Estimation of net surface shortwave radiation over the tropical Indian Ocean using geostationary satellite observations: Algorithm and validation



[1] This paper presents the development of a methodology to estimate the net surface shortwave radiation (SWR) over tropical oceans using half-hourly geostationary satellite estimates of outgoing longwave radiation (OLR). The collocated data set of SWR measured at 13 buoy locations over the Indian Ocean and a Meteosat-derived OLR for the period of 2002–2009 have been used to derive an empirical relationship. The information from the solar zenith angle that determines the amount of solar radiation received at a particular location is used to normalize the SWR to nadir observation in order to make the empirical relationship location independent. As the relationship between SWR and OLR is valid mostly over the warm-pool regions, the present study restricts SWR estimation in the tropical Indian Ocean domain (30°E–110°E, 30°S–30°N). The SWR estimates are validated with an independent collocated data set and subsequently compared with the SWR estimates from the Global Energy and Water Cycle Experiment-Surface Radiation Budget V3.0 (GEWEX-SRB), International Satellite Cloud Climatology Project-Flux Data (ISCCP-FD), and National Centers for Environmental Prediction (NCEP) reanalysis for the year 2007. The present algorithm provides significantly better accuracy of SWR estimates, with a root-mean-square error of 27.3 W m−2 as compared with the values of 32.7, 37.5, and 59.6 W m−2 obtained from GEWEX-SRB, ISCCP-FD, and NCEP, respectively. The present algorithm also provides consistently better SWR compared with other available products under different sky conditions and seasons over Indian Ocean warm-pool regions.

1. Introduction

[2] The shortwave radiative flux absorbed by the ocean surface is a key factor that influences air-sea interactions as well as the atmospheric and oceanic circulations. There is little information about the large-scale and high spatiotemporal variability of shortwave radiation (SWR) over the tropical Indian Ocean. The estimate of SWR is of prime importance for the ocean- modeling community, particularly for climate studies. The accurate information of SWR would lead to reduced uncertainty in the simulation of air-sea temperature differences, which represents a major problem in assessing the heat balance of the warm-pool region [Lukas, 1989]. This would also result in significantly improved representation of the intra-annual oceanic surface processes and long-term surface variability. In recent years there have been efforts to establish an accurate estimate of the surface shortwave radiation budget (SRB) over the tropical ocean sand to assess its impact on climate change. The diurnal variations in the sea surface temperature (SST) are driven by surface solar radiation that is modulated primarily by the presence of cloudiness.

[3] The tropical oceans, where the SST is usually greater than 28°C, form a major part of the largest warm pool on the Earth. The variations of surface insolation over the warm-pool regions are primarily due to the variations in cloudiness, which are manifestations of deep convection [Shinoda et al., 1998; Sengupta et al., 2001]. The long-term surface measurements of radiative fluxes are limited primarily to the continental regions [Ohmura and Gilgen, 1991]. The satellite estimates of cloudiness and top-of-the-atmosphere fluxes have been used in conjunction with radiative transfer models to produce global estimates of the SRB over land and ocean [Li, 1995; Rossow and Zhang, 1995; Whitlock et al., 1995; Gupta et al., 1997]. The most common algorithms used in computing SRB estimates from satellite data are described by Pinker and Laszlo [1992] and Darnell et al. [1992].

[4] Global analyses, such as those of the National Center for Environmental Prediction (NCEP) [Kalnay et al., 1996], NASA Data Assimilation Office (NASA/DAO) [Schubert et al., 1993], and the European Centre for Medium-Range Weather Forecasts Reanalysis [Uppala et al., 1999] projects have provided climate researchers with alternative estimates of the Earth's SRB. Bony et al. [1997] used data from the NASA/DAO and NCEP reanalysis for the period 1987–1988 to compare SWR at the surface with satellite estimates in the Tropics (30°S–30°N). NCEP found the annual mean bias in SWR over the tropical ocean ranging from −10 to −30 W m−2, while NASA/DAO SWR biases ranged from −50 W m−2 in subsidence regions of the subtropics to −25 W m−2 in convective regions near the equator.

[5] Shinoda et al. [1998] proposed a simple empirical relationship (referred hereafter as SH98) that is valid over warm-pool oceanic regions to estimate the daily averaged SWR from the daily averaged outgoing longwave radiation (OLR) products obtained from the NOAA polar orbiting satellite. Shahi et al. [2010], proposed the use of high-temporal-resolution data of OLR to improve the diurnal sampling in order to match that of the buoy SWR in computing daily averaged estimates. They demonstrated a significant improvement in the SH98 by using a daytime average of OLR obtained from half-hourly geostationary satellite observations. The rationale behind using the daytime OLR as opposed to daily (24-h average OLR) is that the daily averaged SWR is affected primarily by the presence of daytime clouds, and therefore the algorithm performs better than that used in SH98. The empirical relationship developed by Shahi et al. [2010], however, was developed for a single-buoy location and is not valid over other locations, particularly over different latitude regions. The present study is an extension of the study by Shahi et al. [2010] to develop a location-independent algorithm valid over a wide Indian Ocean region. In the present study, a methodology has been developed using a large collocated data set of the spatially well-distributed buoy-measured SWR and the half-hourly estimates of OLR from geostationary satellite observations over the Indian Ocean region. The generalization of the empirical relationship has been achieved by normalizing the SWR values to nadir locations by making use of the solar zenith angle information. The SWR estimates using the present methodology have been tested with the independent data set. The daily and monthly averaged values of SWR are compared with the corresponding values from NCEP, International Satellite Cloud Climatology Project-Flux Data (ISCCP-FD), and Global Energy and Water Cycle Experiment-Surface Radiation Budget V3.0 (GEWEX-SRB) data sets for the year 2007 over these buoy locations. The present algorithm is computationally inexpensive and provides near-real-time and high-spatial-resolution accurate estimates of SWR over warm-pool regions of the Indian Ocean from geostationary satellite products of OLR.

2. Data

[6] The SWR on hourly and daily scales has been available from several moorings across the tropical Pacific and Atlantic Oceans since 1997 and more recently from the Indian Ocean [McPhaden et al., 1998, 2008; Bourlès et al., 2008]. The SWR is computed as the product of the albedo (0.055) and the downwelling SWR measured by the Research Moored Array for African-Asian-Australian Monsoon Analysis and Prediction (RAMA) buoy. The SWR is represented as the net shortwave flux into the ocean. Daily average SWR is computed as a 24 h average. Figure 1 shows the locations of the 13 RAMA buoys in the warm-pool region of the Indian Ocean. Data for the period of 2002–2009 have been used for the analysis. Table 1 shows the availability of the data during this period at different buoy locations.

Figure 1.

Locations of RAMA buoys in Indian Ocean overlaid on a sample Meteosat OLR coverage.

Table 1. RAMA Buoy Locations and SWR Data Availability
Buoy NumberBuoy LocationData Length (days)
15°S, 95°E651
215°N, 90°E273
312°N, 90°E478
44°N, 90°E414
51.5°N, 90°E95
60°N, 90°E1118
71.5°S, 90°E1256
81.5°N, 80.5°E387
90°N, 80.5°E529
101.5°S, 80.5°E273
114°S, 80.5°E22
128°S, 80.5°E343
138°S, 67°E350

[7] The OLR products have been obtained from the European Organisation for the Exploitation of Meteorological Satellites (EUMETSAT) for the Indian Ocean coverage of the Meteosat geosynchronous satellites, Meteosat-5 (subsatellite point at 57°E) up to December 2006 and Meteosat-7 (subsatellite point at 63°E) beyond January 2007. Meteosat-derived OLR is estimated from the half-hourly observations in the thermal infrared (10.5–12.5 μm) and water vapor absorption (5.7–7.1 μm) channels at a spatial resolution of 5 km, using the algorithm developed by Schmetz and Liu [1988] with coefficients for Meteosat-5/7 obtained from EUMETSAT. Based on radiative transfer simulations for a set of diverse atmospheric profiles, the accuracy of the Meteosat OLR product is shown to be ∼3 W m−2 [Schmetz and Liu, 1988]. Singh et al. [2007] also found that the accuracy of OLR products derived for Kalpana satellite using a narrowband-to-broadband conversion method is ∼2.5 W m−2. The validation of the OLR product derived from the GOES with broadband OLR observations from the Clouds and the Earth's Radiant Energy System (CERES) onboard polar orbiting satellites Tropical Rainfall Measuring Mission (TRMM) and EOS-Terra reveal that the accuracy of the narrowband-to-broadband flux computation is ∼7 W m−2 on a daily time scale [Ba et al., 2003].

[8] A collocated data set of daily averaged SWR (24 h average) and daytime averaged OLR (12 h average) is generated for each of the buoy locations. The daytime averaged OLR is obtained using the average of the half-hourly observations between sunrise and sunset (6 A.M. to 6 P.M.), as suggested by Shahi et al. [2010]. A training data set, using collocated observations from each of the buoys that represents a complete annual cycle, is created for the algorithm development. The remaining data set is treated as the independent testing data set. The number of collocated pairs obtained for training data set is 4462 and that for the testing data set is 1072.

[9] The SWR estimated using the algorithm developed in the present study has also been compared with the net SWR for the year 2007 available from following well-known sources, viz.: (1) Global Energy and Water-Cycle Experiment Surface Radiation Budget (GEWEX-SRB SRF) project of NASA World Climate Research Programme (WCRP), (2) International Satellite Cloud Climatology Project Surface Radiative Flux Data set (ISCCP-FD SRF), and (3) National Center for Environmental Prediction (NCEP) analysis.

[10] The GEWEX-SRB (version3) data set contains a 3 h global field of surface radiative shortwave flux parameters derived using the Pinker-Laszlo shortwave algorithm [Pinker and Laszlo, 1992]. These 3 h temporal resolution data are averaged to obtain the daily averaged net surface SWR, using a normalization correction to account for the discretization of the solar cycle. The spatial resolution of the GEWEX data set is 1°. The Pinker-Laszlo algorithm uses extensive data sets from various sources such as cloud parameters from ISCCP, atmospheric profiles from NASA Goddard Space Flight Center (GSFC) four-dimensional (4-D) data assimilation system, and ozone measurements from the Total Ozone Mapping Spectrometer (TOMS) archive.

[11] The ISCCP-FD SRF data set [Zhang et al., 2004] is computed using an advanced radiative transfer model from the Goddard Institute for Space Studies (GISS) with improved observations of surface, atmosphere, and clouds based on the ISCCP data set. The flux data set is available at 3 h intervals over the entire globe during the period July 1983 through December 2007 on a 280 km equal-area grid.

[12] The NCEP-National Center for Atmospheric Research (NCAR) reanalysis project uses a state-of-the-art analysis-forecast system to perform data assimilation using past data from 1948 to the present [Kalnay et al., 1996]. The downward shortwave radiation flux at the ocean surface obtained from the NCEP analysis is multiplied by the albedo of the ocean (= 0.055) to obtain the corresponding SWR values. Products from the NCEP analysis are available at the spatial resolution of 1.875°.

3. Algorithm Development

[13] The algorithm proposed by Shahi et al. [2010] demonstrated that the use of high-temporal-resolution observations of OLR from geostationary satellites could improve the SWR estimates compared with those obtained using low-temporal-resolution polar satellite observations using the SH98 algorithm and assessed the performance of these two algorithms. The improved diurnal sampling of satellite observations helped improve the accuracy of the SWR products by approximately 20%. This also resulted in an increase in the coefficient of variability (R2) between the daily averaged SWR and OLR from 0.51 to 0.63, when the daytime averaged OLR was used instead of the 24 h average OLR. However, the algorithm was developed using a single buoy located in the warm-pool region of the Indian Ocean that may not be valid at other latitudes. In this paper, the scope of the earlier algorithm [Shahi et al., 2010] has been extended to a wider region of the tropical Indian Ocean.

[14] SH98 showed that the coefficient of correlation between the daily averaged OLR and SWR is significant only in the warm-pool region (10°S–10°N, 60°E–180°E) and suggested that this is due to the fact that the variations in the cloudiness is captured by the OLR and is indicated by the variations of surface shortwave fluxes. We looked into various aspects of the SWR variations related to the region and season to understand the reasons for poor correlations away from the equatorial warm pool. We arrived at the conclusion that the reduction in the coefficient of correlation from the equator to the higher latitudes may be due to the fact that the solar zenith angle increases drastically toward the higher latitudes and also due to the fact that the length of the day is nearly uniform over an equatorial region whereas it varies at higher latitudes. This necessitates a correction factor in the net SWR values before they are correlated with the OLR values. In this paper, this has been achieved by using a normalized value for the SWR, which is suitably taken as the solar zenith angle corrected value given by

equation image

where β is the solar zenith angle at the time of the peak insolation (i.e., at the local noon).

[15] To test this logic, we have analyzed the scatterplots of the OLR versus SWR (referred to hereafter as SWR1) and that of the OLR versus SWR2 at two buoys located at different latitudes, i.e., at the equator and at 12°N. Figures 2a and 2b show the scatterplots of OLR with SWR1 as independent variables for the two buoy locations, respectively. The coefficient of variability (R2) at the equatorial buoy location is very high, R2 = 0.81, and it is small at higher latitudes, R2 = 0.61. The analysis has been repeated for normalized SWR, i.e., SWR2, as an independent variable and is shown in Figures 2c and 2d. The coefficient of variability remains nearly unchanged for the equatorial buoy, whereas it has improved significantly to R2 = 0.76 for the buoy located at 12°N. This improvement in the relationship is a result of the normalization of SWR with respect to the nadir incidence of the solar radiation. We have further experimented with the normalization procedure, and here we use the square of the secant of the solar zenith angle with the net SWR to obtain a new variable:

equation image
Figure 2.

Scatterplots for daily averaged OLR versus daily averaged net shortwave radiations: (a, b) SWR1, (c, d) SWR2, and (e, f) SWR3 for Buoy at 0°N, 90°E (Figures 2a, 2c, and 2e), and 12°N, 90°E (Figures 2b, 2d, and 2f).

[16] The relationship of the OLR to SWR3 as an independent variable has further improved the R2, which increased to 0.83 for the buoy located at 12°N (Figure 2e), whereas it remains almost constant for the equatorial buoy (Figure 2f). It is interesting to note that the differences in the slope of the best fit over two buoy locations has decreased from SWR1 (1.35 and 0.94) to SWR3 (1.47 and 1.29). The improvement in the relationship between OLR and SWR from SWR2 to SWR3 may be due to the fact that the additional factor of sec(β) is able to normalize the duration of the daylight while moving from an equatorial location to the higher latitudes. Since the duration of the daylight that exhibits latitudinal and seasonal variations can affect the daily averaged SWR, it is desirable to include a suitable parameter in the algorithm to account for the same. Similar analyses have been carried out for the remaining buoy locations. The values of R2 at different buoy locations for SWR1, SWR2, and SWR3 have been summarized in Figure 3. The x axis shows the buoy locations ordered in latitudes starting from equator to N/S latitudes. It may be clearly observed that the differences in R2 are small for SWR1, SWR2, and SWR3 over buoys that are near the equator (between 5°N and 5°S). As we move farther away from the equator, the R2 value decreases for SWR1, which restricted its use only over the narrow tropical warm-pool region, as concluded by Shinoda et al. [1998]. The R2 values show an increase in SWR2 as compared with SWR1 over higher latitudes, which further improved in SWR3. The values of R2 are very high (∼0.80) over all the buoy locations, indicating that the relationship between daily averaged OLR and normalized SWR (SWR3) can be used to estimate the net SWR using geostationary satellite estimates of OLR over larger tropical regions.

Figure 3.

Coefficient of variations (R2) of daily averaged OLR with SWR1, SWR2, and SWR3 for various buoy locations.

[17] The pooled training data set (N = 4462) was used to establish a relationship between different forms of SWR, i.e., SWR1, SWR2, and SWR3, with daytime averaged OLR estimated from geostationary satellite observations. It may be seen in Figure 4a that R2 between SWR1 and OLR for the pooled training data set is 0.73, which increases to 0.79 in the case of SWR2 and further increases slightly to 0.80 for SWR3. The improvement in R2 from SWR1 to SWR2 and SWR3 does not look very high, as seen in Figures 2b, 2d, and 2f for the buoy at 12°N, as the pooled data set contains the majority of the data points close to the equator and relatively smaller numbers over higher latitudes. On close inspection of Figure 4c, it is clear that the relationship between OLR and SWR3 is slightly nonlinear, and therefore a polynomial fit would best describe the relationship. Figure 4d is similar to Figure 4c but with a second-degree polynomial describing the best fit (referred to as SWR4). The value of R2 increases slightly to 0.81 as compared with 0.80 as in the linear fit for SWR3. The final equation used for the estimation is given by the polynomial equation between SWR4 and OLR:

equation image
Figure 4.

Scatterplots along with the best fit equations for different SWR estimates: (a) SWR1, (b) SWR2, (c) SWR3, and (d) SWR4, with daily averaged OLR for pooled data set.

[18] Since SWR4 is equal to the product [(SWR)sec2(β)], we have

equation image

or, finally, in general form,

equation image

[19] This equation is used for estimating the SWR from the daytime averaged OLR value using the solar zenith angle at the time of peak insolation (β) computed from the location and day of the observation. This algorithm (SWR4) to estimate the net SWR is referred to hereafter as the Space Applications Centre (SAC) algorithm.

4. Assessment of SWR Algorithms

[20] The assessment of the SWR estimates by different algorithms, i.e., SWR1, SWR2, SWR3, and SWR4, is carried out with both the training and the testing data sets. Table 2 shows the detailed statistics of the SWR estimates compared with those measured by the buoys. For comparison purposes we also show the statistics for the SWR estimates generated using the SH98 algorithm. It may be noted that the statistics obtained from the training as well as testing data sets are consistent. The mean SWR values estimated from different algorithms are close to the observed mean, 208.3 and 208.5 W m−2 in the training and the testing data sets, respectively, except for SH98, which is 222.3 and 221.3 W m−2. The standard deviation (SD) of the estimated SWR from SWR4 for the training and the testing data sets with values of 61.2 and 61.4 W m−2, respectively, are closest to those of the observed SWR, 66.8 and 67.6 W m−2. The corresponding SDs for SH98 are significantly smaller, with values of 44.5 and 45.1 W m−2. The SD in SWR4 also shows improvement over SWR1, which has corresponding values of 57.1 and 57.7 W m−2. The root-mean-square error (RMSE) computed for different SWR estimates from the buoy observations shows that the SAC algorithm (SWR4) is the best, with a RMSE of 28.9 W m−2 in the training data set and 28.3 W m−2 in the testing data set, with the corresponding values for SWR1 as 34.6 and 33.8 W m−2. It may be noted that the RMSE values for SH98 are 39.4 and 38.5 W m−2, which are consistent with the daily RMSE value of 40.1 W m−2 obtained by Shinoda et al. [1998] from Improved Meteorology sensor system (IMET) mooring data by Weller and Anderson [1996]. The algorithm by Shinoda et al. [1998] was an improvement over Reed's formula [Reed, 1977], which provided SWR estimates with a RMSE of 45.8 W m−2. In terms of the RMSE, the improvement in the SWR estimates from the present SAC algorithm compared with SWR1 [Shahi et al., 2010] is about 16%, and 26% from that of SH98.

Table 2. Statistics of the SWR Estimates (W m−2) Using Different Algorithms
SWR ProductTraining Data Set (N = 4462)Testing Data Set (N = 1072)
Observed SWR208.366.8- 208.567.6--

[21] Figure 5a shows the histogram of the absolute error in the SWR estimates obtained using different equations. It may be noted that SWR1 has the smallest frequency toward lower absolute errors and the highest frequency toward the higher absolute errors, whereas it is the opposite in the case of SWR4. It implies that SWR4 provides higher occurrences of smaller absolute errors and lower occurrences of higher absolute errors, thereby proving its superiority over the others. The frequency of absolute errors smaller than 10 W m−2 is about 30% for SWR4, but is merely ∼20% for SWR1. Further assessment of the SWR estimates is carried out only for SWR4 (i.e., SAC algorithm). Figures 5b and 5c show the scatterplots between the SWR estimated using the SAC algorithm (SAC-SWR) and the buoy-measured SWR (Buoy-SWR) for the training and the testing data sets, respectively. Most of the data pairs lie along the one-to-one line, with R2 at 0.81 for the training data set and at 0.83 for the testing data set. The comparison analysis suggests that the SAC algorithm given by equation (3) provides the best estimate of SWR.

Figure 5.

(a) Statistics of SWR estimates obtained using different methods, and scatterplots of actual and estimated SWR using the SAC algorithm for (b) training data set, and (c) testing data set.

5. Comparison and Validation

[22] The SWR estimates from the SAC algorithm (SAC-SWR) have been validated with buoy observations for 2007 and compared with the similar products available from well-known sources, GEWEX, ISCCP, and NCEP analyses. Figures 6a6d show the scatterplots for the SWR estimates from SAC, GEWEX, ISCCP, and NCEP, respectively. These figures show that the SAC-SWR is in the best agreement with the Buoy-SWR (R2 = 0.80) followed by GEWEX-SWR (R2 = 0.76) and ISCCP (R2 = 0.64). The NCEP product of SWR has a poor match with the buoy-observed SWR, with values of R2 at 0.10. Figure 7a shows the histogram of SWR values from buoy measurements and its comparison with those from SAC, GEWEX, ISCCP, and NCEP estimates of SWR products during 2007 over Indian Ocean buoy locations. The frequency of occurrence of SWR values in the SAC product matches very well with the observed SWR, followed closely by the GEWEX and ISCCP SWR products. The NCEP SWR products show only a limited range of values (100–300 W m−2), thus failing to match the frequency distribution with the observations. Further, to examine the ability of different products to provide accurate estimates of the SWR products for different sky conditions, the RMSE values are computed for different ranges of observed SWR values. The low ranges of SWR values depict cloudy conditions, whereas the high ranges of SWR values depict the clear-sky conditions. It may be observed from Figure 7b that the RMSE for SAC-SWR products is lowest for the entire range of SWR values, indicating that the accuracy of the SAC product is best under all sky conditions. The accuracy of the GEWEX SWR is close to that of the SAC-SWR for the entire range of SWR except for the 150–200 W m−2 range, where the SAC product shows an improvement of about 45% over GEWEX SWR. The RMSE in the ISCCP is higher than GEWEX and very high for NCEP products, particularly for the lower values of the SWR products, i.e., under cloudy conditions. This indicates that the SAC product, closely followed by the GEWEX product, is capable of providing accurate estimates of SWR even under cloudy-sky conditions, whereas the NCEP product fails to provide better estimates of SWR under cloudy conditions.

Figure 6.

Comparison of various SWR estimates with buoy-measured SWR for the year 2007: (a) SAC, (b) GEWEX, (c) ISCCP, and (d) NCEP.

Figure 7.

Statistics of different SWR estimates for 2007: (a) histogram of SWR and (b) RMSEs for different bins of actual buoy-measured SWR values.

[23] The detailed statistics of the comparison of different SWR products are provided in Table 3. The RMSE of 27.3 W m−2 for the SAC product is the smallest, followed by 32.7 W m−2 for the GEWEX and 37.5 W m−2 for ISCCP. The improvement in the accuracy for the SAC-SWR products over that of the GEWEX product is about 17%. The NCEP product has very poor accuracy of the SWR products, with RMSE of 59.6 W m−2. The mean biases in all these products are very low, with values of 0.1, −2.1, −8.5, and 9.0 W m−2 for SAC, GEWEX, ISCCP, and NCEP, respectively. It is interesting to note that the range of the SWR values, shown by the minimum and maximum of the SWR values in the entire data set, is very well captured by the SAC product, 7.1–310.9 W m−2, as compared with 4.7–311.1 W m−2 in the buoy-measured SWR. The GEWEX product shows a higher value of the maximum SWR, 326.0 W m−2 as compared with 311.1 W m−2 in the buoy observation. The NCEP product has the poorest range of SWR values, 95.3–289.2 W m−2, far from the observed range.

Table 3. Statistics of the Different SWR Products With Respect to the Buoy-Observed SWR for 2007 (N = 1720)
 SWR Products
RMSE (W m−2)-27.332.737.559.6
Bias (W m−2)-0.1−2.1−8.59.0
Minimum SWR (W m−2)
Maximum SWR (W m−2)311.1310.9326.0317.4289.2

[24] To examine the seasonal behavior of the SWR products we have computed the monthly statistics of different SWR products for the year 2007. Figure 8a shows that the accuracy of the SAC-SWR product is the best and is consistent throughout the year, with RMSE values close to 25–30 W m−2 for all the months. The accuracy of GEWEX SWR product follows the SAC-SWR products closely, except for the winter season, when the RMSE in GEWEX SWR is as high as 40 W m−2. The ISCCP SWR has higher RMSE compared to both the SAC and GEWEX products particularly during January–June. The NCEP product shows very high RMSE throughout the year. The monthly estimates of the bias in SWR products, shown in Figure 8b, indicate that the bias is smallest in the SAC-SWR followed by that of the GEWEX and ISCCP SWR. There is a large variation in the bias for NCEP over different months.

Figure 8.

Monthly statistics of different SWR estimates (W m−2) for 2007 computed from buoy observations: (a) RMSE and (b) bias.

[25] In order to show the day-to-day variations, we analyzed the 7 day running mean of different SWR products for the year 2007 at a buoy location (4°N, 90°E) that has complete data over the whole year. From Figure 9, it is clear that the day-to-day variations of SWR are very well captured by the SAC, GEWEX, and ISCCP products. The NCEP product shows very small day-to-day variations in SWR as compared with the observed SWR. These figures clearly show that the SAC algorithm provides accurate estimates of SWR throughout the year for all the seasons. Another important aspect of comparison of 7 day running-mean SWR products is that the estimate of the standard error on weekly time scales is much smaller than that on daily time scales, with values of 11.8, 13.4, 15.3, and 32.8 W m−2 for SAC, GEWEX, ISCCP, and NCEP products, respectively. This shows that high-spatiotemporal-resolution SWR products from the present algorithm could be utilized to improve the accuracy of the product by averaging at various spatial and temporal scales.

Figure 9.

Time series of the 7-day running mean of SWR at a buoy location (4°N, 90°E) from different products and their comparison with the buoy-observed SWR.

[26] One of the prime reasons for the poorer agreement between ISCCP-FD and buoy SWR may be due to its coarser resolution (280 km) and the location of the buoy over the convectively active region of the Bay of Bengal. To explain this we have computed the standard deviation of SWR using high-spatial-resolution SAC-SWR (5 km resolution) at different spatial scales, 1°, 2°, and 2.5°, which correspond to the spatial scales of the GEWEX, NCEP, and ISCCP, respectively. Figure 10 shows that the standard deviation of SWR increases with the increase of spatial resolution. The standard deviation of SWR over most of the Indian Ocean region is about 10–20 W m−2 at GEWEX product resolution, whereas it is about 15–30 W m−2 for NCEP and 20–35 W m−2 for ISCCP product resolutions. The standard deviations are highest over the convectively active region of the Bay of Bengal. These are the regions where most of the buoys are located. Although GEWEX and ISCCP-FD products of SWR are based on cloud information derived from the satellite, the poorer agreement of ISCCP SWR with in situ buoy observations may primarily be attributed to the high spatial variability of cloudiness at its resolution. Besides, the GEWEX algorithm is known to be technically superior.

Figure 10.

Standard deviation (W m−2) of the SWR computed from high-spatial-resolution (5 km) Meteosat SWR at different spatial scales: (a) 1°, (b) 2°, and (c) 2.5°, corresponding to the GEWEX, NCEP, and ISCCP, respectively, during June–July 2007.

[27] The SWR products for the Indian Ocean region have been generated from daily averaged OLR products computed from Meteosat's Indian Ocean coverage. The SAC will soon host the daily average SWR products computed using the present algorithm with OLR products drawn from Kalpana and Meteosat satellites. Figure 11 shows a sample image of the monthly average SWR products for January, April, July, and October of 2007. Although the GEWEX products for SWR have accuracies close to those of the SAC products, the former is available only up to 2007, whereas the latter will be routinely available for various applications, such as an input to the ocean circulation models and climate studies. The monthly maps show that the patterns of the SWR products match very well in the SAC and GEWEX products, whereas they are far from realistic in the ISCCP and NCEP products. The advantage of the SAC products of SWR, besides being more accurate than the GEWEX products, is that they are available in near real time from the simple algorithm using satellite data. The other major advantage of the SAC product is the high spatial resolution of 5 km at the subsatellite point over the equator, as opposed to 1°, 2.8°, and 1.875° from GEWEX, ISCCP, and NCEP, respectively, thus providing higher spatial information of SWR. The SAC-SWR products have shown the best accuracy among the presently available SWR products under different sky conditions and various seasons throughout the year. The only weakness of the SAC algorithm is that the products are valid only over warm-pool oceanic regions and have been validated only over the Indian Ocean region. The algorithm needs to be tested over other parts of the global oceans for its validity. In contrast the GEWEX, ISCCP, and NCEP products are available over the entire globe.

Figure 11.

Monthly mean maps of SWR (W m−2) over the tropical Indian Ocean for January, April, July, and October from different data sets: (a) SAC, (b) GEWEX, (c) ISCCP, and (d) NCEP.

6. Conclusions

[28] Net shortwave radiation (SWR) is the most important parameter for radiation budget studies as well as for ocean circulation models, provided accurate products are available under both clear- and cloudy-sky conditions. Geostationary satellites provide an ideal platform to estimate near-real-time SWR at high temporal and spatial resolutions. The algorithm proposed in this paper is an extension of the work of Shahi et al. [2010] to provide the estimates of daily SWR over the entire warm-pool regions of the Indian Ocean. This is achieved by using normalized SWR values at nadir solar zenith angle to establish an empirical relation in the form of a second-order polynomial function with the daytime averaged OLR values obtained from geostationary satellite observations. A pooled data set of collocated pairs of satellite-derived OLR and buoy-measured SWR is generated from 13 buoy locations over warm-pool regions of the Indian Ocean for the period 2002–2009. The entire data set is divided into a training data set and a testing data set. The training data (N = 4462) contains those buoy measurements that cover approximately the entire annual cycle. The remaining observations (N = 1072) are used as the independent testing data set. The RMSE of the SWR estimates obtained using the SAC algorithm is 28.9 W m−2 in the training data set, whereas it is 28.3 w/m−2 in the independent testing data set. The SAC algorithm developed for SWR estimations shows an improvement of about 16% over the algorithm proposed by Shahi et al. [2010] and about 27% improvement over that of Shinoda et al. [1998]. The SAC product of SWR shows a close match in the mean and standard deviation of the SWR with that obtained from the buoy- observed SWR.

[29] The validation of SAC-SWR products with buoy-observed SWR for the year 2007 and its comparison with the GEWEX, ISCCP, and NCEP products suggest that the SAC products have the best accuracy. The RMSE of the SAC-SWR product is 27.3 W m−2 as compared with 32.7, 37.5, and 59.6 W m−2 obtained from GEWEX-SRB, ISCCP-FD, and NCEP, respectively. The SAC algorithm shows an overall improvement of 17% over the GEWEX product, which is the best among the presently available sources for SWR. The SAC product shows better accuracy for the entire range of SWR values, indicating that this product is equally good for the entire range of sky conditions. Interestingly, the SAC-SWR product outperforms ISCCP-FD and NCEP SWR products and is slightly better than GEWEX under the deep cloudy conditions, represented by very low SWR values. The SAC-SWR product has also been shown to be consistently better than other available products for different seasons.

[30] The algorithm developed in this paper has an advantage over other products because of its simplicity for operational implementation and availability in near real time from the geostationary satellite observations. The high-spatial-resolution products at 5 km will be available through MOSDAC using Indian Ocean coverage of the Meteosat satellite and at 10 km resolution from Indian geostationary satellites. These are expected to provide improved information as compared with the state-of-the-art GEWEX products at 1° resolution (available up to 2007). The SWR product generated by the SAC algorithm is expected to be useful for studies on air-sea interaction, diurnal cycle of SSTs, ocean general circulation, and climate modeling. In this paper, the algorithm has been developed and tested for the Indian Ocean region. Future work will address applications of the present algorithm over other parts of the warm-pool regions of the global oceans. The validity of present algorithm is, however, limited to only the warm-pool oceanic regions, where variations of surface insolation are primarily due to the variations in cloudiness, a manifestation of deep convection.


[31] The authors would like to especially thank R. R. Navalgund, Director, Space Applications Centre, and J. S. Parihar, Deputy Director, EPSA/SAC, for their constant encouragement. The authors would like to acknowledge EUMETSAT for Meteosat data, PMEL/NOAA for RAMA buoy data, NASA Goddard Institute for Space Studies for providing ISCCP FD SRF data sets (, NASA Langley Research Center Atmospheric Science Data Center for GEWEX SRF SRB data (, and NCEP/NCAR for NCEP reanalysis data. Authors would like to acknowledge J. Vialard and P. Kumar for discussions related to ISCCP data. We thank anonymous reviewers for the valuable suggestions to improve the manuscript.