and correlation coefficient of the satellite derived D26 and TCHP are calculated using the 24 683 available observational profiles, are shown in Fig. 8. The bias of the satellite derived D26 ranges from -4.17 to 11.11 m and the RMSE ranges from 9.37 to 16.63 m. The monthly average correlation coefficient between the satellite derived and observed D26 is 0.80, ranging from 0.68 to 0.89. The bias of the satellite derived TCHP ranges from -6.27 to 7.45 kJ/cm ², which indicates that the satellite derived TCHP might be generally underestimated from December to May and overestimate form June to November. The RMSE of the satellite derived TCHP ranges from 11.91 to 19.32 kJ/cm ² with a maximum value in January and minimum in September. The correlation coefficient between the satellite derived and observed TCHP is >0.8, except in August. The F-test demonstrated that the correlations between the satellite derived D26(TCHP) and observed D26(TCHP)in each month are statistically significant at the 99% confidence level.

5 CASE STUDY OF A TC

We investigated the rapid intensify TC Gonu(2007)to study the relationship between intensification and TCHP. Gonu emerged near the eastern Arabian Sea(14.4°N, 70.9°E)on June 1, 2007, where a warm upper ocean was favorable for TC formation and development. Gonu started to intensify on June 2 and reached its peak with winds of 145 knots on June 4. While moving towards the northwest, Gonu traveled close to the Gulf of Oman, which was a region with cooler ocean temperatures and thus the intensity of Gonu decreased until it dissipated by June 7. In pervious study, rapid intensification of a TC is defined as a central pressure reduction >10 hpa in 6 hours or an increase in wind speed >30 knots in 24 hours(Bender and Ginis, 2000). Accordingly, Gonu exhibited a period of rapid intensification from 12:00 UTC June 3 to 12:00 UTC June 4. The track of Gonu overlaid by the SST, SSHA, satellite derived TCHP, and D26 is shown in Fig. 9. Gonu formed over a region of warm water in the eastern Arabian Sea with TCHP >60 kJ/cm ² and gradually strengthened until 06:00 UTC June 3. During the subsequent 24 hours, Gonu intensified rapidly and reached Category 5 level. The horizontal distribution of SSHA on June 2 is displayed in Fig. 9c, indicating that Gonu entered a region with a warm and thick layer at 12:00 UTC June 3, and the relevant TCHP actually shows high values around the center of Gonu(Fig. 9b). When Gonu reached its maximum intensity, the ambient TCHP exceeded 90 kJ/cm ² . Eventually, Gonu entered a colder region where TCHP was ＜30 kJ/cm ², which subsequently weakened the cyclone.

Along Gonu’s route, there are six observational profiles along the track and therefore the satellite derived T/S profiles can be verified during the cyclone’s existence(Fig. 10). The biases of the satellite derived and observed temperature profiles range from -2.52 to 1.41°C with the maximum biases located at depths of 50–200 m. The profile prior to the development of the cyclone(5/31)have lower bias, while the biases of the profiles during the cyclone’s existence(6/2–6/3)are larger. During the period of the cyclone, the satellite derived temperature display some over-estimations compared with the observed results at depths >100 m.

The variations of five parameters at 6-hourly intervals, i.e. SST anomalies(ΔSST), D26 anomalies(ΔD26), TCHP anomalies(ΔTCHP)(the anomaly data are calculated by the subtraction of WOA09 from the variables, respectively), central pressure, and maximum sustained wind speed during the rapid intensification of Gonu, are displayed in Fig. 11. The values of ΔSST, ΔTCHP, and ΔD26 were 1°×1° averages around the central position of Gonu at the corresponding time. The values of ΔSST and ΔTCHP exhibit a homologous variation with wind speed during the period of rapid intensification from 00:00 UTC June 3 to 12:00 UTC June 4. During this period, ΔSST at 18:00 UTC June 3 is smaller than the monthly climatology for June, but ΔTCHP is larger, indicating that the subsurface T/S variability dominated the TCHP variation.

The relationship between the change in wind intensity and TCHP/SST are investigated for the period of rapid intensification. Scatter diagrams of TCHP/SST and wind speed increment are shown in Fig. 12. The previous 6-hourly TCHP(pre-TCHP) and previous 6-hourly SST(pre-SST)time series have greater correlation with the changing rate of wind speed compared with the simultaneous correlation. Energy exchange between the upper ocean and the cyclone is a gradual and cumulative process; therefore, the effect of TCHP on cyclone enhancement is gradual. Meanwhile, the value of R ² between the pre- TCHP and change of wind intensity is 0.66, which is much larger than that(0.32)between the pre-SST and change of wind intensity. This indicates that the change of wind intensity is more sensitive to TCHP than SST.

6 CONCLUSION

Based on historical T/S profiles from 1900–2010, this study used the least square method to establish the regression relationship between SST, steric height, and subsurface information in the North Indian Ocean. T/S fields from 1993–2012 were derived from satellite sensing of SST and SSHA determined by employing the regression relationship. This satellite derived T/S dataset had daily temporal resolution, a spatial resolution of 0.25°×0.25°, and 26 vertical layers in the upper 1 000 m. Independent GTSPP data were used to validate the satellite derived T/S dataset. The results showed that the satellite derived T/S profiles agreed well with the observations. The RMSE of temperature was no more than 1°C in the upper 50-m depth or below the thermocline. The maximum RMSE occurred near the thermocline at 100–150 m depth, and ranged from about 1–2.5°C. The vertically averaged RMSE and correlation coefficient were 0.83 and 0.87, respectively. In addition, satellite derived TCHP and D26 were also validated against the observed data, for which the correlations of the monthly averages were 0.86 and 0.80, respectively, which were statistically significant at the 99% confidence level by the F-test. Finally, the period of rapid intensification of TC Gonu was examined as a case study using the satellite derived TCHP. The results demonstrated that the effect of TCHP on the rapid intensification was more significant than SST. It was found that there was a delay in the effect of TCHP on the intensity change of Gonu. The correlation between wind speed change and pre-TCHP reached 0.81 during the period of rapid intensification, which explained 66% of the intensity change variance and higher than the simultaneous result. This phenomenon suggested a lag feature in the response of the inner core of the TC to the ocean. In this study, we only investigated the intensification process of the Gonu case related to TCHP. Further study will systematically analyze the influence of the upper ocean thermohaline structure on the TC intensity change in the North Indian Ocean.