Assessing Coastal Plumes in a Region of Multiple Discharges:...
8
Assessing Coastal Plumes in a Region of Multiple Discharges: The U.S. -Mexico Border SUNG YONG KIM,* ,† ERIC J. TERRILL, † AND BRUCE D. CORNUELLE ‡ Marine Physical Laboratory, Scripps Institution of Oceanography, University of California, San Diego, La Jolla, California 92093, and Climate, Atmospheric Science, and Physical Oceanography, Scripps Institution of Oceanography, University of California, San Diego, La Jolla, California 92093 Received March 17, 2009. Revised manuscript received June 22, 2009. Accepted July 16, 2009. The San Diego/Tijuana border region has several environmental challenges with regard to assessing water quality impacts resulting from local coastal ocean discharges for which transport is not hindered by political boundaries. While an understanding of the fate and transport of these discharged plumes has a broad audience, the spatial and temporal scales of the physical processes present numerous challenges in conducting assessment with any fidelity. To address these needs, a data- driven model of the transport of both shoreline and offshore discharges is developed and operated in a hindcast mode for a four-year period to analyze regional connectivity between the discharges and the receiving of waters and the coastline. The plume exposure hindcast model is driven by surface current data generated by a network of high-frequency radars. Observations provided by both boat-based CTD measurements and fixed oceanographic moorings are used with the Roberts-Snyder-Baumgartner model to predict the plume rise height. The surface transport model outputs are compared with shoreline samples of fecal indicator bacteria (FIB), and the skill of the model to assess low water quality is evaluated using receiver operating characteristic (ROC) analysis. Introduction Impaired coastal water quality from point source discharges is a well-known concern for human health, degradation of marine ecosystems, and negative impacts to coastal econo- mies when beaches are closed. At Imperial Beach located in southern San Diego, a significant number of beach closures by the San Diego County Department of Environmental Health (DEH) results from the presence of high concentra- tions of fecal indicator bacteria (FIB) exceeding California health standards - Assembly Bill 411 (AB411, Statutes of 1997, Chapter 765). Similarly, high FIB levels are periodically reported on the Tijuana shoreline in the sparse sampling that takes place within 10 miles north and south of the Tijuana River mouth. Not only is the incidence of bacterial con- tamination and associated beach closures a problem, but time lags between sampling of the coastal water and completion of the analysis likely result in situations when beach waters may be clean when closed, and not clean when open. The beach closure has similar economic consequences regardless of FIB levels. Possible sources of bacterial contamination responsible for beach closures in this region include the Tijuana River (TJR), the South Bay International Water Treatment Plant Ocean Outfall (SBIWTP, called herein SBO), northward flow from the Punta Bandera treatment plant discharged at San Antonio de los Buenos (PBD), and local runoff from Imperial Beach. While there is political and public interest in coastal water quality, especially in the cross-border context on which this paper is based, the ocean monitoring of the physics responsible for the transport of a discharge plume has historically not been conducted at the appropriate time and space scales. This is principally due to the high cost of continuous measurements of ocean conditions and the relatively short decorrelation length scale of a point measurement of ocean current. While the TJR flow is a likely cause for impaired water quality during the rainy season, little is known about the offshore and alongshore extent of the river plume. Likewise, impaired water quality events can occur during the dry season, indicating that plume water from one of the other two discharges may impact the coastline. Stormwater and wastewater discharges in the Southern California region have been addressed using other obser- vational methodologies, including applications of traditional in situ oceanographic observations, analysis of historical data gathered as part of the National Pollutant Discharge Elimi- nation System (NPDES) permit, and interpretation of in- termittently obtained remote sensing data on the river discharge in the coastal area and the ocean outfall plume. The dynamics of both river and outfall plumes have a history of experimental studies which provide a basis for the approach described herein. The river plume models depend on the width of the river discharge, bottom topography, wind speed and direction, ambient current field, input flow rate, and inertial forcing (1, 2). The vertical rise and dilution of a sewage plume discharged at the seafloor can be addressed through models based on hydraulic principles of buoyant jets in stratified and unstratified fluids, with the Roberts- Snyder-Baumgartner (RSB) model (3) serving as a well documented model in use by the U.S. Environmental Protection Agency (EPA) for evaluating near-field discharge plume behavior. The approach adopted here to understand the plume transport in the coastal zone is to integrate surface current data measured by an array of high-frequency (HF) radars and produce a Lagrangian analysis of the plume trajectory. A primary distinction of this study from previous studies is the four-year duration of a continuous hindcast analysis conducted on an hourly interval. This paper is organized into three sections. First, water quality, rainfall, surface currents, and ocean stratification are summarized to provide a description of the study region conditions (see Supporting Information (SI) for more details). Then a hindcast analysis of three regional discharges using a random walk model and the exposure kernels are described. Lastly, we present the model results and discuss underlying assumptions used in the model. The summer (dry) and winter (wet) seasons discussed in this paper indicate April to September and October to March, respectively. Summary of Observations and a Plume Exposure Hindcast Model Water Quality Sampling. Water quality standards in the State of California Health and Safety Code (AB 411) establish a set * Corresponding author phone: +1 (858) 822-5201; fax: +1 (858) 534-7132; e-mail: [email protected]. † Marine Physical Laboratory. ‡ Climate, Atmospheric Science, and Physical Oceanography. Environ. Sci. Technol. 2009, 43, 7450–7457 7450 9 ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 43, NO. 19, 2009 10.1021/es900775p CCC: $40.75 2009 American Chemical Society Published on Web 08/21/2009 Downloaded by UNIV OF CALIFORNIA CDL ACQUISITIONS on September 29, 2009 | http://pubs.acs.org Publication Date (Web): August 21, 2009 | doi: 10.1021/es900775p
Transcript of Assessing Coastal Plumes in a Region of Multiple Discharges:...
Assessing Coastal Plumes in aRegion of Multiple Discharges: TheU.S.-Mexico BorderS U N G Y O N G K I M , * , † E R I C J . T E R R I L L , †
A N D B R U C E D . C O R N U E L L E ‡
Marine Physical Laboratory, Scripps Institution ofOceanography, University of California, San Diego, La Jolla,California 92093, and Climate, Atmospheric Science, andPhysical Oceanography, Scripps Institution of Oceanography,University of California, San Diego, La Jolla, California 92093
Received March 17, 2009. Revised manuscript receivedJune 22, 2009. Accepted July 16, 2009.
The San Diego/Tijuana border region has several environmentalchallenges with regard to assessing water quality impactsresulting from local coastal ocean discharges for which transportis not hindered by political boundaries. While an understandingof the fate and transport of these discharged plumes has abroad audience, the spatial and temporal scales of the physicalprocesses present numerous challenges in conductingassessment with any fidelity. To address these needs, a data-driven model of the transport of both shoreline and offshoredischarges is developed and operated in a hindcast mode fora four-year period to analyze regional connectivity betweenthe discharges and the receiving of waters and the coastline.The plume exposure hindcast model is driven by surfacecurrent data generated by a network of high-frequency radars.Observations provided by both boat-based CTD measurementsand fixed oceanographic moorings are used with theRoberts-Snyder-Baumgartner model to predict the plumerise height. The surface transport model outputs are comparedwith shoreline samples of fecal indicator bacteria (FIB), andthe skill of the model to assess low water quality is evaluatedusing receiver operating characteristic (ROC) analysis.
IntroductionImpaired coastal water quality from point source dischargesis a well-known concern for human health, degradation ofmarine ecosystems, and negative impacts to coastal econo-mies when beaches are closed. At Imperial Beach located insouthern San Diego, a significant number of beach closuresby the San Diego County Department of EnvironmentalHealth (DEH) results from the presence of high concentra-tions of fecal indicator bacteria (FIB) exceeding Californiahealth standards - Assembly Bill 411 (AB411, Statutes of 1997,Chapter 765). Similarly, high FIB levels are periodicallyreported on the Tijuana shoreline in the sparse samplingthat takes place within 10 miles north and south of the TijuanaRiver mouth. Not only is the incidence of bacterial con-tamination and associated beach closures a problem, buttime lags between sampling of the coastal water andcompletion of the analysis likely result in situations when
beach waters may be clean when closed, and not clean whenopen. The beach closure has similar economic consequencesregardless of FIB levels.
Possible sources of bacterial contamination responsible forbeach closures in this region include the Tijuana River (TJR),the South Bay International Water Treatment Plant OceanOutfall (SBIWTP, called herein SBO), northward flow from thePunta Bandera treatment plant discharged at San Antonio delos Buenos (PBD), and local runoff from Imperial Beach.
While there is political and public interest in coastal waterquality, especially in the cross-border context on which thispaper is based, the ocean monitoring of the physics responsiblefor the transport of a discharge plume has historically not beenconducted at the appropriate time and space scales. This isprincipally due to the high cost of continuous measurementsof ocean conditions and the relatively short decorrelation lengthscale of a point measurement of ocean current. While the TJRflow is a likely cause for impaired water quality during therainy season, little is known about the offshore andalongshore extent of the river plume. Likewise, impairedwater quality events can occur during the dry season,indicating that plume water from one of the other twodischarges may impact the coastline.
Stormwater and wastewater discharges in the SouthernCalifornia region have been addressed using other obser-vational methodologies, including applications of traditionalin situ oceanographic observations, analysis of historical datagathered as part of the National Pollutant Discharge Elimi-nation System (NPDES) permit, and interpretation of in-termittently obtained remote sensing data on the riverdischarge in the coastal area and the ocean outfall plume.The dynamics of both river and outfall plumes have a historyof experimental studies which provide a basis for theapproach described herein. The river plume models dependon the width of the river discharge, bottom topography, windspeed and direction, ambient current field, input flow rate,and inertial forcing (1, 2). The vertical rise and dilution of asewage plume discharged at the seafloor can be addressedthrough models based on hydraulic principles of buoyantjets in stratified and unstratified fluids, with the Roberts-Snyder-Baumgartner (RSB) model (3) serving as a welldocumented model in use by the U.S. EnvironmentalProtection Agency (EPA) for evaluating near-field dischargeplume behavior.
The approach adopted here to understand the plumetransport in the coastal zone is to integrate surface currentdata measured by an array of high-frequency (HF) radarsand produce a Lagrangian analysis of the plume trajectory.A primary distinction of this study from previous studies isthe four-year duration of a continuous hindcast analysisconducted on an hourly interval.
This paper is organized into three sections. First, waterquality, rainfall, surface currents, and ocean stratificationare summarized to provide a description of the study regionconditions (see Supporting Information (SI) for more details).Then a hindcast analysis of three regional discharges usinga random walk model and the exposure kernels are described.Lastly, we present the model results and discuss underlyingassumptions used in the model. The summer (dry) and winter(wet) seasons discussed in this paper indicate April toSeptember and October to March, respectively.
Summary of Observations and a Plume ExposureHindcast ModelWater Quality Sampling. Water quality standards in the Stateof California Health and Safety Code (AB 411) establish a set
* Corresponding author phone: +1 (858) 822-5201; fax: +1 (858)534-7132; e-mail: [email protected].
† Marine Physical Laboratory.‡ Climate, Atmospheric Science, and Physical Oceanography.
Environ. Sci. Technol. 2009, 43, 7450–7457
7450 9 ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 43, NO. 19, 2009 10.1021/es900775p CCC: $40.75 2009 American Chemical SocietyPublished on Web 08/21/2009
Dow
nloa
ded
by U
NIV
OF
CA
LIF
OR
NIA
CD
L A
CQ
UIS
ITIO
NS
on S
epte
mbe
r 29
, 200
9 | h
ttp://
pubs
.acs
.org
P
ublic
atio
n D
ate
(Web
): A
ugus
t 21,
200
9 | d
oi: 1
0.10
21/e
s900
775p
of criteria for identifying if a human health risk is present.These criteria are based on both a single sample result forTotal Coliform (TC), Fecal Coliform (FC), and Enterococcus(ENT), as well as the ratio of FC to TC. The exceedancerelations are posed as a water quality indicator (g) with abinary state: clean (C) or contaminated (D).
The FIB sample data at the shoreline stations for fouryears (April 2003-March 2007) (Figure 1 and Table SI-S1)are provided by San Diego County DEH and the City of SanDiego Wastewater District as part of their AB411 samplingsand ref 4. They are indicated with dots (C) and triangles (D)(Figure 2a). The D condition is highly correlated with thewet season (winter), which is consistent with other studies(5, 6).
Rainfall in San Diego and Tijuana River flow. The hourlycumulative rainfall at San Diego Lindbergh Field (SAN) andthe hourly TJR flow rate are shown in Figure 2b and c,respectively. The rainfall in southern San Diego is concen-trated during the wet season, and leads to the TJR outflowswith heavy sediment and debris loads, which are potentialcauses of impaired coastal water quality.
Surface Currents. An array of HF radars provides hourlysurface current maps with 1 km spatial resolution (see SI).Prior to their usage for Lagrangian trajectory computations,surface currents are objectively mapped using a samplecovariance matrix computed from four-year hourly data tofill in missing data (7). The uncertainty of the estimatedsurface current field is about 8.6 cm s -1, which is roughlyconsistent with reported root-mean-square (rms) errors: 1-7cm s-1 (8, 9).
For a typical rain event, surface currents are generallyheading north at the start of the TJR flow since storm eventsapproach from the south, a pattern that was found to beconsistent for over 80% of rain events when the dailyaveraged TJR flow rate is 103 m3 s-1 or more.
Vertical Ocean Temperature Structure. Temperatureobservations for approximately 3 years at the SBO (A2 inFigure 1) provide the water column stratification and allowus to predict the plume rise height which discharges at a
depth of 28 m. Surfacing of the plume water occurs duringweak stratification allowing the plume’s buoyancy to over-come the trapping effect of the density stratification. Inaddition, salinity is found to be a weak influence on changesin stratification in shallow southern California waters (10)including San Diego.
Regional Discharges. Three candidate discharges re-sponsible for high FIB levels in southern San Diego aresummarized (Table 1). The source locations (A1, A2, and A3in Figure 1) are used as initial positions for the particletrajectory calculation in the subsequent hindcast analysis.The TJR outflows just north of the border and drainsprecipitation from a watershed of 4,480 km2, two-thirds ofwhich is in Mexico (A1 in Figure 1). The SBO is located 3.5miles offshore west of the border and has dischargedadvanced primary-treated wastewater through an underwateroutfall year round since 1999 (A2 in Figure 1). The City ofTijuana discharges pretreated waters directly into the beachat San Antonio del los Buenos (6 miles south of the border)from the PBD (A3 in Figure 1) (6, 11, 12).
A Plume Exposure Hindcast Model. The forward particletrajectory in the finite time domain is calculated as:
where x(t)) [x(t)y(t)]† and u(t)) [u(t)v(t)]† denote the locationof the particle and the surface currents at the particle locationat a given time (t), respectively (t0 is the initial time of thesimulation and † denotes the matrix transpose). εu and εv arethe random variables with zero mean and rms of ε.
In Lagrangian stochastic models, the random walk modelinherits the similarity of the Lagrangian statistics of thepassive tracer in the coastal region compared to the random
FIGURE 1. Study domain of the water quality monitoring in southern San Diego. Shoreline water quality stations (C0, C2-C13,and C15-C18 in Table SI-S1) are indicated as dots. Three potential sources of the impaired water quality are marked astriangles, and they are releasing locations of the particle trajectory model: Tijuana River (A1, TJR), the South Bay InternationalWastewater Treatment Plant (A2, SBO), and Punta Bandera treatment plant discharge (A3, PBD) at San Antonio de los Buenoapproximately 6 miles south of the U.S.-Mexico Border (black solid line). The southern effluent outfall of the SBO is theactive effluent, and the temperature profile is observed at A2. The nearcoast cell is defined as the area within 1 km from thecoastline (shaded area).
x(t) ) ∫t0
t(u(t′) + εu)dt′ + x(t0) ≈ ∑
k
(u(tk) + εku)∆t + x(t0)
(1)
y(t) ) ∫t0
t(v(t′) + εv)dt′ + y(t0) ≈ ∑
k
(v(tk) + εkv)∆t + y(t0)
(2)
VOL. 43, NO. 19, 2009 / ENVIRONMENTAL SCIENCE & TECHNOLOGY 9 7451
Dow
nloa
ded
by U
NIV
OF
CA
LIF
OR
NIA
CD
L A
CQ
UIS
ITIO
NS
on S
epte
mbe
r 29
, 200
9 | h
ttp://
pubs
.acs
.org
P
ublic
atio
n D
ate
(Web
): A
ugus
t 21,
200
9 | d
oi: 1
0.10
21/e
s900
775p
flight model (13, 14). However, the random flight model hasbeen used in studies of marine larvae spreading (15–18) forthe active tracer simulation. The random walk model is usedin this study to preserve the shape of the power spectrumof the original current field and to simulate the pollutantsas the passive tracer. The diffusion parameter (εu and εv)represents unresolved velocities as the uncertainty in the HF
radar measurements (ε ) 5 cm s-1). A large number ofparticles are released and tracked with each time step so thattheir statistical distribution can be used to infer an exposureprobability for a given discharge.
For this study, all discharges are assumed to be passivewith no dynamical impact on the flow, allowing the mappedsurface currents to be the initial current field into which thedischarge occurs. Fifty particles are released hourly at eachsource location (A1, A2, and A3 in Figure 1) and tracked forthree days, which is consistent with the estimated lifetimeof the FIB (19–21), regardless of the time-dependent natureof the discharge flow rate. In case the particle crosses overthe coastline boundary in a given current field, the trajectoryis recalculated by applying the alongshore currents, whichare simply the along-coast projection of currents measured1 km offshore, to constrain the particle to follow the coastline.No time-dependent decay of the FIB is used for the analysissince the goal is to examine the plume water exposureprobability as opposed to concentration prediction and decayrate of FIBs in marine waters are not well documented.
A snapshot of the particle trajectory and the histogramof particles within the nearcoast cellsan alongshore bandextending 1 km from the coast (shaded area in Figure 1)sisshown in Figure 3, which is a part of the near real-time TJRplume tracker (http://www.sccoos.org/data/tracking/IB/).This histogram helps to understand upcoast and downcoasttransports in time.
Exposure Kernels and ROC Analysis. The coastal expo-sure kernel (CEK, P) indicates the relative probability to the
FIGURE 2. (a) Water quality sampling data along the coastline in southern San Diego. The triangles and dots, respectively, indicatethe contaminated (D) and clean (C) conditions based on the FIB criteria. (b) Hourly cumulative rainfall (cm) at SAN. (c) Hourly TJRflow rate (m3 s-1, log scale).
TABLE 1. Potential Sources of Bacterial Contamination inSouthern San Diego: Tijuana River (TJR), South Bay Inter-national Wastewater Treatment Plant (SBO), and PuntaBandera Discharge (PBD) and Number of Days (δ) for FourYears That Each Source Is Active
potential sources of bacterial contamination
location
sourcelongitude
(W)latitude
(N) discharge typeflow rate
[m3s-1 (MGD)]
TJR 32.5556 117.1369 wet (winter) season ∼2.9 (66)SBO 32.5373 117.1835 plume surfacing ∼0.9 (20)PBD 32.4336 117.1100 continuous 1-1.5 (22-35)
number of days each source is active
source all season summer winter
TJR 369 71 298SBO 420 2 418PBD 1461 732 729
7452 9 ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 43, NO. 19, 2009
Dow
nloa
ded
by U
NIV
OF
CA
LIF
OR
NIA
CD
L A
CQ
UIS
ITIO
NS
on S
epte
mbe
r 29
, 200
9 | h
ttp://
pubs
.acs
.org
P
ublic
atio
n D
ate
(Web
): A
ugus
t 21,
200
9 | d
oi: 1
0.10
21/e
s900
775p
number of particles at source location (or maximum number)resulting from surface transports:
where F(x,y) denotes the number of particles in space. In asimilar way, the shoreline exposure kernel (SEK, p) can bedefined in the nearcoast cell:
where f(l) denotes the number of particles along the coast.The CEK and SEK are normalized to a maximum of 100%allowing the kernel to represent a statistical assessment ofthe resulting exposure during a discharge. They are intendedfor the rapid examination of how a discharge may exposeitself, but should be used with caution if relative impacts ofdifferent sources are desired since their flow rates andconcentrations of contaminants can vary.
For a simple comparison between sources, the SEK is scaledby the percentage of time it is active, providing a time integratedstatistic that can be used to assess how often a particular sourcemay be transported to a particular shoreline location relativeto another. A scaled shoreline exposure kernel (SSEK, s) isdefined as the SEK scaled by the number of days (δ) that thesource is active over the entire period (T). T is recommendedto be either annual or seasonal (summer/winter):
A technique in the statistical decision theorysreceiveroperating characteristic (ROC) analysis (22) used in binaryclassifier systemssis applied for the model evaluation. Thediagnosis and event correspond to model outputs andobservations, respectively. All possible cases fall into fourcategories in the contingency table (Table 2). When thediagnosis and the event agree, true-positive (TP) and true-negative (TN) are claimed, and if they disagree, false-positive
(FP) and false-negative (FN) are claimed. The beach closureis evaluated with the diagnosis based on the random walkmodel (f) and the event of the FIB level (g ) C or D) [FigureSI-S6].
The TP proportion (sensitivity, R) and the FP proportion(1-specificity, �) are as a function of the threshold value (λ)for the probability density function (h):
and
Each point of the ROC curve represents a pair of R and� in a given threshold. The area under the ROC curve indicateshow well the diagnosis system distinguishes between positiveand negative cases, and is equal to 0.5 for the binary classifiersystem with random variables.
ResultsA time-averaged synthesis of the particle trajectories is usedto construct three probability kernels for the surface transportof regional discharges: CEK, SEK, and SSEK (eqs 3-5). Thebin sizes of exposure kernels are a square box of 0.2 × 0.2km2 (CEK) and a rectangular box of 0.2 × 1 km2 (SEK andSSEK), respectively (see SI).
FIGURE 3. Snapshot of the TJR plume track model (upper panel) and the histogram of the particles within the nearcoast cell (lowerpanel). The hourly released particles at the TJR mouth (black plus mark) are tracked for three days, and the color of particlerepresents the age of the particle since it was released. The shoreline water quality stations in the map and histogram arehighlighted with red if any particles exist within the nearcoast cell.
P(x, y) ) F(x, y)max[F(x, y)]
× 100 (3)
p(l) ) f(l)max[f(l)]
× 100 (4)
s(l) ) p(l)δT
(5)
TABLE 2. Contingency Table for Two Events and TwoDiagnoses (Positive and Negative Represent the D and CConditions, Respectively)
eventpositive negative
diagnosis positive true-positive (TP) false-positive (FP)negative false-negative (FN) true-negative (TN)
R(λ) ) TPTP + FN
) h(g ≡ D|f e λ) (6)
�(λ) ) FPFP + TN
) h(g ≡ C|f e λ) (7)
VOL. 43, NO. 19, 2009 / ENVIRONMENTAL SCIENCE & TECHNOLOGY 9 7453
Dow
nloa
ded
by U
NIV
OF
CA
LIF
OR
NIA
CD
L A
CQ
UIS
ITIO
NS
on S
epte
mbe
r 29
, 200
9 | h
ttp://
pubs
.acs
.org
P
ublic
atio
n D
ate
(Web
): A
ugus
t 21,
200
9 | d
oi: 1
0.10
21/e
s900
775p
To examine the typical surface transport pattern insouthern San Diego, the CEK is considered while each sourceis active with two casessyear-round (column a in Figure 4),season (summer and winter for columns b and c in Figure4, respectively)sas well as the SEK (Figure 5).
Coastal Exposure Kernel (CEK). The plume exposurehindcast model starts and stops according to the type ofthree local discharges (Table 1). Both CEK and SEK areconditionally computed only with each discharge sourceactive. Although the TJR flow rate is nearly zero, the residualflow after a flood event may last several days and the dischargemay take several days to be flushed out of the estuary. Forthis study, we assume the TJR is active for seven days afterrain event. The surfacing of the SBO serves as the trigger forthe SBO CEK. The number of days for four years that eachsource is active is summarized in Table 1.
The TJR plume statistics exhibit a broad distribution nearthe coast, with the influence of the eddy flows presenting in thenorthern and southern edges of the domain (Figure 4Aa). TheSBO surface transport is slightly oval due to relatively strongeralongshore currents than cross-shore currents. The higherprobabilities in the nearcoast cell result from a trapping of plumewater at the coastline in the simulation (Figure 4Ba). Most tracersfrom the PBD plume water are transported south and spreadalong the coastline. Upcoast transports from the PBD cross theU.S.-Mexico border intermittentlys56 times for 234 days(∼16%) over four yearssand their probabilities are less than2.5% compared with those in the PBD area (Figure 4Ca). The
plume probability decays exponentially in the cross-shoredirection with 1-2 km length scale.
The summer CEK has more alongshore spreading ofthe plume water, and the winter CEK shows a greateroffshore extent of the plume. For example, 10% contoursof the seasonal TJR CEKs cover approximately 20 and 12miles centered by the TJR for summer and winter,respectively (Figure 4Ab,Ac). Due to the sparse rain andstrong ocean stratification during summer, the winter SBOCEK is similar to the year SBO CEK (Figure 4Ba,Bc). Thewinter PBD CEK exhibits the dominant offshore stretch ofplume statistics (Figure 4Cc). The seasonal upcoast eventsin which the PBD plume crosses the border occur 17 timesduring 66 days (summer) and 28 times during 120 days(winter).
Shoreline Exposure Kernel (SEK). Two shoreline sources(TJR and PBD) exhibit slightly different distribution (Figure5 column a): nearly similar shape centered by the TJR (TJRSEK) and the abrupt decay of the concentration in thenorth of the PBD (PBD SEK). These differences result fromspatial variations of the circulation observed in the region.The offshore source generates nearly uniform probabilityalong the coast (SBO SEK). SBO SEKs in both seasons arevery similar except the probability in the northern area ofC13 station.
The annual probabilities of the SBO and PBD at theU.S.-Mexico border are almost the same, and the annualprobabilities from the three sources in the north of the
FIGURE 4. Coastal exposure kernels (CEKs) while each local discharge is active: (A) TJR, (B) SBO, (C) PBD. (a) All season, (b)summer, (c) winter. (Dots in the coastline indicate FIB sampling locations.)
7454 9 ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 43, NO. 19, 2009
Dow
nloa
ded
by U
NIV
OF
CA
LIF
OR
NIA
CD
L A
CQ
UIS
ITIO
NS
on S
epte
mbe
r 29
, 200
9 | h
ttp://
pubs
.acs
.org
P
ublic
atio
n D
ate
(Web
): A
ugus
t 21,
200
9 | d
oi: 1
0.10
21/e
s900
775p
border increase in size from TJR, to SBO, to PBD (Figure5Ab). For the dry season, the influence of the SBO isapproximately zero, so two sources (TJR and PBD) may
contribute most of the exposures by the plume water(Figure 5Bb). In the wet season, the probability of the TJRplume water is more dominant than the other sources,
FIGURE 5. Shoreline exposure kernels (SEKs) and scaled shoreline exposure kernels (SSEKs) while each local discharge is active (TJR,SBO, and PBD): (A) all season, (B) summer, (C) winter. (a) SEK, (b) SSEK (zoomed in). The vertical dot line indicates the source location.
FIGURE 6. Example of the plume exposure hindcast model evaluation using ROC analysis. (a) The time series of the SEK superposedwith FIB samplings. Red and blue triangles indicate contaminated (D) and clean (C) conditions, respectively, and the number ofparticles in the near-coast cell is presented with gray scale. (b) Hourly TJR flow rate (m3s-1). (c) ROC curve.
VOL. 43, NO. 19, 2009 / ENVIRONMENTAL SCIENCE & TECHNOLOGY 9 7455
Dow
nloa
ded
by U
NIV
OF
CA
LIF
OR
NIA
CD
L A
CQ
UIS
ITIO
NS
on S
epte
mbe
r 29
, 200
9 | h
ttp://
pubs
.acs
.org
P
ublic
atio
n D
ate
(Web
): A
ugus
t 21,
200
9 | d
oi: 1
0.10
21/e
s900
775p
and the SBO yields the almost evenly spread probabilitywith local maximum near the C13 station (Figure 5Cb).
Receiver Operating Characteristic (ROC) Analysis. TheROC analysis is applied to several rain events during thewet season. The number of particles within the nearcoastcell is plotted against the observed FIB levels. FP and TPare calculated as a function of the number of particles (λin Figure SI-S6). An example of ROC analysis is shown inFigure 6. The area under the curve is about 0.72. For fouryears of water quality samples compared to model outputduring rain events, the average accuracy of the plumeexposure hindcast model is about 70%, which is areasonable level of performance.
DiscussionA framework to assess the role of coastal discharges andthe relationship with FIBs is presented. An integratedcoastal observation system using surface currents, tem-perature profiles, CTD cast data, satellite imagery, rainfall,and river flow measurements is applied to develop a plumeexposure hindcast statistical model for coastal water qualityprediction. A Lagrangian particle trajectory (random walk)model using hourly surface current maps applied to threedischarge sites provides regional plume water exposurekernels as well as shoreline exposure kernels for twoshoreline sources (TJR and PBD) and one offshore outfall(SBO). The data-driven model is evaluated with FIB sampledata using ROC analysis for several rain events, and foundto have 70% accuracy. The hindcast analysis uses con-tinuously observed surface current maps for four years asthe primary resource, and will provide an improved qualityassessment of the prediction model and increase theconfidence of the surface transport, data-driven modelwarning system. Moreover, it plays an important role formanagement of resources to mitigate the environmentalproblems. However, as the plume is trapped and trans-ported along the coast, the model ignores the importantrole of wave-driven currents in the surfzone and theexchange of water across the breaking surf. A logical nextstep to improve the fidelity of this hindcast approachpresented is to include effects of surf zone currents andsurf zone exchange.
The surface transport of the passive tracer is simulatedwith the random walk model rather than the random flightmodel, which preserves the shape of the power spectrumof the current field. The Lagrangian trajectory modelframework using HF radar-derived surface currents hasbeen applied to other applications including tracking acoastal river plume, water quality, oil spill, search andrescue, and biological larvae spreading (17, 23). The nearreal-time TJR plume tracking model is now used in theSan Diego County DEH for decision making and for guidingthe posting of beach advisories (24).
AcknowledgmentsAuthors are sponsored by the COCMP, NOAA, ONR, andIBWC. Observations are from SCCOOS, NCDC, IBWC, andSan Diego County DEH. A MATLAB compatible version ofthe RSB model developed by P. J. Roberts in GIT wasgraciously provided by B. Jones of the USC. We thank J.Bowen, P. Reuter, M. Otero, L. Hazard, and T. Cook atCORDC in SIO, and C. Clifton at the San Diego CountyDEH.
Note Added after ASAP PublicationThis article was released ASAP on August 21, 2009 withminor errors in the text. The correct version was postedon September 1, 2009.
Supporting Information AvailableComplement description of observations, regional dis-charges, notes on exposure kernels and ROC analysis,Figures SI-S1 to SI-S6, Tables SI-S1 to SI-S2, referencesfor SI. This material is available free of charge via theInternet at http://pubs.acs.org.
Literature Cited(1) Hill, A. E. Buoyant effects in coastal shelf seas. In The Sea: The
Global Coastal Ocean; Brink, K. E., Robinson, A. R., Eds.; JohnWiley & Sons, Inc.: New York, 1998; Vol. 11, Chapter 2 , pp 21-62.
(2) Washburn, L.; McClure, K. A.; Jones, B. H.; Bay, S. M. Spatialscales and evolution of stormwater plumes in Santa MonicaBay. Mar. Environ. Res. 2003, 56, 103–125.
(3) Roberts, P. J. W.; Snyder, W. H.; Baumgartner, D. J. Ocean outfalls.I: Submerged wastefield formation. J. Hydr. Engr. ASCE 1989,115, 1–25.
(4) California Ocean Plan, Water quality control plan for oceanwaters of California., State Water Resources Control Board,Sacramento, CA, 1990.
(5) Kim, J. H.; Grant, S. B.; Mcgee, C. D.; Sanders, B. F.; Largier, J. L.Locating sources of surf zone pollution: A mass budget analysisof fecal indicator bacteria at Huntington Beach, California.Environ. Sci. Technol. 2004, 38, 2626–2636.
(6) Svejkovsky, J.; Jones, B. Detection of coastal urban stormwaterand sewage runoff with synthetic aperture radar satelliteimagery. Eos Trans. Am. Geophys. Union 2001, 82, 621, 624-625, and 630.
(7) Kim, S. Y.; Terrill, E. J.; Cornuelle, B. D. Objectively mapping HFradar-derived surface current data using measured and idealizeddata covariance matrices. J. Geophys. Res. 2007, 112, C06021;doi: 10.1029/2007JC003756.
(8) Kim, S. Y.; Terrill, E. J.; Cornuelle, B. D. Mapping surface currentsfrom HF radar radial velocity measurements using optimalinterpolation. J. Geophys. Res. 2008, 113, C10023; doi: 10.1029/2007JC004244.
(9) Ohlmann, C.; White, P.; Washburn, L.; Terrill, E.; Emery, B.;Otero, M. Interpretation of coastal HF radar-derived surfacecurrents with high-resolution drifter data. J. Atmos. OceanicTechnol. 2007, 24, 666–680.
(10) Winant, C. D.; Bratkovich, A. W. Temperature and currents onthe southern California shelf: A description of the variability.J. Phys. Oceanogr. 1981, 11, 71–86.
(11) San Diego County, California Regional Water Quality ControlBoard. San Diego Region, Order No. 2000-129; City of SanDiego, CA, 2000.
(12) Orozco-Borbon, M. V.; Rico-Mora, R.; Weisberg, S. B.; Noble,R. T.; Dorsey, J. H.; Leecaster, M. K.; McGee, C. D. Bacteriologicalwater quality along the Tijuana-Ensenada, Baja Calilfornia,Mexico shoreline. Mar. Pollut. Bull. 2006, 52, 1190–1196.
(13) Griffa, A.; Owens, K.; Piterbarg, L.; Rozovskii, B. Estimates ofturbulence parameters from Lagrangian data using a sto-chastic particle model. J. Mar. Res. 1995, 53, 371–401.
(14) Griffa, A. Applications of stochastic particle models to oceano-graphic problems. In Stochastic Modeling in Physical Ocean-ography; Adler, R. J., Muller, P., Rozovskii, B., Eds.; Progress inProbability; Birkhauser: Cambridge, MA, 1996; pp114-140.
(15) Siegel, D. A.; Kinlan, B. P.; Gaylord, B.; Gaines, S. D. Lagrangiandescriptions of marine larval dispersion. Mar. Ecol.: Prog. Ser.2003, 260, 83–96.
(16) Isaji, T.; Spaulding, M. L.; Allen, A. A. Stochastic particle trajectorymodeling technique for spill and search and rescue models.Estuarine Coastal Modell. 2005, 537–547.
(17) Ullman, D. S.; O’Donnell, J.; Kohut, J.; Fake, T.; Allen, A. Trajectoryprediction using HF radar surface currents: Monte Carlosimulations of prediction uncertainties. J. Geophys. Res. 2006,111, C12005; doi: 10.1029/2006JC003715.
(18) Spaulding, M. L.; Isaji, T.; Hall, P.; Allen, A. A. A hierarchy ofstochastic particle models for search and rescue (SAR): Ap-plication to predict surface drifter trajectories using HF radarcurrent forcing. J. Mar. Env. Eng. 2006, 8, 181–214.
(19) Noble, R. T.; Leecaster, M. K.; McGee, C. D.; Moore, D. F.; Orozco-Borbon, V.; Schiff, K.; Vainik, P.; Weisberg, S. B. SouthernCalifornia Bight 1998 Regional Monitoring Program Volume III:Storm event shoreline microbiology; Technical Report, SouthernCalifornia Coastal Water Research Project (SCCWRP): CostaMesa, CA, 2000; p 65.
(20) Ackerman, D.; Weisberg, S. B. Relationship between rainfalland beach bacterial concentrations on Santa Monica Baybeaches. J. Water Health 2003, 85–89.
7456 9 ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 43, NO. 19, 2009
Dow
nloa
ded
by U
NIV
OF
CA
LIF
OR
NIA
CD
L A
CQ
UIS
ITIO
NS
on S
epte
mbe
r 29
, 200
9 | h
ttp://
pubs
.acs
.org
P
ublic
atio
n D
ate
(Web
): A
ugus
t 21,
200
9 | d
oi: 1
0.10
21/e
s900
775p
(21) Noble, R. T.; Lee, I. M.; Schiff, K. C. Inactivation of indicator micro-organisms from various sources of fecal contamination in sea waterand freshwater. J. Appl. Microbiol. 2004, 96, 464–472.
(22) Swets, J. A. Measuring the accuracy of diagnostic systems. Science1988, 240, 1285–1293.
(23) Spaulding, M. L.; Anderson, E. L.; Isaji, T.; Howlett, E. Simulationof the oil trajectory and fate in the Arabian Gulf from the MinaAl Ahmadi spill. Mar. Environ. Res. 1993, 36, 75–115.
(24) Cllfton, C. C.; Kim, S. Y.; Terrill, E. J. Using real time observingdata for public health protection in ocean waters. Proceedingsin Coastal Zone 07, 2007; available at http://www.csc.noaa.gov/cz/2007/Coastal_Zone_07_Proceedings/PDFs/Tuesday_Abstracts/2901.Clifton.pdf.
ES900775P
VOL. 43, NO. 19, 2009 / ENVIRONMENTAL SCIENCE & TECHNOLOGY 9 7457
Dow
nloa
ded
by U
NIV
OF
CA
LIF
OR
NIA
CD
L A
CQ
UIS
ITIO
NS
on S
epte
mbe
r 29
, 200
9 | h
ttp://
pubs
.acs
.org
P
ublic
atio
n D
ate
(Web
): A
ugus
t 21,
200
9 | d
oi: 1
0.10
21/e
s900
775p
