Using fluorescence for rapid wastewater presence detection in sources of drinking water

Size: px
Start display at page:

Download "Using fluorescence for rapid wastewater presence detection in sources of drinking water"

Transcription

1 Using fluorescence for rapid wastewater presence detection in sources of drinking water Nicolas Peleato, Robert C. Andrews, Ray L. Legge* Department of Civil Engineering, University of Toronto and *Department of Chemical Engineering, University of Waterloo October 29, 2014

2 Source Water Impacts Increased stresses on surface waters (particularly urban areas) Climate change/rainfall patterns Increased urban developments Planned and unplanned impacts of wastewater need to be addressed Drinking water source waters also typically are wastewater receiving bodies Wastewaters treated to reduce public health risk from pathogens and limit organic loading 2

3 3

4 Source Water Impacts 4

5 Source Water Impacts Hard to address impacts on drinking water source waters: Treatment failures Bypasses Changes in surface water currents Low probability/high impact events Anthropogenic contaminants Consumer confidence Need tools to readily quantify level of potential wastewater impacts Ideally Inexpensive, rapid, and in real time 5

6 Monitoring for Impacts What surface water quality assessment tools do we have? Total organic carbon (TOC) ph Conductivity Turbidity Nitrate Biochemical oxygen demand (BOD) Chemical oxygen demand (COD) Many, many more! Studies have identified TOC and conductivity as appropriate for possible use in real time impact detection Detection limits between 1% and 4% wastewater by volume 6

7 Monitoring for Impacts On-line monitoring optical technologies are attractive Non-destructive Usually inexpensive Accurate Limited to no sample preparation needed Typically quick results Optical technologies relevant to water: Ultraviolet absorbance (UVA) Fluorescence 7

8 UVA and Fluorescence Light source Light intensity detector Amount of light emitted: Fluorescence Wavelength selector Sample Wavelength selector Light intensity detector Amount of light intensity lost: UVA 8

9 Extinction or absorbance Ultraviolet Absorbance Usually measured at 254 nm Indicator of organic content Many wavelengths can be measured making a full spectrum or water fingerprint Scans over a day (right) show rapid changes to water quality using UVA 1 1 Langergraber, vand den Broeke, J., Lettl, W., Weingartner, A., Real-time detection of possible harmful events using UV/vis spectrometery. Spectroscopy Europe 18.4,

10 Emission (nm) Intensity Fluorescence - Water Quality Monitoring Fluorescence intensity values collected Produce a full spectrum or fluorescence excitationemission matrix (FEEM) Excitation: nm (10 nm steps) Emission: nm (1 nm steps) Peak locations tell us: compound type Peak intensities tell us: linear relation to concentration Excitation (nm) 0 10

11 Fluorescence - Water Quality Monitoring Use of distinct peak locations has been shown to be a useful technique for natural organic matter (NOM) characterization Typically can identify several organic compounds: Humic acids Fulvic acids Proteins through aromatic amino acids (tryptophan, tyrosine, phenylalanine) Products of microbial metabolism Possible identification of polyaromatic hydrocarbons (PAH) 11

12 Assumptions & Hypothesis Assumptions: Biological treatment of wastewater will increase protein concentrations Tryptophan and tyrosine fluorescence - well correlated with protein concentrations Elevated fraction of NOM as proteins is unique to wastewaters Hypothesis: Protein-like peaks identified in FEEMs can identify presence of wastewater 12

13 Results Lake Ontario Sampling: Lake Ontario (Toronto) Secondary effluent (SE) from Toronto wastewater facility (discharges to Lake Ontario) Secondary effluent (SE) mixed in lab with raw water at known concentrations (by volume) 2%, 1%, 0.16%, 0.08%, 0.02% Lake Ontario water with the wastewater also analyzed (0% SE) 7 sampling periods Grab samples only 13

14 Emission (nm) Intensity Emission (nm) Intensity Results Lake Ontario 600 Raw water 0.16% wastewater (1:625) a b e Excitation (nm) c d a, humic acid b, fulvic acid c, protein-like Excitation (nm) d, protein-like e, protein-like

15 Feature extraction FEEMs present a complex analysis problem since the dimensionality of each sample is very high (6669 variables) Feature extraction: transform data into a set of features that best differentiate samples (i.e. dimensionality reduction) Peak picking: intensity at pre-specified peak locations Regional integration: volume contained within certain boundaries around pre-specified peak locations Principal component analysis (PCA): dimensionality reduction based on orthogonal directions of maximum variance (two-way) Parallel factors analysis (PARAFAC): dimensionality reduction based on three-way factor analysis Ratios: indicate compositional shifts. F 450 /F 500 (@ 370 nm excitation) has been shown to indicate microbial activity 15

16 Intensity Results: changes to peaks 6 peaks were identified in the spectra Correlation between peak intensities and SE concentrations 100 Peak Slope R C A A (humic acid) B (fulvic acid) B C (protein-like) E D F D (protein-like) E (protein-like) Secondary effluent concentration (% by volume) F (protein-like)

17 Results: PCA & PARAFAC PCA 0% SE 1% SE 0.08% SE 2% SE 0.16% SE 0.02% SE PARAFAC Regions represent quartiles of the point distributions 17

18 Classification Measures provide separation of samples based on secondary effluent concentration High degree of overlap between concentrations Want to develop a multivariate model which can accurately categorize influenced and non-influenced samples Popular classification models assessed: Support vector machine (SVM) K-nearest neighbours (KNN) Linear discriminant analysis (LDA) Peak picking with non-linear SVM Accuracy > 93% False positive/negatives < 2% 18

19 Emission (nm) Intensity Emission (nm) Intensity Results Lakefield, Ontario Fluorescence spectra are expected to differ for varying wastewater quality/type and source waters Need to test high organic backgrounds (rivers) and several wastewater qualities Lakefield wastewater is treated by a large lagoon system and discharges into the Otonabee River (high organic) Excitation (nm) Otonabee River water Excitation (nm) Lakefield wastewater 0 19

20 Results Continuous Monitoring All experiments have been only grab samples the intention is for fluorescence to give real time information! We have developed and are testing a set up able to take a sample every 10 minutes 20

21 Peak a (protein-like) integration (protein-like) Results Continuous Monitoring TOC difference < 0.1 mg/l Black: raw water; Red: 0.5% wastewater Sample time (minutes) 21

22 Protein-like score Results Continuous Monitoring Addition of Wastewater (0.5%) Sample time (minutes) 22

23 Conclusions Increased protein peak intensities observed for low levels of wastewater Several feature extraction methods and classification methods were analyzed Optimal combinations so far have a > 93% accuracy False positive and negative rates < 2 % Wastewater treatment and receiving water quality impact method accuracy Continuous monitoring has been established with promising initial results 23

24 Acknowledgments Toronto Water Abhay Tadwalkar, David Scott, R.L. Clark staff, Vanessa Wilson Peterborough and Lakefield staff Kevan Light, John Armour, Chris Norman 24

25 NSERC Chair Partners: City of Barrie Durham Region GE Water & Process Technologies Lake Huron and Elgin Area Primary Water Supply Systems (London) Peterborough Utilities Commission Ontario Clean Water Agency Regional Municipality of York Regional Municipality of Peel Toronto Water 25