WMO/CAS/WWW

FIFTH INTERNATIONAL WORKSHOP on TROPICAL CYCLONES


TOPIC 0.4e: Special focus – Satellite data assimilation plans at selected forecast centers

Satellite Data Assimilation Plan in CMA

Presentor : Jishan Xue
Chinese Academy of Meteorological Sciences
Chinese Meteorological Administration
46 South Zhongguancun Street, 100081 P.R.China
E-mail: jsxue@cams.cma.gov.cn
Fax: 86-10-62173952
Abstract:
The use of satellite observations is the best way for solving the data voids problem encountered in forecast of tropical cyclones over the western Pacific and the South China Sea where few conventional soundings and surface observations are available. For this reason, the assimilation of satellite observations is set to be the first priority in a five year project GRAPeS which was launched in China last year in order to develop its new generation NWP system. The use of satellite data is implemented in the framework of variational assimilation. In the first phase of the project, a three dimensional variational assimilation (3DVar) scheme is developed, and will be upgraded to 4DVar in the second phase. The basic satellite data to be used include ATOVS radiances from polar orbiting satellites and cloud drift winds from geo-stationary satellites. Sea surface winds retrieved from satellite-borne radar are also very useful information and expected to be used. Case study show positive impact of combined use of these data on improving the large scale analysis. In addition, satellite data are found to be useful in defining the structure of tropical cyclones. New methods combining satellite derived winds and moisture information with dynamic features of tropical cyclones are being developed. It is expected that the progress in assimilation of satellite information will obviously improve the operational forecast of tropical cyclones.
0.4.1 Introduction
All tropical cyclones landing Chinese coast originate in the Western Pacific or the South China Sea where few conventional soundings or surface data are available. With sparse observations over the oceans, it is impossible to define the large scale flow patterns and the inner structure of tropical cyclones, and such inaccurate analyses cause big errors in both numerical prediction and statistical forecasts of tropical cyclones. So the data voids problem is the most serious challenge faced by Chinese Scientists and forecasters in operational centers.
The use of satellite observations is likely the best way for solving this data voids problem. In fact, the cloud images from geo-stationary satellites have been used for the forecast of tropical cyclones since the early 1970s. But up to now, the main application of satellite information in forecast of tropical cyclones is positioning the cyclones and estimating their intensities with these satellite images. There are two ways of extracting more observational information from satellites to improve the analysis of large scale flow patterns and the structures of tropical cyclones themselves, i.e. using retrieval data, such as atmospheric temperature and humidity, and the direct assimilation of satellite radiances.
Whichever methodology is adopted, an advanced variational data assimilation system is required. In 2001, Chinese Meteorological Administration (CMA) launched a five-year project to develop new generation Global/ Regional Assimilation and Prediction System (GRAPeS). The goal of the project is to improve short term and medium range numerical weather prediction, especially for those high impact weather events such as typhoon and heavy rains in summer monsoon, and 3DVar and 4DVar focusing on application of satellite data are among the first priorities of the project.
0.4.2 GRAPeS project and its variational data assimilation
The whole project of GRAPeS consists of two phases. The object of the first phase (2001-2003) is development of a non-hydrostatic model dynamic framework and 3 dimensional variational data assimilation system emphasizing on the use of satellite radiances and other derived products from both geo-stationary and polar orbiting satellites. The object of the second phase (2004-2005) in terms of data assimilation is upgrading 3DVar developed in the first phase to 4DVar and the usage of meso-scale data e.g. Doppler radar observations.
The scientific design and coding of GRAPeS 3DVar (G-3DVar hereafter for short) were completed in the end of 2001. It is a grid point system in contrast to spectral systems used in most NWP centers and a recursive filter is used instead of matrix algebraic computation. Special preconditioning is introduced in order to speed up the convergence of iteration in minimizing the cost function and to reduce the scale of computation by transforming model variables to control variables which are physically independent to each other.
Some of products derived from the satellites, such as cloud drift wind and moisture derived from the geo-stationary satellite images may be used in the same way as the conventional data with different observational error statistics. To assimilate satellite radiances, a special observation operator, which converts the atmospheric profile and surface characteristics to radiation energy observed by the satellites must be introduced. In G-3DVar, RTTOV6 developed by ECMWF is adopted. RTTOV6 is an advanced software package including the fast radiance transfer model and its adjoint. An experimental system consisting of G-3DVar and a meso-scale model has been set up and real case tests are undertaken.
The further development of G-3DVar in the future includes improvements of error statistics of both observation and background field, introduction of more sophisticate algorithms for cloud detection and classification, and study on the scheme dealing with the impact of surface emissivity of different categories of land surface. Another important issue relevant to the satellite data assimilation and application is the quality control of radiances from the Chinese FY series satellites. This will be one of main focuses of research and development in the field of satellite meteorology in China.
0.4.3 Application of satellite data for prediction of large scale circulation related to tropical cyclones
With G-3DVar and G-4DVar in addition to cloud images, more satellite information will be used to improve the analysis and forecast of large scale flow pattern relevant to tropical cyclones.

  1. Satellite derived wind
The satellite-derived winds are of great importance because of good spatial coverage, so they give valuable information of large scale flow pattern, which are the background of tropical cyclones, in upper and lower troposphere over oceans. As mentioned in last section, the wind derived from cloud motion is used by G-3DVar as conventional data except that different error statistics should be adopted. The error due to the inaccurate assignment of elevation of cloud should also be taken into account.
In addition to this rather traditional way, another scheme is also being considered. Special attention will be paid to the winds derived by tracking the images of water vapor channel. It has been found that confluence and diffluence of upper troposphere flow detected by monitoring consecutive geo-stationary satellite images of water vapor channel reflect special vertical circulation patterns which are important for tropical cyclones. A scheme based on the so-called potential vorticity retrieval is proposed but more research work must be done. Satellite wind is also useful to define the inner structure of tropical cyclones. This will be discussed in next section about improving bogus model.
  1. Advanced TIROS Operational Vertical Sounder (ATOVS) Radiances
ATOVS radiances contain 40 channels: 20 of infrared (HIRS) and 20 of microwave (AMSU-A and AMSU-B). Most channels may be used to improve the detection of atmospheric temperature and humidity. Some of them may be used for estimation of surface parameters. Direct assimilation of AMSU-A, AMSU-B and HIRS radiances in the framework of G-3DVar and G-4DVar has the potential to improve the analysis over the oceans. So it is valuable for the prediction of tropical cyclones. Case studies show that combined use of satellite derived wind and ATOVS radiances within G-3DVar framework produces much better analysis for forecast of tropical cyclones. But due to the complexity of the problem, more research works and careful tuning of the scheme still should be done.
Radiances in infrared bands are contaminated by clouds. So HIRS radiances are used only in clear sky area and the main usage of radiances is to improve the large-scale analysis. Theoretically microwave radiance is free of cloud contamination, so it will be useful even in cloudy area, but it has been shown that thick cloud with precipitation will have some impact on microwave radiances. The usage of microwave radiances within tropical cyclones will be investigated.
Temperature and moisture may be retrieved from ATOVS radiances firstly. After the first step the retrieved information is used by the assimilation scheme in the same way as for conventional sounding data. Even though, experiences of some NWP centers show that the use of retrieved data are not as good as the direct use of radiances, but the assimilation of retrieved data has some benefits such as less cost of computer resources, so parallel experiment with retrieved data and direct assimilation of ATOV data is undertaken. The real operation scheme in the future may be some kind of mixture of the two algorithms depending on available computer resources and other requirements.
  1. Sea surface winds from microwave sensors or satellite-borne radar
Sea surface winds observed from microwave sensors or satellite borne radar are now available for near real-time use. Some experiments are going on to investigate the impact of sea surface winds from Quick Scatterometer (Quikscat) on forecast of tropical cyclones. The Quikscat is a mission of ocean-observing satellite of NASA. The sea surface winds are derived by measuring the backscattered or echoed radar pulses bounced back to the satellite. The data cover global oceans. Preliminary results show that the position of the center of tropical cyclone is defined more accurately from the sea surface winds from microwave sensor than from first guess and conventional surface data, so the forecast error of track may be reduced. Fig.0.4.1 shows the first guess (left panel) and the analysis (right panel) of wind field on 1000hPa Aug.18 2002. There were no surface observations over the ocean except Quikscat winds in this case. The impact of Quikscat winds on the circulation around the tropical storm is obvious. However big errors in these data are found in cases of strong winds. Special preprocessing of the data and quality control are necessary when the data are used routinely. This is one issue being studied in both the Institute of Tropical and Oceanic Meteorology Guangzhou and the Institute of Typhoon Shanghai. Their experiences are expected followed by other operational forecast centers in China.


  1. first guess b) analysis

Fig.0.4.1: wind field on 1000 hPa Aug.18, 2002


  1. Moisture from cloud images
An algorithm to retrieve vertical profile of atmospheric moisture with multi-channels cloud images from geo-stationary satellite has been developed in the National Satellite Meteorological Center (NSMC) CMA. This algorithm is a potential candidate to replace the simple method of assigning the relative humidity according to vertical extension of clouds. Use of geo-stationary images has the advantages of high temporal resolution and consistency with cloud patterns. The latter may be important for initializing the model physics such as convection.
0.4.4 Application of satellite information for improving inner structure of tropical cyclones
The structures of tropical cyclones have impact on both track and intensity of tropical cyclones. Due to sparsity of data, bogus model of tropical cyclones is introduced in current operational NWP system in China. The satellite data will help to modify the bogus cyclone so that it may be closer to its reality or eventually take place of bogus cyclones.
  1. Satellite winds
Both upper level and sea surface winds from satellite around the centers of tropical cyclones are valuable information of the inner structure of the tropical cyclones. Due to low vertical resolution and difficulties encountered in accurately assigning the height of observation, the detail of the structure of tropical cyclone could not be defined only with satellite winds. Development of techniques to retrieve the structure of the tropical cyclones is included in the whole GRAPeS data assimilation project. Two methodologies are being considered. One follows the technique of potential vorticity retrieval and the other intends to modify the bogus pattern to fit the observations.
  1. Total column water vapor(TCWV)
There are several space-borne instruments able to detect the total column water vapor or precipitable water vapor. These data could help to depict the asymmetry of cloud bands within tropical cyclones. The TCWV will be tested in the frame work of GRAPeS 3DVar in the near future and in the frame work of GRAPeS 4DVar when it is completed.
0.4.5 Summary
The assimilation of satellite data is the first priority in Chinese project of development of new generation NWP system. A variational assimilation system is being developed. With this new system, ATOVS radiances and satellite derived winds, including both sea surface winds and upper winds, may be directly assimilated and become the main data sources over the oceans. This will greatly alleviate the data voids problem over oceans. In addition to the usage in analysis of large scale flow pattern relevant to tropical cyclones, the satellite information, such as winds on different levels and TCWV, are also helpful to defining the structure of tropical cyclones. So traditional bogus model will be improved or eventually replaced by real time satellite observations. All progress in assimilation of satellite data are expected to make improvement in operational forecast of tropical cyclones.