Advances in Space Research 33 (2004) 1120 1124 www.elsevier.com/locate/asr The HIRS outgoing longwave radiation product from hybrid polar and geosynchronous satellite observations Hai-Tien Lee a, *, Andrew Heidinger b, Arnold Gruber c, Robert G. Ellingson d a Cooperative Institute for Climate Studies (CICS)/ESSIC-NOAA, University of Maryland, 224 Computer and Space Science Bldg, Rm 4115, College Park, MD 20742, USA b NOAA/NESDIS, CIMSS, University of Wisconsin, 1225 W. Dayton St., Madison, WI 53706, USA c NOAA/NESDIS, CICS/ESSIC, University of Maryland, 224 CSS Bldg Rm. 2207, College Park, MD 20742, USA d Department of Meteorology, Florida State University, 404 Love Bldg, Tallahassee, FL 32306, USA Received 2 December 2002; received in revised form 28 February 2003; accepted 14 March 2003 Abstract Traditionally, earth radiation budget studies use polar orbiting satellites to estimate the global distribution of outgoing longwave radiation (OLR). However, the two-pass per day per satellite orbit significantly limits the accuracy of the daily mean OLR due to diurnal variations. Geosynchronous satellites, on the other hand, have high temporal resolution but have less spatial coverage. To increase the accuracy of daily mean of the High Resolution Infrared Radiation Sounder (HIRS) OLR, one could blend the OLR from polar orbiters with the diurnal information provided by the geosynchronous satellite observations. We demonstrate a scheme to generate a hybrid OLR product by integrating the HIRS observations on board of the US National Oceanic and Atmospheric Administration polar satellites and the Geostationary Operational Environment Satellite Imager observations. This kind of product can be utilized to avoid the day/night bias that may appear when broadband longwave is derived from spectral subtraction of a SW signal from an unfiltered total channel with errors in calibration of the short wave part of the spectrum. The paper investigated the errors in the HIRS daily mean OLR at averaging domains of various temporal and spatial scales. Ó 2003 COSPAR. Published by Elsevier Ltd. All rights reserved. 1. Introduction The outgoing longwave radiation (OLR) is a crucial parameter for studying many areas in the atmospheric science. The OLR has continued being observed or estimated from various satellite instruments and algorithms since the very early era of meteorological satellites. Besides the broadband instruments that are dedicated for observing the OLR, e.g., Earth Radiation Budget Experiment (ERBE), Scanner Radiometer for Radiation Budget (ScaRab), and Cloud and EarthÕs Radiant Energy System (CERES), there are many algorithms that estimate OLR by converting the narrowband radiance observations into broadband flux quantities. Schmetz and Liu (1998)developed an OLR * Corresponding author. Tel.: +1-301-405-0494; fax: +1-301-314-1876. E-mail address: lee@atmos.umd.edu (H.-T. Lee). algorithm that uses window and water vapor channels observations from Meteosat 2. Cheruy et al. (1991) also developed an OLR algorithm for Meteosat using window and water vapor channels radiances. Minnis et al. (1991) developed an algorithm to estimate OLR using the Geostationary Operational Environment Satellite (GOES) Imager window channel with additional water vapor information from analysis. US National Oceanic and Atmospheric Administration (NOAA) has operationally maintained an OLR estimate derived primarily from the Advanced Very High Resolution Radiometer (AVHRR) observations on board of the NOAA polar orbiters since 1974 (Gruber and Krueger, 1984). The OLR algorithm developed for the High Resolution Infrared Radiation Sounder (HIRS) by Ellingson et al. (1989) was implemented by NOAA NESDIS in experimental mode since 1992 and has become operational since September 1998 (John Sapper, personal communication). The HIRS 0273-1177/$30 Ó 2003 COSPAR. Published by Elsevier Ltd. All rights reserved. doi:10.1016/s0273-1177(03)00750-6
H.-T. Lee et al. / Advances in Space Research 33 (2004) 1120 1124 1121 OLR algorithm consists of weighted sum of four HIRS channel radiances that are sensitive mainly to the near surface and/or cloud top temperature, lower tropospheric water vapor content, and the upper tropospheric temperature and water vapor content. NOAA OLR products are regularly used in research and operational diagnostic purposes in, e.g., calibration of NOAA National Centers for Environmental Prediction NCEP medium range forecast model, NOAA Climate Prediction CenterÕs (CPC) Climate Diagnostic Bulletin for ENSO monitoring and forecast, and CPC global precipitation products. Among the narrowband to broadband conversion algorithms, one major advantage of the sounder-based OLR estimation over the imager-based ones is its accuracy because the sounder can provide more complete information about the surface and atmospheric conditions. Ellingson et al. (1994) compared the HIRS OLR to approximately, 1,00,000 collocated ERBE observations and their results show that the HIRS technique provides instantaneous OLR estimates that agree with the ERBE observation to within about 5 W m 2 RMS errors; and more than 99% of the variance of ERBE observations for both day and night observations can be explained. Ba et al. (2003) adapted the HIRS technique to the GOES Sounder and validated GOES Sounder OLR with collocated Tropical Rainfall Measurement Mission (TRMM) and Terra CERES OLR. Their results show instantaneous RMS agreement to within about 7 W m 2 for 1 1 homogeneous scenes. Polar orbiting satellites can only provide two observations a day for a given location, except for higher latitudes where the orbits overlap. This is a major limitation in observing the OLR diurnal variations and therefore a primary error source for calculating the daily mean. Geosynchronous satellites can provide observations in high temporal resolution; however, this spatial coverage is limited to the extent of the earth-view full disk. GOES SounderÕs spatial coverage is even more limited; it only covers approximately the North America domain due to the limitations in data handling capacity. While GOES Imager can estimate OLR for the full disk, it is less accurate and may have biases under certain conditions due to its limited available channels. Therefore we demonstrated in this paper an approach to combine the GOES and HIRS OLR to create a hybrid product that would bear the accuracy quality of HIRS OLR and meanwhile provide accurate diurnal variation information. This is similar to the steps that CERES is taking in time and space averaging processes (Young et al., 1998; Takmeng Wong, personal communication). The HIRS OLR is considered as an improvement over the AVHRR OLR in the NOAA operational OLR products; it may also act as the bridge for current AVHRR OLR users for adapting to the CERES OLR when it becomes operational in the National Polar- Orbiting Operational Environmental Satellite System (NPOESS) era. In the following sections, we will first describe the GOES Imager OLR algorithm, following by the explanation of the hybrid GOES and HIRS OLR product, proceeding to analyses of errors related to the OLR diurnal variation with the use of this hybrid product as a reference, and in the last section, summary and discussions. 2. Goes imager OLR algorithm We developed GOES Imager OLR algorithm following similar procedures of the HIRS OLR algorithm (Ellingson et al., 1989) and GOES Sounder OLR algorithm (Ba et al., 2003). The OLR can be estimated by the linear combination of radiances of the GOES ImagerÕs window and water vapor channels as OLR ¼ a 0 ðhþþ X a i ðhþn i ðhþ; ð1þ i where the aõs are regression coefficients whereas the constant term, a 0, is in the unit of W m 2 and the remainder a i Õs are in the unit of cm 1 sr, h is the satellite zenith angle, and N is the observed radiance in the unit of W m 2 (cm 1 sr) 1. We used GOES 8 Imager data in this study and OLR was estimated with its channel 3 (water vapor channel) and channel 5 (window channel) radiances. Table 1 lists the regression coefficients, the regression RMS error and explained variances (R 2 )as functions of satellite zenith angle (SZA). Coefficients a 1 and a 2 are for GOES Imager channels 3 and 5, respectively. The spectral locations of GOES 8 Imager channels 3 and 5 are in the wavelengths ranges of 6.5 7.0 lm and 11.5 12.5 lm, respectively. The GOES 12 Imager has some changes in the spectral specifications. The main changes include a broader water vapor channel that ranges from 5.8 to Table 1 OLR regression coefficients and statistics for GOES 8 Imager SZA ( ) a 0 (W m 2 ) a 1 ((W m 2 cm 1 sr) 1 ) a 2 ((W m 2 cm 1 sr) 1 ) RMS (W m 2 ) R 2 0 74.24 8482.3 1333.3 4.38 0.989 21.48 73.58 8454.3 1356.6 4.13 0.990 47.93 71.10 8083.4 1470.2 3.07 0.995 53.00 70.54 7864.5 1509.2 2.87 0.995 70.73 71.05 5816.7 1716.1 5.32 0.984
1122 H.-T. Lee et al. / Advances in Space Research 33 (2004) 1120 1124 7.3 lm compared to the original 6.5 to 7.0 lm, and a new 13.3 channel lm (designated as channel 6). These changes have positive effects in increasing the accuracy of the OLR estimation with this multi-spectral algorithm. The broader water vapor channel provides information about the upper tropospheric humidity and the 13.3 lm channel provides lower tropospheric temperature information.incorporating radiance observations from channels 3, 4 and 6, GOES 12 Imager can estimate OLR with regression RMS errors reduced by about 20% compared to GOES I-L Imagers. This study uses only GOES 8 Imager data. 3. Hybrid OLR product Since the GOES Imager observations cannot provide the atmospheric temperature and water vapor structure information, the OLR estimates may present biases over certain climatologically persistent features. This is similar to the problems in AVHRR OLR where Gruber et al. (1994) have found systematic biases, e.g., associated with in-sensitivity to inversion and water vapor variations over sub-tropical oceans. We devised a method to correct the GOES Imager OLR by adjusting it to the collocated HIRS estimates for each area each day. Fig. 1 illustrates the process of combining the GOES Imager and HIRS OLR. For each of the collocated area, a daily mean bias in GOES Imager OLR is determined as the average deviation of GOES OLR estimates to that of the HIRS. A daily mean OLR determined by averaging over the bias-removed GOES Imager OLR estimates is thus considered a better representation. This quantity is used as the reference to analyze the errors in HIRS daily mean OLR in the next section. In this study, we have chosen to apply this bias-removal procedure region by region and day by day, it may be preferable to apply it in a continuous time series fashion to prevent discontinuities in the instantaneous OLR diurnal record. Fig. 2 shows the scatter plots of the original GOES 8 Imager OLR and NOAA 16 HIRS OLR collocated within 30 min at 1 1 resolution over the GOES full disk domain for August 7, 2002. The horizontal spatial resolution for HIRS is about 17 km at the nadir. GOES Imager horizontal spatial resolutions at the nadir are 4 and 8 km depending on the channels. The mean difference is about 1 W m 2 with an RMS difference of about 10 W m 2. When HIRS is compared to the biasremoved GOES Imager OLR, there is nearly no mean difference, as expected, and it has an RMS difference to within about 7 W m 2. The GOES Imager OLR explains more than 97% of variances of the HIRS, and there is a near perfect relationship, a slope of 1.000. This indicates that the bias removal procedure has successfully integrated the HIRS and GOES Imager OLR products and, considering that there are errors attributed from temporal and spatial sampling differences as well as the observing zenith angle differences, we have quite strong confidence that the hybrid product would have a quality near the HIRS or the GOES Sounder. Both HIRS and GOES Imager radiance data used herein were processed using the operational calibration coefficients. There are possible errors, and especially biases, present in those coefficients such that they can contribute to the biases shown between the two products. The effects from errors in calibration coefficients may be determined when this hybrid product is validated with other measurements, e.g., CERES, and this is left for future studies. Fig. 1. Schematic for the bias removal procedure for GOES OLR. Solid dots connected by a solid line are the original GOES OLR estimates. The star symbols indicate the HIRS OLR estimates from which the daily mean bias of GOES OLR is determined as the average of the differences between HIRS and GOES OLR at the HIRS observing times within a given day, shown as d 1 and d 2 for examples. GOES OLR is compensated by this daily mean bias (as a parallel shift of the solid curve up or down) to remove the bias. The daily mean GOES OLR is the average through the bias-removed GOES OLR that are shown as the dots connected by the dashed line. The daily mean HIRS OLR is the average of the HIRS OLR estimates. Fig. 2. Original GOES 8 Imager OLR and NOAA 16 HIRS OLR collocated within 30 min at 1 1 resolution over the GOES full disk domain for August 7, 2002.
H.-T. Lee et al. / Advances in Space Research 33 (2004) 1120 1124 1123 4. OLR diurnal variation The two-pass per day HIRS OLR estimate cannot accurately take into account of the effects of diurnal variation, especially wherever there are convective cloud systems that tend to be relatively transient. We compare the daily mean determined from the hybrid OLR products, i.e., the bias-removed GOES Imager OLR, to that of the HIRS. For experimental purposes, we performed different averages for various grid sizes and lengths of period. The purpose is to investigate the errors in HIRS daily mean OLR associated with different time and space averaging processes. We chose to experiment with four equal-angle grid sizes: 1, 2.5, 5 and 10 over a period of total fifteen days of data. Fig. 3 shows the RMS differences between HIRS daily mean OLR and the reference daily mean, i.e., the daily mean OLR determined from the bias-removed GOES Imager OLR. Points shown in Fig. 3 are average values over the full disk domain for various grid sizes as functions of various lengths of averaging period, denoted as the n-day average. It shows that for oneday average, the RMS errors in HIRS daily mean range from about 4 W m 2 to about 12 W m 2. For 1 1 grid size, the RMS errors decrease with the increasing lengths of averaging period, from about 12 W m 2 to about 3 W m 2, corresponding to 1-day average and 15- day average, respectively. When the results were presented in terms of the explained variances, the HIRS daily mean OLR explains about 91 99% of variances, for 1-day average and 15-day average, respectively, for the 1 1 grid size. When the temporal averaging period length extends to 15 days, the un-explained variances by HIRS daily mean OLR for the four spatial averaging domains all fall within about 1%. Fig. 3. RMS differences between HIRS daily mean OLR and the reference daily mean. Points shown are average values over the full disk domain for various grid sizes and are shown as functions of various lengths of averaging period. These results indicate that, in general, the OLR errors in daily averaging resulted from two-pass a day sampling rate will be largely reduced when longer period average is performed, such as monthly average, or larger grid size is considered. However, one should note that such reduction of errors is only valid for areas with nonpersistent features. Some areas, e.g., in tropical convection zones and over the land, large errors can still be seen. For geographically sensitive applications such as precipitation estimate, such error must be dealt with to prevent biases even in monthly time scale. 5. Summary and discussions We adapted the HIRS and GOES Sounder OLR techniques to the GOES Imager. Using the GOES Imager window and water vapor channels radiance observations, we can estimate the OLR with regression errors of about 4 W m 2. We compared GOES Imager OLR to the collocated HIRS OLR that were observed within 30 min in 1 1 grid over the GOES full disk domain, the RMS differences is about 10 W m 2. After we removed the biases in GOES OLR by adjusting it to the HIRS OLR estimates, the RMS difference reduced to about 7Wm 2 with essentially no mean differences and a near perfect relationship, a slope of 1.000. We felt confident that the hybrid product would have a quality near that of the HIRS or the GOES Sounder. We investigated the HIRS daily mean OLR errors associated with the diurnal variation. We chose the hybrid OLR product to act as a reference in determining the daily means. It shows that for one-day average, the RMS errors in HIRS daily mean range from about 4 W m 2 to about 12 W m 2. These errors reduce with increasing length of averaging period. When averaged through 15 days, the HIRS daily mean OLR can explain about 99% of variances of the reference OLR for all four spatial grid sizes, ranging from 1 1 to 10 10. Although these averaging processes can largely reduce the RMS errors that associated with random errors, one should be cautioned that there are errors that neither the spatial nor the temporal averaging can remove, for examples, for areas with some systematic convection systems that peak twice a day, in the afternoon and in the early morning. For those areas, neither the morning nor the afternoon polar satellite can produce a representative daily mean, and daily means estimated from either satellite observations will be systematically biased. Thus to study OLR in relative short time scale, or regional problems, one must be careful that the OLR estimates do provide accurate representation pertaining to the diurnal variations. The integration of HIRS OLR with GOES Imager OLR demonstrated that by taking the advantages of both products, one could provide an OLR product that
1124 H.-T. Lee et al. / Advances in Space Research 33 (2004) 1120 1124 satisfies such requirements. By combining all the available geosynchronous satellites, one could even produce such hybrid product for a near global coverage as CE- RES is currently attempting to accomplish. This OLR data may be beneficiary to synoptic scale applications such as numerical model calibration and large-scale precipitation estimation. Acknowledgements The authors gratefully acknowledge the assistance of the ASR editorial board members, the COSPAR publication committee members and the two anonymous referees who provided critical evaluation and constructive suggestions. We thank Michael Chalfant and Americo Allegrino at NOAA NESDIS Satellite Data Service Office for providing the HIRS OLR data. We also thank Rachel Pinker, Xu Li and Wen Meng for their assistance in the use of McIDAS system for the GOES data. This study was supported by the NOAA/NESDIS Office of Systems Development (OSD) Ground System grant to the Cooperative Institute for Climate Studies (CICS) at the University of Maryland. References Ba, M.B., Ellingson, R.G., Gruber, A. Validation of a technique for estimating OLR with the GOES Sounder. J. Atmos. Ocean. Tech. 20, 79 89, 2003. Cheruy, F., Kandel, R.S., Duvel, J.P. Outgoing longwave radiation and its diurnal variation from combined ERBE and Meteosat observations. 1. Estimating OLR from Meteosat data. J. Geophys. Res. 96, 611 622, 1991. Ellingson, R.G., Lee, H.-T., Yanuk, D., Gruber, A. Validation of a technique for estimating outgoing longwave radiation from HIRS radiance observations. J. Atmos. Ocean. Tech. 11, 357 365, 1994. Ellingson, R.G., Yanuk, D.J., Lee, H.-T., Gruber, A. A technique for estimating outgoing longwave radiation from HIRS radiance observations. J. Atmos. Ocean. Tech. 6, 706 711, 1989. Gruber, A., Ellingson, R.G., Ardanuy, P., Weiss, M., Yang, S.-K., Oh, S.N. A comparison of ERBE and AVHRR longwave flux estimates. Bull. Am. Meteor. Soc. 75, 2115 2130, 1994. Gruber, A., Krueger, A.F. The status of NOAA outgoing longwave radiation data set. Bull. Am. Meteor. Soc. 65, 958 962, 1984. Minnis, P., Young, D.F., Harrison, E.F. Examination of the relationship between outgoing infrared window and total longwave fluxes using satellite data. J. Climate 4, 1114 1133, 1991. Schmetz, J., Liu, Q. Outgoing longwave radiation and its diurnal variation at regional scales from Meteosat. J. Geophys. Res. 93, 11192 11204, 1998. Young, D.F., Minnis, P., Doelling, D.R., Gibson, G.G., Wong, T. Temporal interpolation methods for the Clouds and the EarthÕs Radiant Energy System Experiment (CERES). J. Appl. Meteor. 37, 572 590, 1998.