Title: Sources and Variability of Cryptosporidium in the Milwaukee River Watershed: An Application of Autom
1Sources and Variability of Cryptosporidium in the
Milwaukee River Watershed An Application of
Automatic Sampling for Pathogens
Steve Corsi, Rob Waschbusch, John Walker, Jon
Standridge Cooperator Water Environment
Research Foundation
2Introduction
- The public has come to expect a safe source of
drinking water - 1993 Cryptosporidium outbreak in Milwaukee
- 403,000 people fell ill
- 150 died
- This incident made Waterborne Cryptosporidium an
international water quality issue - Original source of cryptosporidium was not
identified - Many sources exist in the Milwaukee River
Watershed
3Objectives
- Define relative magnitude and variability of
Cryptosporidium from major sources - Urban land use
- Agricultural land use
- Public wastewater discharges (POTW)
- Characterize contributions by environmental
factors - Hydrograph timing
- Climatic effects
- Seasonal variations
- Develop a model to predict the probability of
high Cryptosporidium concentrations
4Milwaukee River Watershed
Monitoring Locations
5Crypto Sampling Approach
- 2 years of monitoring
- Fixed-interval sampling every 4 weeks
- Event sampling precipitation and snowmelt runoff
events - 2 events sampled per season each year at
subwatershed sites - 6 months per POTW (3-4 events FI)
6Automatic sampling Techniques Pros and Cons
- Pros
- Unattended Sampling
- Less storm chasing (a better nights sleep)
- Flow weighted or time based options
- Hydrograph coverage
- Represent short term variability
- Long-term composite
- Loading computations (Relative source
contributions) - Event-triggered sample collection
- Flow, turbidity, rain, . . .
- Telemetry for remote access
- Reduced labor
- Cons
- Start-up equipment and installation costs
- Fixed location for equipment
- Point samples rather than integrated over cross
section - Some additional QA/QC requirements
7Monitoring Station Schematic
8Additional Instrumentation for other studies
- Meteorological sensors (wind, solar radiation,
rainfall) - Wave height
- Fluorometer for dye tracer studies
- Many other sensors . . .
- All can be interfaced to datalogger system
- Used to trigger sampling
- Used to complement other data in final
analysis/modeling
9Monitoring Station
10Extra Details for Clean Sampling
11Two years of Data Collection
- Cryptosporidium concentrations
- Streamflow
- Precipitation
- Turbidity
- Land-use data
- Water temperature, pH, conductivity (Underwood
Creek)
12Cedar Creek Sampling Periods
13Underwood Creek Sampling Periods
14Event Sampling Based onHydrograph Position
15Results
- Data preparation (Censored data estimates)
- Exploratory data analysis
- Site by site comparison
- Event loadings
- Effect of hydrograph position
- Interevent variability
- Seasonality of data
- Probabilistic regression model
16Percent Occurrence
17Censored data analysis
- Using simple substitution can bias the data
- For this project
- Estimated values for censored data
- Assumption of a single statistical distribution
for each site (lognormal) - Procedures defined by Helsel and Cohn (Water
Resources Research, 1988, Vol 24, No 12, p.
1997-2004)
18Censored Value Estimates Underwood Creek
19Site by Site ComparisonConcentrations
100000
78
28
71
28
16
9
o
10000
x
1000
x
x
Cryptosporidium (oocysts/100L)
100
10
Event FI Cedar Creek
Event FI Underwood Creek
Event FI POTW 1
1
20Event loadings Cryptosporidium Unit Area Event
Loads
16
16
1000
100
Million Cryptosporidium per square mile
10
1.0
Cedar Underwood Creek
Creek
21Hydrograph Position Underwood Creek
100000
14
16
21
16
32
o
10000
1000
Cryptosporidium (oocysts/100-L)
100
10
1
Base First Rise Peak Falling Flow Flush
22Hydrograph position at Other Sites
- No differences were detected due to hydrograph
position at - Cedar Creek
- POTW 1
23Cedar Creek Cryptosporidium Concentration for
Individual Runoff Events
24Underwood Creek Cryptosporidium Concentration
for individual runoff Events
25POTW 1 Cryptosporidium Concentration for
individual runoff Events
26Seasonal effects
- Cedar creek had slightly higher concentrations in
the summer (plt0.05) - Underwood Creek and POTW sites did not show a
seasonal trend - No general conclusions on seasonality can be made
from this study
27Seasonal Effects From Previous Studies
- Greater concentrations during
- Fall and early spring (LeChevallier et al. 2002)
- Fall and winter (Bodley-Tickell et al. 2002)
- Winter and summer (Rouquet et al., 2000)
- Winter and spring (Atherholt et al. 1998)
28Statistical Model
- Discriminant function analysis
- To help determine which variables may explain
variability - Logistic Regression
- To predict the probability of a high
Cryptosporidium concentration - Performed on
- combination sites
- Individual sites
- Baseflow samples
- Event samples
29Independent Variables Used in Statistical Analysis
- Flow
- Turbidity
- Precipitation depth
- Precipitation intensity (max 5, 15, and 30 min)
- Antecedent precipitation (1, 3, 5 day)
- USLE erosivity index
- Hydrograph position
- Water temperature, pH, conductance (Underwood
Creek) - Annually cyclical term for seasonality Cos(T),
Sin(T)
30Logistic Regression Results
- Two different equations were most successful in
predicting high Cryptosporidium concentrations
(gt300 oocysts/100L) - Baseflow periods for both Cedar and Underwood
Creeks - Event periods for Cedar and Underwood Creeks
31Baseflow Periods
- Probability of high cryptosporidium increases
with increasing - Turbidity
- Antecedent precipitation
32Event periods
- Probability of high concentrations increases
- Precipitation intensity
- Antecedent precipitation
- During November and January
- Probability of high concentrations decreases on
- Falling limb of hydrograph
33Prediction Accuracy for High Cryptosporidium
Concentraions (gt300 oocyst/100L)
34Conclusions
- Concentrations and Unit Area Loads were greater
at the urban site than the rural or POTW sites
- Concentrations in subwatersheds were greater
during runoff periods than during baseflow periods
- Fixed interval sampling did not represent the
overall magnitude and variability - event samples (more than one) are necessary
35Conclusions (Continued)
- Discriminant function analysis and logistic
regressions can provide predictions of the
occurrence of high concentrations
36Daze