Implementation of All-Sky Assimilation of Microwave Humidity Sounding Channels in Environment Canada’s Global Deterministic Weather Prediction System

Maziar Bani Shahabadi aMeteorological Research Division, Environment and Climate Change Canada, Victoria, British Columbia, Canada

Search for other papers by Maziar Bani Shahabadi in
Current site
Google Scholar
PubMed
Close
and
Mark Buehner aMeteorological Research Division, Environment and Climate Change Canada, Victoria, British Columbia, Canada

Search for other papers by Mark Buehner in
Current site
Google Scholar
PubMed
Close
Open access

Abstract

Cloud-affected microwave humidity sounding radiances were excluded from assimilation in the hybrid four-dimensional ensemble–variational (4D-EnVar) system of the Global Deterministic Prediction System (GDPS) at Environment and Climate Change Canada (ECCC). This was due to the inability of the current radiative transfer model to consider the scattering effect from frozen hydrometeors at these frequencies. In addition to upgrading the observation operator to RTTOV-SCATT, quality control, bias correction, and 4D-EnVar assimilation components are modified to perform all-sky assimilation of Microwave Humidity Sounder (MHS) channel 2–5 observations over ocean in the GDPS. The input profiles to RTTOV-SCATT are extended to include liquid cloud, ice cloud, and cloud fraction profiles for the simulation and assimilation of MHS observations over water. There is a maximum (35%) increase in the number of channel 2 assimilated MHS observations with smaller increases for channels 3–5 in the all-sky experiment compared to the clear-sky experiment, mostly because of newly assimilated cloud-affected observations. The standard deviation (stddev) of difference between the observed global positioning system radio occultation (GPSRO) refractivity observations and the corresponding simulated values using the background state was reduced in the lower troposphere below 9 km in the all-sky experiment. Verifications of forecasts against the radiosonde observations show statistically significant reductions of 1% in the stddev of error for geopotential height, temperature, and horizontal wind for the all-sky experiment between 72- and 120-h forecast ranges in the troposphere in the Northern Hemisphere domain. Verifications of forecasts against ECMWF analyses also show small improvements in the zonal mean of stddev of error for temperature and horizontal wind for the all-sky experiment between 72- and 120-h forecast ranges. This work was planned for operational implementation in the GDPS in fall 2023.

For information regarding reuse of this content and general copyright information, consult the AMS Copyright Policy (www.ametsoc.org/PUBSReuseLicenses).

Corresponding author: Maziar Bani Shahabadi, maziar.banishahabadi@ec.gc.ca

Abstract

Cloud-affected microwave humidity sounding radiances were excluded from assimilation in the hybrid four-dimensional ensemble–variational (4D-EnVar) system of the Global Deterministic Prediction System (GDPS) at Environment and Climate Change Canada (ECCC). This was due to the inability of the current radiative transfer model to consider the scattering effect from frozen hydrometeors at these frequencies. In addition to upgrading the observation operator to RTTOV-SCATT, quality control, bias correction, and 4D-EnVar assimilation components are modified to perform all-sky assimilation of Microwave Humidity Sounder (MHS) channel 2–5 observations over ocean in the GDPS. The input profiles to RTTOV-SCATT are extended to include liquid cloud, ice cloud, and cloud fraction profiles for the simulation and assimilation of MHS observations over water. There is a maximum (35%) increase in the number of channel 2 assimilated MHS observations with smaller increases for channels 3–5 in the all-sky experiment compared to the clear-sky experiment, mostly because of newly assimilated cloud-affected observations. The standard deviation (stddev) of difference between the observed global positioning system radio occultation (GPSRO) refractivity observations and the corresponding simulated values using the background state was reduced in the lower troposphere below 9 km in the all-sky experiment. Verifications of forecasts against the radiosonde observations show statistically significant reductions of 1% in the stddev of error for geopotential height, temperature, and horizontal wind for the all-sky experiment between 72- and 120-h forecast ranges in the troposphere in the Northern Hemisphere domain. Verifications of forecasts against ECMWF analyses also show small improvements in the zonal mean of stddev of error for temperature and horizontal wind for the all-sky experiment between 72- and 120-h forecast ranges. This work was planned for operational implementation in the GDPS in fall 2023.

For information regarding reuse of this content and general copyright information, consult the AMS Copyright Policy (www.ametsoc.org/PUBSReuseLicenses).

Corresponding author: Maziar Bani Shahabadi, maziar.banishahabadi@ec.gc.ca

1. Introduction

Satellite radiances are the dominant observation types for numerical weather prediction (NWP) and are heavily relied upon to produce the initial state of the atmosphere, especially in remote areas where there is less coverage from conventional observations. Traditionally, satellite observations have only been assimilated in clear-sky condition due to limitations in the forecast model and radiative transfer model to simulate cloud-affected radiances.

In the past decade, NWP centers dedicated much effort to make use of cloudy radiances with success. The European Centre for Medium-Range Weather Forecasts (ECMWF) implemented all-sky microwave radiance assimilation from the Special Sensor Microwave Imager/Sounder (SSMIS) and the Advanced Microwave Scanning Radiometer for the Earth Observing System (AMSR-E) in 2009 (Bauer et al. 2010). The so-called all-sky framework is based on the use of a symmetric observation error model to increase the observation error standard deviation (stddev) as a function of the cloud amount for assimilation of cloud-affected radiances. All-sky assimilation of observations from four Microwave Humidity Sounder (MHS) satellites and one SSMIS satellite improved the dynamical forecasts in tropical and midlatitude regions up to 6 days in the 4D-Var system at ECMWF (Geer et al. 2014). All-sky assimilation of cloud-affected MHS observations in the 4D-Var system at the Met Office improved forecasts of humidity and wind fields in the extratropical regions during boreal winter and summer seasons (Candy and Migliorini 2021). Lee et al. (2020) reported 1.11% reduction in the specific humidity root-mean-square error (RMSE) of the 6-h forecasts with all-sky assimilation of MHS observations in the hybrid four-dimensional ensemble–variational (4D-EnVar) system at the Korean Integrated Model (KIM) forecast system.

At Environment and Climate Change Canada (ECCC), our first all-sky attempt was the assimilation of Advanced Microwave Sounding Unit-A (AMSU-A) temperature channels 4–5 for nonprecipitating scenes over ocean (Shahabadi and Buehner 2021). This was implemented in the 4D-EnVar system of the Global Deterministic Prediction System (GDPS) at ECCC’s third Innovation Cycle (IC-3) in fall 2021. At microwave humidity sounding frequencies, the scattering from hydrometeors becomes an important mechanism that needs to be included in the observation operator. Including this process in the observation operator requires additional cloud and humidity information, which could lead to a more complex observation operator and an increase in the size of the control vector during the assimilation stage. Up until now, the microwave cloud-affected humidity sounding radiances were excluded from assimilation in the 4D-EnVar system due to these complexities. This study describes the modifications to overcome the above challenges and demonstrates the results from extending the all-sky framework to utilize cloud-affected MHS humidity channels 2–5 over ocean.

The current use of MHS observations in the operational GDPS is described in section 2. Section 3 describes the changes made to the current operational system for the all-sky assimilation of MHS observations. Section 4 discusses the results from all-sky assimilation experiments. Conclusions and future work plans are presented in section 5.

2. Usage of MHS observations in the operational Global Deterministic Prediction System

MHS observations are assimilated in clear-sky mode in the current 4D-EnVar system of the GDPS. Table 1 provides details on the satellites with onboard MHS instrument and the channel-specific information. MHS channels are treated similarly between all the satellites, except for NOAA-19 where only channel 2 is assimilated. This choice is based on continuous data monitoring at ECCC and is beyond the scope of this work. The RTTOV model, version 13 (Saunders et al. 2018), is used for the simulation and assimilation of MHS observations. Background state fields are interpolated in time and space to the observation time and location to construct the input profiles to RTTOV. At ECCC, the slant-path interpolation for satellite radiances (Shahabadi et al. 2018, 2020) was implemented in the 4D-EnVar system of the GDPS during IC-3 and is included in this study.

Table 1.

Satellites (MetOp-1, MetOp-2, MetOp-3, NOAA-18, and NOAA-19) with onboard MHS instruments used in this work, characterizations of different MHS channels, and the status of assimilation. Polarization: V = vertical; H = horizontal.

Table 1.

a. Quality control

As the focus of this study is the all-sky assimilation of MHS observations over water, the quality control of the MHS observations over open water is described. Channel 2–5 MHS observations are assimilated in the 4D-EnVar system: channels 2 and 5 are assimilated only over open water, whereas channels 3 and 4 are assimilated over land and water. The scattering index (SI; K) is used for quality control of the MHS observations and is computed as follows over open water:
SI=BT1BT2(39.2010+0.1104×θ),
where BT1 and BT2 are channel 1 and 2 brightness temperatures at 89 and 150 GHz, and θ is the observation zenith angle. If SI over water is greater than 15, channels 2–5 are rejected. In addition to the check for SI, the rogue check is performed for quality control of the MHS observations: for any of channels 2–5, if the absolute difference between the observed and brightness temperatures simulated from the background state (OMB) is greater than a constant inflation factor (IFrogue) times a historical global estimate of stddev(OMB) (σhist), the observation is rejected. Table 2 shows the IFrogue and σhist values used for MHS MetOp-1 in operational data assimilation. Compared to the tropospheric humidity channels 4–5, a smaller IFrogue was initially assigned to window channel 2 to assimilate this channel more conservatively in the operational system. As will be discussed later, IFassim is used to estimate the observation error stddev at the analysis step. In addition, if channel 2 abs(OMB) > 5 K, channel 2–5 MHS observations are rejected.
Table 2.

Constant inflation factors for rogue check (IFrogue), assimilation (IFassim), and historical global estimate of the stddev(OMB) (σhist) used for MHS MetOp-1 rogue check in operational data assimilation. IF and σhist are channel and instrument dependent.

Table 2.

The spatial thinning is applied to the set of observations from each satellite that are not flagged for rejection in the quality control stage. Similar to the treatment of other satellite radiance observations, the observations are grouped together in each 15-min interval within the 6-h assimilation window and the spatial thinning is applied separately for each group increasing the minimum distance between observations within each group to 150 km.

b. Bias correction

At ECCC, dynamic bias correction coefficients are used to remove biases from satellite radiances. As described in Buehner et al. (2015), the bias correction component involves assimilation of nonradiance observations in a simplified variational analysis system to generate unbiased analyses. A linear fit of the difference between these analyses and the quality controlled and thinned radiance observations of the previous 7 days to a set of bias model predictors determines the bias correction coefficients. The bias correction coefficients are computed and applied for each instrument and channel separately. The bias model predictors for satellite radiance observations are 1000–300-, 200–50-, and 50–5-hPa geopotential height thicknesses and a scan-dependent bias.

c. 4D-EnVar assimilation system

The GDPS uses the 4D-EnVar assimilation system (Buehner et al. 2015). The background state and analysis increment horizontal resolutions are 15 and 39 km, respectively. The forecast model and 4D-EnVar assimilation system run on 85 hybrid levels, extending from the surface to 0.1 hPa. The analyzed variables are the noncloud state variables including horizontal winds, temperature, humidity, surface pressure, and surface skin temperature. Since IC-3, a background state liquid cloud variable is added, but only used for the simulation and all-sky assimilation of AMSU-A temperature channels 4–5 for nonprecipitating scenes over ocean in the GDPS (Shahabadi and Buehner 2021). There is no outer loop in the 4D-EnVar system, and because the background state clouds are largely overestimated at ECCC, a 0.5 global multiplicative factor is applied to the liquid cloud profile for the simulation and assimilation of AMSU-A observations in all-sky mode. The choice of 0.5 cloud scaling factor was based on sensitivity experiments performed during IC-3 by running multiple assimilation cycles with different cloud scaling factors. The factor of 0.5 resulted in the smallest low-tropospheric mean temperature and humidity biases.

The 4D-EnVar background error covariances are a blend of flow-dependent and homogeneous and isotropic components. The flow-dependent component is obtained from ensembles of short-term forecasts produced by a 256-member operational global local ensemble transform Kalman filter (LETKF; Buehner 2020) since IC-3. The homogeneous and isotropic component is from the so-called National Meteorological Center (NMC) method (Parrish and Derber 1992). With the introduction of scale-dependent localization in the GDPS in IC-3 (Caron and Buehner 2022), the flow-dependent and homogeneous isotropic weights of the background error covariances are 1 and 0 below 60 hPa, 1 and 0.25 above 20 hPa, and changing linearly between 60 and 20 hPa, respectively. Since no analysis increment is produced for liquid cloud as a result of all-sky assimilation of AMSU-A observations, liquid cloud is not included in the ensembles of short-term forecasts, nor is it part of the homogeneous and isotropic background covariance matrix.

Approximately 13 million observations per day are assimilated in the current GDPS, which includes radiosondes, aircrafts, land stations, ships and buoys, scatterometers, atmospheric motion vectors, satellite-based radio occultation, ground-based GPS instruments, and microwave and infrared satellite sounders and imagers. The length of the assimilation window is 6 h.

The observation error stddev for MHS and other clear-sky radiances at the analysis step is IFassim × σhist (Table 2), which is channel and instrument dependent. The observation error stddev for all-sky assimilation of AMSU-A channel 4–5 observations is the corresponding IFassim times the cloud-dependent stddev(OMB) [σall in Eq. (3) of Shahabadi and Buehner 2021]. A new model will be used to derive the observation error stddev for all-sky assimilation of MHS channels, which will be discussed in further detail in the following section.

3. Changes for all-sky assimilation of MHS observations

The current all-sky implementation treats all the satellites with onboard MHS instrument similarly. Simulation and assimilation of MHS observations are made using the new RTTOV-SCATT version 13 observation operator, which accounts for the multiple scattering from hydrometeors at microwave frequencies (Bauer et al. 2006; Geer et al. 2021). RTTOV-SCATT could compute scattering and absorption from multiple hydrometeors such as liquid cloud, ice cloud, rain, snow, and graupel. For the work presented here, the effects from rain, snow, and graupel are ignored. The input profiles to RTTOV-SCATT are extended to include liquid cloud, ice cloud, and cloud fraction for the simulation and assimilation of MHS observations over water. Consistent with all-sky assimilation of AMSU-A channels 4–5 at ECCC (Shahabadi and Buehner 2021), the cloud profiles are scaled by the 0.5 global multiplicative factor before they are used in the observation operator.

a. New observation error model

Geer and Bauer (2011) used a state-dependent observation error model as a function of symmetric cloud amount to define the observation error stddev. The symmetric cloud amount is the averaged amount of cloud from model and observations. The SI, defined as the difference between MHS brightness temperature of channels 1 and 2 (BT1 − BT2), was used to determine the cloud amount. Geer et al. (2014) suggested removing the water vapor absorption contribution for computing the SI over water:
SI=(BT1BT2)(BT1clrBT2clr).
The first term in parentheses is computed using either simulated or observed brightness temperatures. The second term is the water vapor absorption contribution and is computed from clear-sky model profiles. We adopt the same methodology: SI from observed and simulated brightness temperatures is computed using Eq. (2) and averaged to obtain symmetric SI¯. Similar to Geer et al. (2014), the piecewise linear fit to the stddev(OMB) binned by SI¯ is defined as
σall={σ1(SI¯SI1)σ1+λ(SI¯SI1)(SI1<SI¯<SI2)σ2(SI¯SI2),
where
λ=σ2σ1SI2SI1.
SI1 and SI2 are the clear and cloudy symmetric cloud amount thresholds and σ1 and σ2 are the clear and cloudy stddev(OMB) used to define linear fit. The values for these parameters are shown in Table 3.
Table 3.

Observation error model parameters used to define piecewise linear fit to the stddev(OMB) in MHS all-sky assimilation.

Table 3.

An inflation factor times piecewise linear fit σall is used as the observation error stddev for quality control and assimilation. Figure 1 shows the stddev(OMB) (in blue) and the linear fit (in red) as functions of SI¯ for MHS channels 2–5. Data are collected for the period from 15 June to 1 August 2019 with the green line showing the number of observations used to compute the stddev(OMB) in each SI¯ bin. In Table 3, the choice of SI1 and SI2 values is guided by the linear fit shown in Fig. 1. The constant σall is prescribed at large SI¯ values to ensure observations with large SI¯ are not assimilated since the sample size (i.e., green line in Fig. 1) used to construct the error model is not large enough. This decision is to focus on the bulk of observations with lower SI¯ values for the best performance of the observation error model.

Fig. 1.
Fig. 1.

The stddev of OMB (K; blue) for the period from 15 Jun to 1 Aug 2019 as a function of SI¯ (K) for MHS channels (a) 2, (b) 3, (c) 4, and (d) 5. The SI¯ bin width of 5 K is used to generate the plots. The linear fit (red) and data count in each bin (green) are also shown.

Citation: Monthly Weather Review 152, 4; 10.1175/MWR-D-23-0227.1

The SI¯ can range from few kelvins that indicates the presence of some scattering due to ice particles to 50 K that is associated with deep convection.

b. Quality control

MHS observations with large SI¯ values are usually associated with large OMB values. These observations are often rejected in the rogue check test. For additional safety to avoid assimilating MHS observations with extreme SI¯ values, only observations within 5.0K<SI¯<20.0K range (vertical dashed line in Fig. 1) are allowed for assimilation in all-sky mode.

The rogue check test is also modified for all-sky assimilation: if the observation abs(OMB) is greater than the clear-sky inflation factor used operationally (IFrogue in Table 2) times the fit to the stddev(OMB) [σall in Eq. (3)], the observation is rejected. The clear-sky quality control for rejection of MHS observations based on channel 2 [reject channels 2–5 if channel 2 abs(OMB) > 5 K] is relaxed in all-sky quality control: if the rogue check test rejects channel 2 [abs(OMB) > IFrogue × σall], channel 2–5 MHS observations are rejected.

c. Bias correction

Geer et al. (2018) showed that the accurate results from all-sky assimilation of MHS were achieved without considering the cloud-dependent bias correction for radiances. This is also consistent with all-sky assimilation for AMSU-A channels at ECCC (Shahabadi and Buehner 2021) and all-sky assimilation for MHS channels at the KMA assimilation system (Lee et al. 2020). Hence, the bias correction coefficients for MHS observation are calculated using only the clear-sky MHS observations. A profile is marked as clear sky if 5K<SI¯<5K.

d. The 4D-EnVar assimilation system

The observation error stddev used for MHS all-sky assimilation is the clear-sky error inflation factor (IFassim in Table 2) times the fit to the stddev(OMB) [σall in Eq. (3)].

Similar to all-sky assimilation for AMSU-A channels, cloud profiles (liquid cloud, ice cloud, and cloud fraction) are from the background state and are not modified during the assimilation. Since no analysis increment is produced for the cloud fields as the result of all-sky assimilation of MHS observations, cloud fields are not part of the ensembles, nor the homogeneous and isotropic background covariance matrix. Compared to the clear-sky assimilation, the changes in the analyzed state from the all-sky assimilation of MHS observations are due to mapping the cloud-affected MHS radiances to the noncloud state variables through the Jacobians of the observation operator.

4. Results

a. Experiment setup

Global all-sky assimilation experiments were conducted for the two periods of 13 June–31 August 2019 (summer 2019) and 12 December–29 February 2020 (winter 2020). For each period, the control experiment is ECCC’s GDPS operational configuration where MHS observations are assimilated in clear-sky mode. The clear-sky and all-sky experiments are launched from same initial conditions.

b. Impact on data count and fit of background state to the observations

The MHS assimilated observations in the all-sky experiment provide more uniform spatial coverage of the globe than the clear-sky experiment, where many MHS observations are rejected in the tropical and subtropical cloudy regions. This is demonstrated in Fig. 2 for channel 5 MHS observations in 2° × 2° latitude–longitude grid from 13 to 26 June 2019.

Fig. 2.
Fig. 2.

Number of assimilated MHS channel 5 observations in 2° × 2° latitude–longitude grid in (a) clear-sky and (b) all-sky experiments from 13 to 26 Jun 2019. The total number of assimilated MHS channel 5 observations during this period for each experiment is also shown.

Citation: Monthly Weather Review 152, 4; 10.1175/MWR-D-23-0227.1

The percentage change in the number of assimilated observations over open water in the all-sky experiment compared to the clear-sky experiment is shown for different channels of MHS (Fig. 3a), AMSU-A (Fig. 3b), and Advanced Technology Microwave Sounder (ATMS; Fig. 3c) in green curve. The plots also show the stddev(OMB) of assimilated observations over water in all-sky experiment normalized by the corresponding quantity in the clear-sky experiment in black curve. The stddev values are rounded to the nearest 10−4 causing the apparent quantization of the plotted normalized values. The statistical significance (shown as filled black dots) is computed with the F test and 95% confidence level. The results are for summer 2019, but winter 2020 yields similar results (not shown).

Fig. 3.
Fig. 3.

The normalized stddev(OMB) [stddev(OMB)allSky/stddev(OMB)clearSky × 100; black] and number of assimilated observations (numObsallSky/numObsclearSky × 100; green) over open water for (a) MHS, (b) AMSU-A, and (c) ATMS. The filled black dots denote the statistical significance computed with the F test at 95% confidence level.

Citation: Monthly Weather Review 152, 4; 10.1175/MWR-D-23-0227.1

For MHS, there is a maximum (35%) increase in the number of channel 2 assimilated observations with smaller increases for channels 3–5 in the all-sky experiment compared to the clear-sky experiment over open water (Fig. 3a, green curve) mostly because of newly assimilated cloud-affected observations. The normalized stddev(OMB) of 220% for MHS channel 2 (Fig. 3a, black curve) means there is 120% increase in the stddev(OMB) in the all-sky experiment. The change in the normalized stddev(OMB) in the all-sky experiment is statistically significant for all MHS channels. There are many changes from clear-sky to all-sky framework that can affect the fit of background state to the observations [e.g., substantial increase in the subset of assimilated observations where these additional observations are predominantly assumed to be cloudy, resulting in an increase in the stddev(OMB) in the all-sky experiment], which makes it difficult to derive conclusions from the change of the fit of background state to the observations between the clear-sky and all-sky experiments (Geer et al. 2012; Zhu et al. 2016; Shahabadi and Buehner 2021). As we see later, the assimilation of the new cloud-affected MHS observations have a positive impact on the forecast in the all-sky experiment even though the stddev(OMB) of the assimilated MHS observations seems to be increased.

For low-peaking AMSU-A temperature channels 4–5, there is a slight reduction in the number of assimilated observations over open water and reduction in the stddev(OMB) in the all-sky experiment (Fig. 3b). For AMSU-A channels 8 and 9, there is a slight increase and decrease in the stddev(OMB), respectively, in the all-sky experiment. For ATMS temperature channels 5–12, there is a small 0.2% increase in the stddev(OMB) for channel 5 observations and a small 0.1% decrease in the stddev(OMB) for the rest of the temperature channels 6–12. For ATMS humidity channels 17–22, there is up to 0.3% increase in the number of assimilated observations over open water and up to 0.8% increase in the stddev(OMB) in the all-sky experiment. This large addition of ATMS observation for the water vapor channels is a strong indication that the model state has moved closer to these observations, again assumed to have some cloud impact, and has allowed for their assimilation.

There is no statistically significant change in the fit of the background state to radiosonde observations between clear-sky and all-sky experiments (results not shown). This is expected since most of the radiosondes are over land and do not sample ocean region where the new MHS observations are assimilated.

The stddev(OMB) for the global positioning system radio occultation (GPSRO) refractivity observations over open water are compared between clear-sky and all-sky experiments: the stddev(OMB) of the assimilated GPSRO refractivity observations in the all-sky experiment is normalized by the corresponding quantity in the clear-sky experiment for summer 2019 (Fig. 4a) and winter 2020 (Fig. 4b) evaluation periods. The observations are grouped in 3-km-thick layers between 0- and 40-km altitudes above mean sea level (MSL). There are around 150 000 GPSRO observations below 3 km, while there are 260 000–290 000 in the layers between 3 and 6 km and between 6 and 9 km. The statistical significance (shown as filled black dots) is computed with the F test and 95% confidence level.

Fig. 4.
Fig. 4.

The normalized stddev(OMB) [stddev(OMB)allSky/stddev(OMB)clearSky × 100] of the assimilated GPSRO refractivity observations over water for (a) summer 2019 and (b) winter 2020 evaluation periods. The observations are grouped in 3-km-thick layers between 0- and 40-km altitudes above MSL. The filled black dots denote the statistical significance computed with the F test at 95% confidence level.

Citation: Monthly Weather Review 152, 4; 10.1175/MWR-D-23-0227.1

There are significant reductions in the stddev(OMB) for the GPSRO refractivity observations in the lower troposphere for the two evaluation periods: 99.5% and 99.4% normalized stddev(OMB) in summer 2019 and winter 2020 periods below 9 km.

c. Impact on forecasts

For each evaluation period, the forecasts were launched from the analyzed initial condition at 0000 and 1200 UTC and were evaluated against the radiosonde observations and ECMWF analyses. The verifications against the radiosondes do not show differences between clear-sky and all-sky experiments in lead times smaller than 48 h, but there are statistically significant improvements in the all-sky experiment between 72- and 120-h forecast ranges. Figure 5 shows the stddev of the radiosondes minus forecast for geopotential heights in the all-sky experiment normalized by the corresponding value from the clear-sky experiment in the Northern Hemisphere domain for the 72- and 120-h forecast ranges for winter 2020 period. The stddev values are rounded to the nearest 10−2. The filled black dots show statistically significant changes above 90% confidence level.

Fig. 5.
Fig. 5.

The stddev of radiosondes minus forecasts for geopotential heights in the all-sky experiment normalized by the corresponding value from the clear-sky experiment [stddev(geopotentialHeight)allSky/stddev(geopotentialHeight)clearSky × 100] in the Northern Hemisphere domain at (a) 72- and (b) 120-h forecast ranges. The filled black dots denote the statistical significance with 90% confidence level. The experiments are from winter 2020 evaluation period.

Citation: Monthly Weather Review 152, 4; 10.1175/MWR-D-23-0227.1

There is 1% reduction in the stddev in 72- and 120-h forecast ranges in the troposphere for geopotential height in the all-sky experiment. Similar magnitudes of reductions in the stddev in 72- and 120-h forecast ranges in the troposphere exist for temperature (Fig. 6) and horizontal wind (Fig. 7) in the all-sky experiment for winter 2020. The comparisons with radiosondes showed a consistent positive influence in a broad tropospheric region for geopotential height, temperature, and horizontal wind.

Fig. 6.
Fig. 6.

As in Fig. 5, but for temperature.

Citation: Monthly Weather Review 152, 4; 10.1175/MWR-D-23-0227.1

Fig. 7.
Fig. 7.

As in Fig. 5, but for horizontal wind.

Citation: Monthly Weather Review 152, 4; 10.1175/MWR-D-23-0227.1

Comparing the forecasts against radiosonde humidity observations does not show statistically significant differences between the experiments, and the results are not shown. Verifications against radiosondes during summer 2019 period show similar results (not shown). The errors in temperature field can occur over broad areas and be long-lived. In contrast, errors in humidity fields are typically present at smaller scales and are more short-lived. The radiosondes over land pick up the signal of the long-lived advected temperature improvements from all-sky assimilation of MHS observations over open water, whereas they have difficulty showing statistically significant signal of the short-lived noisy advected humidity fields. For forecast verifications against the analyses, the scorecard approach, where the relative changes in RMSE with respect to the control experiment for different variables, levels, and forecast ranges are graphically shown, is used. For verifications against ECMWF analyses, the scorecards for global domain do not show statistically significant differences between clear-sky and all-sky experiments (not shown). The zonal mean of stddev difference against ECMWF analyses show improvements between 48- and 120-h forecast ranges in the all-sky experiment. Figure 8 shows the normalized difference in the zonal mean of stddev of error for temperature between the clear-sky and all-sky experiments against ECMWF analyses, defined as [mean(σclearSky) − mean(σallSky)]/mean(σclearSky) × 100, for winter 2020. The improvements and degradations in the all-sky experiment are shown in red and blue, with dots showing statistically significant results above 90% confidence level.

Fig. 8.
Fig. 8.

Normalized difference in the zonal mean of stddev of error for temperature between clear-sky and all-sky experiments compared against ECMWF analyses, defined as [mean(σclearSky) − mean(σallSky)]/mean(σclearSky) × 100, at (a) 24-, (b) 48-, (c) 72-, and (d) 120-h forecast ranges during winter 2020 evaluation period. The red (blue) color shading indicates that the all-sky (clear-sky) experiment has better performance. The dots denote statistically significant results above 90% confidence level from the F test.

Citation: Monthly Weather Review 152, 4; 10.1175/MWR-D-23-0227.1

There are 3%–7% statistically significant reductions in the zonal mean of the stddev of error for temperature in tropospheric northern and southern extratropical regions in the all-sky experiment at 48-, 72-, and 120-h forecast ranges for winter 2020. Figure 9 shows similar forecast improvements in the all-sky experiment for horizontal wind.

Fig. 9.
Fig. 9.

As in Fig. 8, but for horizontal wind.

Citation: Monthly Weather Review 152, 4; 10.1175/MWR-D-23-0227.1

There are some degradations in 60°–90°S and 80°–90°N in 120-h forecasts for temperature (Fig. 8) and horizontal wind (Fig. 9). Comparing the forecasts against ECMWF humidity analyses does not show statistically significant differences between the experiments, and the results are not shown. Similar results were obtained for verifications against ECMWF analyses during summer 2019 period (not shown). It is evident that the forecast verifications against ECMWF analyses showed a neutral to positive impact with a predominantly positive signal when excluding the higher latitudes. It is important to note that there is spatial averaging done for verifications against analyses that could contribute to different impacts seen here than in verifications against radiosondes.

5. Summary and future work

This study presents the results from extending the assimilation of MHS observations to include cloud-affected observations in the 4D-EnVar system in the Global Deterministic Prediction System (GDPS) at Environment and Climate Change Canada (ECCC). The method is referred to as “all-sky” assimilation and was implanted during ECCC’s third Innovation Cycle (IC-3) in fall 2021 for AMSU-A temperature channels 4–5 for nonprecipitating scenes over ocean. The focus of this study is to utilize the cloud-affected MHS humidity channels 2–5 over ocean. This involves modifications of the radiance observation operator, observation error statistics, quality control, bias correction, and the 4D-EnVar assimilation system in the GDPS.

For simulation and assimilation of MHS observations in all-sky mode, 1) the RTTOV-SCATT observation operator is used to include the scattering effect from cloud hydrometeors and 2) in addition to the existing noncloud state variables, background state cloud variables, including liquid cloud, ice cloud, and cloud fraction, are used. The observation error model introduced in Geer and Bauer (2011) was adopted to define the observation error stddev for the quality control and assimilation of MHS observations in all-sky mode.

Global all-sky assimilation experiments were conducted for the two periods of 13 June–31 August 2019 (summer 2019) and 12 December–29 February 2020 (winter 2020). For each period, the control experiment is ECCC’s GDPS operational configuration where MHS observations are assimilated in clear-sky mode.

The MHS assimilated observations in the all-sky experiment provide more uniform spatial coverage than the clear-sky experiment in the tropical and subtropical cloudy regions. This leads to a maximum (35%) increase in the number of channel 2 assimilated MHS observations with smaller increases for channels 3–5 in the all-sky experiment compared to the clear-sky experiment over open water, mostly because of newly assimilated cloud-affected observations. The stddev of the difference between the observed MHS channel 2–5 brightness temperatures and the corresponding simulated values using the background state [stddev(OMB)] were increased in the all-sky experiment [normalized stddev(OMB) of 220% for channel 2]. There are many changes from clear-sky to all-sky framework that can affect the fit of background state to the observations (e.g., substantial increase in the subset of assimilated observations), which makes it difficult to derive conclusions from the change of the fit of background state to the observations between the clear-sky and all-sky experiments. There is also a statistically significant reduction in the stddev(OMB) for the GPSRO refractivity observations over water in the lower troposphere for the two evaluation periods: 99.5% and 99.4% normalized stddev(OMB) in summer 2019 and winter 2020 periods below 9 km.

For each evaluation period, the forecasts were launched from the analyzed initial condition at 0000 and 1200 UTC and were evaluated against the radiosonde observations and ECMWF analyses. Forecast verifications against the radiosondes show 1% statistically significant reduction in the stddev of error for the all-sky experiment between 72- and 120-h forecast ranges for geopotential height, temperature, and horizontal wind in the troposphere in the Northern Hemisphere domain. The zonal mean of stddev differences of temperature and horizontal wind in northern and southern extratropical regions against ECMWF analyses show small improvements between 48- and 120-h forecast ranges in the all-sky experiment. Forecast verifications against radiosonde humidity observations and ECMWF analyses in global domain do not show statistically significant differences between the experiments.

At ECCC, all-sky assimilation of MHS observations is over open water only and it is expected to see much of the impact over water. Since the spatial coverage of the radiosondes is mostly over land, it is not surprising that verifications against the radiosondes show less differences between clear-sky and all-sky experiments for analyses and lead times smaller than 48 h. Verifications against the GPSRO observations in Fig. 4 are over water and show the positive signal for the regions directly affected by the all-sky assimilation of MHS observations.

The results of this study are in contrast with a similar work from ECMWF where all-sky assimilation of MHS observations resulted in improved short-range forecasts but with little impact in medium range. Some of the factors that may contribute to this difference are as follows: it is possible that the 4D-Var assimilation system at ECMWF with a 12-h window is more effective at extracting wind information from all-sky assimilation of MHS observation than the 4D-EnVar assimilation system at ECCC with only a 6-h window; the cloud fields used for simulation and assimilation of MHS observations at ECCC are largely overestimated; the 4D-Var system at ECMWF diagnoses the cloud fields from dynamic and humidity fields within the minimization, whereas at ECCC, clouds are taken from the background state and not modified during the assimilation; and all-sky assimilation of MHS observations at ECMWF is over both land and ocean grid points, whereas at ECCC, it is only over ocean.

For future work, we aim to implement an outer loop in the 4D-EnVar system to reduce the dependency on the background state clouds and take steps to eliminate the application of global cloud scale factor. There is an ongoing effort to improve the quality of cloud fields of the forecast model.

This work was planned for the operational implementation in the GDPS during ECCC’s fourth Innovation Cycle (IC-4) in fall 2023. The extension of all-sky assimilation to ATMS temperature channels 5–6 over ocean was recently tested, following the methodology presented in Shahabadi and Buehner (2021) for all-sky assimilation of AMSU-A temperature channels 4–5. All-sky assimilation of ATMS temperature channels 5–6 over ocean will also be part of the implementation package proposed for the GDPS during IC-4. With the implementation of all-sky assimilation functionally for microwave humidity channels, we are now in the position to convert temperature and humidity channels of individual microwave instruments from clear-sky to all-sky mode. Tests for all-sky conversion of ATMS and Microwave Humidity Sounder-2 (MWHS2) humidity channels are planned for the near future.

Acknowledgments.

The authors thank Sylvain Heilliette for the implementation of the RTTOV-SCATT observation operator in our data assimilation system. They wish to thank Josep Aparicio for internal review of the preliminary version of this manuscript.

Data availability statement.

GDPS forecasts, analyses, and observations are generated in ECCC in-house file formats. The space required to store all data for each experiment is approximately 40 TB. These data are archived and will be kept for a period of 5 years. Upon request, the data could be converted to user-friendly format (e.g., netcdf and sqlite) and provided to interested parties.

REFERENCES

  • Bauer, P., E. Moreau, F. Chevallier, and U. O’Keeffe, 2006: Multiple-scattering microwave radiative transfer for data assimilation applications. Quart. J. Roy. Meteor. Soc., 132, 12591281, https://doi.org/10.1256/qj.05.153.

    • Search Google Scholar
    • Export Citation
  • Bauer, P., A. J. Geer, P. Lopez, and D. Salmond, 2010: Direct 4D-Var assimilation of all-sky radiances. Part I: Implementation. Quart. J. Roy. Meteor. Soc., 136, 18681885, https://doi.org/10.1002/qj.659.

    • Search Google Scholar
    • Export Citation
  • Buehner, M., 2020: Local ensemble transform Kalman filter with cross validation. Mon. Wea. Rev., 148, 22652282, https://doi.org/10.1175/MWR-D-19-0402.1.

    • Search Google Scholar
    • Export Citation
  • Buehner, M., and Coauthors, 2015: Implementation of deterministic weather forecasting systems based on ensemble–variational data assimilation at Environment Canada. Part I: The global system. Mon. Wea. Rev., 143, 25322559, https://doi.org/10.1175/MWR-D-14-00354.1.

    • Search Google Scholar
    • Export Citation
  • Candy, B., and S. Migliorini, 2021: The assimilation of microwave humidity sounder observations in all-sky condition. Quart. J. Roy. Meteor. Soc., 147, 30493066, https://doi.org/10.1002/qj.4115.

    • Search Google Scholar
    • Export Citation
  • Caron, J., and M. Buehner, 2022: Implementation of scale-dependent background-error covariance localization in the Canadian global deterministic prediction system. Wea. Forecasting, 37, 15671580, https://doi.org/10.1175/WAF-D-22-0055.1.

    • Search Google Scholar
    • Export Citation
  • Geer, A. J., and P. Bauer, 2011: Observation errors in all-sky data assimilation. Quart. J. Roy. Meteor. Soc., 137, 20242037, https://doi.org/10.1002/qj.830.

    • Search Google Scholar
    • Export Citation
  • Geer, A. J., P. Bauer, and S. J. English, 2012: Assimilating AMSU-A temperature sounding channels in the presence of cloud and precipitation. ECWMF Tech. Memo. 670, 43 pp., https://www.ecmwf.int/sites/default/files/elibrary/2012/9514-assimilating-amsu-temperature-sounding-channels-presence-cloud-and-precipitation.pdf.

  • Geer, A. J., F. Baordo, N. Bormann, and S. English, 2014: All-sky assimilation of microwave humidity sounders. ECMWF Tech. Memo. 741, 59 pp., https://doi.org/10.21957/obosmx154.

  • Geer, A. J., and Coauthors, 2018: All‐sky satellite data assimilation at operational weather forecasting centres. Quart. J. Roy. Meteor. Soc., 144, 11911217, https://doi.org/10.1002/qj.3202.

    • Search Google Scholar
    • Export Citation
  • Geer, A. J., and Coauthors, 2021: Bulk hydrometeor optical properties for microwave and sub-millimetre radiative transfer in RTTOV-SCATT v13.0. Geosci. Model Dev., 14, 74977526, https://doi.org/10.5194/gmd-14-7497-2021.

    • Search Google Scholar
    • Export Citation
  • Lee, S., H.-J. Song, H.-W. Chun, I.-H. Kwon, J.-H. Kang, and S. Lim, 2020: All-sky microwave humidity sounder assimilation in the Korean Integrated Model forecast system. Quart. J. Roy. Meteor. Soc., 146, 35703586, https://doi.org/10.1002/qj.3862.

    • Search Google Scholar
    • Export Citation
  • Parrish, D. F., and J. C. Derber, 1992: The National Meteorological Center’s spectral statistical-interpolation analysis system. Mon. Wea. Rev., 120, 17471763, https://doi.org/10.1175/1520-0493(1992)120<1747:TNMCSS>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Saunders, R., and Coauthors, 2018: An update on the RTTOV fast radiative transfer model (currently at version 12). Geosci. Model Dev., 11, 27172737, https://doi.org/10.5194/gmd-11-2717-2018.

    • Search Google Scholar
    • Export Citation
  • Shahabadi, M. B., and M. Buehner, 2021: Toward all-sky assimilation of microwave temperature sounding channels in environment Canada’s global deterministic weather prediction system. Mon. Wea. Rev., 149, 37253738, https://doi.org/10.1175/MWR-D-21-0044.1.

    • Search Google Scholar
    • Export Citation
  • Shahabadi, M. B., J. M. Aparicio, and L. Garand, 2018: Impact of slant-path radiative transfer in the simulation and assimilation of satellite radiances in environment Canada’s weather forecast system. Mon. Wea. Rev., 146, 43574372, https://doi.org/10.1175/MWR-D-18-0126.1.

    • Search Google Scholar
    • Export Citation
  • Shahabadi, M. B., M. Buehner, J. Aparicio, and L. Garand, 2020: Implementation of slant-path radiative transfer in environment Canada’s global deterministic weather prediction system. Mon. Wea. Rev., 148, 42314245, https://doi.org/10.1175/MWR-D-20-0060.1.

    • Search Google Scholar
    • Export Citation
  • Zhu, Y., and Coauthors, 2016: All-sky microwave radiance assimilation in NCEP’s GSI analysis system. Mon. Wea. Rev., 144, 47094735, https://doi.org/10.1175/MWR-D-15-0445.1.

    • Search Google Scholar
    • Export Citation
Save
  • Bauer, P., E. Moreau, F. Chevallier, and U. O’Keeffe, 2006: Multiple-scattering microwave radiative transfer for data assimilation applications. Quart. J. Roy. Meteor. Soc., 132, 12591281, https://doi.org/10.1256/qj.05.153.

    • Search Google Scholar
    • Export Citation
  • Bauer, P., A. J. Geer, P. Lopez, and D. Salmond, 2010: Direct 4D-Var assimilation of all-sky radiances. Part I: Implementation. Quart. J. Roy. Meteor. Soc., 136, 18681885, https://doi.org/10.1002/qj.659.

    • Search Google Scholar
    • Export Citation
  • Buehner, M., 2020: Local ensemble transform Kalman filter with cross validation. Mon. Wea. Rev., 148, 22652282, https://doi.org/10.1175/MWR-D-19-0402.1.

    • Search Google Scholar
    • Export Citation
  • Buehner, M., and Coauthors, 2015: Implementation of deterministic weather forecasting systems based on ensemble–variational data assimilation at Environment Canada. Part I: The global system. Mon. Wea. Rev., 143, 25322559, https://doi.org/10.1175/MWR-D-14-00354.1.

    • Search Google Scholar
    • Export Citation
  • Candy, B., and S. Migliorini, 2021: The assimilation of microwave humidity sounder observations in all-sky condition. Quart. J. Roy. Meteor. Soc., 147, 30493066, https://doi.org/10.1002/qj.4115.

    • Search Google Scholar
    • Export Citation
  • Caron, J., and M. Buehner, 2022: Implementation of scale-dependent background-error covariance localization in the Canadian global deterministic prediction system. Wea. Forecasting, 37, 15671580, https://doi.org/10.1175/WAF-D-22-0055.1.

    • Search Google Scholar
    • Export Citation
  • Geer, A. J., and P. Bauer, 2011: Observation errors in all-sky data assimilation. Quart. J. Roy. Meteor. Soc., 137, 20242037, https://doi.org/10.1002/qj.830.

    • Search Google Scholar
    • Export Citation
  • Geer, A. J., P. Bauer, and S. J. English, 2012: Assimilating AMSU-A temperature sounding channels in the presence of cloud and precipitation. ECWMF Tech. Memo. 670, 43 pp., https://www.ecmwf.int/sites/default/files/elibrary/2012/9514-assimilating-amsu-temperature-sounding-channels-presence-cloud-and-precipitation.pdf.

  • Geer, A. J., F. Baordo, N. Bormann, and S. English, 2014: All-sky assimilation of microwave humidity sounders. ECMWF Tech. Memo. 741, 59 pp., https://doi.org/10.21957/obosmx154.

  • Geer, A. J., and Coauthors, 2018: All‐sky satellite data assimilation at operational weather forecasting centres. Quart. J. Roy. Meteor. Soc., 144, 11911217, https://doi.org/10.1002/qj.3202.

    • Search Google Scholar
    • Export Citation
  • Geer, A. J., and Coauthors, 2021: Bulk hydrometeor optical properties for microwave and sub-millimetre radiative transfer in RTTOV-SCATT v13.0. Geosci. Model Dev., 14, 74977526, https://doi.org/10.5194/gmd-14-7497-2021.

    • Search Google Scholar
    • Export Citation
  • Lee, S., H.-J. Song, H.-W. Chun, I.-H. Kwon, J.-H. Kang, and S. Lim, 2020: All-sky microwave humidity sounder assimilation in the Korean Integrated Model forecast system. Quart. J. Roy. Meteor. Soc., 146, 35703586, https://doi.org/10.1002/qj.3862.

    • Search Google Scholar
    • Export Citation
  • Parrish, D. F., and J. C. Derber, 1992: The National Meteorological Center’s spectral statistical-interpolation analysis system. Mon. Wea. Rev., 120, 17471763, https://doi.org/10.1175/1520-0493(1992)120<1747:TNMCSS>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Saunders, R., and Coauthors, 2018: An update on the RTTOV fast radiative transfer model (currently at version 12). Geosci. Model Dev., 11, 27172737, https://doi.org/10.5194/gmd-11-2717-2018.

    • Search Google Scholar
    • Export Citation
  • Shahabadi, M. B., and M. Buehner, 2021: Toward all-sky assimilation of microwave temperature sounding channels in environment Canada’s global deterministic weather prediction system. Mon. Wea. Rev., 149, 37253738, https://doi.org/10.1175/MWR-D-21-0044.1.

    • Search Google Scholar
    • Export Citation
  • Shahabadi, M. B., J. M. Aparicio, and L. Garand, 2018: Impact of slant-path radiative transfer in the simulation and assimilation of satellite radiances in environment Canada’s weather forecast system. Mon. Wea. Rev., 146, 43574372, https://doi.org/10.1175/MWR-D-18-0126.1.

    • Search Google Scholar
    • Export Citation
  • Shahabadi, M. B., M. Buehner, J. Aparicio, and L. Garand, 2020: Implementation of slant-path radiative transfer in environment Canada’s global deterministic weather prediction system. Mon. Wea. Rev., 148, 42314245, https://doi.org/10.1175/MWR-D-20-0060.1.

    • Search Google Scholar
    • Export Citation
  • Zhu, Y., and Coauthors, 2016: All-sky microwave radiance assimilation in NCEP’s GSI analysis system. Mon. Wea. Rev., 144, 47094735, https://doi.org/10.1175/MWR-D-15-0445.1.

    • Search Google Scholar
    • Export Citation
  • Fig. 1.

    The stddev of OMB (K; blue) for the period from 15 Jun to 1 Aug 2019 as a function of SI¯ (K) for MHS channels (a) 2, (b) 3, (c) 4, and (d) 5. The SI¯ bin width of 5 K is used to generate the plots. The linear fit (red) and data count in each bin (green) are also shown.

  • Fig. 2.

    Number of assimilated MHS channel 5 observations in 2° × 2° latitude–longitude grid in (a) clear-sky and (b) all-sky experiments from 13 to 26 Jun 2019. The total number of assimilated MHS channel 5 observations during this period for each experiment is also shown.

  • Fig. 3.

    The normalized stddev(OMB) [stddev(OMB)allSky/stddev(OMB)clearSky × 100; black] and number of assimilated observations (numObsallSky/numObsclearSky × 100; green) over open water for (a) MHS, (b) AMSU-A, and (c) ATMS. The filled black dots denote the statistical significance computed with the F test at 95% confidence level.

  • Fig. 4.

    The normalized stddev(OMB) [stddev(OMB)allSky/stddev(OMB)clearSky × 100] of the assimilated GPSRO refractivity observations over water for (a) summer 2019 and (b) winter 2020 evaluation periods. The observations are grouped in 3-km-thick layers between 0- and 40-km altitudes above MSL. The filled black dots denote the statistical significance computed with the F test at 95% confidence level.

  • Fig. 5.

    The stddev of radiosondes minus forecasts for geopotential heights in the all-sky experiment normalized by the corresponding value from the clear-sky experiment [stddev(geopotentialHeight)allSky/stddev(geopotentialHeight)clearSky × 100] in the Northern Hemisphere domain at (a) 72- and (b) 120-h forecast ranges. The filled black dots denote the statistical significance with 90% confidence level. The experiments are from winter 2020 evaluation period.

  • Fig. 6.

    As in Fig. 5, but for temperature.

  • Fig. 7.

    As in Fig. 5, but for horizontal wind.

  • Fig. 8.

    Normalized difference in the zonal mean of stddev of error for temperature between clear-sky and all-sky experiments compared against ECMWF analyses, defined as [mean(σclearSky) − mean(σallSky)]/mean(σclearSky) × 100, at (a) 24-, (b) 48-, (c) 72-, and (d) 120-h forecast ranges during winter 2020 evaluation period. The red (blue) color shading indicates that the all-sky (clear-sky) experiment has better performance. The dots denote statistically significant results above 90% confidence level from the F test.

  • Fig. 9.

    As in Fig. 8, but for horizontal wind.

All Time Past Year Past 30 Days
Abstract Views 112 112 0
Full Text Views 345 345 102
PDF Downloads 245 245 64