Key Words:
trophic state index, reservoir, random field simulation, remote sensing

Abstract:
Reservoir water quality is traditionally monitored and evaluated based on field data collected at limited locations. Whether the limited field data represent the overall trophic state of a vast water body is often disputed. In this study we utilize Landsat TM data to evaluate the overall trophic state of Te-Chi Reservoir in Central Taiwan. Three water quality parameters: chlorophyll-a (Chla) concentrations, total phosphorous (TP) measurement, and secchi disk depth (SDD), are found to have high correlations with transformed spectral features derived from bands 1, 2, 3 and 4 of TM data. Therefore, TM data are used to yield a trophic-state-index (TSI) map of the reservoir. Ranges of Chla, TP, and SDD measurements in Taiwan's reservoirs are generally much larger than that of Minnesota's lakes which original TSI model was developed; therefore, a modified TSI model was proposed for Techi Reservoir. Based on measurements of water quality parameters, TSI cutoff values of 66 and 78 are used to signify the reservoir trophic state of eutrophication and hypereutrophication. In order to provide a confidence level for reservoir trophic state evaluation, we employ a two-dimensional random field simulation technique to generate a large pool of TSI realizations. Using these realizations, probability distributions of the trophic states of specific reservoir cross-sections and the reservoir as a whole can be estimated. A probabilistic approach of reservoir trophic state evaluation is proposed as follows: the overall reservoir trophic state is considered eutrophic or hypereutrophic if the probability that the overall TSI exceeds the cutoff value (66 or 78) is greater than a predetermined level, say 0.9. Based on this criterion, the trophic state of Te-Chi reservoir is found to be eutrophic in summer and meso-eutrophic in winter.

1. Introduction
Traditionally the quality and trophic state of water in impoundments have been assessed using limnological methods and laboratory analyses of field-sampled data. Single- and multi-parameter indices were developed for trophic classification of lakes (Brezonik and Shannon, 1971; Beeton and Edmonson, 1972). Carlson (1977) proposed a trophic state index (TSI), that retains the