1. Preface
The Tibetan Plateau is mainly formed of tall mountain and highland wihr an average altitude of 4000 meters above sea level. It is the highest plateau in the world. The Tibetan plateau is with a topography of slow depressed from west to east and sharp precipitation in the north and south sides. In the plateau, the Mountain of Altun Shan ~ Qilian Shan, Kunlunshan, Karakorum Shan~Tanggula Shan, Gangside Shan~ Nyainqen Tanglha Shan, Himalayas arrange with the west-eastward direction. As a highland, the Tibetan Plateau has an important impact on the weather system and climate change over China and even East Asia. Because of the complicated topography, the conventional observation stations over there just have a small coverage of the Tibetan Plateau which can not satisfy the needs of weather forecast and climate research. Due to the large coverage and complicated topographic structure, 4 zones with the boundary of 900 E and 360 N over the plateau were divided for improving the retrieval accuracy. From 1991 to 1994, the mean square deviation of the maximum shelter temperature observation in the July is 3.1~4.30C over the 4 zones as the samples were not even. The TOVS instrument on board NOAA satellite has a powerful remote sensing ability and with more observation bands. In the paper, it takes the MSU (microwave sounder) of TOVS instrument, with its advantages of penetrating cloud, to obtain the continuos meteorological parameters that has more representation over the plateau comparing with the conventional data.

2. Data and Data Processing

2.1 NOAA satellite data and its processing
The NOAA satellite orbit of 14.30 (local time) is close to the time of the occurrence of the daily maximum shelter temperature. Table 1 shows the wavelength of related channels of MSU and HIRS/2, contributing layer of peak energy, and the representing name of data. Before the preprocessing of retrieval, there is a limb darken and water vapor attenuation correction for the channel 8 of HIRS/2,as well as a bi-directional reflectance correction for the visible channel. The limb darken correction, land surface reflectivity correction and the correction of antenna lobe effect also will be done for the MSU.

2.2 Conventional data

The conventional data come from 2 datasets. The dataset of January has 2117 records that obtained in January from 1991 to 1994, it includes the daily maximum shelter temperature of the months. The datasets of July contains 1877 records that obtained in July from 1991 to 1994,it also includes the daily maximum shelter temperature of each July.

2.3 Above sea level height of satellite observation point The meteorological elements are related to the height above sea level, as the temperature would decrease when the height increases. The Tibetan Plateau is not only with a high height point above sea level Is 3000 meters while the highest point the highest point above sea level is the plateau is more than 8000 meters. Taking the above sea level height as a factor in the model, it can undoubtedly increase the accuracy of forecast. The result indicates that with this factor in the model, the calculation accuracy of ternary linear can increase at least 0.5°C and this is impressive in the satellite data retrieval. If the known above sea level height of observation station is accepted in the process of regression analysis and calculation coefficients, the satellite observation point would be random in the modeling calculation of temperature once the model is setup. But at present the satellite products still can not provide more accurate data for the above sea level height. For using the factor to simulate topographic height of Tibetan Plateau, the zonal division assignment is given by longitude and latitude control program (Table 2). The key assignment is later put to the projecting topography in the zone for having the corresponding height value of every observation point. After the process, it is almost identical for the simulated above the sea level height and the specific one. In order to improve calculation accuracy, the dividing of the topographic height would be the finer the better.

2.4 the matching of data among HIRS/2, MSU and surface observation
The satellite sub-point resolution for HIRS/2 and MSU is 17.4 km and 109.3 km respectively. To interpolate MSU data to the corresponding point of HIRS/2 for setting MSU data in each HIRS/2 data point, and then match the HIRS/2 data with ground observation data[1].

3. Cloud Detection
The radiation of underlying surface in the channel 1 of MSU is related to the dielectric constant of the surface. As in the way to the satellite, microwave radiation just suffers a weaker attenuation from atmosphere and condensation water vapor, it is able to penetrate cloud in the cloudy day. The microwave remote sensing has the advantage of having a little affect from cloud. But its emissive energy is lower than long wave and is less sensitive to the change of temperature comparing with long wave radiation.

The channel 8 of HIRS/2 is in the position with the maximum thermal radiation of the earth-atmosphere system^[2]. Under the clear sky condition, the satellite obtained radiation is directly, the satellite obtained radiation is directly from land surface thus the TM8 is high than TM1 for the same point. With this, we can identify the point having TM8 lower than TB1 as the cloudy point otherwise, it is point without cloud. More detailed information can be found from the paper "Monitoring Snow Cover With TOVS Data"^[3].

4. The Setup of Model
The Tibetan Plateau is a region with complicated topographical structure and wide coverage. Due to the statistical result is closely related to the representativeness of sample, the plateau is then divided into 4 zones with the boundary of 900E and 300N. The zone exceeding 90oE and not reaching 300N is as the zone 1; the zone exceeding both 900 E and 300N is as the zone 2; the zone not coming up to both 900E and 300N is as the zone 3; and the zone not reaching 900E but exceeding 300N is as the zone 4; the independent factors with major contribution for the ternary linear regression equation are screened from both MSU and HIRS/2 in each zone. The setting-up of model is based on the maximum composite correlation (R) and minimum mean square deviation (RMS). The equation of linear regression is as below:

In the equation, bo is the intercept, bi regression bi represents coefficient of model and xi id the independent variable.

The model contains winter season (January) and summer season (July), as well as types of clear sky and cloudy day. There are totally 16 modes to be set for the 4 zones. In the table 3, it presents the independent variable, number of sample, coefficient of composite correlation, mean square deviation, and F statistics of the 16 modes. Except the height above the sea level, in table 3, other two factors are mainly from microwave, which occupy about 85% of the independent t variable in the 16 modes. Figure 1 is the correlation dissemination chart of the 4 modes in zone 1, it shows the dissemination between simulated result and real temperature. One can find in figure 1 that the results from the clear day are better than the ones from cloudy day no matter in summer or winter. In the cloudy day model, most of the simulation values are normal just except a few low temperature data. It indicates in table 3 that the correlation dissemination of other zones is better than zone 1, the result of zone 1 id the worst among the 4 zones.