Abstract
Evaportranspiration each land cover was estimated from vegetation index, NDIV. Water budget method and Penman formula lead to the daily evaportranspiration from watershed while the satellite data, SPOT, gave NDVI each land cover. Then, their correlation gave the regression lines each land cover. On the other hand, the time series of NDVI each land cover except water surface the other hand, the time series of NDVI each land cover except water surface changes seasonally and fitted very well with logistic urves for less than 300 Julian day. Especially, the time series of NDVI for forest and grass also fitted very well with the logistic curves modified with seasonal periodicity (r2=0.997 and 0.971, respectively). Moreover, NDVI for forest gave good agreement with the time series of air temperature. Using these regression curves for NDVI, the temporal and spatial distributions of evapotranspiration from watershed can be estimated.

1.Introduction
Water budget analysis has been required for forest and agricultural fields from the viewpoint of global environment. Especially, evaportranspiration from vegetationdeermines water cycle and affects on not only agricultural production but also natural disasters. Traditionally evaportranspiration has been calculated with water budget method, meteorological approach (Brutsaert, 1982), and vegetation index in satellite data (Running and Nemani, 1988). In this study, from vegetation index in satellite data, the spatial distributions and time series of evaportranspiration each land cover were estimated from Tade river watershed in the Saga Plains. First, annual evaportranspiration for Tade river watershed was calculated with water budget method, while daily evaportranspiration each land cover was obtained with Penman formula (Brutsaert, 1982). Both the results lead to daily evaportranspiration each land cover. Second, vegetationindex, NDVI, each land cover was obtained from satellite data, and combined with the corresponding daily evaportranspiration by linear regression, with which the spatial distributions of evaportranspiration are calculated. Third, in order to obtain the time series of evaportranspiration each land cover, the seasonal fluctuation of NDVI was expressed with logistic curves from SPOT data in 1997. Moreover, fluctuation of NDVI was expressed with logistic curves from SPOT data in 1997. Moreover, the correlation between NDVI and meteorological data was calculated, which showed NDVI correlates with air temperature very well. With the results, the interpolation for time series of NDVI was obtained and the seasonal fluctuation for evaportranspiration was estimated. Therefore, using the above methods, the spatial distributions and time series of evaportranspiration can be estimated from satellite data.

2.Material and Methods

2.1 Data used
For calculation, we used hydrological data (precipitation and discharge), meteorological data (air temperature, humidity, surface temperature, wind speed, short-wave radiation), multi-temporal SPOT data (Table 1), and digital elevation maps with 1:25,000 scale and 50-m spatial resolution. The satellite data were used for mainly land cover classification and NDVI calculation.

Table 1 Satellite data used for this study

Satellite	Date
SPOT	1997/01/17,03/05,05/27,06/17,07/23,09/13,12/05

2.2 Observation Sites
A discharge observation site locates at Hiromaru Bridge over Trade River (catchment area: 28.37 km2) and a rainfall observation site locates at Saga Local Meteorological Station. Another observation site is at Agricultural Research Center of Saga Prefecture.

2.3 Water Budget Method
For long term observation, evaportranspiration was obtained from rainfall and discharge for the catchment. Water budget is given by

dS/dt = P-Q-E (1)
E = aE^p (2)
Q = kS^p (3)
where S is the catchment storage, t time, P precipitation, Q discharge, E evaportranspiration, a a coefficient (<1), Ep potential evaporation, k a coefficient, and p the exponent.

2.4 Penman Method
Daily evaportranspiration each land cover can be calculated from meteorological data with penman formula defined as (Brutsaert, 1982)

E =

D --------- D + g

Qn +

g --------- D + g

f(u)(ea-ea) (4)

Q_n = R_n/L_l  (5)

R_n = R_s(1-a_s) +e_sR_ld -R_lu (6)

R_ld = e_asT_a⁴ (7)

e_a = 1.24(e_a/T_a)^1/7 (8)

R_lu = e_ssT_s⁴ (9)
Where D = (de*/dT)^T, y =c_pP/el, Q_n is the available energy flux density in mm/day, f(u) the wind function in mm/day, u the mean wind speed in m/sec, e_a* the saturation vapor pressure in hpa,e_a the vapor pressure in hPa, R_n the net radiationin W/m², L_l the latent heat of vaporization (=2.454 x 10⁶ J/kg), R_s the short-wave radiationin W/m^2,, a_s the albedo of the surface, R_ld the downward long-wave radiation in W/m², R_lu the upward long-wave radiation in W/m², e_s the emissivity of the surface m(=0.97), e_a the atmospheric emissivity, s the Stefan-Boltzmann constant (5.67 x 10^-8 Wm^-2K^-4), T^a the air temperature in K, and T^s is the surface temperature in K.

2.5 Vegetation Index and Evapotranspiration
Vegetatin Index, NDVI, relates with the biomass and correlates with evaportranspiration (Nemani and Running, 1989; shin and Sawamoto, 1996; Sqwamoto and Shin, 1997). NDVI is defined as (Rouse et al., 1973; 1974)

NDVI =

IR - R ---------- IR + R

(10)

Where IR is the infrared pixel value and R is the red pixel value.

Here, since evaportranspiration for vegetation depends on leaf area index (LAI), the air temperature, and precipitation, the next equation may be assumed.

T_p = LAI f (T,P) (11)
Where Tp is the transpiration, LAI the leaf area index, and f(T,P) a function of T and P.LAI is expressed by (Running and Nemani, 1988)

LAI = a . exp (kNDVI) (12)
Where a and k are constants.

On the other hand, relationship between evaportranspiration and transpiration is given by (Campbell, 1985)

E = T_p / [1-exp(0.82LAI)] (13)
Thus, from Eq. (11), (12), and (13), the evaportranspiration can be expressed with NDVI as

E = b . exp (kNDVI) . f (T,P) / [1-exp(0.82a exp(kNDVI)] (14)
Where b is a constant

Therefore, the evaportranspiration increase exponentially with NDVI. However, as it increase almost lineally with NDVI within a certain range, the next linear regression was applied for this study.

E = A NDVI +b (15)
Where a and b are constants.