Abstract
Nine estimators of pollutant loading were tested for robustness to sample size reduction in a small suburban watershed of the New York City, USA drinking supply area. Discharge data were recorded from August 2001 to September 2002, which overlapped with collection of 30 baseflow and 35 storm samples for ten constituents (Ca, Cl, DOC, Mg, Na, NH4, NO3, SO4, TN, and TP). Loading estimators were: simple average, period average, linear interpolation average, discharge frequency, Beale Ratio, simple regression, quasi‐maximum likelihood (QMLE) regression, smear regression, and maximum variance unbiased estimator (MVUE) regression methods. Sample size reduction explored uniform random removal without replacement down to 10% of the original sample as well as preferential random and non‐random reduction of baseflow and storm flow. At 10% random sample reduction, the median departure for estimator loads was 2% from the full‐sample prediction, while at 90% of the original sample the median departure was at 35%. Preferential reduction was most robust when storm events were removed, and baseflow retained. In this scenario, median estimator departure remained at 10% given a 75% reduction of storm event samples. Regression estimator methods were generally best when storm event data were not fully removed, and averaging, interpolation, and frequency based estimator methods were best when only baseflow data were retained.