Dataset 1: 21 years (Oct 1973–Sep 1994)

Dataset 2: Same dates and a Monte Carlo simulation method used

Dataset 3: 10 years of reservoir operation (Oct1997–Sep2007).

Dataset 4: Total reservoir input and output values

Training = 80% of dataset

Test = 20% of Dataset

200 cases were used for training
52 cases were used for testing

number of errors for the 200 cases was calculated as 3 (1.5%)

A diffusion matrix was formed

Test results only 3 cases were wrongly estimated(5.8%)

200 cases (months) were used for training
52 cases were used for testing

The number of errors for the 200 cases was calculated as 3 (1.5%)

test results, 3 cases were incorrectly estimated (5.8%)

When the 2 methods were compared, in 252 cases (months) a clear difference was seen in only 2 case

Error rates of the rules were significantly low

results may not be applied over all dry and wet periods

errors for the 96 cases was only 1 (1%)

testing data, 24 cases were used to test the
4 rules derived from 96 training cases

3 cases (12.5%) were incorrectly estimated

error rates of the rules were relatively high

results may not be applicable over all dry and wet periods

96 cases were used for training
24 cases were used for testing

number of errors in 96 cases was calculated as 21 (21.9%)

testing data, 11 of 24 cases were accurately estimated