How good are ocean buoy observations of radiative fluxes?



[1] Satellite-derived surface shortwave radiation (SWR) has been extensively evaluated against ground observations over land but to a lesser extent over oceans that cover larger portion of the Earth surface. In this study new surface radiative flux estimates from the Moderate Resolution Imaging Spectro-radiometer (MODIS) are evaluated against buoy measurements of downwelling SWR in the tropical oceans. As a benchmark for achievable accuracies from satellites, similar evaluation is performed over land against the Baseline Surface Radiation Network (BSRN) observations, believed to be of highest available quality. Comparable accuracy of the new satellite SWR data is found over land and over oceans.

1. Introduction

[2] Many attempts have been made to estimate surface shortwave radiation (SWR) from satellite-observed radiances and atmospheric and surface variables, on both regional and global scales [Pinker and Ewing, 1985; Ramanathan, 1986; Raschke et al., 1991; Pinker and Laszlo, 1992; Zhang et al., 2004; Gupta et al., 1999; Mueller et al., 2004; Rigollier et al., 2004]. Most methods are implemented with observations from geostationary satellites that capture the diurnal variability of clouds but have coarser spectral and spatial resolution than low temporal resolution polar orbiting satellites. Observations from the International Satellite Cloud Climatology Project (ISCCP) [Rossow and Schiffer, 1999] (version D1) are widely used for estimating surface radiative fluxes at global scale. They are provided at a nominal resolution of 2.5° at 3-hourly time intervals and use sampled pixels at 30 km spatial resolution. The inference schemes that utilize the ISCCP data require auxiliary information from independent sources that are not consistent in space and time with the satellites observations used to derive the SWR.

[3] In this study we use three years (2003–2005) of global scale estimates of SWR and Photosynthetically Active Radiation (PAR) (0.4–0.7 μm) at 1° × 1° grid cells as derived from MODIS observations from Terra and Aqua satellites (H. Wang and R. T. Pinker, Shortwave radiative fluxes from MODIS: Model development and implementation, submitted to Journal of Geophysical Research, 2009). They are evaluated against a comprehensive set of buoy observations over the oceans (SWR only) and ground observations over land (SWR and PAR).

2. Satellite Estimates of Surface SWR Fluxes

2.1. Satellite Observations

[4] The MODIS instrument is a state-of-the-art sensor with 36 spectral bands with an onboard calibration of both solar and infrared bands. The wide spectral range (0.41–14.24 μm), frequent global coverage (one to two days revisit), and high spatial resolution (250 m for two bands, 500 m for five bands and 1000 m for 29 bands), permit global monitoring of atmospheric profiles, column water vapor amount, aerosol properties, and clouds, at higher accuracy and consistency than previous Earth Observation Imagers [King et al., 1992]. The largest uncertainties in satellite estimates of surface SWR are due to inadequate information on cloud properties such as cloud fraction, cloud optical depth, and cloud thermodynamic phase. Hence, improvements in estimating SWR greatly depend on better detection of cloud properties.

[5] The MODIS cloud mask is applied globally at single pixel resolution; it uses 17 spectral bands (visible at 250 m and infrared at 1000 m resolution) to improve cloud detection and to mitigate past difficulties experienced by sensors with coarser spatial resolution or with fewer spectral bands [Ackerman et al., 1998]. Three bands in the infrared window regions of 8–11 μm and 11–12 μm are used to differentiate cloud phase. Spectral bands, including window regions in the visible and near-infrared, as well as 1.6, 2.1 μm shortwave infrared bands and 3.7 μm mid-wave infrared bands are used for the retrieval of cloud optical depth and cloud particle effective radius. Seven channels are designed to measure aerosol properties over oceans and three bands over land.

[6] We use Level-3 MODIS Atmosphere Daily Global Product (MOD08_D3, MYD08_D3) model input parameters that include: Optical Depth for Land and Ocean, Cloud Top Pressure Day, Cloud Optical Thickness Liquid, Cloud Optical Thickness Ice, Cloud Effective Radius Liquid, Cloud Effective Radius Ice, Cloud Effective Radius Undetermined, Cloud Fraction Liquid, Cloud Fraction Ice, Cloud Fraction Undetermined, Cloud Optical Thickness Undetermined, Total Ozone, Atmospheric Water Vapor. Clouds with undetermined phase are treated as water clouds in the computation of radiative fluxes.

[7] Since the MODIS atmospheric water vapor is retrieved only when at least 9 out of 25 Field of Views are cloud free, precipitable water from the National Centers for Environmental Prediction Reanalysis Data [Kistler et al., 2001] is used for conditions with large cloud fraction. Missing aerosol optical depths under cloudy conditions and over arid areas are filled with information from the Multi-angle Imaging Spectro-Radiometer Component Global Aerosol Product [Martonchik et al., 1998, 2002] and monthly global aerosol climatology, derived from ‘typical-year’ aerosol transport model results [Kahn et al., 2001]. Spectral surface albedo on 1° × 1° grid cells is taken from the Filled Land Surface Albedo Product provided by the MODIS project [Moody et al., 2005]. Monthly mean sea ice extent on 1° × 1° grids from the Special Sensor Microwave/Imager is provided by the NOAA National Climate Data Center [Ferraro et al., 1996]. Daily snow cover data are from MODIS/Terra Snow Cover Daily Global 0.25 Degree Geographic Climate-Modeling Grid [Hall et al., 2002; Hall and Riggs, 2007]. Surface albedo of ice over oceans in visible and near-infrared is set to 0.77 and 0.33, respectively. The surface elevation on 1° × 1° grid is calculated from the global Digital Elevation Model adopted in the Penn State University/National Center for Atmospheric Research meso-scale model (known as MM5). The daily average is computed by assuming that MODIS observations from Terra at 10:30 local time and from Aqua at 13:30 local time represent the atmospheric conditions from sunrise to local noon and from local noon to sunset, respectively. The diurnal variation of incident fluxes will be dictated only by the incident solar flux at top of the atmosphere which is determined by the cosine of the solar zenith angle. The daily integration of radiative fluxes is reduced to the integration of the cosine of the solar zenith angle (Wang and Pinker, submitted paper, 2009).

2.2. Ground Observations

2.2.1. Ground Observations Over Oceans

[8] In-situ measurements from the Pilot Research Moored Array in the Tropical Atlantic (PIRATA) moorings in the tropical Atlantic [Bourlès et al., 2008], and the Tropical Atmosphere Ocean/Triangle Trans-Ocean Buoy Network (TAO/TRITON) moorings in the tropical Pacific Ocean [McPhaden et al., 1998] are used for comparisons with the satellite estimates. Downwelling SWR is detected by the Eppley Laboratory pyranometers that have nominal resolution 0.4 W m−2 and relative accuracy of ±2% in the 0–1600 W m−2 interval in laboratory conditions [Cronin and McPhaden, 1997].

2.2.2. Ground Observations Over Land

[9] Over land we used eighteen BSRN sites [Ohmura et al., 1998] from the United States, Europe and Africa. Since 1992 the BSRN stations provide data for the calibration of satellite-based measurements of radiative fluxes and other climate applications such as validation of radiative fluxes from numerical models. As of June 2008 there are 43 stations contributing to the BSRN. The BSRN stations over the United States (previously known as SURFRAD [Augustine et al., 2005]) also use independent instruments to measure PAR. These observations were used to evaluate the MODIS based PAR estimates.

3. Methodology

[10] A forward inference scheme was developed to derive surface, Top of the Atmosphere (TOA), and atmospheric spectral shortwave radiative fluxes at global scale using satellite based information (Wang and Pinker, submitted paper, 2009). The model takes into account all atmospheric constituents and a multi-layered structure of the atmosphere that accounts for surface elevation effects and the vertical distribution of radiative fluxes. Spectral fluxes such as PAR and near-infrared radiation can also be estimated. The model is implemented with products from the Moderate Resolution Imaging Spectro-radiometer (MODIS) sensor both on Aqua and Terra at 1° × 1° grid cells on global scale, for a three year period (2003–2005). The new model (UMD/SRB_MODIS) takes into account both water and ice clouds, the variation of cloud particle effective radius, and their corresponding optical properties by adopting a new cloud parameterization scheme. In the near-infrared band three spectral intervals are used so that characterization of spectral change of water absorption and cloud optical properties could be incorporated. The parameterization of water vapor absorption uses a scheme based on an extensive spectroscopic database and incorporates a water vapor continuum model [Edwards and Slingo, 1996]. The aerosol models used include several aerosol types following Hess et al. [1998]. In addition, the layered structure of the new inference scheme facilitates the treatment of surface elevation effects. The forward approach of the new inference scheme avoids intermediate steps such as narrow to broadband conversions and angular corrections needed if the original satellite radiance were to be used; instead, we utilize geophysical parameters derived independently from MODIS.

[11] Information on Photosynthetically Active Radiation (PAR) (0.4–0.7 μm), is needed for applications dealing with biogeochemical processes such as net primary productivity and ecosystem modeling [Running et al., 1999; Platt, 1986; Prentice et al., 1992, Nemani et al., 2003]. When implemented with MODIS products the UMD/SRB_MODIS model can estimate SWR fluxes in several spectral intervals, one of which is the PAR region.

[12] Ground measurements of PAR are very limited. There are far fewer ground stations measuring PAR than total SWR. Satellite observations are the only source of consistent information on PAR. MODIS observations have been previously applied to estimate PAR using simplified radiative transfer models over land and over oceans [Van Laake and Sánchez-Azofeifa, 2004]. Those simplified models are limited in their capability to deal with multiple scattering by clouds and aerosols and lack transferability to global scale.

4. Results

[13] To illustrate products derived from MODIS observations we show the global distribution of the time-mean surface SWR on a 1° × 1° grid averaged over 2003–2005 (Figure 1a), while the illustration for the PAR (Figure 1b) focuses on the July mean (averaged over the three Julys) that is the peak growing season for the northern Hemisphere (where the ground truth comparison is done). These three year data sets have been evaluated against the best available observations both over tropical oceans and over land.

Figure 1.

(a) Annual mean downwelling SWR derived from MODIS observations at 1° × 1° grid cells as averaged over a three year period (2003–2005). (b) July mean surface PAR derived from MODIS observations at 1° × 1° grid cells as averaged over a three year period (2003–2005).

[14] In Figure 2 we show results for daily mean surface SWR fluxes against PIRATA and TAO/TRITON buoy observations during January 2003 to December 2005. Small fraction of data is eliminated from the comparisons (1.1% for PIRATA and 1.3% for TAO/TRITION). The removal of extreme outliers was done by eliminating cases where the difference between estimated and observed flux was larger than three times the standard deviation of the differences. After averaging for each buoy network and averaging in time over the three years of records the bias of the time-space mean of daily DSWR is 3 and 1 W/m2 for the tropical Atlantic and the tropical Pacific, respectively. The difference between instantaneous daily satellite and buoy data is better characterized by corresponding RMSE that is 25 and 36 W/m2, respectively. A summary is presented in Table 1. If monthly average data is used instead of daily average data, the bias remains low at 2 and −1 W/m2 for the Atlantic and Pacific, correspondingly, while the RMSE is reduced to 13 and 11 W/m2 which is about 5% from the observed mean values, as summarized in Table 2.

Figure 2.

Evaluation of daily mean downwelling SWR estimated by UMD/SRB_MODIS (2003–2005) against PIRATA and TAO/TRITON buoy observations. PIRATA and TAO/TRITON data are available since 1997/1979, respectively. Cases eliminated: 1.1% (PIRATA); 1.3% (TAO/TRITON).

Table 1. Summary of Evaluation Results for Daily Mean Surface SWR and PAR Against Tropical Ocean Buoy and BSRN Observations for January 2003–December 2005
RegionMean of Observations (W/m2)Correlation CoefficientBias (%) (W/m2)RMSE (%) (W/m2)Number of Observations
Atlantic2320.893 (1)25 (11)8478
Pacific2120.861 (0)36 (17)16317
Land/SWR2010.98−3 (2)21 (11)11439
Land/PAR810.960 (0)11 (14)4391
Table 2. Summary of Evaluation Results for Monthly Mean Surface SWR and PAR Against Tropical Ocean Buoy and BSRN Observations for January 2003–December 2005
RegionMean of Observations (W/m2)Correlation CoefficientBias (%) (W/m2)RMSE (%) (W/m2)Number of Observations
PIRATA2320.902 (1)13 (5)295
TAO/TRITON2190.94−1 (0)11 (5)776
Land/SWR1980.99−3 (2)10 (5)381
Land/PAR810.990 (0)5 (6)146

[15] In Figure 3 we present evaluation of daily mean surface SWR fluxes estimated with the UMD/SRB_MODIS model against measurements at 18 BSRN stations over land for the same time period. For daily average data the bias is −3 W/m2 while the RMSE is 21 W/m2. Both biases and RMSE are similar for the ocean and land comparisons for downwelling SWR. Similar good comparison is found for daily mean surface PAR fluxes against the SURFRAD (BSRN) measurements over the United States for January, 2003 to December, 2005 (Figure 4). No data were excluded from the analysis. The PAR estimates agree well with ground measurements, with a correlation coefficient of 0.96, a bias of 0 W/m2 and a RMSE of 11 W/m2 (14%) for the daily time scale and coefficient of 0.99, a bias of 0 Wm2 and a RMSE of 5 W/m2 (6%) on a monthly time scale.

Figure 3.

Evaluation of daily mean downwelling SWR flux estimated by UMD/SRB_MODIS against BSRN measurements over land (January 2003–December 2005). Cases eliminated: 1.6%.

Figure 4.

Evaluation of daily mean surface downwelling PAR fluxes estimated with the UMD/SRB_MODIS model against the SURFRAD (BSRN) measurements over the United States for January 2003 to December 2005.

5. Summary

[16] The need for improving air-sea fluxes in atmospheric re-analyses products, which are widely used to study climatic changes in the ocean and the atmosphere, has been recognized (e.g., US-CLIVAR Program). A major component of these fluxes are radioactive. The advantages of using reanalysis fields in climate research are that they are readily available and that they provide basin-wide coverage, but their quality must always be tested by independent means. Radiative fluxes are particularly difficult to obtain from global atmospheric analysis models, since their cloud schemes often are unrealistic, especially when relatively high temporal and spatial resolution is desired (days and 50 to 100 km scales). Quality assessment of the reanalysis fluxes involves comparisons with in situ and satellite observations. In situ observations are sparse, however, leaving satellite observations as the only quality-control tool of the fluxes on a basin scale. Satellite observations provide nearly global coverage and routine sampling. The objective of this study was to evaluate comprehensively satellite based estimates of radiative fluxes over the oceans using the most advanced buoy observational systems. It was demonstrated the there are no significant differences between the validation results over oceans when compared to land. It was also shown that PAR results over land are even in better agreement with ground observations than the SWR fluxes. Therefore, it is reasonable to assume that similar results can be obtained over oceans where PAR measurements are as yet not available.


[17] We thank NFS for providing support for this work under grant ATM0631685 to the University of Maryland. The work also benefited from support under NASA grant NNG05GB35G to the University of Maryland. Support for SG is provided by the NASA OWVST. We are grateful to the NASA Goddard Earth Sciences Data and Information Services Center (GES DISC) Giovanni web system for the support in providing the MODIS data and to the various teams that produced information used in this study. We acknowledge the TAO Project Office of NOAA/PMEL for providing data from the Pilot Research Moored Array in the Tropical Atlantic (PIRATA) and from the Tropical Atmosphere Ocean/Triangle Trans-Ocean Buoy Network (TAO/TRITON) in the Tropical Pacific Ocean that were used in this study. We thank the Baseline Surface Radiation Network (BSRN) for providing the ground observations over land. We also thank the anonymous reviewers for their sincere effort to improve this contribution.