Retrieval of atmospheric water vapor properties from the Visible Infrared Imaging Radiometer Suite (VIIRS) satellite sensor on the Suomi National Polar-orbiting Partnership (S-NPP) and the National Oceanic and Atmospheric Administration-20 (NOAA-20) platforms is challenging due to the absence of infrared (IR) water vapor absorption bands. Fortunately, measurements in the missing spectral region are available on the Crosstrack Infrared Sounder (CrIS), a hyperspectral IR sensor also on the same platforms. Spectral measurements in these IR absorption bands can be constructed for VIIRS through fusion of the imager and sounder data. Weisz et al. (2017) demonstrated a fusion method to construct IR water vapor and carbon dioxide absorption band radiances for VIIRS at 750 m spatial resolution. With the addition of the missing spectral bands to VIIRS on S-NPP, this study evaluates total column precipitable water vapor (TPW) and upper tropospheric humidity (UTH) in clear skies through comparison to the Moderate Resolution Imaging Spectroradiometer (MODIS) MYD07 (Borbas et al., 2017) and MYD08 (Platnick et al., 2017) Collection 6.1 and version 1.0 VIIRS (Borbas et al., 2019a–d) atmospheric products. The MYD07 is a level-2 swath product that provides temperature and water vapor profiles at 5 km spatial resolution, while the MYD08 provides water vapor on a daily, 8 d and monthly global grid at 1∘×1∘ resolution. Through comparison to the MYD07/MYD08 products, we will demonstrate that the VIIRS water vapor product shows better agreement when these constructed band radiances are included, with the major improvement being found in the tropics.

While VIIRS has a wide scanning swath, high horizontal resolution, a nearly constant pixel size across the scan, and a day/night band (DNB), its spectral complement lacks thermal infrared (IR) absorption bands necessary to accurately retrieve tropospheric moisture content as well as cloud properties that rely on those spectral measurements. In particular for moisture retrievals, VIIRS does not take measurements in the broad 6.7 µm water vapor band that are measured by the MODIS (Seemann et al., 2003). Fortunately, the missing IR spectral bands can be gleaned from measurements on the companion hyperspectral CrIS sensor on the same platform.

Here, we denote the instantaneous field of regard as field of view (FOV) for the sounder and pixel for the imager exclusively to minimize confusion between the two sensors. To achieve TPW and UTH at imager pixel resolution, this study employs the innovative data fusion approach of Weisz et al. (2017) that constructs MODIS-like water vapor and CO2 sensitive radiances directly at the imager resolution through use of collocated VIIRS and CrIS radiances. In this study, the data fusion method provides MODIS-like IR absorption band radiances at the VIIRS M-band spatial resolution (750 m). The VIIRS+CrIS fusion radiances are available for the entire record of both the S-NPP and NOAA-20 platforms (Baum et al., 2019a).

The availability of these IR-band radiances for VIIRS at 750 m pixel resolution makes it possible to retrieve a cloud mask and moisture properties using algorithms developed and tested using the full MODIS spectral band suite (Borbas et al., 2011). The goal of this study is to determine the impact of supplementing VIIRS with imager-resolution VIIRS+CrIS fusion bands on retrieving TPW and UTH and establish the feasibility of extending the MODIS TPW and UTH into the future with those derived from VIIRS+CrIS fusion. The continuation of such a high-spatial-resolution product will maintain the benefit to observe high-spatial-scale weather phenomena, detect an urban heat island (Hu and Brunsell, 2015), or determine atmospheric correction for high-spatial-resolution remote sensing products, such as the MODIS land surface temperature products (Proud et al., 2010; Hulley et al., 2017; Wan, 2010).

This work is a clear-sky moisture companion to the cloud parameter fusion paper of Li et al. (2020), wherein they reported on the improvement in VIIRS detection and characterization of clouds through the use of additional infrared fusion radiances.

This paper is organized as follows: Sect. 2 discusses data and fusion method, Sect. 3 summarizes the moisture retrieval method and presents results, and a summary of the findings is provided in Sect. 4.

The VIIRS sensor is a 22-band scanning radiometer that is currently flying on the NASA S-NPP and the NOAA-20 platforms. VIIRS has 16 bands scanning a 3000 km swath at 750 m resolution (medium resolution, or M), five bands at 375 m resolution (imaging, or I), and a day/night band. For this investigation, the focus is on using the bands at M resolution. The data used in this study include the standard level-1B VIIRS data for both the S-NPP and NOAA-20 platforms made available by the Atmosphere Science Investigator-led Processing System (A-SIPS) located at the University of Wisconsin – Madison Space Science Engineering Center (SSEC).

CrIS is a Fourier transform spectrometer with 1305 spectral channels in normal spectral resolution (NSR) and 2211 channels in full spectral resolution (FSR) over three wavelength ranges: longwave infrared (LWIR) (9.14 to 15.38 µm); midwave-infrared (MWIR) (5.71 to 8.26 µm); and shortwave infrared (SWIR) (3.92 to 4.64 µm). CrIS scans a 2200 km swath width (±50 ∘), with 30 Earth-scene views. Each view consists of nine FOVs from a 3×3 array that have a nadir spatial resolution of approximately 14 km.

The fusion method requires an accurate collocation between the high-spatial-resolution imager data (with pixels at 750 m) and the lower-spatial-resolution sounder data (with FOVs at about 14 km). The fusion method described in Weisz et al. (2017) consists of two steps for each imager pixel: (a) search nearby neighbors to find the five FOVs that best match the split-window (i.e., 11 and 12 µm) imager pixel radiances averaged over the FOV to an individual pixel's split-window measurements – this is accomplished using a k-d (or multi-dimensional) tree search algorithm (Bentley, 1975) on both high-spatial-resolution (M-band 11 and 12 µm data) and low-spatial-resolution (M-band 11 and 12 µm data averaged over the CrIS FOV) imager radiances. (b) Convolve the high-spectral-sounder radiances (at low spatial resolution) to the desired IR broadband, then average the convolved sounder radiances associated with the selected five nearest neighbors to construct the desired spectral band for each imager pixel. Spectral radiance convolution refers to the process of converting high-spectral-resolution (narrowband) to broadband radiance measurements by applying a spectral response function (SRF) of a given broadband. Here, SRFs associated with the spectral bands of the MODIS sensor on the NASA Earth Observation System (EOS) Aqua platform are applied to CrIS measurements. The VIIRS+CrIS fusion IR absorption band radiances are available for the entire records of S-NPP and NOAA-20 at the Level-1 and Atmosphere Archive and Distribution System (LAADS) Distributed Active Archive Center (DAAC) at the NASA Goddard Space Flight Center (Baum et al., 2019ab).

VIIRS+CrIS fusion radiances alongside observed radiances for MODIS bands 25 (4.5 µm), 27 (6.7 µm), and 35 (13.9 µm), repeated from Weisz et al. (2017), are shown in Fig. 1. The fusion results for band 27 show more inaccuracies, because H2O-sensitive spectral bands sense different tropospheric regions than split-window spectral bands. Also, small-scale and narrow spatial features in moisture (e.g., dry slots and cloud edges), which are not captured by the sounder due to its large spatial resolution, are more problematic for the fusion process. Furthermore, the results at the edge of the imager granule (i.e., outside the sounder swath) should be used with caution since they do not account for “limb darkening” and hence tend to be less accurate. Results shown in Fig. 1 for the VIIRS+CrIS fusion radiances compared well in a qualitative sense with the observed MODIS radiances, even in the more challenging water vapor band.

To assess the viability of the moisture products to provide continuity with similar products from MODIS, we perform a comparison with collocated measurements (i.e., matchups) with Aqua MODIS. For this study, the collocation process requires the VIIRS 750 m pixel to be fully contained within the MODIS 1 km pixel; the scene must be high confidence clear (as identified by the MODIS cloud mask MYD35); and the scan angles for the matching pair must be less than 50∘ so that it is within the swath of the CrIS sensor. Figure 2 shows the results of tens of thousands of instances of collocated MODIS and VIIRS+CrIS fusion radiances that are converted to brightness temperatures (BTs) in two water vapor and four carbon dioxide bands for the month of April 2018. It can be seen that the mean clear-sky brightness temperature differences (BTDs) between VIIRS+CrIS fusion and original MODIS data are less than 0.5 K for MODIS H2O bands 27 and 28 (6.7 and 7.3 µm) and MODIS CO2 bands 33 to 36 (13.3, 13.6, 13.9, and 14.2 µm) for 11 µm BTs ranging from 200 to 280 K. Root mean square scatter (not shown) about these mean values is found to be 1.1 K for the H2O bands and 0.5 K for the CO2 bands. MODIS radiance comparisons with respect to Infrared Atmospheric Sounding Interferometer (IASI) over 6 years found that the water vapor bands showed scatter up to 1.0 K in the H2O bands and 0.5 K in the CO2 bands (Moeller et al., 2014). The fusion comparison results are similar. Thus, it can be summarized that an order of magnitude spatial resolution (from 14 km to 750 m) has been added at the cost of introducing measurement offsets of 0.25 to 0.5 K and noise of 0.5 to 1.0 K. Results for all 12 months in 2018 (not shown) are similar; in fact, results for the entire S-NPP archive show comparably positive fusion results.

Our retrieval of TPW and UTH from selected IR measurements adopts a statistical regression algorithm (Seemann et al., 2003, 2008; Li et al., 2000; Smith and Woolf, 1988; Hayden, 1988) performed using clear-sky radiances (and BTs) measured over land and ocean for both day and night. The regression is developed with the SeeBor training database (Borbas et al., 2005) that consists of over 15 000 atmospheric profiles globally and seasonally well distributed. The water vapor retrieval algorithm has two parts: first, the regression coefficients are calculated using radiative transfer calculations, and then the regression retrieval is performed. The radiative transfer calculation of the MODIS-like radiances of bands 25, 27, 28, and 30–36 is performed using the forward model called Radiative Transfer for TOVS (RTTOV) version 12 (Saunders et al., 2018). The regression relationships between the calculated BTs and retrieved moisture products are generated for four (and three) different BT zones over land and ocean, respectively, and 60 sensor zenith angles from nadir to 60∘. The only other ancillary information required is the surface pressure, which is provided by National Centers for Environmental Prediction (NCEP) reanalysis data (Saha et al., 2010). TPW and UTH are determined for clear-sky radiances measured by VIIRS and calculated from VIIRS+CrIS fusion. The retrieval approach is similar to that adopted for MODIS. There is a strong reliance on radiances from 6.7, 11, and 12 µm. The operational VIIRS cloud mask (called CLDMSK_L2_VIIRS_SNPP.001; Ackerman et al., 2019) is applied to VIIRS to characterize the probability of cloud cover.

Figure 3 shows CrIS TPW and UTH at the sounder FOV resolution; they outline the tropospheric moisture gradients at coarse (∼14 km) resolution for clear and partly cloudy skies. The soundings are obtained using the dual regression method (Smith et al., 2012; Weisz et al., 2013), which is a computationally fast, physically based method that retrieves profiles as well as surface and cloud properties from high-spectral-resolution radiances measured in both clear- and cloudy-sky conditions at single-FOV resolution. TPW represents the total column integration of the moisture profile, while UTH is the integration from 400 hPa to the top of the atmosphere. Also shown are the regression retrieval results for the VIIRS+CrIS fusion spectral band radiances (created using the MODIS-like IR spectral response functions) at higher spatial resolution (∼750 m) in pixels deemed to be clear in the VIIRS cloud mask. They display more refined features and improve the coverage but show higher values of TPW off the coast of Baja and miss some of the UTH features in Wyoming and Colorado that were suggested in the CrIS soundings. While the results are derived from two independent algorithms, this example illustrates a challenge for the VIIRS split-window search for nearby FOVs; the search will rely primarily on low-level temperature and moisture features and less on mid-level to upper-level moisture gradients.

The absence of water vapor and CO2 absorption IR spectral bands on the VIIRS imager on the S-NPP and NOAA-20 polar-orbiting platforms limits the capability for tropospheric moisture retrievals, especially for upper tropospheric moisture. This study shows the advantage of using IR absorption bands 4.5, 6.7, 7.3, 13.3, 13.6, 13.9, and 14.2 µm, which are constructed at VIIRS spatial resolution (750 m) using a data fusion approach using both sounder (CrIS) and imager (VIIRS) measurements following the approach in Weisz et al. (2017). The positive impact of adding the constructed fusion spectral bands on TPW and UTH retrievals is demonstrated. The moisture retrievals are based on the MODIS MYD07 Collection 6.1 algorithm package. Evaluations of the resulting moisture products are performed through comparisons to the operational MODIS Collection 6.1 and VIIRS (VIIRS-only and VIIRS+NUCAPS) version 1.0 moisture products.

Improvements in VIIRS+CrIS products, enabled by addition of fusion radiances, over the VIIRS-only and VIIRS+NUCAPS products are observed for TPW when quantitatively compared to the MYD08 products. In our 1-month study for January 2017, the global mean of the TPW derived from the VIIRS+CrIS fusion radiances is 0.2 mm higher with a scatter of 1.4 mm when compared to the MYD08 TPW; without the fusion radiances (VIIRS-only product), the mean is 1.1 mm too high with a scatter of 2.7 mm with most of the overestimation occurring in the tropics. The VIIRS+CrIS fusion TPW also demonstrates improvement over the VIIRS+NUCAPS TPW (with 0.3 mm mean and 2.0 mm scatter with respect to the MYD08 product). Similar TPW results are also found for 1 month in each season of 2017. VIIRS+CrIS UTH, now possible with the addition of the fusion radiances, is found to be within 10 % of the MYD08 UTH in mean and scatter for the same 4 months.

The results in this study are limited to a VIIRS sensor scan angle of 50∘ to minimize the impact of the CrIS swath being less than that of the imager. These findings are limited in scope but clearly demonstrate the potential in the use of the fusion IR absorption spectral bands in generating moisture products and continuing the moisture record from MODIS and the previous generations of polar-orbiting satellite sensors. In future work, we plan to extend this evaluation to longer time periods, undertake a global comparison of VIIRS+CrIS fusion moisture products with ground-based measurements (Bedka et al., 2010; Roman et al., 2016), and possibly replace the operational VIIRS+NUCAPS moisture products with the VIIRS+CrIS fusion-derived moisture products.