Top of the Atmosphere Reflected Shortwave Radiative Fluxes from ABI on GOES-18

Abstract

In this study, we describe the derivation and evaluation of Top of the Atmosphere (TOA) Shortwave Radiative (SWR) Fluxes from the Advanced Baseline Imager (ABI) sensor on the GOES-18 satellite. The TOA estimates use narrowband observations from ABI that are transformed to broadband (NTB), based on simulations and adjusted to total fluxes using Angular Distribution Models (ADMs). Subsequently, the GOES-18 estimates are evaluated against the Clouds and the Earth’s Radiant Energy System (CERES) data, the only observed SWR broadband flux dataset. The importance of agreement at the TOA is that most methodologies to derive surface SWR start with the satellite observation at the TOA. Moreover, information needed to compute radiative fluxes at both boundaries (TOA and surface) is needed for estimating the energy absorbed by the atmosphere. The methodology described was comprehensively evaluated, and possible sources of errors were identified. The results of the evaluation for the four seasonal months indicate that by using the best available auxiliary data, the accuracy achieved in estimating TOA SWR at the instantaneous scale ranges between 0.55 and 17.14 W m⁻² for the bias and 22.21 to 30.64 W m⁻² for the standard deviation of biases (differences are ABI minus CERES). It is believed that the high bias of 17.14 for July is related to the predominantly cloudless sky conditions, when the used ADMs do not perform as well as for cloudy conditions.

1. Introduction

Our climate is controlled by the surface energy budget on a global scale and is dominated by radiative fluxes. The incoming shortwave radiation (SWR) from the sun that reaches the Earth’s surface is a dominant component of this budget. Strides in the technical capabilities of satellites make them a primary source for information on large scales. Satellites, originally polar orbiters, have become an integral part of most programs to monitor the climate. The limitation and impact of poor representation of the diurnal cycle from polar orbiters has been recognized, and interest in geostationary satellites has increased. Yet, satellite-based estimates differ from each other and from those provided by numerical models. Major differences are related to the quality of satellite observations, such as the frequent changes in satellite observing systems, degradation of sensors, restricted spectral intervals, viewing geometry of sensors, and changes in the quality of atmospheric inputs that drive the inference schemes. Achievable accuracy is also related to the lack of maturity of basic information needed in the implementation process, such as reliable cloud-screened products, which are frequently in the process of development and modifications (Heidinger [1]). Reducing differences among the satellite-based estimates requires, among others, updates to inference schemes so that the most recent auxiliary information can be fully utilized.

Two products, the reflected SWR at the top of the atmosphere and the downward SWR at the surface, are routinely generated at the NOAA from the Advanced Baseline Imager (ABI) onboard the GOES-R series of US satellites (Laszlo et al. [2,3]). Both products require a shortwave broadband top of atmosphere (TOA) albedo converted from six ABI narrowband reflectance (

Table 1. Wavelength, type, and resolution of the six ABI bands used in this study.

ABI Band	Central Wavelength (μm)	Type	Best spatial Resolution (km)
1	0.47	Visible	1
2	0.64	Visible	0.5
3	0.86	Near-IR	1
4	1.37	Near-IR	2
5	1.6	Near-IR	1
6	2.2	Near-IR	2

). Critical elements of an inference scheme for TOA radiative flux estimates from satellite observations are (1) transformation of narrowband quantities into broadband ones; (2) transformation of bi-directional reflectance into albedo by applying Angular Distribution Models (ADMs). The SWR ADMs are scene-type-dependent and are selected by the instantaneous cloud and surface conditions observed by the imager. The narrow-to-broadband (NTB) conversion used in this study was derived from simulated broadband and narrowband reflectance, as detailed in Pinker et al. [4] for the ABI on GOES-16 and 17. Observation-based broadband ADMs derived primarily from the Clouds and the Earth’s Radiant Energy System (CERES) were used to complete the derivation of the TOA radiative flux. The “ground truth”, namely, the CERES observations, is also undergoing adjustments and recalibration. As such, an evolutionary process can be expected, as demonstrated in this study by emphasizing changes made in our approach when GOES-18 is compared to GOES-16 and 17. New spectral characteristics of ABI onboard GOES-18 were applied to the simulated narrowband fluxes. The procedure of matching GOES-18 observations with those from CERES was updated. We used the radiance from the ABI product, converted it to reflectance, and applied NTB conversions and ADMs before comparing it to CERES fluxes. We used the radiances in six relevant ABI bands (out of sixteen), as shown in Table 1.

Each band has a different resolution. Before use, we resampled them to 2 km. Subsequently, these 2 km data were remapped to the 20 km CERES SSF instantaneous footprint. Previously, we remapped the CERES observations to those of GOES-16 and 17. Now we remapped the high-resolution observations from GOES-18 to those of CERES. In Section 2, we describe the entire process, including a brief description of methodology, data used, matching of observations between ABI/GOES-18 and CERES Single Scanner Footprint (SSF), and averaging the ABI data to match the CERES pixel; in Section 3, we present the results, and in Section 4, we discuss and summarize the findings.

2. Approach

2.1. Methodology

We briefly describe the physical basis and the development of the NTB transformations of satellite observed radiances and the bi-directional corrections to be applied to the broadband reflectance to obtain broadband TOA albedo. The Advanced Baseline Imager (ABI) observations onboard the NOAA GOES-R series of satellites provide reflectance in six narrow bands in the shortwave spectrum. There is a need to transform these narrowband observations to broadband values. The following is done: (1) calculation of TOA high-resolution spectral and broadband (total shortwave) reflectance with MODTRAN-4.3 (Berk et al. [5]). This requires input information on scene-dependent land use classifications from the International Geosphere-Biosphere Programme (IGBP) (Hansen et al. [6]) using the MODTRAN built-in reflectance model corresponding to IGBP surface types and 100 profiles selected from the SeeBorV5.1 global atmospheric database from the Cooperative Institute for Meteorological Satellite Studies (CIMSS) (https://cimss.ssec.wisc.edu/training_data/data/, accessed on 10 August 2025), which is given for 33 layers in the vertical. In the simulations, 10 solar bins and 48 viewing bins (that include viewing zenith angles and azimuth angles) were specified. Variability of atmospheric aerosols was also accounted for. For cloudy sky simulations, we used four surface types derived from 12 IGBP types (Pinker et al. [4]) and simulated six types of default clouds as provided in MODTRAN-3.7 (cumulus, altostratus, stratus, stratus, stratocumulus, nimbostratus, and cirrus, which are stratified by CODs); (2) convolution of the high-resolution spectral reflectance to simulate ABI narrowband reflectance; (3) regression of the simulated ABI reflectance against the broadband reflectance calculated in step 1. The ADMs from CERES (Loeb et al. [7]) were augmented with theoretical simulations (Niu and Pinker [8]). This was done to extend the observational CERES database that was undersampled in certain directions (angular bins). Surface conditions were one of the primary inputs into the MODTRAN simulations. The International Geosphere-Biosphere Programme (IGBP) land classification (Hansen et al. [6]; Loveland et al. [9]) dataset is at 1/6° resolution and includes 18 surface types. We converted the 1/6° (~18.5 km) resolution to the ABI 2 km grid using the nearest grid method. A detailed description of the methodology and issues related to matching the theoretical ADMs with those of CERES can be found in Niu and Pinker [8] and applications to GOES-16 and 17 in Pinker et al. [4]. Modifications to these procedures when applied to GOES-18 will be described in Section 2.3 and Section 2.4.

2.2. Data Used

L1B GOES-18 radiance data files for the CONUS domain (file names OR_ABI-L1b-RadC-) were downloaded from the NOAA Comprehensive Large Array-Data Stewardship System (https://www.aev.class.noaa.gov/, accessed on 10 August 2025) and the ABI Spectral Response Function (SRF) from NOAA’s Center for Satellite Application and Research Calibration Center website (https://ncc.nesdis.noaa.gov/GOESR/ABI.php. A cloud mask described in Heidinger et al. [1] was used. In addition to the cloud mask, cloud phase (ice, water, mixed) and cloud optical depth were used as well.

Not all the required angular information needed for the implementation of regressions was available online and had to be recomputed (such as solar geometry). Reference data for CERES observation were downloaded from (https://cmr.earthdata.nasa.gov/search/concepts/C7460991-LARC_ASDC.html, accessed on 10 August 2025). Data file name is “CERES_SSF_Terra-XTRK_Edition4A_Subset”.

The resolution of the instantaneous CERES SSF product footprint is about 20 km. In addition to the cloud mask, cloud phase (ice, water, mixed) and cloud optical depth were used, as well as the following datasets:

• ABI L2 Clear Sky Mask (example file name: OR_ABI-L2-ACMC-M6_G18_s20223291901176_e20223291903549_c20223291905088.nc)
• ABI L2 Cloud Top Phase (example file name: OR_ABI-L2-ACTPC-M6_G18_s20223291901176_e20223291903549_c20223291905408.nc)
• ABI L2 Cloud Optical Depth at 640 nm (example file name: OR_ABI-L2-CODC-M6_G18_s20223291901176_e20223291903549_c20223291907099.nc)

For each 2 km ABI pixel, the channel radiances were converted to SWR fluxes using cloud properties and a climatology of aerosols based on geographic surface type (Rural, Maritime, Urban, Desert, Tropospheric).

2.3. Matching Observations Between ABI/GOES-18 and CERES SSF

ABI on GOES-18 is a geostationary sensor at nadir longitude of 137° W. The Pacific U.S. (PACUS) sector coverage of the GOES-West satellite is 12° N–60° N, 90° W–175° W, as shown in Figure 1 (for details, see https://www.star.nesdis.noaa.gov/goes/index.php, accessed on 10 August 2025).

The ABI GOES-18 data are provided as images for each observation time on a fixed grid whose coordinate values are the sensor scanning angle in units of radians relative to the satellite sub-point location. The fixed grid was rectified to an ellipsoid defined by the Geodetic Reference System 1980 (GRS80) Earth model. The navigation between the fixed grid coordinates (scanning angle in east/west and elevation angle in north/south) and geodetic latitude and longitude was determined by the geostationary satellite projection or fixed grid projection (FGP). Details can be found in “GOES R SERIES PRODUCT DEFINITION AND USERS’ GUIDE” (https://www.goes-r.gov/users/docs/PUG-main-vol1.pdf, accessed on 10 August 2025).

Below (Figure 2) is an example of the matchup between CERES and ABI for day 327 in 2022, 21:21:17 UTC.

The CERES data are given in a time series format. Every CERES pixel’s longitude, latitude, and time are provided along with the flux data. To find the matchups, we looped over each ABI image time and searched in the CERES time series to find the CERES pixels that fall in the ABI domain within a certain period. For example, given an ABI image timed at 1 December 2022 17:15:00 with area coverage from 15° N to 50° N latitude and from 170° W to 80° W longitude, we found that there were 502 CERES pixels that fell within the area in a time interval of ±5 min around the ABI time, as shown in Figure 3a.

The number of overlapping pixels ranges from 0 to 18,000 for Terra and from 0 to 16,000 for Aqua. For all ABI images available during the period from 23 November to 1 December 2022, the number of images containing overlapping pixels with CERES data was counted and is shown in

Table 2. Number of available ABI images with pixels overlapping with CERES data within the time period from 23 November to 1 December 2022.

# of Overlap Pixels	CERES/Terra	CERES/Aqua
>500	128	78
>1000	121	72
>5000	87	43
>10,000	47	21
>12,000	36	16
>15,000	13	7

.

As an example, if we chose only those ABI images with more than 12,000 overlap pixels for CERES/Aqua, we obtain sixteen matchups between CERES and ABI at a specific ABI time. Next, we processed these ABI cases to obtain TOA fluxes and performed a comparison with the co-located CERES data.

2.4. Averaging ABI Data to Match the CERES Pixel

Due to the fixed grid projection format (FGF) used to store the ABI data, one can easily transform the map coordinates between the longitude/latitude format and the ABI image col/row format. This analytic way to calculate the location of the matchup pixel in the ABI image saves time compared to searching through the whole ABI image to find the location of the matchups.

ABI flux data have 2 km nadir resolution. CERES Field of View (FOV) is about 16 × 32 km (the 20 km is a nominal spatial resolution that is roughly the same size as 16 × 32 km). When matching ABI with CERES, we first calculated the col/row location of the CERES pixel in the ABI image. Around that location, a 16 × 32 km block of ABI data was extracted and averaged to match the CERES SSF data (these are the values at nadir; however, in this study, we used the same size for all CERES FOVs). Each ABI pixel was weighted by the inverse of its distance to the center of the CERES footprint to estimate the CERES point spread function. An example is shown in Figure 3b.

Since ABI data are available about every 5 min for the CONUS domain, there might be more than one ABI image that overlaps with a CERES pixel within the 10 min time interval. In such a situation, the ABI data from all available times within the 10 min time interval and within the 16 × 32 block were averaged to match the CERES data.

2.5. Experiments with New ADMs

For GOES-16 and 17, we used ADMs as described in Loeb et al. [7]. The next-generation ADMs that were developed for Terra and Aqua using all available CERES rotating azimuth plane radiance measurements (Su et al. [10]) are not yet available to the public and therefore not used in this study. The CERES observed radiances were converted to flux using the Edition 4 ADMs, which were selected by scene type, where the scene types were based on the associated cloud, surface, and atmospheric retrievals.

As reported in Su et al. [11], significant differences between the new and the older ADMs were for clear-sky scene and polar scene types. Over clear land, the ADMs were developed for every 1° latitude × 1° longitude region for every calendar month. It is claimed that compared to Loeb et al.’s [6] ADMs (hereafter CERES2003 ADM), the new ADMs change the monthly mean instantaneous fluxes by up to 5 W m⁻² on a regional scale of 1° latitude × 1° longitude, but the flux changes are less than 0.5 W m⁻² on a global scale.

Matchups between ABI and CERES SSF TOA fluxes in 23 November 2022 00Z to 1 December 2022 23Z were used. There were twenty-one cases for Aqua and forty-seven cases for Terra (Appendix A).

As seen in Figure 4, compared to the CERES TOA fluxes, the ABI fluxes are larger in the low range (CERES TOA fluxes less than about 100 W m⁻²) and smaller in the high range (CERES TOA fluxes larger than about 250 W m⁻²). The low and high ranges generally correspond to cloud-free scenes (and or low solar zenith angles) and cloudy scenes (and or high solar zenith angles), respectively. In the mid-range (approximately between 100 and 250 W m⁻²), there is a large scatter of data points, several of them with significant negative ABI-CERES TOA flux differences. However, most of them are clustered around the one-to-one line. As a result, the frequency distribution of the ABI-CERES TOA flux differences peaks around zero with a large negative tail. Another peak at about a positive 30 W m⁻² is present that comes from the low range of CERES fluxes, as shown in the density scatterplot. We conducted several experiments to identify the reason for this secondary peak.

2.5.1. Experiment 1

We applied the old approach (Pinker et al. [4]) to the ADMs (Figure 5), and the secondary spike disappeared. The old approach is based on a synergy between the original CERES ADMs described in Loeb et al. [7] and simulations, as described in Niu and Pinker [8]. The bias increased, but there was some reduction in the standard deviation of biases.

2.5.2. Experiment 2

In the second experiment, we used the CERES2003 ADM for all surface types except for clear ocean. For clear ocean, we replaced it with our original ADMs that combined the CERES2003 ADMs and our simulations (Niu and Pinker [8]). This resulted in lower ABI fluxes in the mid range, which in turn increased the number of points below the one-to-one line and reduced the overall bias, but with practically no change in the standard deviation (Figure 6).

2.5.3. Experiment 3

Based on the improvement seen in Experiment 2, we have redone all the cases for November 2022 using this approach (for clear ocean). The cases used are detailed in Appendix A. A summary of the results is shown in Figure 7 and detailed in Appendix B. As is evident, substantial improvement has been achieved.

3. Results

3.1. Results for January, April, and July After Removing Outliers

Outliers are defined as data points with std greater than three. At least for some of the outliers, the large differences are from locations where there are only a few valid pixels available in the corresponding CERES box. In this case, the block average may not be representative of the real condition of the CERES box. The scatter plots and statistics after removing the outliers are shown below (Figure 8, Figure 9 and Figure 10, corresponding to Juanary, April and July 2023, respectively). The ADM used is the CERES2003 ADMs with ocean ADM values replaced with the old, synthesized ADMs.

3.2. Investigation of the July Case

To better understand why the results for July agree less with CERES than those for the other three months, we investigated the cloudy conditions during the selected months. For example, we plotted the conditions for each day in July and April 2023. We illustrate each month with one case. The notations in the following figures are:

(a)

Matchup between ABI and CERES swaths. Red pixels are CERES FOVs.

(b)

ABI flux image in 2 km resolution matched to CERES swath.

(c)

ABI flux image re-gridded to CERES resolution. Matched to CERES swath.

(d)

CERES flux image.

(e)

Density scatter plot between CERES and ABI.

(f)

Histogram and statistics.

(g)

ABI cloud fraction.

(h)

ABI 640 nm cloud optical depth.

In the following figure, we illustrate the cloud conditions for 14 July 2023. The figures show TOA fluxes that are considered a proxy for the presence of clouds. Illustrated are also the results of the evaluation as specified above for Figure 11a–f. Figure 11g,h show the ABI cloud fraction and the 640 nm cloud optical depth, respectively.

It was established, and shown in Figure 11 specifically for 14 July 2023, that most July cases have clear conditions or thin clouds over the ABI-CERES overlap region. The higher bias comes mainly from the clear pixels.

We also investigated cloud conditions during April 2023. In April, cloudy cases are more frequent than in July. A case for April is shown in Figure 12. We see a lower bias in April than in July.

4. Discussion and Summary

Most satellite observations provide information at TOA only and observe in narrow spectral bands. To derive the total flux at the TOA as well as at the surface, there is a need to transform such observations into broadband fluxes. This is accomplished primarily by simulations that require validation against broadband observations. The primary source of broadband observations at the TOA comes from CERES. However, CERES radiances require empirical correction using Angular Distribution Models (ADMs) to convert radiance into a flux. The ADMs require proper scene identification based on associated imager clouds and surface conditions.

As such, there is no absolute “truth” available for evaluations. In this study, we used narrowband observations from ABI on GOES-18 that were transformed to broadband based on simulations (NTB) and adjusted to total fluxes using an ADM that is a combination of the CERES2003 ADM and the Niu2011 ADM for ocean to estimate the broadband flux. Subsequently, the GOES-18 estimates were evaluated against the Clouds and the Earth’s Radiant Energy System (CERES) data, the only available source of such information. The importance of agreement at the TOA is that most methodologies to derive surface SWR fluxes start with the satellite observation at the TOA. Moreover, information on radiative fluxes at both boundaries (TOA and surface) is needed for estimating the energy absorbed by the atmosphere. The methodology described to make such comparisons was evaluated, and possible sources of error were identified.

We completed an evaluation of shortwave SWR fluxes at the top of the atmosphere (TOA) as derived from GOES-18 against CERES for four months from November 2022 and beyond, representing different seasons. This is the first effort to evaluate ABI for GOES-18 for such an extensive period. In contrast to the evaluation of GOES-16 and 17, we remapped ABI observations to CERES. Previously, the mapping was from CERES to ABI. Previous experiments have shown that mapping procedures do have an impact on the results. The results of the evaluation indicate that the accuracy achieved in estimating TOA SWR fluxes ranges between 0.55 and 17.14 for the bias and 22.21 and 31.20 for the std. It is believed that the high bias of 17.14 for July is related to the predominantly clear sky conditions, when the used ADMs may have some problems.