An Observing System Simulation Experiment (OSSE) was designed and developed to assess the potential benefit of the Infrared Sounding on the Meteosat Third Generation (MTG-IRS) geostationary meteorological satellite system to regional forecasts. In the proposed OSSE framework, two different models, namely, the MM5 and WRF models, were used in a nature run and data assimilation experiments, respectively, to reduce the identical twin problem. The 5-day nature run, which included three convective storms that occurred during the period from 11 to 16 June 2002 over US Great Plains, was generated using MM5 with a 4 km. The simulated “conventional” observations and MTG-IRS retrieved temperature and humidity profiles, produced from the nature run, were then assimilated into the WRF model. Calibration experiments showed that assimilating real or simulated “conventional” observations yielded similar error statistics in analyses and forecasts, indicating that the developed OSSE system worked well. On average, the MTG-IRS retrieved profiles had positive impact on the analyses and forecasts. The analyses reduced the errors not only in the temperature and the humidity fields but in the horizontal wind fields as well. The forecast skills of these variables were improved up to 12 hours. The 18 h precipitation forecast accuracy was also increased.
Remotely sensed satellite observations play an important role in modern data assimilation and forecast systems [
The European Organisation for the Exploitation of Meteorological Satellites (EUMETSAT) is preparing for the next European operational geostationary meteorological satellite system named the Meteosat Third Generation (MTG). The MTG series will comprise six satellites, with the first spacecraft likely to be ready for launch in 2017. This program will provide space-acquired meteorological data until at least the late 2030s. Details can be found at the EUMETSAT website (
Assimilating the MTG-IRS observations into a mesoscale model is expected to improve regional numerical weather forecast skills. Recent studies suggest that the accurate representation of the low-level water vapor and temperature is crucial for the quantitative precipitation forecast [
The OSSE method is regarded as an efficient way to test the impact of future observing systems and new data assimilation techniques on weather analysis and prediction [
The OSSE system may overestimate benefit of the observations when the same numerical weather model is used in the nature run and forecast experiment (i.e., the identical twin problem). To alleviate this problem, a model with different physics schemes is often used in regional OSSE to simulate model errors [
This paper describes the framework design of the OSSE system and summarizes the preliminary results of the potential benefit of MTG-IRS on regional forecasts of three convective storms occurred over the US Great Plains in June 2002. The OSSE system is characterized by a nature run using the MM5 model with convective system resolving grid size of 4 km, (rapid update) cycling data assimilation and forecast experiments using the WRF model with 12 km resolution, and a pair of calibration experiments. The overview of selected cases is in Section
Three convection cases that occurred between 11 and 16 June 2002 during the International H2O Project (IHOP 2002) are selected. These three cases include dry line, convective storms, and a severe mesoscale convective system (MCS). On 11 June 2002, a dry line formed in the Oklahoma panhandle in the late afternoon. Three storms developed on this particular day in regions close to the border of Colorado and Oklahoma. On 12 June 2002, a northeast-to-southwest-oriented squall line from the Kansas and Oklahoma border to the Texas Panhandle was initiated at around 2100 UTC 12 June 2002. At the same time, isolated convections formed from the Kansas and Oklahoma border to Texas. They gradually strengthened along most parts of the dry line, and a severe storm developed near a triple point that was formed by the dry line and a cold air outflow. The 15-16 June case is a severe MCS that occurred over the US central and southern plains in the late afternoon 15 June 2002. The MCS produced severe weather including several tornadoes in southern Kansas, and a swath of wind damage reports through central Oklahoma southward through central Texas as it propagated south-southeastward. More detailed descriptions of three cases were reported by Wechwerth et al. [
The OSSE flowchart is illustrated in Figure
Flowchart of experiments.
In previous published OSSE studies, for example, Masutani et al. [
The MM5 model version 3.6 [
The data assimilation system is the WRF Variational Data Assimilation System (WRF-Var) [
The model configuration for the nature run in this study employs 505 × 505 grid points with 4 km horizontal resolution and 35 vertical levels. The model top is at 50 hPa. The domain covers the central areas of the United States (Figure
Model domain superposed with topography (unit: m) for both MM5 and WRF. The states and Rocky mountains are also indicated.
Given the true state, simulated observations can be generated. Conventional observations such as radiosonde, pibal, surface station observations, and aircraft report are simulated to provide the basic simulated observing system for the reference data assimilation experiment. The satellite wind observations are generated as well. For simplicity, the term “conventional” observation is defined to include all the observation types (e.g., satellite wind) listed above. The types and positions of the real observations are obtained from the NCEP Automated Data Processing observations. Readers are referred to see websites
Simulated MTG-IRS temperature and moisture profiles are generated hourly from the nature run. The package to perform these simulated retrievals has been used for the Geostationary Operational Environmental Satellite-R Series (GOES-R) Hyperspectral Environmental Suite (HES) trade-off studies at Cooperative Institute for Meteorological Satellite Studies (CIMSS) and described in various publications (e.g., [
(i)
(ii)
The radiance dataset calculated for the training dataset was used to calculate empirical orthogonal functions (EOF) using a standard singular value decomposition of the covariance matrix. Values for the radiometric noise are added in the generation of the regression coefficients from a correlation of the EOFs with the training dataset. The actual PCR retrieval is applied to the simulated MTG-IRS radiances. First these radiances are projected onto the EOFs derived from the training dataset. Values for temperature, moisture, and surface temperature are then obtained after the application of the regression relations. After the PCR retrieval, a nonlinear iterative procedure is applied to the radiative transfer equation to further improve the profiles [
Currently, the MTG-IRS retrieval algorithm can only provide temperature and humidity profiles over the clear sky regions. In addition, the retrieval algorithm cannot deal with complicated land surfaces over the Rocky Mountains. Thus, there are no retrievals over the Rocky Mountain. But in other clear sky regions in the model domain, the retrieval algorithm worked well.
The MTG-IRS retrieved profiles are assimilated in the presence of other simulated conventional observations to assess their added values. The experiments are listed in Table
List of experiments.
Experiment | Cycling period | Resolution (km) | Initial condition and assimilated data |
---|---|---|---|
NoDA | |||
Nature | No | 4 | ETA 40 km analysis |
Control | No | 12 | GFS analysis and forecast, no observations |
| |||
3DVar | |||
ROP | 6 h | 12 | Background (BG)+ real conventional observations |
SOP-6hc | 6 h | 12 | BG+ simulated conventional observations (SOP) |
SOP-RPtq-6hc | 6 h | 12 | BG + SOP + retrieved profiles (RP) of temperature |
SOP-RPq-6hc | 6 h | 12 | BG + SOP + retrieved profiles of humidity (RPq) |
SOP-1hc | 1 h | 12 | BG + SOP |
SOP-RPtq-1hc | 1 h | 12 | BG + SOP + RPtq |
POP | 6 h | 12 | BG + perfect observation (POP) |
The simulated data with high spatial density are thinned before data assimilation. Satellite winds are thinned to 36 km grids, and their errors are then assumed to be independent. The MTG-IRS retrievals have high horizontal error correlations, and they are thinned to 36 km grids for data assimilation. Vertical correlation in retrievals is not considered in data assimilation. However, retrievals are thinned to the 35 vertical pressure levels in WRF model.
Note that in all the control and assimilation experiments, the initial and boundary conditions of the WRF model are interpolated from the 1-degree resolution NCEP GFS analyses and subsequent 5-day forecasts at 0000 UTC 11 June 2002. The physics schemes chosen in the WRF simulations include the Noah land surface model, the WSM6 microphysics scheme [
The background error covariance is generated by the National Meteorological Center (NMC) method [
The control run is performed without data assimilation. In the control experiment, the WRF model is initialized from the GFS analysis at 0000 UTC 11 June 2002 and integrated for 5.5 days. Its 18 h forecast valid at 1800 UTC 11 June 2002 serves as the background (BG) fields for the first cycle of other data assimilation experiments. The control run also serves as the benchmark for intercomparison with other experiments with observations assimilated.
The added value of MTG-IRS retrievals is assessed in a cycling data assimilation and forecast mode. In a cycling mode, a previous forecast is used as the first guess for the current analysis. The data assimilation experiments, SOP-6hc and SOP-1hc, in which only simulated conventional observations are assimilated, are references experiment for quantifying MTG-IRS data impact. To validate the performance of the designed OSSE, we also conducted an experiment, ROP, in which the real conventional observations are assimilated. In OSSE, the assimilation of real observations will yield similar error statistics in analyses and forecasts compared to the assimilation of corresponding simulated observations.
To assess the added value of the retrieved MTG-IRS temperature and moisture profiles, the simulated conventional observations are assimilated in all data assimilation experiments. For example, in the experiment SOP-RPq-6hc, only the MTG-IRS humidity profiles are assimilated in addition to the simulated conventional data every 6 hours, while the experiment SOP-RPtq-6hc uses both MTG-IRS temperature and humidity profiles. The experiment SOP-RPtq-1hc is the same as SOP-RPtq-6hc except the assimilation cycling period is 1 hour instead of 6 hours.
The POP experiment is designed to assess the maximum impact of the perfect observations. The temperature, moisture, and wind profiles obtained directly from the truth at every 36 km × 36 km grid point are assimilated in this experiment.
To objectively evaluate the impacts of MTG-IRS data on the regional scale analysis and forecast, traditional skill scores such as the root-mean-square (RMS) error between an experiment and the “truth” are computed.
The impact on precipitation forecast is quantified in terms of Equitable Threat Score (ETS) and frequency bias. The ETS is defined as
The frequency bias is the ratio of the forecast frequency to the observed frequency, that is,
A 5-day nature run covering the selected three IHOP 2002 convective events is completed first. Figure
6 h accumulated precipitation in the observation (a), (c), and (e) and in the nature run (b), (d), and (f) valid at 0600 UTC 12 (a) and (b), 0600 UTC 13 (c) and (d) and 0600 UTC 16 (e) and (f) June 2002. Note that, the color scales and map projections are different between the observation and the simulation.
The numbers of upper air observations including radiosondes and pibals vary from cycle to cycle, more at 0000 UTC and 1200 UTC and less at 0600 UTC and 1800 UTC. The upper air observations at 0000 UTC and 1200 UTC are about 30. For surface observations, the numbers remain almost unchanged throughout the 6 h cycles. The typical number of surface observation is about 650. The numbers and positions of the satellite wind and MTG-IRS retrievals can change dramatically due to weather conditions.
The temperature and water vapor mixing ratio profiles over clear air regions are retrieved. By comparing the retrievals with the “truth,” it is found that the retrievals faithfully represent the relatively large-scale patterns and some mesoscale details of the real temperature field, especially in the middle atmosphere (figures not shown). The bias and RMS errors are shown in Figure
Bias and RMS error of (a) the retrieved profiles of
To validate the performance of the designed OSSE, we first conducted a pair of calibration experiments ROP and SOP-6hc to compare the impacts from assimilating real conventional observations to assimilating corresponding simulated observations. If the designed OSSE system works well, the assimilation of real observations will yield similar error statistics in analyses and forecasts compared to the assimilation of simulated observations. Figure
Bias and RMS error in temperature ((a), (b)), water vapor ((c), (d)),
The WRF 3D-Var analyses were performed in cycling mode starting from 1800 UTC 11 to 1200 UTC 15 June. The analysis at every 6 h is used to initialize a 24 h forecast. To better understand the impact of MTG-IRS retrieved profiles, three types of analysis differences are computed: (1) the analysis of assimilating simulated observation profiles (SOP) minus the first guess; (2) the results of assimilating simulated observations and MTG-IRS retrievals (SOP-RPtq or SOP-RPq) minus the results of SOP; (3) the analysis minus the “truth.” The observation number of MTG-IRS retrieval profiles is much larger than that of the conventional observations in SOP. We expect that the MTG-IRS observations will have a greater impact on analysis. The analysis should be more accurate when both simulated conventional observations and MTG-IRS retrievals are assimilated compared to the analysis produced by assimilating only simulated conventional observations, given that the MTG-IRS retrievals are of high qualities.
The analysis at the beginning of the cycle at 1800 UTC 11 June is shown in detail since all experiments have the same background at that time. Figures
Differences of the 850 hPa temperature
Differences of the 850 hPa water vapor mixing ratio
Differences of the 850 hPa
In the background, there is a positive temperature bias along the Rocky Mountains (Figure
The MTG-IRS has a larger impact than conventional observations in humidity analysis compared Figure
MTG-IRS retrievals have little impact on wind analysis comparing with SOP-6hc. In SOP-6hc, the
The analyses at other times have also been checked. Analyses show that assimilation of simulated conventional observations improves analyses of
Using the nature run as the “truth,” the RMS errors of the analysis and the forecast are computed. In each 6-hourly cycling experiment, there are 16 analyses and 16 forecasts during the cycling period. The averaged RMS errors at analysis time are shown in Figure
Averaged RMS errors for experiments POP, SOP-RPq-6hc, SOP-RPtq-6hc, SOP-6hc, and control at analysis time. (a) Zonal wind component
The above results indicate the assimilation of MTG-IRS temperature and humidity profiles significantly reduces the temperature and the water vapor analysis error in the low-level atmosphere. Past research shows that the realistic mesoscale details of the horizontal variations in low-level moisture and temperature in analyses would help to improve forecast skills for convective events [
The averaged RMS errors of the 12 h forecast are shown in Figure
Same as Figure
The maximum potential impact of perfect observations of MTG-IRS on analyses can be found in POP in which the true soundings of temperature, humidity, and
Since high temporal resolution MTG-IRS retrievals can be obtained, it is useful to test their impact on the regional forecasts. Two 1-hourly cycling experiments have been carried out. The RMS errors of the experiments (figures not shown) indicate that increasing assimilation frequency yields only slight improvements on the analysis compared to 6-hourly cycling experiments. The 12 h forecast accuracy of temperature and moisture are also improved, while a neutral impact on the wind forecast is observed. However, it is found that the hourly cycling has a noticeable positive impact on precipitation forecasts, as will be shown in Section
For the first case, the equitable threat scores (ETS) and frequency bias for 18 h accumulated rainfall with thresholds of 10 mm, 30 mm, 50 mm, 65 mm, 80 mm, and 100 mm are shown in Figure
(a) ETS and (b) frequency bias of 18 h accumulated precipitation for experiments control, SOP-6hc, SOP-RPq-6hc, and SOP-RPtq-6hc, valid at 1200 UTC 12 June.
For the second convective event, ETS and frequency bias for 18 h accumulated rainfall with thresholds of 10 mm, 30 mm, 50 mm, and 65 mm are shown in Figure
(a) ETS and (b) frequency bias of 18 h accumulated precipitation for experiments control, SOP-6hc, SOP-1hc, SOP-RPq-6hc, SOP-RPtq-6hc, and SOP-RPtq-1hc, valid at 1200 UTC 13 June, 2002.
For the third convective event, the threat scores (figures not shown) for 18 h accumulated rainfall show that MTG-IRS temperature and moisture retrievals (SOP-RPtq-6hc) have positive impacts on precipitation forecast at threshold 10 mm. The rainfall forecast frequency for experiments SOP-RPtq-6hc and SOP-RPtq-1hc is better than experiments assimilating only the simulated conventional data.
The averaged ETS and frequency bias of all 16 forecasts of 18 h accumulated rainfall with thresholds of 10 mm, 30 mm, 50 mm, 65 mm, 80 mm, and 100 mm during the entire cycling period are shown in Figure
(a) Averaged ETS and (b) averaged frequency bias of 16 forecasts of 18 h accumulated precipitation for experiments control, SOP-6hc, SOP-1hc, SOP-RPq-6hc, SOP-RPtq-6hc, and SOP-RPtq-1hc.
An OSSE system was designed and conducted to document the added value of temperature and water vapor observations derived from the MTG-IRS to regional forecasts, especially to precipitation prediction. The OSSE system was characterized by the nature run using the MM5 model with 4 km resolution, and the 12 km resolution 6/1-hourly cycling data assimilation and forecast experiments using the WRF model, and a pair of calibration experiments. The nature run and data assimilation and forecast experiments employed two different mesoscale models to reduce the identical twin problem. The calibration run was conducted to show the performance of the OSSE system. The OSSE calibration showed that assimilation of real or simulated conventional observations gives similar error statistics in analyses and forecasts, indicating that the developed OSSE system performed reasonably well.
Three convective cases over US Great Plains were selected to show MTG-IRS’s impact on high-impact weather forecasts. The results showed that the MTG-IRS retrievals have positive impacts on the analyses and the subsequent forecasts. The forecast skills for
Although the preliminary results from case studies are promising, we note that the current OSSE configuration may overestimate the impact of MTG-IRS retrievals. For conventional observations, the same observation statistics are used in generating simulated observations and data assimilation procedure. For MTG-IRS retrievals, the “true” observation error (against nature run) is used. Currently, only the conventional data are assimilated in control experiments. These can lead to an overoptimistic assessment of the observation impact as well. Other issues, for example, sensitivity to forecast model resolution, need to be further studied as well. In future work, more cases, radar observations, and other types of satellite data (e.g., microwave radiance and infrared images) will be simulated in the OSSE system to give a more realistic estimation of MTG-IRS data on regional forecasts.
The authors of the paper do not have a direct financial relation with the commercial identities mentioned in the paper.
The authors would like to thank Stephen A. Tjemkes and Rolf Stuhlmann (EUMETSAT) for providing MTG-IRS retrievals and their valuable suggestions and comments on this work. NCAR is sponsored by the National Science Foundation. Any opinions, findings, and conclusions or recommendations expressed in this publication are those of the authors and do not necessarily reflect the views of the National Science Foundation.