Chris Velden (USA) and John LeMarshall (Australia)

0.4a Imminent uses and data assimilation
Expected contributions of satellite observation data
to the operational numerical prediction of typhoons.

Presentor: Hideo Tada
Numerical Prediction Division, Forecast Department
Japan Meteorological Agency
1-3-4 Ote-machi, Chiyoda-ku, Tokyo 100-8122, Japan

E-mail: hidetada@npd.kishou.go.jp
Fax: +81-3-3212-8407


In order to extract useful information from satellite observations, sophisticated methods are essential for data assimilation processing. Satellite data are very useful due to their wide geographical coverage. However, characteristics of satellite data are different from those of traditional observations. Hence, extensive development is required in the numerical weather prediction (NWP) systems.

The Japan Meteorological Agency (JMA) has operated a three-dimensional variational assimilation (3D-Var) method for the analysis of the Global Spectral Model (GSM) since September 2001, and a four-dimensional variational (4D-Var) method for the Meso-Scale Model (MSM) since March 2002. A direct assimilation of NOAA/ATOVS is planned to be introduced for the global 3D-Var analysis by the end of 2002. A positive impact on the typhoon track prediction was reported from preliminary tests. Meanwhile, experiments of MSM with 4D-Var have shown improvements of the predicted typhoon structures by assimilating the precipitation data derived from TRMM/TMI and DMSP/SSMI.

Important keys of data processing can be put forward as quality controls, statistical monitoring, management of huge data volumes, and so on. Considerable efforts are required to manage each item. Therefore, useful information on these treatments have to be exchanged among NWP centers. A long-term perspective should be established on satellite data utilization in the operational system.

JMA is planning to replace the operational NWP system in 2006. In the next system, the JMA aim is to utilize advanced sounder data, precipitation data from Global Precipitation Measurement (GPM), occultation data of navigation satellites such as Global Positioning System (GPS), and so on. JMA expects a great impact on the tropics by assimilating these data comprehensively. However, developing efficient assimilation schemes for various satellites is expected to be a hard work. It's important to create a framework to be able to introduce new schemes smoothly into the operational system by maintaining balance between independent and cooperative developments with other forecast centers.

0.2.1 Introduction

Satellite data are indispensable to improve the accuracy of the initial state of NWP models. However, characteristic features of satellite data such as reliability or error conditions are different from those of traditional observations, which will force us to do extensive developments in operational NWP systems. Due to rapid developments in computing technology, NWP models have been significantly improved. On the other hand, many satellites with various kinds of new instruments are planned. Development of analysis schemes to assimilate these huge observation data sets into elaborate NWP models is important.

To improve prediction of tropical cyclones, techniques that utilize observations of water vapor, precipitation, and winds that related directly to the phenomena have to be developed. In addition, comprehensive information of satellite data is very important to analyze the environment around tropical cyclones over the oceans where the traditional observations are sparse.

NWP systems are complex with many subsystems such as observations, quality control, and data assimilation. Hence, well-balanced developments of these subsystems are important. In this paper, I will explain the JMA's tentative problems and future plans for satellite data utilization.

0.2.2 Developments of the JMA data assimilation schemes

The objective analyses of JMA had adopted an optimum interpolation (OI) method for a long time. This method is cheap and has a strong statistical interpolation framework. Objective analyses of the observations are used to correct the error in the first-guess fields produced by NWP models. The information for this correction is extracted as a departure of the observed value relative to the first-guess value. However, OI schemes only accept departures of the same variables as those of the prognostic variables (e.g., temperature, wind (u,v)components...). The physical constraints allowed in the OI are also limited to the simple ones such as a geostrophic constraint.

In the variational assimilation framework, various types of satellite data such as radiances or precipitation can be assimilated as long as they can be calculated in terms of the prognostic variables. JMA introduced a three-dimensional variational assimilation (3D-Var) method for the Global Spectral Model (GSM) in September 2001, and a four-dimensional (4D-Var) scheme for the Meso-Scale Model (MSM) in March 2002. These developments are the operational foundations to cope with an future increases of satellite data and the effective utilization of them.

0.2.3 Satellite data utilization in the JMA operational NWP system

The JMA's current satellite data utilization procedures have been developed under the framework of OI analyses. JMA are still using temperature and humidity retrieved from satellite radiances by other satellite centers even after the global analysis was updated to 3D-Var. This procedure was necessary due to the considerable delay for technology transition at JMA compared with other major NWP centers.

Several problems arise in using the data retrieved by other satellite centers. One major problem is that the retrieval algorithms tend to be a "black box" so the error characteristics are hard to determine quantitatively. An additional problem is that the background information (e.g., first-guess) used in the retrieval algorithm at the other centers would indirectly affect the accuracy of the assimilated results. Under ordinary circumstances, quality of observational data is independent of the performance of NWP models. So it's not desirable to have the data retrieved by other centers to contain some kind of information dependent on the other NWP models.

To overcome these problems, JMA is now developing direct assimilation schemes of raw satellite data via the variational framework. If extra retrievals are avoided, data assimilation systems become immune to additional conversion errors. Treatments of the observational error also become easier if raw information from the instruments is used. As the first step of the development, JMA plans to introduce the direct assimilation scheme of the ATOVS vertical sounding data from NOAA satellites in December 2002 for the global 3D-Var analysis.

The utilization of other satellite data is in preliminary experimental stage, such as ocean-surface wind vectors by the microwave scatterometer on the QuikSCAT satellite, precipitable water vapor derived by the microwave imagers on the DMSP and TRMM satellites, and so on.

0.2.4 Recent topics

a) Improvement of JMA forecast skill with the global 3D-Var scheme

Figure 1 shows the time sequence of the forecast error of 250hPa wind by the operational global model. A significant reduction of the JMA errors (thick lines) after the 3D-Var was introduced in September 2001 can be seen.

Figure 1: Sequence of monthly mean Root-Mean-Square Errors (RMSE) of 250hPa wind forecasts in the Northern Hemisphere at 24h(left) and 48h(right). The thick lines correspond the sequence of JMA global model while the other lines are for other NWP centers. The 3D-Var assimilation was introduced on September 2001.

b) An impact of NOAA/ATOVS direct assimilation on the typhoon track prediction

The JMA global model has some problems in the prediction of subtropical high pressure systems that have been related to a low temperature bias over the tropics. Improvements of cumulus convection scheme have been tried from various aspects. At the same time, JMA has kept using bogus data of relative humidity vertical profiles estimated from black body temperature (TBB) observations from geostationary satellites to cope with the humidity data shortage over tropical oceans. The introduction of NOAA/ATOVS direct assimilation was tested with an improved cumulus convection scheme and removal of the bogus humidity data.

Figure 2 shows the predicted tracks of Typhoon 0104 (UTOR). A significant improvement can be seen with the new scheme. It is probable that the impact was affected by the improvement of tropical ~ subtropical condition by the new cumulus convection scheme plus the improvement of the initial condition of atmosphere surrounding the typhoon by the NOAA/ATOVS direct radiance assimilation.

Figure 2: The predicted track of Typhoon 0104 (UTOR) from initial date of 12UTC of 1 July 2001. The line with typhoon-marks shows the "best-track" while the other lines correspond to various options of the prediction.

c) An improvement of the typhoon structure with the precipitation data from TRMM/TMI and DMSP/SSMI

This is one of the examples of the on-going developments with the JMA Meso-Scale Model. Figure 3 shows a result of the experiment of 4D-Var that assimilated the precipitation data retrieved from TRMM/TMI and DMSP/SSMI. An increase of the typhoon intensity can be seen by assimilating precipitation. This 4D-Var scheme is still in a preliminary experimental stage. JMA is planning to promote utilization of precipitation data through the subsequent implementations of the 4D-Var scheme into the Regional and Global Spectral Models.

d) A development of creating "observation-type" typhoon bogus data

The center position of typhoon is a very important factor for the track prediction. In terms of a provision of disaster information, it's highly critical that the center position be correct in the initial state. Therefore JMA has applied a "grid-type" typhoon bogus implanted in the first-guess field. To assimilate typhoon information more properly, an "observation-type" typhoon bogus has been developed and tested with the meso-scale 4D-Var scheme.

Figure 4 shows an example of new bogus data. An axial-symmetric wind component is derived based on the statistical research, while an asymmetric component is calculated from the typhoon structure in the first-guess field. Many options such as data distribution or variables have been tested to represent the proper typhoon structure in the initial state.

Figure 3: An example of the impact of assimilating precipitation data for the Typhoon 0115 (Danas). The left panel shows the observation, the middle panel shows the operational prediction (ft=6h) and the right panel is the corresponding 6h prediction after assimilating the precipitation retrieved from TRMM/TMI and DMSP/SSMI.

Figure 4: An example of the "observation-type" typhoon bogus data (star-marks).

0.2.5 Critical issues on satellite data utilization

a) Quality control procedures

Whenever new kinds of observations are to be used, the characteristic features of their reliability have to be documented through statistical research. Quality control procedures should be designed with sufficient knowledge of the data error characteristics. Just one strange datum may ruin the processing even for highly sophisticated schemes.

Especially, operational real-time processing is a fight against time. Huge volumes of data have to be processed in the structured schedule. Under these circumstances, processing schemes are often evaluated based on cost / performance basis. Operational processing schedules do not permit adoption of sophisticated schemes that may be used in academic research. Reducing the degrees of freedom in the processing while maintaining the total effects as high as possible is an important key. Therefore, sufficient knowledge of observation error and background error characteristics is necessary. Collaborations with satellite/data processing centers should be encouraged on a routine basis.

b) Statistical monitoring of observational data

The various types of errors in observational data include measurement error, bias (calibration) error, conversion error, transmission error, etc. In addition, an error may be caused by observational representativeness. Quantitative knowledge of various types of errors is achieved by statistical monitoring. Whereas developments of NWP model and data assimilation schemes tend to be given highest importance, unspectacular work such as quality control monitoring should be more emphasized that may also improve NWP performance.

Procedures for statistical monitoring or scheme regulation are usually hard to get published. These kinds of information should be exchanged more openly. However, it seems to be difficult to achieve the communication level given the absence of a strong sense of obligation to share such information.

d) Management of huge volumes of data

Rapid increases in the satellite data volume is also a severe problem in the operational processing. The more you process raw data, the larger the data volume tends to be. This data volume problem was not large when only conventional observations were used. Now, JMA is planning to actively promote utilization of raw satellite data. Thus, the data volume is expected to be ten times or more relative to the present level. Therefore, how to manage huge volumes of data with limited operational resources will be a great factor in the design of NWP systems. There are many stages and phases in data processing: quality monitoring and controls, data assimilation, validation of model outputs, and so on. You have to have a clear concept of satellite data utilization and storage management from the long-time point of view.

0.2.6 JMA's future plans of satellite data utilization

JMA is addressing major topics on the NWP developments in the next computer system starting in 2006. Some of the important factors on satellite data utilization that relate closely to the NWP performance are given here.

  1. Advanced sounder data (interferometer)

Advanced sounders have many observing channels to be able to extract detailed vertical structures of temperature or water vapor. On the other hand, it's difficult to properly treat the observation error correlation among channels. Optimum selection of the channels will be a great issue for operational data assimilation.

Because infrared (IR) sounders are affected by clouds, simultaneous use with microwave instruments is required to cover the whole weather conditions. Quality controls for cloud detection or treatments of precipitation are also necessary.

b) Precipitation measurement

As described in section 4, JMA is conducting Meso-Scale Model experiments with the 4D-Var scheme to assimilate the precipitation retrieved from TRMM/TMI and DMSP/SSMI. In the GPM (Global Precipitation Measurement) framework, the core satellite with precipitation radar and the eight constellation satellites with microwave radiometers cover the whole globe and provide distribution of global precipitation every three hours. JMA is developing the 4D-Var schemes for all the major NWP models to effectively assimilate precipitation data from the GPM. This development will be completed for implementation in the next computer system.

(c) Direct assimilation of raw radiances

To predict precipitation phenomena with high precision, atmospheric conditions surrounding the precipitation system must be observed and predicted. Water-vapor-related elements (e.g., precipitable water) will play an important role in the data assimilation. JMA is developing direct assimilation schemes for raw radiances observed by microwave radiometers to analyze the environmental elements around precipitation systems. Combining all of the information about precipitation (water vapor and environmental structure) will lead to a great impact on the prediction of tropical cyclones.

d) GPS occultation data

Utilization of the occultation data from the navigation satellites such as GPS is also a key objective for the next computer system. The low-earth-orbit (LEO) satellites observe atmospheric refraction of microwave emitted by GPS satellites, and thus provides temperature and humidity profile information related to a refraction factor. The high vertical resolution of the data is attractive in terms of providing additional information relative to advanced sounder data. JMA is currently developing an assimilation scheme to trace radiowave. One of the present problems is the small geographical coverage. In the near future, various LEO satellites with GPS receivers are planned to be launched by various institutes. An increase of the data is expected to bring a larger impact on NWP processing.

e) Assimilation of clouds

To promote utilization of raw radiances observed by satellites, improved radiative-transfer models are required. Most of the current major models are designed for clear-sky radiance conditions. Direct assimilation of radiances needs to be expanded by the effects of clouds and precipitation. Direct assimilation of clouds is thus a big objective at JMA in the future.

f) Requirement from operational aspects

From an operational perspective, a stable provision of the data stream is necessary in terms of instrument performance stability and also through successive satellite plans.

A time-delay in the data dissemination is also a critical factor in providing NWP outputs to forecasters within the limited time schedule. The global model presently waits for two and a half hours for the initial analysis for the routine prediction, while the meso-scale model waits only 50 minutes due to the requirement for providing quantitative information for disaster prevention. Developing schemes that assimilate more raw data while avoiding any needless conversion process is required to save computation time.

An additional problem is the huge size of the data stream. Effective data communication systems including an upgrade of the GTS should be considered.

0.2.7 Summary

a) Long-term perspective on satellite data utilization

The technology for acquiring satellite data is significantly changing. Various satellites with various instruments are planned to be launched. NWP systems are also developing day by day. In these circumstances, a long-term perspective and strategy of satellite data utilization in the operational systems is essential.
As described in section 5, the issues relating to the satellite data utilization include various aspects such as quality monitoring and controls, data assimilation, validation of NWP outputs, storage management, etc. Whether each NWP center can treat all of issues independently is unknown. Cooperative developments seems to be one option. However, to implement the cooperative achievement into your own system, you have to adopt them to match your system features. Tunings by other institutes are usually not applicable. Thus, sharing schemes may not be so easy as it seems. Keeping the balance between independent and cooperative developments will be important unless independent developments are possible.

b) To improve the precision of tropical forecast

JMA is looking forward to the expansion of the GPM framework as described in section 5. A global coverage of precipitation systems will have a great impact on prediction in the tropics. Of course, developments of more effective schemes to utilize such satellite data, and improvements of the NWP models to achieve small biases in the tropics are indispensable. However, even if errors in assimilation and prediction reduced substantially, there still remains a predictability problem. Hence JMA is also exploring the possibilities of an ensemble prediction on typhoons.

The key future information will be related with water vapor measurements. In the traditional NWP (e.g., OI) era, observations of geopotential heights had a great impact to control the wind field, which advects other meteorological elements. In 4D-Var assimilation, continuous observations of water vapor will affect the dynamical field more effectively through dynamics of NWP models.

Based upon the above arguments, JMA will promote comprehensive developments for the effective utilization of satellite observations.