2.Materials and Methods

2.1 The Study Area
A small catchment test area (11 km²) was chosen to exemplify the essential points of this research. The study area represents a typical tropical environment prone to high degree of rain-induced erosion. It covers the eastern portion of Ishigaki, an island south of the main Okinawa Island Prefecture in Japan. It is located about 24º23' N latitude, 124º14' E longitude, with topography varying from flat, undulating to hilly terrain currently being subjected to intense agricultural cultivation. Crops consists mainly of tobacco, sugarcane, rice and pineapple. The area is drained through the Todoroki River with outlet to the coastal area, where a special concern is focused on mitigating the effects of sediment discharge on coral reef zones (Dikou and Takeaki, 1999).

2.2 Remote Sensing and Field Observation Data
Field campaigns were conducted on three occasions: early June, early August and early September with various inland data-logging instruments to measure river discharge, turbidity, depth and rainfall in different locations inside the watershed. Spectral signatures for both soil and vegetation were also gathered to parametrize reflectances from 320 to 1080 mm range. Water samples have been previously obtained for turbidity-sediment concentration calibration of the instruments. Soil samples were processed in the laboratory to obtain various hydrologic parameters such as hydraulic conductivity, soil moisture, porosity, grain size and typing. Vegetation structure measures such as plant spacing, layering and leaf dimensions were also recorded.

Due to the absence of remotely-sensed data, both spaceborne and airborne, within the period of field observations, the land cover captured from the aerial photographs taken in 1995 was assumed to be the prevailing land surface condition. The use of an asynchronous data can be justified since the dates of the aerial photography lie within the same season of the year and that no major changes in cropping patterns and infrastructure developments has occurred since.

2.3 Spectral Mixture Modeling of Vegetation and Soil
Spectral unmixing is a deconvolution technique that aims to decompose the mixed reflectance spectrum, R_i(l) of a ground element in an imagery into landcover components based on the LSMM (Settle and Drake, 1993) given by:

and


where R_f(l) is reflectance of end-member spectrum of a homogenous land cover for each band i of n land cover types, and ei is the residual noise. The objective is to estimate , for each land