2.1.

Satellite Instruments

Table 1 provides some of the main features of the CrIS, VIIRS, and ABI sensors. CrIS and VIIRS are together on the S-NPP platform launched on October 28, 2011, and on the NOAA-20 platform launched November 18, 2017. Both are in a low Earth Sun-synchronous (or polar) orbit with an afternoon equatorial crossing time near 1330 UTC. ABI has been launched into GEO on November 19, 2016, on GOES-16 and placed the east position at 75.2°W. The second ABI was launched on March 1, 2018, on GOES-17 and placed in the west position at 137.2°W.

Table 1

CrIS, VIIRS, and ABI sensor characteristics.

2.2.

Product Fusion Method

The imager/sounder radiance fusion method, which is described in more detail in Refs. 1 and 6, can be summarized as follows. To infer sounder radiances and products at imager spatial resolution, the fusion process requires two steps. In the first step, imager data are used as the input to a nearest neighbor search—specifically a multidimensional k-d tree search—to find the best pairing of imager single pixel radiances (usually in the split window at 11 and $12\mu \mathrm{m}$) with a selected number (usually five) of nearby coarse resolution imager radiances. The latter are imager split window radiances averaged over collocated sounder FOVs. The collocation of VIIRS and ABI pixels within the CrIS FOV follows the process described in Ref. 7. The search finds, for each imager pixel, the closest, in radiance and geometrical space, low spatial resolution FOVs. In the second step, the mean of the sounder data (radiances or retrievals) at these selected FOVs is computed to provide the final fused product at imager pixel resolution. In the case of radiance fusion, the hyperspectral sounder data are first integrated over a selected SRF to produce a convolved radiance (and repeated for other desired spectral bands) before the nearest neighbor averaging is performed. These fusion radiances can then be used to infer profile, cloud and surface parameters; when MODIS SRFs are used, the heritage algorithms (such as those in the International MODIS and Atmospheric Infrared Sounder Processing Package, or IMAPP, described in Ref. 8) can be applied to the fusion radiances. Similarly, when performing product fusion, the sounder retrieval products (i.e., instead of convolved radiance data) from the FOVs selected in step one are averaged (for a given imager pixel) to create the new sounding products at imager pixel resolution; only CrIS soundings for FOVs determined to be clear of clouds by the sounder profile retrieval algorithm are averaged and displayed for imager pixels determined to be cloud-free by the imager cloud mask. These retrieval products, since they originate from hyperspectral retrievals of high vertical resolution and have been transferred to high spatial resolution through the fusion method, provide improved spatial mesoscale detail in temperature, moisture, and cloud features to the benefit of both hyperspectral sounder and imager applications, including severe weather nowcasting/prediction.

In this paper, CrIS retrieval profiles are derived using the UW hyperspectral retrieval system, which is based on the dual-regression (DR) method, although any other retrieval algorithm can be used. DR (Refs. 910.–11) is a 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. The retrievals of surface skin temperature, atmospheric temperature, moisture, and ozone profiles (down to the surface in clear skies, above opaque clouds, and down below thin or broken clouds) are achieved with known vertical resolution and accuracy. Applying direct fusion to these CrIS retrieval products provides products now at high spatial (imager) resolution while maintaining high vertical resolution features.

3.1.

VIIRS and CrIS Product Fusion

Although our focus here is on the direct product fusion (i.e., the k-d tree search results are directly applied to sounder retrievals), retrievals from the fusion radiance estimates are also presented to compare soundings at imager resolution produced in the two different ways: (1) VIIRS/CrIS fusion is initiated to generate broadband ${\mathrm{CO}}_2$ and ${\mathrm{H}}_2\mathrm{O}$ sensitive IR radiances at imager resolution (that are MODIS-like) and then a simple regression is used to produce retrievals with those radiances and (2) VIIRS/CrIS product fusion is performed, where DR CrIS high spectral resolution profile retrievals are directly transferred via the fusion process to high spatial imager resolution. The regression retrieval method used in (1) is based on the operational MODIS algorithm developed for IMAPP.⁸ Figures 1 and 2 show the temperature and ${\mathrm{H}}_2\mathrm{O}$ retrieval results at selected pressure levels for two CrIS granules on April 10, 2018, at 0936 and 0942 UTC covering the western USA.

Fig. 1

Temperature (K) is shown at 300, 500 and 850 hPa in the top, middle and bottom panels, respectively. (a) CrIS DR retrievals at sounder resolution, (b) regression retrievals derived from VIIRS/CRIS fusion radiances at imager resolution, and (c) VIIRS/CrIS fusion products at imager resolution. On April 10, 2018, at 0936 and 0942 UTC (CrIS granule start times).

Fig. 2

${\mathrm{H}}_2\mathrm{O}$ mixing ratio (g/kg) is shown at 500 and 850 hPa in the top and bottom panels, respectively. (a) CrIS DR retrievals at sounder resolution, (b) regression retrievals derived from VIIRS/CrIS fusion radiances at imager resolution, and (c) VIIRS/CrIS fusion products at imager resolution. On April 10, 2018, at 0936 and 0942 UTC (CrIS granule start times).

CrIS soundings at sounder resolution outline the tropospheric temperature and moisture gradients at coarse ($\sim 14\mathrm{km}$) resolution for clear and partly cloudy skies. Soundings from broad MODIS-like IR spectral bands at fine resolution ($\sim 750\mathrm{m}$) in pixels deemed to be clear in the VIIRS cloud mask (MODIS/VIIRS cloud mask or MVCM discussed in Ref. 12) display more refined features and improve the coverage, but miss some of the mid (500 hPa) and low (850 hPa) level attributes suggested in the CrIS soundings. It should be mentioned that cloudy imager pixels found by the VIIRS MVCM are not displayed in the fusion results shown in Figs. 1 and 2. As can be seen in Figs. 1 and 2, the DR clear/cloudy sky detection with CrIS compares well with the MVCM cloud mask.

CrIS soundings mapped directly to fine resolution by the fusion process delineate mid- and low-level moisture features more in agreement with the CrIS soundings than the broadband retrievals. Fusion of the CrIS soundings directly to imager resolution captures more of the information content of the high spectral resolution measurements than fusion to broadband radiances followed by sounding retrievals. Although the fusion results are shown here only at 850, 500, and 300 hPa pressure levels, any of the 101 pressure levels provided in the CrIS profile retrievals could be shown.

CrIS DR retrievals and the product fusion results for another scene, namely the Chesapeake Bay area on May 19, 2018, around 18 UTC, are shown in Figs. 3 and 4. This case study will also be discussed in Secs. 3.2 and 3.3 with respect to LEO/GEO fusion. Overall, the fusion temperature products (Fig. 3) show increased coverage and fine-scale variations while maintaining the same features depicted by the hyperspectral retrievals.

Fig. 3

Temperature (K) provided by the CrIS DR retrieval (top) and the VIIRS/CrIS product fusion (bottom) is shown at the 300, 500, and the 850 hPa pressure level in panels (a), (b), and (c), respectively. On May 19, 2017, at 1748 and 1752 UTC (CrIS granule start times).

Fig. 4

${\mathrm{H}}_2\mathrm{O}$ mixing ratio (g/kg) provided by the CrIS DR retrieval (top) and the VIIRS/CrIS product fusion (bottom) at the 500 and 850 hPa pressure level in panels (a) and (b), respectively. On May 19, 2017, at 1748 and 1752 UTC (CrIS granule start times).

Similar conclusions can be drawn for low-level ${\mathrm{H}}_2\mathrm{O}$ [Fig. 4(b)], but the mid-level humidity fusion product [Fig. 4(a)] shows some deviations from the hyperspectral retrieval products. These differences are due to the fact that the VIIRS split-window bands, which are not sensitive to mid-level moisture, are used in the k-d tree search.

3.2.

ABI and CrIS Product Fusion

We now explore expanding the k-d tree search to include more imager spectral bands (when they are available). Using GEO and LEO data, specifically ABI data from 18 UTC and CrIS data from 1748 and 1754 UTC on May 19, 2017, we compare fusion results from k-d tree searches with just the ABI split window bands (11.2 and $12.3\mu \mathrm{m}$) to those from k-d tree searches using eight out of the ten ABI IR bands. The center wavelengths of the eight bands are 6.2, 7.0, 7.3, 8.5, 10.3, 11.2, 12.3, and $13.3\mu \mathrm{m}$, which correspond to ABI bands 8 through 11, and 13 through 16 (i.e., band 7 at $3.9\mu \mathrm{m}$ sensitive to reflected solar radiation and band 12 at $9.6\mu \mathrm{m}$ sensitive to stratospheric ozone are excluded).

Figures 5 and 6 show the results for the 300, 500, and 850 hPa temperature and 500 and 850 hPa ${\mathrm{H}}_2\mathrm{O}$ mixing ratio products. Empty (white) areas in Figs. 5 and 6 delineate cloudy regions; the ABI baseline cloud mask (based on the MVCM of Ref. 12) is applied to the final fusion results to remove cloudy ABI pixels.

Fig. 5

Temperature (K) is shown at 300, 500, and 850 hPa in the top, middle, and bottom panels, respectively; (a) CrIS DR retrievals at sounder resolution, (b) ABI/CrIS fusion results using a two-band (split-window) k-d tree search, and (c) ABI/CrIS fusion results using an eight-band k-d tree search. On May 19, 2017, ABI data are from 1800 UTC and CrIS data are from 1748 and 1754 UTC.

Fig. 6

${\mathrm{H}}_2\mathrm{O}$ mixing ratio (g/kg) is shown at 500 and 850 hPa in the top and bottom panels, respectively; (a) CrIS DR retrievals at sounder resolution, (b) ABI/CrIS fusion results using a two-band (split-window) k-d tree search, and (c) ABI/CrIS fusion results using an eight-band k-d tree search. On May 19, 2017, ABI data are from 1800 UTC and CrIS data are from 1748 and 1754 UTC.

The extra bands in the k-d tree search, especially the ${\mathrm{H}}_2\mathrm{O}$ sensitive 6.2- and $7.0\text{-}\mu \mathrm{m}$ bands, enable better depiction of mid-level (Fig. 6, top) dry and moist features. For example, the two moist pockets seen in the DR retrievals [Fig. 6(a)] are successfully included when using the eight bands [Fig. 6(c)] in the k-d tree search but not when using only the split-window bands [Fig. 6(b)]. Note that the ABI/CrIS relative humidity result at 500 hPa from the split window k-d tree search [Fig. 6(b)] differs somewhat from the VIIRS/CrIS fusion results [Fig. 4(a), bottom] since the split windows for the two instruments are different (the ABI split window has more sensitivity to low-level moisture, which dominates the search). These moisture pockets are somewhat visible in Fig. 4, which shows the VIIRS/CrIS fusion results for the same case, but are less pronounced than the 8-band ABI/CrIS results. The low-level (850 hPa) ${\mathrm{H}}_2\mathrm{O}$ (Fig. 6, bottom) shows minor differences between the split-window versus eight band k-d tree search, implying that the information near the surface is mostly contained in the split-window bands (but the other window bands at 8.5 and $10.3\mu \mathrm{m}$ do contribute some additional information regarding surface features).

3.3.

ABI and CrIS Product Fusion Time Sequence

The potential benefits of using hyperspectral sounder retrievals in the evaluation of atmospheric stability and real-time monitoring of extreme weather events are discussed in Ref. 13. Utilizing hyperspectral retrieval products transferred to high spatial resolution by means of the imager/sounder product fusion method as well as employing a time series of prior and subsequent measurements may further contribute to the improvement of various severe weather forecasting and warning operations. With respect to the ABI/CrIS fusion process, opportunity now arises to perform additional k-d tree searches using ABI radiances every 15 min while connecting to the same fixed time CrIS retrieval product. The high spectral (which translates to high vertical) resolution retrieval products can then be generated not only at high spatial but also at high temporal resolution.

ABI radiances for the moisture sensitive band 8 ($6.3\mu \mathrm{m}$) at 16, 17 18, and 20 UTC on May 19, 2017, are shown in Fig. 7(a), illustrating the movement of mid-level moisture and intensification of clouds from the western to the northeastern part of the region. Figure 7(b) shows the mid-level (500 hPa) humidity ABI/CrIS product fusion results, which have been derived from combining the ABI radiances at these four times with the CrIS soundings at 1748 and 1754 UTC. Figure 7(c) shows the NCEP 13-km rapid refresh (RAP) model analysis for the humidity at 500 hPa for the same four times. The ABI/CrIS product fusion is spanning 4 h, which precede and follow the sounder overpass; it suggests the moisture increase as well as adjacent drying at the 500-hPa pressure level from 16 to 17 UTC (moistening in panel b is outlined by a red frame, drying by the black frame). The mid-level drying is followed by convective activity in eastern Virginia found independently in the ABI image at 20 UTC [convection is highlighted by the black frame in Fig. 7(a)] and in the RAP (although displaced somewhat to the north). Low-level moistening and mid-level drying are used in nowcasting as indicators of preconvective conditions, hence, the rapid changes in 500 hPa ${\mathrm{H}}_2\mathrm{O}$ mixing ratio are of interest in this time sequence. Finally, Fig. 7(d) shows the associated low-level (850 hPa) humidity ABI/CrIS product fusion results. The low-level moisture fields have more missing pixels due to cloud interference than the mid-level moisture results. But they show the low-level moistening from 16 to 17 UTC (eastern box outlined in black) connected with mid-level drying [shown in Fig. 7(b)] that presages convective development; the low-level drying to the west (outlined in red) shows where convection updrafts have already occurred and moisture has been transported to the mid-levels. The color scale has been adjusted to emphasize the low-level-moisture gradients over land. The RAP hourly 850 hPa ${\mathrm{H}}_2\mathrm{O}$ mixing ratio field is shown in Fig. 7(e); small scale features are not readily discernable but the general movement of low-level moisture eastward is apparent.

Fig. 7

ABI band 8 ($6.3\mu \mathrm{m}$) radiance ($\mathrm{mW}{\mathrm{m}}^{-2}{\mathrm{sr}}^{-1}\mathrm{cm}$) is shown in (a), ABI/CrIS 8-band product fusion ${\mathrm{H}}_2\mathrm{O}$ (g/kg) at 500 hPa and 850 hPa is shown in (b) and (d), respectively, whereas the 13-km RAP analysis ${\mathrm{H}}_2\mathrm{O}$ product at 500 and 850 hPa is shown in (c) and (e), respectively. The high (low) moisture feature at 17 UTC inside the red (black) box in (b) and (d) corresponds to the existing cloud development in the red box (new convection in the black box) outlined at 20 UTC in (a).

It should be mentioned that the model analysis incorporates the radiosondes as well as the available satellite data. The errors associated with the retrieval are documented in the literature (see, for example, Ref. 11) and those associated with imager sounder radiance fusion are cited in Ref. 1. The comparison of the moisture gradients with model analysis highlights that both (i.e., model analysis and product fusion) describe the same synoptic situation but the imager sounder fusion provides more small-scale features that are meteorologically consistent with developing weather events.

The results of pursuing fusion of the CrIS tropospheric stability retrieval with the ABI radiances are shown in Fig. 8. The CrIS lifted index (LI) derived for 1748 and 1754 UTC [panel (a)] is projected back to 16 UTC as a fusion product using the ABI radiances [panel (b)]. The operational ABI LI (derived with the time interpolated GFS model first guess) is shown in panel (c) of Fig. 8. The 16 UTC LI fusion product enhances the depiction of the unstable region (black box), where convection occurs 4 h later. The indication from the ABI only product is much less pronounced. This example suggests that the imager/sounder fusion can be accomplished across different times, and therefore can provide the high temporal resolution requisite for predicting the onset and the evolution of severe convection.

Fig. 8

(a) CrIS DR LI (°C) retrievals derived at CrIS measurement times 1748 and 1754 UTC on May 19, 2017. (b) ABI/CrIS fusion LI (°C) from 2 h earlier (using the ABI images at 16 UTC) (with the red and black boxes delineating more stable and less stable atmospheres associated with those in Fig. 7). (c) ABI only LI (°C) at 16 UTC.