Using Optic Sensors for Near-Continuous Water Quality Monitoring of Stormwater Runoff

Highlights Spectrophotometers enabled water quality monitoring for 96% of the runoff volume recorded during the study at a high frequency of 4 minutes, providing insights into pollutant dynamics. Local calibration of spectrophotometers proved more challenging in the stormwater context than the previously reported perennial streams due to inherent flashy flows and concentrations, resulting in limited representation of high-concentration local samples and an unstable relationship between the water color matrix and concentrations. Using instrument-reported to lab-reported concentration regressions (SLR), the annual nitrate, DOC, and TSS loads were estimated within an uncertainty range of ±10%, and within ±16% using absorbance to lab-reported concentration regressions (PSLR). Event load estimation using spectrophotometers had uncertainty estimated within ±10% around the mean for nitrate and DOC loads using the best calibrations for 90% of the events. Much higher uncertainties, between 40 and 60%, were observed for TSS. ABSTRACT. This paper presents methods, challenges, and uncertainties associated with using spectrophotometers for continuous stormwater water quality monitoring. Spectrophotometer sensors, proven insightful in other fields of hydrological research, were tested for monitoring the inflow to a constructed stormwater wetland. These sensors allowed for water quality monitoring at a 4-minute frequency, providing insight into pollutant dynamics, and enabled 95% of total recorded inflow volume to be sampled over 20 months of study. The main challenge of using spectrophotometers in non-sewer stormwater was instrument calibration due to inherently flashy flows and generally low concentration ranges of nitrate. To calibrate the sensor, local samples were collected by autosamplers and analyzed in the laboratory. Two local calibration methods were tested: simple linear regressions (SLR) between instrument and lab concentrations and partial least square regressions (PLSR) between the absorbance spectra and lab concentrations. The resulting annual and event loads were used to estimate the uncertainties associated with each regression model. The observed uncertainties were due to the (1) regression type, (2) uniformity of concentration range represented in calibration samples, and (3) presence or absence of stratification in the calibration samples. Stratified samples yielded lowest uncertainty for both calibration models. SLR model offered lower uncertainty ranges than PLSR for estimation of both annual and event loads for stormwater.