Confidence interval in estimating solute loads from a small forested catchment

HR: 08:45h
AN: H51K-04 [Abstracts]
TI: Confidence interval in estimating solute loads from a small forested catchment
AU: * TADA, A
EM: atada@kobe-u.ac.jp
AF: Graduate School of Agricultural Science, Kobe University, 1-1, Rokkodai, Nada, Kobe, 6578501, Japan
AU: TANAKAMARU, H
EM: tanakam@kobe-u.ac.jp
AF: Graduate School of Agricultural Science, Kobe University, 1-1, Rokkodai, Nada, Kobe, 6578501, Japan
AB: The evaluation of uncertainty in estimating mass flux (load) from catchments plays the important role in the evaluation of chemical weathering, TMDLs implementation, and so on. Loads from catchments are estimated with many methods such as weighted average, rating curve, regression model, ratio estimator, and composite method, considering the appropriate sampling strategy. Total solute loads for 10 months from a small forested catchment were calculated based on the high-temporal resolution data and used in evaluating the validity of 95% confidence intervals (CIs) of estimated loads. The effect of employing random and flow-stratified sampling methods on 95% CIs was also evaluated. Water quality data of the small forested catchment (12.8 ha) in Japan was collected every 15 minutes during 10 months in 2004 to acquire the “true values” of solute loads. Those data were measured by the monitoring equipment using FIP (flow injection potentiometry) method with ion-selective electrodes. Measured indices were sodium, potassium, and chloride ion in the stream water. Water quantity (discharge rate) data were measured continuously by the V-notch weir at the catchment outlet. The Beale ratio estimator was employed as the estimation method of solute loads because it was known as unbiased estimator. The bootstrap method was also used for calculating the 95% confidence intervals of solute loads with 2,000 bootstrap replications. Both flow-stratified and random sampling was adopted as sampling strategy which extracted sample data sets from the entire observations. Discharge rate seemed to be a dominant factor of solute concentration because the catchment was almost undisturbed. The validity of 95% CIs were evaluated using the number of inclusion of “true value” inside CIs out of 1,000 estimations derived from independently and iteratively extracted sample data sets. The number of samples in each data set was set to 5,500, 950, 470, 230, 40, and 20, equivalent to hourly, 6-hourly, 12-hourly, daily, weekly, and biweekly sampling intervals respectively. As a result, 95% CIs worked properly only at the random sampling which contained 5,500 samples out of 34,000 entire observed data. In other case, 95% CIs could give much less confidence for both random and flow-stratified sampling. The width of 95% CIs seemed to be appropriate but only 50% to 80% of 95% CIs out of 1,000 estimations contained true value inside. It was concluded that the Beale ratio estimator made slightly biased estimation of solute loads and those bias and failure in deriving proper 95% CIs came mainly from sampling method. The difference between random and flow-stratified sampling was also insignificant except sodium which had a strong non-linear relationship between loads and discharge rates. Flow-stratified sampling was not enough to provide the i.i.d. (independent and identically distributed) sample data sets for the purpose of solute loads estimation.
DE: 0496 Water quality
DE: 1806 Chemistry of fresh water
DE: 1873 Uncertainty assessment (3275)
SC: Hydrology [H]
MN: 2007 Fall Meeting