Abstract
We have been developing the model for NNP estimation using climate data and ADEOS-11\GLI data. This model is based on the relationship between the photosynthetic quantum efficiency and the new vegetation index obtained from satellite data by the pattern decomposition method. The new vegetation index is VIPD. VIPD reflects all multi-spectral data from visible to near-infrared wavelength in contrast to NDVI which is defined as a combination of red and near-infrared data. Therefore, VIPD is effective for the analysis of GLI data which has 36 wavelengths. We applied the model to Landsat\TM data and evaluated NPP based on the model.

1.Introduction
The net primary production (NPP) is defined as the net production of organic matter by the vegetation the balance between the gross primary production by photosynthesis and the respiration of vegetation. We have been developing the model of NPP estimation for ADEOS-11\GLI data. ADEOS-11 satellite will be launched in 2000 and GLI sensor has hyper-multi-spectral bands, namely 36 bands. GLI data covers global area for each 4 days. We will make a global NPP map based on this model using GLI data.

By now, we have mainly used data by NOAA/AVHRR which observes the global area frequently to obtain the global NPP map . AVHRR data has only two wavelength bands from visible to near-infrared region, namely red and near-infrared bands. However, the vegetation state affects on reflectance's from visible to near-infrared wavelength.Therefore in order to estimate NPP, it is more effective to use all GLI bands data.

The pattern decomposition method, that we developed for multi-spectral data, trans-forms all multi-spectral data to three coefficients data. The three coefficients are linear to a land cover ration in a pixel. The characteristics is important to analyze the satellite data, in particular, with low spatial resolution.We defined a new vegetation index based on the three coefficients, which includes all information from original data. It is called the Vegetation Index based on Pattern Decomposition, namely VIPD. VIPD was demonstrated to be linear to the vegetation cover ration and the photosynthetic quantum efficiency. We can estimate the production of organic matter by photosynthesis using these relationships.

In order to validate the model, we made NPP maps for Nara basin in Japan and the rain forest in Malaysia using Landsat/TM data and compared the results with that the by other models and the ground measurement data. Furthermore, we estimated NPP for rice fields of Nara basin in Japan using Landsat/TM data. Because we can consider that the rice yield is the net primary production of rice from the plantation to the harvest, it is possible to validate the model with more precise accuracy.

2. Pattern decomposition method and new vegetation index
The pattern decomposition method was developed as the analysis method of multi-spectral data [1, 2]. The concept of the method is that the spectral pattern composed with the multi-spectral data of a certain sample is able to be described by a combination of three standard patterns, namely water, vegetation and soil patterns. In other words, the method is based on the spectral mixing analysis, but the unique characteristic of the method is to fix the standard patterns as three. As a result, the multispectral data is reduced into the three decomposition coefficients (water, vegetation and soil). In addition to that, each coefficients includes the physical parameters associated with each other land cover type. Therefore, we can utilize the method for the extraction of the signature from multi-spectral data.

In fact, we can obtain the three decomposition coefficients from multi-spectral data using the pattern decomposition methods as follows.
A₁ = P_w1 X C_w + P_v1 X C_v + P_s1 X C_s + R₁
……..
A_n = P_wn X C_w + P_vn X C_v + P_sn X C_s + R_n           (1)
These equations are obtained for n-bands data. Here, A_i indicates the reflectance of band No.i from 1 to n bands. P_wi, P_vi and P_si mean the standard patterns for water, vegetation and soil, respectively. We determine the three decomposition coefficients C_w, C_v and C_s so that the remainders R_i is as small possible using the least square method.