Impacts of an Early Morning Low Earth Orbit Observing Platform in a Future Global Observing Network Scenario

2.1 Nature Run

The Nature Run used for this version of the GMAO OSSE is the GEOS-5 Nature Run, commonly referred to as the G5NR. This Nature Run is a two year long free forecast using a version of the GEOS model similar to that used for the MERRA-2 reanalysis (Putman, 2014). The spatial resolution of the G5NR is C1440 on the cube-sphere which is approximately 7 km horizontal resolution, with 72 vertical levels and 30 minute temporal output.

The G5NR has been extensively validated (Gelaro et al., 2014) to evaluate the realism of the model forecast. The overall performance of the NR is good, with some differences in climatology compared to the real world that have been identified. For example, there are some unrealistic polar stratospheric clouds located over the winter pole. These differences from reality are taken into account as much as possible when generating the simulated observations and interpreting the OSSE results.

2.2 Forecast Model

A version of the GEOS model that was quasi-operational in 2022–2023 (version 5.29.4, Rienecker et al. (2008)) was used as the forecast model for these experiments. The resolution of the GEOS model was C360 (25 km) in the horizontal with 72 vertical levels. This model has substantially different physics and dynamics from the version of GEOS used to generate the G5NR including updated advection schemes, boundary layer and land model physics, new convective and radiative schemes, and different damping schemes. Privé et al. (2023a) illustrated the improved performance of this version of the GEOS model in the OSSE framework due to these differences with the Nature Run as insufficient model error is a common problem in OSSEs that affects many aspects of OSSE performance.

The hybrid 4-dimensional ensemble variational (4DEnVar) GSI is used for data assimilation. This version of the GSI was also in quasi-operational use at the GMAO in 2022–2023. The ensemble size for the DAS is 32 members and this version of GSI employs an incremental analysis update (IAU) scheme.

2.3 Synthetic Observations

A “future” global observing network scenario is used as the basis for these experiments. In this scenario the main orbital platforms are NOAA-20, MetOp-B, and a ring of three geostationary platforms with hyperspectral infrared sounders (GEO IRS). While two platforms are most often used in the same LEO orbit for redundancy, the use of single platforms represents a minimum acceptable scenario. Additional remote sensing instrument platforms include GMI GPM, AMSR2 GCOM W1, and SSMIS F17. Conventional data types include atmospheric motion vectors (AMVs), rawinsondes, aircraft, surface types, and GNSS-RO. AMVs are from the Himawari, Meteosat, and GOES geostationary platforms as well as polar MODIS data. Radio occultations are simulated for MetOp-A,-B, and -C, TerraSar-X, TanDEM-X, KOMPSAT-5, and COSMIC-2 (E1, E2, E4, E5, and E6) with approximately 6000 soundings per day. The Future Control case includes these data types but omits the 0530 orbit. A full list of the observation types included in the future global network cases is given in Table 2.

For those observing types that are currently in operational use the synthetic observations were based on archived real observations from June–September 2020 as described fully in Errico et al. (2013), Errico et al. (2017), and Privé et al. (2023a). Spatiotemporal locations for data types aside from AMVs were based on the locations of real observations with cloud contamination of clear-sky types determined by the location of clouds and precipitation in the NR using probabilistic methods (Errico et al., 2020). For infrared clear-sky radiance types, probabilistic functions based on the fraction of low, medium, and high altitude clouds were used to determine if a thick cloud affects each radiance observation. If so, the cloud-top temperature was assigned to be the same as the atmospheric temperature at the same elevation and the CRTM was informed that the effective radiative surface occurred at the cloud-top level. The goal of this was to effect the removal of infrared radiances in areas of dense clouds by the quality control of the DAS so as to result in statistically similar quantities of ingested infrared observations. The effects of thin clouds on infrared radiances were accommodated by the addition of simulated observation errors. For clear-sky microwave radiance types contamination by precipitation is handled similarly, using probabilistic functions with the NR precipitation fields for stratiform, convective, and anvil clouds. All-sky microwave observations were simulated for MHS, AMSR-2, and GMI using the NR cloud and moisture fields. For radiance observations the Community Radiative Transfer Model (CRTM, Han et al. (2006); Ding et al. (2011)) version 2.2.3 was used to generate radiances from profiles taken from the G5NR.

The synthetic observations were calibrated to match characteristics of the corresponding real observations. Simulated observation errors were applied to the synthetic observations with partially correlated errors applied to different data types so that the observation correlations matched the correlations of real data (Errico et al. (2013), Errico et al. (2017)). The probability thresholds used to determine cloud contamination for radiances were adjusted to match the counts of ingested simulated observations to that of real observations for each channel or level. The magnitude of simulated observation errors was adjusted so that the standard deviations of observation innovation (O-B) for the ingested simulated observations matched that of real observations. While most data types were calibrated in a bulk global fashion GNSS-RO had observation errors that were latitude-based in magnitude (Privé et al., 2022). These simulated errors were intended to compensate for the lack of sufficient representativeness error, operator error, instrument error, and also slight cloud contamination for radiances. A fraction of the total error added to the simulated observations was set to be correlated with the remainder being uncorrelated. Horizontally correlated errors were added to AMSU-A, ATMS, MHS, SSMIS, AMVs; vertically correlated errors were added to rawinsondes, GNSS-RO, and AMVs, and channel correlated errors were added to AIRS, IASI, and CrIS-FSR. Simulated biases were not added to the simulated observations but the GSI bias correction was allowed to act normally. A multi-week spinup period was implemented during calibration for spinup of the radiance bias coefficients.

The GEO IRS observations were simulated as described by McGrath-Spangler et al. (2022) and McGrath-Spangler et al. (2024). The three GEO IRS platforms had subpoints at 0°, 105°W, and 140°E with 53 temperature sounding channels and 15 water vapor channels assimilated by the DAS. The channel selection used the MTG-S IRS as a baseline with a triangular spectral response function. Hourly full-disk GEO IRS observations were simulated with cloud contamination tuned to the same probabilistic settings as IASI. Simulated observation errors were not added to the synthetic GEO IRS observations but testing has indicated that the standard deviation of the intrinsic representativeness error of the simulated GEO IRS observations ranges from approximately 0.25 K in the upper troposphere to 1–2 K for mid and lower tropospheric levels.

2.4 Calibration

A validation run (“Real”) was performed ingesting the real observations from 2020 into the same version of the GEOS and GSI in order to provide the target statistics for calibration. A 2020 Control OSSE case was run using the simulated 2020 observing network for validation. The resulting calibration and validation of this version of the GMAO OSSE framework is described in full by Privé et al. (2023a).

FSOI estimates of observation impact for the 2020 Control, Future Control, and Real cases are shown in Figure 3. For the Real case, FSOI is calculated using self-analysis for verification while the OSSE Control cases use the NR for verification which results in a small difference in the impacts but generally does not change the relative ranking of data types (Privé and Errico, 2019). In general, observation impacts in the OSSE tend to be weaker than for real observations due to insufficient model error in the OSSE framework, resulting in less work available for the observations to perform. However, the relative ranking of data types is well represented overall. In the 2020 Control case ATMS ranks relatively high in terms of impact but is slightly underweighted in terms of ranking overall. CrIS-FSR has weaker impacts than ATMS but is accurately ranked in the OSSE. AMVs are overweighted in the OSSE while AMSU-A, AIRS, scatterometer winds, and rawinsondes have weaker relative rank than for Real.

Figure 3

In the Future Control the GEO IRS has a moderately high ranking, comparable to GNSS-RO. The GEO IRS impacts in this study should be considered as the maximum possible limit for this configuration of the instruments with impacts likely to be somewhat reduced with the addition of observation errors. Despite the loss of one platform ATMS remains highly ranked, possibly compensating for the loss of most AMSU-A platforms. Some conventional observations such as rawinsondes, aircraft, and AMVs see a modest increase in impact in the Future Control compared to the 2020 Control along with GNSS-RO.

2.5 0530 Orbit Simulation

In order to simulate a new platform at 0530 the archived 2020 NPP orbit (1330 LTAN) for ATMS and CrIS-FSR was temporally shifted by eight hours when generating the observations. The cloud thinning and simulated error parameters used to calibrate the CrIS-FSR NPP and ATMS NOAA-20 simulated observations were used for simulations of the 0530 observations. These platforms were chosen because they showed the best calibration compared to real data in prior validation of the 2020 OSSE framework. Channel correlated errors were added to the CrIS-FSR observations and horizontally correlated errors were added to the ATMS observations. Both CrIS-FSR and ATMS were assimilated for clear-sky only in accordance with the use of real observations in this version of the DAS. Simple geometric thinning was used with thinning box sizes of 145 km for ATMS and 180 km for CrIS-FSR.

Figure 4 shows calibration of CrIS-FSR and ATMS data on the 0530 orbit compared to the Real validation case. CrIS-FSR has relatively good matching of ingested observation counts in the lower troposphere but lower counts in the upper levels (channels 1–50) and some mismatches in counts in mid-troposphere. The deficiency in counts for stratospheric channels is possibly due to the pre-thinning methods used for simulated data (Errico et al., 2017). ATMS has somewhat lower counts of ingested observations than Real which is thought to be due in part to difficulties in simulating the super-obbing methods used with real ATMS data. The ATMS ingested observation counts for the 0530 orbit are slightly lower than the ATMS-NOAA20 counts which could reflect differing cloud coverage at the respective overpass times. Both CrIS-FSR and ATMS have relatively close matching of observation innovation statistics.

Figure 4

2.6 Experiment Setup

Four experiment cases were performed, each for a three-month period from July to September. The cases were initialized on 1 July after a spinup period in June for the OSSE framework during which the satellite bias coefficients and the background state were allowed to adjust. The full ensemble for the hybrid 4DEnVar was not run for each experiment but was instead taken from a prior OSSE run for the same time period of the G5NR (Privé et al., 2023a). Once-daily extended forecasts were performed, initialized at 0000 UTC, and extended for a 7-day forecast.

The four experiments consist of the Future Control case which did not include a 0530 orbit platform, the CrIS-Only case in which CrIS-FSR on the 0530 orbit was included in addition to the Future Control observation network, the ATMS-Only case in which ATMS on the 0530 orbit was included in addition to Future Control, and the CrIS+ATMS case in which both CrIS-FSR and ATMS on the 0530 orbit were included in addition to Future Control. The observations used are shown in Table 2 for all four cases.

3.1 Analysis Impacts

Analysis impacts are calculated using the root-temporal-mean-square analysis errors (RTMSE), using four times daily analysis states at 0000, 0600, 1200, and 1800 UTC. In order to put the impacts into context with the magnitude of the Future Control analysis errors the impacts are normalized by the RTMSE for the Future Control (RTMSE_Control):

where RTMSE_N is the normalized analysis error impact for a particular experiment case with non-normalized RTMSE_Exp. Therefore a normalized analysis impact of –0.01 corresponds to a 1% reduction in the analysis error compared to the Future Control analysis error.

Figure 5 compares the zonal mean normalized analysis temperature impact versus Future Control of the 0530 orbit platform instruments for the three experiments. The largest beneficial (blue) temperature impacts are seen for the CrIS+ATMS case with impact magnitudes of greater than 5% in some regions such as near the south pole and the upper equatorial troposphere. There is a small region of degradation (orange) near the surface over Antarctica; note that the vertical axis is in native model η level equivalent pressure which is surface-following. Modest beneficial impacts on the order of 0.5–2% are found throughout most of the troposphere.

Figure 5

Comparing Figures 5b and 5c, the ATMS 0530 instrument has overall greater beneficial impacts than CrIS-FSR, especially in the southern hemisphere extratropics. However, CrIS-FSR has a more strongly beneficial impact in the tropical upper troposphere. CrIS-FSR is not assimilated over land or sea ice regions which contributes to the weaker impacts compared to ATMS especially near the south pole as there are minimal CrIS-FSR observations ingested poleward of 60°S.

The specific humidity analysis impacts are shown in Figure 6 with beneficial impacts due to the addition of the 0530 platform except for in the lower stratosphere. ATMS shows the greatest impacts in the lower troposphere and in the southern hemisphere extratropics while CrIS-FSR has the largest beneficial impacts in the upper tropical troposphere and the lower troposphere. The pattern of humidity impacts is somewhat similar to that seen with temperature impacts with the exception of better humidity impacts from ATMS in the boundary layer and degradation in the lower stratosphere. Stratospheric degradation of humidity fields is commonly seen in the GEOS/GSI system from the addition of any new data in the real world, indicating that this may be a feature of the forecast system and not particular to the 0530 ATMS and CrIS-FSR.

Figure 6

Zonal wind analysis impacts are shown in Figure 7. For ATMS the wind impacts are similar to those seen for temperature, with greatest impacts in the southern hemisphere extratropics. However for CrIS-FSR the zonal wind field does not have the local maximum in beneficial impact in the upper tropical troposphere seen for temperature and humidity. Wind is not directly observed by radiance instruments so wind increments are determined by the covariances employed by the DAS. Near the equator the wind and temperature fields are decoupled so the beneficial temperature impacts for CrIS in this region do not translate directly to wind impacts.

Figure 7

In order to evaluate the redundancy of the CrIS-FSR and ATMS observations on the 0530 platform the impacts of the CrIS-Only and ATMS-Only cases can be compared to the CrIS+ATMS case. While it is not expected that observations with completely orthogonal information content would necessarily have purely additive impacts this comparison can give an estimation of the degree to which the observations are complementary. Figure 8 shows the global mean RMS analysis error impacts for each of the three experiment cases compared to Control as well as the sum of impacts in the CrIS-Only (dashed line) and ATMS-Only cases (dotted line). Comparing the CrIS+ATMS case (heavy solid line) to the sum of impacts in the CrIS-Only and ATMS-Only cases (dot-dashed line) the CrIS+ATMS case impacts are in general only slightly weaker than the sum of the individual instrument impacts, implying that redundancy of the two instruments is small. However, humidity below 750 hPa shows much greater impacts from the sum of individual instruments with ATMS-Only impacts actually more beneficial than in the CrIS+ATMS case. This is possibly due to over-weighting of CrIS-FSR water vapor sensitive channels along with undiagnosed error correlations between CrIS-FSR channels.

Figure 8

In order to investigate the effects of filling in a gap in the LEO constellation a smaller region that is only intermittently overpassed by LEO platforms in the Future Control can be examined. Of particular interest is the conterminous United States region (CONUS), here defined for purposes of calculation as a box between 130°W and 65°W, and 25°N and 50°N, indicated by the black box in Figure 2. The 0530 orbit directly overpasses the CONUS region during the 0000 UTC and 1200 UTC cycle times, clips a corner of the CONUS region during the 0600 UTC cycle, and does not cross over the bounding box during the 1800 UTC cycle.

Figure 9 compares the time mean fractional analysis impacts calculated over the CONUS region for the four different cycle times using root-areal-mean-square forecast error (RAMSE) instead of the RTMSE in order to allow for significance testing over multiple forecasts. Levels with statistically significant differences at the 90% level are indicated with heavy line segments. While it might be expected that the analysis impacts would have the greatest magnitudes during the 0000 (black solid line) and 1200 UTC (blue dot-dash line) cycles when there is a direct overpass, with near zero impacts at other times, this is not the case. In the lower troposphere the impacts for all four cycle times are clustered, showing largest beneficial impacts near 900 hPa and 700 hPa for temperature and humidity including for the 1800 UTC cycle time during which no data from the 0530 orbit is ingested in this region. The mid troposphere shows less clustering although the humidity field near 300 hPa shows a peak in beneficial impact at all cycle times. This result implies that some of the information from the overpass times at 0000 and 1200 UTC is retained to the 0600 and 1800 UTC cycles.

Figure 9

3.2 Forecast Impacts

Extended forecasts were generated once daily, initialized at 0000 UTC and integrated for seven days. Regional forecast impacts were calculated for three different regions: 20°N–90°N (NHEX), 20°S–90°S (SHEX), and 20°S–20°N (Tropics). Normalized forecast impacts were calculated similar to the analysis impacts using the NR fields regridded to the same resolution as the forecast for verification. However, for the calculation of fractional forecast impacts the RAMSE was used in order to determine statistical significance. The calculation for normalized forecast error for each region at each forecast time is the same as for Equation 1 but replacing RTMSE with RAMSE. Significance testing was performed using the Student’s t test at the 90% level and accounting for temporal autocorrelation of forecasts (Wilks, 2011).

Figure 10 shows normalized temperature forecast impacts for the three cases in each of the three regions with negative values indicating a reduction in forecast error when the 0530 orbit was included compared to Future Control. The largest beneficial impacts are seen in the SHEX region with improvements of 1–2% extending out to the seven day forecast – these improvements are largely due to observations from the ATMS instrument. This finding of larger, more durable forecast improvements in the SHEX region has been common to previous studies with the same OSSE framework (Privé et al., 2022, 2023b) and is thought to be due to the combination of the sparsity of in situ observations and the strongly baroclinic nature of error growth in the winter hemisphere extratropics.

Figure 10

Forecast improvements are weaker and of shorter duration in the NHEX region and Tropics although there is some improvement in the mid-troposphere temperature of the NHEX region out to six days. While there are initially substantial improvements in the Tropics region for both ATMS and CrIS-FSR these improvements only persist for one to two days into the forecast and some areas of forecast degradation are observed beginning with the 12-hour forecast period. Both the Tropics and summer NHEX region have weak baroclinic dynamics while convection plays a strong role resulting in error growth that has a short timescale with a substantial component of model error growth. In this convective regime improvements to the initial forecast are quickly overwhelmed by fast-acting physics processes in the model. In highly convective regions a model feedback can occur in which even neutral changes to the analysis state set off deep convection, and leading to a delayed model response in which heating and cooling at different vertical levels can induce a bias in the temperature field in the 12hr to mid-range forecast period as seen near 200–300 hPa in the Tropics region and to a lesser degree in the NHEX region. This has been observed for other instrument types in previous OSSEs.

Specific humidity forecast impacts are shown in Figure 11 and overall these impacts persist for longer duration into the forecast period than the temperature impacts. For the CrIS+ATMS and ATMS-Only cases humidity improvements are statistically significant out to day seven in all three regions while for the CrIS-Only case improvements are seen to day three in the extratropics and day four to five in the Tropics. Both the humidity bias and error variance are improved into the medium range (not shown) with somewhat stronger improvements to the error variance compared to the bias. Improvements of up to 1% are seen overall while some larger improvements of 2% are found in the upper and lower troposphere of the short term forecast. The different forecast response for humidity compared to temperatures implies that the additional information ingested from water vapor channels augments the improvements to the humidity fields that result merely from covariance with the temperature fields. The humidity field is also not affected by convective feedback processes in the same way seen for temperatures in the Tropics and summer hemisphere.

Figure 11

Figure 12 shows zonal wind forecast impacts as in Figures 10 and 11. Impacts are very similar for the CrIS+ATMS and ATMS-Only cases with improvements of up to 1% in the NHEX and Tropics regions and impacts up to 2% in the SHEX region, persisting for 4–7 days. The delayed degradations at some vertical levels in the Tropics and NHEX seen for temperatures are not observed for the wind field. For CrIS-Only improvements are weaker and more short-lived, lasting 2–3 days in all three regions.

Figure 12

3.3 Forecast Sensitivity Observation Impact

FSOI uses a linearized, or adjoint, version of the NWP model to provide estimates of the observation impact for all ingested data types on a single metric of forecast error. Because nonlinear terms are neglected FSOI is generally used for forecast errors at 48 hour or shorter time windows. In these experiments a moist adjoint that includes convective processes is used (Holdaway et al., 2014) with the 24-hour forecast of total wet energy error as the forecast error norm, using the Nature Run for verification when calculating errors. Previous studies have shown that at this forecast time length the FSOI captures approximately 75% of the total nonlinear observation impact (Privé et al., 2020) when the Nature Run is used for verification. FSOI is calculated four times daily, at each cycle time.

FSOI produces complementary information to the data-addition tests previously presented. While data-addition shows the impacts of select observations on all aspects of the analysis and forecast states FSOI estimates the impact of all observation types used simultaneously on a single metric – the 24 hour forecast of total wet energy error. Previous work has shown that data denial experiments and FSOI yield roughly similar impact rankings for each observation type (Gelaro and Zhu, 2009). However, FSOI is sensitive to the optimality of the observation errors assumed by the DAS (Eyre, 2024) and may not accurately reflect observation impacts if the assumed errors are smaller in magnitude than the actual observation errors. FSOI can also be subject to numerical instabilities of the linearized model. For any cycle times where instabilities of the FSOI tool were suspected the FSOI results were excluded from calculations for all data types but these instabilities can be more difficult to identify in regional evaluation of FSOI. These limitations should be noted when considering the following FSOI analysis, particularly for regional and channel results.

Figure 13 shows the mean net observation impact for the three regions calculated per cycle time for the three month experiment period in the CrIS+ATMS case where negative values indicate a reduction in the forecast error due to the observations. In the NHEX region where there is substantial landmass surface coverage and extensive commercial aircraft routes it is seen that rawinsondes (RAOB), aircraft, and atmospheric motion vectors (AMVs) are the most impactful data types. ATMS NOAA-20 is the most beneficial radiance data type with the ATMS 0530 having a somewhat weaker impact. CrIS-FSR ranks between AMSU-A and AMSR2 with less overall impact, in line with the findings from the analysis and forecast impacts discussed previously and with the impacts of these data types in the 2020-based observing network.

Figure 13

In the SHEX region where there are few rawinsondes and aircraft ATMS impacts are second only to AMVs with the ATMS NOAA-20 providing greater impacts than the ATMS 0530. CrIS-FSR instruments have impacts similar to IASI and the GeoXO platform located near the Americas. In the Tropics AMVs and GNSS-RO bending angle observations have by far the greatest observation impacts. ATMS and CrIS-FSR have similar impacts in the Tropics. Differences in the impact of similar instruments on different orbits, such as ATMS on the 0530 orbit compared to the NOAA-20 orbit, are believed to be due to variations in both the background errors related to the diurnal cycle and to the presence of other observation types when an orbit passes over a region where sensitivities may be larger or smaller.

FSOI provides granular information about impacts down to the level of individual observations. The global per-channel impacts of the ATMS and CrIS-FSR data on the 0530 orbit are shown in Figure 14. The per-channel impacts for ATMS and CrIS-FSR on NOAA20 (not shown) are quite similar to those for the same instruments on the 0530 platform. For ATMS, the largest impacts are seen for channels 5–9 which are tropospheric temperature channels. Somewhat weaker beneficial impacts are found for channels 17–20 which are water vapor sensitive channels. The CrIS-FSR per-channel FSOI also has the most beneficial impacts for tropospheric temperature channels. Several water vapor sensitive ATMS and CrIS channels have a positive FSOI indicating degradation rather than improvement. Channels 21–22 on ATMS show slightly positive FSOI while CrIS-FSR channels 152, 153, and 1008 (GSI index 117, 118, and 301, respectively) also show indications of degradation. While these values border on significance, given the suspect CrIS-FSR channels are sensitive below 700 hPa, it is suggestive along with reduced impact in water vapor for the ATMS+CrIS case shown Figure 8 that there may be sub-optimal implementation in the DAS regarding water vapor sensitive channels.

Figure 14

The global FSOI impacts for the 0000 UTC cycle time are compared for the four cases in Figure 15. For the ATMS-Only and CrIS+ATMS cases the impact of ATMS-NOAA20 is substantially reduced due to the addition of the ATMS-0530 instrument. Other data types show a small weakening of impact when the 0530 instruments are included especially for the CrIS+ATMS case. This can be due to either or both redundancy between the observations or reduction in error growth due to the improved initial conditions.

Figure 15

In order to understand short-term impacts over the CONUS region the FSOI is broken down by cycle time in Figure 16, along with the observation counts of ingested data types over the CONUS domain. There are strong temporal differences in the availability of some in situ observation types, in particular rawinsondes which are primarily launched at 0000 and 1200 UTC, and a diurnal cycle in aircraft data, with a minimum occurring for the 0600 UTC cycle. The 0530 orbit overpasses directly over CONUS during the 0000 and 1200 UTC cycles, coinciding with the release of rawindsondes. The NOAA-20 orbit overpasses during the 0600 and 1800 UTC cycles while the geostationary GeoXO platform provides similar observation quantities for all cycle times along with surface in situ observations. Note that the FSOI results only pertain to the 24-hour forecast impacts while longer lead times are likely to be affected by non-local upstream observations. It is also important to focus only on general findings and not analyze small differences too closely when examining more granular FSOI results over a smaller region such as this.

Figure 16

The FSOI for GeoXO provides information on the relative impact of data as the atmospheric state and available observations change at each cycle time. The greatest impacts for GeoXO over CONUS occur at the 0600 UTC cycle when there are few rawinsondes and the smallest number of aircraft observations, although the NOAA-20 platform overpasses at this cycle. The weakest GeoXO impacts occur at the 0000 UTC cycle time when there are large quantities of rawinsondes and aircraft. Comparing the ATMS impacts for the 0530 orbit with the NOAA-20 orbit, the observation impacts peak for overpass cycle times of each instrument. However, the 0530 orbit overpasses coincide with the twice-daily rawinsonde release and also periods of high counts of aircraft observations while the NOAA-20 orbit coincides with the cycle times when there are few rawinsondes and when there is a lull in aircraft activity. This contributes to a larger overall impact for the NOAA-20 platform ATMS and CrIS compared to the same instruments on the 0530 orbit.

For comparison, FSOI for the region of Europe as defined by the red box in Figure 2 is shown in Figure 17. For the European region the main overpass times for the 0530 orbit are 0600 and 1800 UTC with partial coverage during the 0000 UTC cycle and minimal coverage for the 1200 UTC cycle. The largest impact for ATMS 0530 and CrIS-FSR 0530 occurs for the 1800 UTC cycle when observation counts are the highest, however, the second largest impact occurs for the 0000 UTC cycle even though the orbital coverage is only partial. This is likely due to the lack of aircraft observations for the 0000 UTC cycle time. The GEOIRS platform, GEO-MTG, only partially covers the Europe region and plays a weaker role than GeoXO plays over CONUS.

Figure 17

Figure 18 shows the accumulated FSOI for the ATMS and CrIS-FSR on the 0530 orbit over the entire three month forecast period, binned by location, as well as the mean count of ingested radiance observations per bin per cycle. The total accumulated FSOI for all other conventional and remote sensed data types aside from the 0530 orbit and associated observation counts are shown in Figure 18e and f for comparison. Regions of greater magnitude FSOI for the baseline set of observations may indicate a combination of both sensitivity of the forecast to observations as well as coverage by the global observing network. Continental areas that are well-observed by in situ observations, such as the CONUS, Europe, and Australia, may be less sensitive to any particular observation type due to the redundancy of the observing network.

Figure 18

In the tropics and subtropics both ATMS and CrIS-FSR on the 0530 platform have regional areas of enhanced FSOI impact that are similar to the distribution for the baseline observing network – the eastern Pacific off the coast of Mexico, the Atlantic off the coast of West Africa, and the Indian Ocean. The Indian Ocean feature is especially prominent for CrIS-FSR, with a somewhat weaker local extrema in this area for ATMS, and is strongest in July and August (not shown), hinting at possible relation to the Asian monsoon circulation. The Indian Ocean also falls into a gap between the GEO Ring of hyperspectral infrared sounders as can be inferred from Figure 18f. ATMS has stronger FSOI impacts in the Southern Ocean compared to most of the rest of the globe while for CrIS-FSR the FSOI extrema over the subtropical and tropical oceans are stronger than over the Southern Ocean. This difference between the two platforms is likely due to the reduced number of CrIS-FSR observations between 40–60S where IR observations are affected by clouds in active convective regions.