2.6 Evaportranspiration Estimates
Annually evaportranspiration is calculated from precipitation and discharge each catchment with water budget method as mentioned above. The separation of watershed and the calculation of its area were derived from the digital elevation maps. Next, the land cover classification was carried out from SPOT data with cluster classification method and the spatial distribution of NDVI was calculated each land cover as well. At the same time, daily evaportranspiration was obtained from the meteorological data with Penmna method, modified by the result of annual evaportranspiration, and finally determined as actual values.

3. Results and Discussion

3.1 Evaportranspiration
In land cover classification, six kinds of land covers were determined as Table 2. Daily evaportranspiration each land cover was calculated with Penmna method and the annual evaportranspiration was obtained from the daily data in Table 3. Another annual evaportranspiration was calculated with water budget method as well in Table 4. The annual evaportranspiration of this watershed with more than 80% of the area occupied with forest showed 600-700 mm by water budget method and 700-1000 mm by Penman method, which overestimated the annual values for the former. The fluctuation pattern for the precipitation of the watershed was similar to that of the evaportranspiration. The result of Penmna method was though to be almost the same to the observationof pan evaporation. The time series of evaportranspiration each land cover shows the difference of 30% between water surface and bare soils or urban areas, which are the maximum and the minimum, and the similar temporal patterns among them. Finally, daily evaportranspiration each land cover was determined with the result of water budget method. Figure 1 illustrates the time series of NDVI in Trade River watershed, which includes different patterns each land cover. This result is contrary to that of the evaportranspiration, the temporal patterns of which are similar among different land covers. As shown in Eq. 914), evaportranspiration is the exponential function of NDVI or the linear to NDVI within a certain range. Therefore, the time series of evaportranspiration should be reflected by the temporal fluctuation of NDVI. Furthermore, since evaportranspiration depends on precipitation and air temperature strongly as well as land covers, the evaportranspiration may be the function of NDVI, precipitation, and air temperature as Eq. (14).

Table 2 Land cover classification (%)

Catchment	Forest	Paddy	Urban	Grass	Bare Soil	Water
Tade (1993)	81.0	15.2	1.4	1.3	0.6	0.5

Table 3 Estimates of Evaportranspiration from Tade River watershed by penman formula

Year	Forest	Paddy	Urban	Grass	Bare soil	Water	Total
1993	784.9	775.6	587.5	713.2	589.5	827.4	778.9
1997	1038.9	1008.4	813.7	958.4	816.4	1087.7	1029.1

Table 4 Hydrological estimates by water budget method in Tade River watershed

Year	Precipitation(mm/y)	Discharge(mm/y)	Evaportranspiration(mm/y)
1993	2566.5	1852.3	714.2
1997	2352.0	1697.4	654.6

Figure 1. NDVI in Tade River watershed in 1997
3.2 Time Series of NDVI
Apparently, the steady land cover in time such as forest and the unsteady land cover such as agricultural fields should be separated in analysis for the time series of NDVI. The NDVI of forest fluctuates in time with daily air temperature. The correlation between the NDVI and air temperature yielded the next equation.

NDVI = 0.0077T(°C) + 0.2713 (r² = 0.8722)
Therefore, the air temperature may be applied for the interpolation of NDVI for forest through the year.

On the other hand, the agriculture field such as a paddy field changes from grass to bare soils, water surface, crops and bare soils seasonally within three to five months. Especially in double cropping, land covers change through the year. NDVI for paddy fields fluctuates through the logistic curve from rice planting to its earring and approaches the maximum . From the earring to the mature period, the NDVI decreases and yields the minimum in rice reaping (Uchijima, 1976).

As NDVI fluctuates through a continuous curve, it can be interpolated with such a curve. Thus, NDVI for agricultural fields can be estimated through the year using such a curve as a logistic curve. Therefore, the next regression curves were applied for the time series of NDVI.

Air temperature curve :

NDVI = pT + q (16)
where p and q are constants.

Logistic curve:

NDVI =

A -------------- exp(-mt +n) +B

+ D

(17)

Logistic curve with periodicity:

NDVI =

A -------------- exp(-mt +n) +B

-Ct + D

(18)

where t is time, and A,B,C,D, m and n are constants.

As a result, the time series of NDVI could be fitted with these curves very well except water surface as shown in Table 5. Using the air temperature curves as (16), the time series of NDVI was estimated. Eq. (17) gave good agreement with NDVI for less than 300 Julian day. Eq. 918) fitted with NDVI for all land covers almost through the year as shown in Figure 2. Among them, the best fitting of Eq. (18) was obtained for forest (r2=0.997) and grass (r2=0.971) with seasonal periodicity.


Figure 2. Time series of NDVI for each land cover and its regression curve
Table 5 Regression curves and their coefficient of determination R2

NDVI	Forest	Grass	Paddy Field	Water	Urban Area	Bare Soils
Temperature	0.951	0.616	0.659	0.275	0.645	0.603
Logistic Curve	0.648	0.650	0.309	0.002	0.196	0.092
Logistic Curve(Julian Day <300)	0.981	0.990	0.997	0.309	0.985	0.955
Modified Logistic Curve	0.997	0.971	0.726	0.956	0.523	0.904

Table 6 Regression lines between NDVI and Evapotranspiration

Land cover	Regression Lines	R2
Forest	E = 13.89 NDVI - 2.415	0.744
Paddy	E = 9.61 NDVI + 0.884	0.589
Grass	E = 9.63 NDVI -0.124	0.509
Bare Soil	E = 31.9 NDVI -0.406	0.755
Urban Area	E = 9.44 NDVI + 0.749	0.634