Subscriber access provided by Kaohsiung Medical University
Environmental Modeling
Geostatistical prediction of microbial water quality throughout a stream network using meteorology, land cover, and spatiotemporal autocorrelation David Andrew Holcomb, Kyle P Messier, Marc L Serre, Jakob G Rowny, and Jill R Stewart Environ. Sci. Technol., Just Accepted Manuscript • DOI: 10.1021/acs.est.8b01178 • Publication Date (Web): 11 Jun 2018 Downloaded from http://pubs.acs.org on June 11, 2018
Just Accepted “Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted online prior to technical editing, formatting for publication and author proofing. The American Chemical Society provides “Just Accepted” as a service to the research community to expedite the dissemination of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts appear in full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been fully peer reviewed, but should not be considered the official version of record. They are citable by the Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered to authors. Therefore, the “Just Accepted” Web site may not include all articles that will be published in the journal. After a manuscript is technically edited and formatted, it will be removed from the “Just Accepted” Web site and published as an ASAP article. Note that technical editing may introduce minor changes to the manuscript text and/or graphics which could affect content, and all legal disclaimers and ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors or consequences arising from the use of information contained in these “Just Accepted” manuscripts.
is published by the American Chemical Society. 1155 Sixteenth Street N.W., Washington, DC 20036 Published by American Chemical Society. Copyright © American Chemical Society. However, no copyright claim is made to original U.S. Government works, or works produced by employees of any Commonwealth realm Crown government in the course of their duties.
Page 1 of 36
Environmental Science & Technology
1
Geostatistical prediction of microbial water quality
2
throughout a stream network using meteorology,
3
land cover, and spatiotemporal autocorrelation
4
David A. Holcomb,† Kyle P. Messier,‡ Marc L. Serre,† Jakob Rowny,† Jill R. Stewart*,†
5
†Department of Environmental Sciences and Engineering, Gillings School of Global Public
6
Health, University of North Carolina, Chapel Hill, North Carolina 27599-7431, United States
7
‡Department of Civil, Architectural, and Environmental Engineering, University of Texas,
8
Austin, Texas 78712, United States
9
ACS Paragon Plus Environment
1
Environmental Science & Technology
Page 2 of 36
10
TOC Graphic
11
ABSTRACT. Predictive modeling is promising as an inexpensive tool to assess water quality.
12
We developed geostatistical predictive models of microbial water quality that empirically
13
modelled spatiotemporal autocorrelation in measured fecal coliform (FC) bacteria concentrations
14
to improve prediction. We compared five geostatistical models featuring different
15
autocorrelation structures, fit to 676 observations from 19 locations in North Carolina’s Jordan
16
Lake watershed using meteorological and land cover predictor variables. Though stream distance
17
metrics (with and without flow-weighting) failed to improve prediction over the Euclidean
18
distance metric, incorporating temporal autocorrelation substantially improved prediction over
19
the space-only models. We predicted FC throughout the stream network daily for one year,
20
designating locations “impaired”, “unimpaired”, or “unassessed” if the probability of exceeding
21
the state standard was >90%, 10% but 0.1 but < 0.9 were considered unassessed. Additionally,
235
any points with > 0.5 probability of exceeding the standard were considered more likely than not
236
to be impaired.
237
Results and Discussion
238
Geostatistical model comparison. We fit an OLS model, assuming uncorrelated error, and four
239
geostatistical models, distinguished by four different covariance models for the error function, to
240
FC concentration data. Table 1 compares the performance of the five models, which were
241
assessed foremost by minimizing AIC, followed by RMSE and corrected prediction R2. The
242
OLS model and the three spatial models performed similarly, with the OLS model faring slightly
243
worse by all three metrics and the Euclidean space-only model performing slightly better by
244
AIC. The three spatial models all estimated the same proportion of spatial autocorrelation (3% of
245
the total variance) regardless of covariance model structure. Incorporating temporal
246
autocorrelation greatly improved performance, substantially reducing AIC and RMSE and
247
allowing the space/time model to explain over 70% of the variance in FC concentration. The
248
mean function explained the greatest variance portion (59%) in all four geostatistical models; the
249
OLS model also explained 59% of the total variance. The correlated error component of the
250
space/time model accounted for 34% of the total variance, far exceeding the 3% correlated error
251
in the spatial models. RSME and R2 values for space/time model predictions at individual
252
sampling sites ranged from 0.61 – 1.48 ln(cfu)/100 mL and 0.42 – 0.88, respectively (Table S4).
ACS Paragon Plus Environment
14
Page 15 of 36
Environmental Science & Technology
253
Table 1. Comparison of model performance and proportion of variance explained by geostatistical
254
model components.
Performance metrics
Model
Proportion of variance explained Error function
Covariance structure AIC
RMSE*
𝐑𝟐 *
Mean function
[ln(cfu)]
255
Correlated error proportion
Unexplained variance (nugget) proportion
OLS
Nugget
2214
1.21
0.59
0.59
-
0.41
Space-only
Euclidean + nugget
2187
1.19
0.61
0.59
0.03
0.38
Stream distance
Stream distance + Euclidean + nugget
2191
1.19
0.61
0.59
0.03
0.38
Flowweighted
Flowweighted stream distance + Euclidean + nugget
2191
1.19
0.61
0.59
0.03
0.38
Space/time
Euclidean + nugget
2090
1.02
0.71
0.59
0.34
0.07
*Calculated from LOOCV prediction residuals
256
Effects of stream distance and spatiotemporal autocorrelation: error function models.
257
Compared with the OLS model assuming uncorrelated error, we observed a strong improvement
258
in model performance by considering temporal autocorrelation but little improvement when
259
modeling the error function using only spatial autocorrelation (Table 2). Incorporating a stream
260
distance metric had no effect on model performance, as Euclidean distance-based autocorrelation
ACS Paragon Plus Environment
15
Environmental Science & Technology
Page 16 of 36
261
dominated the spatially-correlated error. The longest spatial range was associated with the flow-
262
weighted stream distance component, but the small partial sill suggests limited impact of stream
263
processes in driving FC concentrations—further reflected in the negligible partial sill and
264
reduced spatial range of the stream distance model without flow-weighting.
265
Scale may be an important factor in the relative importance of land and stream-based
266
processes driving microbial contamination; over-land processes appear to dominate in-stream
267
processes in our small watershed, with sampling locations at most 20 km apart. A recent study
268
conducted on a similarly small watershed likewise realized only minor performance gains by
269
modeling correlated error using a flow-weighted stream distance metric, despite observing
270
substantial spatial autocorrelation.59 By contrast, other studies conducted on whole river basins
271
found the stream distance metric (with various flow-weighting schemes) improved FIB
272
estimation, suggesting that in-stream effects may grow dominant at larger scales.30,39
273
Considering larger spatial scales generally increases the number of flow-connected sampling
274
locations analyzed, which may lead to performance gains from flow-weighted and stream
275
distance metrics.60 Several studies predicting other physiochemical characteristics in streams
276
obtained mixed results regarding the relative performance of Euclidean versus (flow-weighted)
277
stream distance metrics.35,57,61–63 The appropriate distance metric for modeling a particular water
278
quality characteristic is dependent on several interacting factors, including the underlying
279
processes driving the characteristic, the spatial scale of the analysis, and the distribution of
280
sampling locations on the network.64
ACS Paragon Plus Environment
16
Page 17 of 36
281
Environmental Science & Technology
Table 2. Error function parameter estimates. Unexplained variance (nugget)
Stream distance covariance
Euclidean covariance
Model Partial sill [ln(cfu)2]
Partial sill [ln(cfu)2]
Spatial range [km]
Temporal range [days]
Partial sill [ln(cfu)2]
Spatial range [km]
OLS
1.48
-
-
-
-
-
Space-only
1.36
0.112
8.8
-
-
-
Stream distance
1.36
0.112
8.9
-
3.4 × 10−4
10.4
Flowweighted
1.36
0.098
9.3
-
0.012
22.7
Space/time
0.24
1.23
12.8
2.1
-
-
282 283
Including time substantially decreased the amount of uncorrelated error, though temporal
284
autocorrelation fully decayed in two days. In contrast to the spatial models, the space/time model
285
only estimates spatial autocorrelation for observations made within this short temporal range,
286
revealing much stronger spatial patterns. The short temporal range, increased spatial range, and
287
large partial sill suggest that localized microbial water quality is dependent on broader
288
environmental conditions that are subject to substantial temporal variability. This is in keeping
289
with previous research at freshwater beaches, in which the day of sampling explained the largest
290
portion of FIB concentration variance between multiple beaches.19 As such, the space/time
291
model may be useful for predicting water quality at unmonitored locations when microbial
292
concentrations are known elsewhere, but is limited for prediction at unmonitored times.
293
Effects of land cover and meteorology: mean function models. We used the standardized
294
predictor variables to specify a full mean function for the flow-weighted model (Table S3). The
ACS Paragon Plus Environment
17
Environmental Science & Technology
Page 18 of 36
295
model with the lowest AIC following backwards elimination retained the following predictors:
296
percent forested upstream watershed area, percent agricultural upstream watershed area, hourly
297
average surface temperature, precipitation in the preceding two days, precipitation in the
298
preceding seven days (excluding the previous two days), and the interaction between the two
299
precipitation terms. The other four models were fit to the same set of predictors. The parameter
300
estimates differed slightly between the five models due to the influence of covariance structure
301
(Table S2), but the mean function for all five models accounted for the same proportion of the
302
total variance in the data (Table 1).
303
The direction of effect estimates (Figure 2) largely coincided with our expectations.
304
Increasing forest cover was protective against microbial pollution,11,14,65,66 while antecedent
305
rainfall and warmer temperatures were associated with increased microbial
306
concentrations.11,13,16,23,25,67 Precipitation in the preceding two days was associated with the
307
largest increase in FC concentration. Agricultural land use also exhibited a strong positive
308
relationship with FC, though it was the least precise parameter estimate and previous studies
309
report conflicting effects of agriculture on FIB.11,66,68 This uncertainty could relate to the
310
geographic distribution of agricultural land on the periphery of the study area: although many
311
sites had agricultural lands upstream, the watersheds generally transitioned to other land uses
312
nearer the sites.
313
Despite previous studies suggesting urbanization increases microbial pollution, we did
314
not observe a significant association between increasing impervious surfaces and FC
315
concentration.11,14,66,69 Though urbanized land was common, most watersheds had low
316
impervious surface percentages, reflecting low-intensity development with potentially diffuse
317
impacts on FC loading. We also failed to detect a significant effect of solar radiation, which is
ACS Paragon Plus Environment
18
Page 19 of 36
Environmental Science & Technology
318
known to inactivate microbes in the environment.65,67,70–72 This may be due in part to the spatial
319
coarseness of the solar radiation measurement, for which a single value measured at the NOAA
320
Durham station was used for all samples collected at a given time. Site-specific topography and
321
canopy cover could have substantially altered the amount of solar radiation reaching the water.
322
To address these limitations, future applications may consider model-based estimation of solar
323
radiation, which has been used effectively in mechanistic FIB models.28,73–75
324
The models presented here provide nuanced estimates of the overall effects of
325
precipitation by considering the interaction between the two precipitation timeframes. More
326
recent precipitation has a stronger effect on microbial concentration than precipitation further
327
removed in time, but the negative interaction estimate suggests that past precipitation mitigates
328
the impact of more recent precipitation (SI). In systems where FC sources accumulate on land
329
before precipitation flushes contamination into water bodies, it follows that past precipitation
330
reduces the stock of contamination available for subsequent rain to wash into the stream.
331
Precipitation intensity may also act as an important driver of contamination in some systems,
332
although it did not appreciably affect the predictive performance of the models presented in this
333
paper (data not shown).
334
Although likely similar to many other mixed-use watersheds, the associations between
335
FC and the predictors considered in this study should be generalized with caution. The study area
336
comprised a small watershed with limited LULC composition, the effects of which were derived
337
essentially from 19 observations for each LULC class (corresponding to the 19 sampling sites).
338
As with the covariance components, scale likely also affects the magnitude of predictor effects.
339
The pronounced sensitivity to precipitation might be tempered in larger systems featuring
340
spatially heterogenous meteorological conditions and a wider range of LULC characteristics.
ACS Paragon Plus Environment
19
Environmental Science & Technology
Page 20 of 36
341
Furthermore, we might expect associations with meteorology to differ in systems dominated by
342
sources with distinctive loading dynamics, such as agricultural watersheds receiving intermittent
343
manure applications.
344
Figure 2. Mean function GLS parameter estimates (points) and 95% confidence intervals (bars)
345
for the space/time model. Estimates are for standardized (mean-centered, standard deviation-
346
scaled) input variables, and represent the expected change in ln(FC) concentration for a 1 standard
347
deviation increase in the input variable with all other variables held constant. Note that the three
348
rain terms are related through the interaction and cannot be interpreted in isolation (SI). The
349
intercept corresponds to the expected ln(FC) concentration when all other predictors equal their
350
respective means.
351
Space/time prediction of microbial water quality. The space/time geostatistical model was
352
chosen to predict FC concentrations at unmonitored locations and times. We predicted universal
353
kriging means and standard errors at each prediction location at 11 a.m. for each of the 365 days
354
between April 1, 2010 and March 31, 2011 and calculated the probability each prediction
355
exceeded the NC standard of 200 cfu/100 mL, from which a corresponding impairment status
356
was assigned. Figure 3 illustrates the spatial and temporal trends in prediction mean, standard
ACS Paragon Plus Environment
20
Page 21 of 36
Environmental Science & Technology
357
error, and impairment status. Prediction error was lower on days FC were measured (Figure 3B),
358
but the short temporal range largely prevented observational data from affecting predictions on
359
consecutive days (Figure 3E). Predictions outside the temporal range were made entirely on the
360
basis of the mean function, which can be observed in the flattening of localized prediction means
361
between days with and without sample data (Figures 3A and 3D).
362
Mean FC concentrations exceeding the standard were predicted on every day of the
363
study; on 36 days (10% of study period), more than 90% of stream miles had predicted FC
364
means above the standard. Assigning impairment status only to predictions with a high
365
probability (< 90%) of being above or below the state standard, some portion of the watershed
366
was considered impaired on 209 days (57%), while unimpaired stream miles were predicted on
367
190 days (52%). We observed no strong spatial patterns of impairment; rather, impairment
368
tended to be event driven, wherein large portions of the watershed briefly became impaired
369
before a relatively rapid return to a generally unassessed state (Figure 3G). While unimpairment
370
likewise displayed no strong spatial pattern, the temporal trends for unimpaired stream miles
371
appeared more seasonally driven, with notable background levels of unimpaired stream miles
372
during the winter months. Conversely, relatively few stream miles were considered unimpaired
373
during months in which impairment events are more common, even when no impairment event
374
was ongoing; in such cases, the majority of the watershed was considered unassessed.
ACS Paragon Plus Environment
21
Environmental Science & Technology
Page 22 of 36
375
Figure 3. Predicted spatial and temporal trends in microbial water quality. The top two rows
376
(panels A-F) map the prediction means (panels A and D), standard errors (panels B and E), and
377
impairment status (panels C and F) for the whole study area on two consecutive days. The bottom
378
row (panel G) is the percentage of stream miles assigned each impairment status by day for the
379
entire one-year study period, using the same scale as the impairment status maps; the dashed line
380
represents the percentage of stream miles more likely impaired than not. Water samples were
381
collected for the first day, which is reflected in the lower standard errors (panel B) compared with
ACS Paragon Plus Environment
22
Page 23 of 36
Environmental Science & Technology
382
the following day (panel E), for which no sample data are available. Areas assigned an impaired
383
status on the first day (panel C) largely correspond to the areas of reduced standard error arising
384
from the observational data (panel B).
385
Role of predictive models and observational data in assessing watershed impairment. Our
386
space/time model predicted that the Jordan Lake near-upstream watershed frequently suffered
387
from unsafe levels of microbial contamination. However, despite 33% of all prediction means
388
exceeding the state standard, only 5% of all predictions could be confidently considered
389
impaired, and only 6% were confidently assessed as unimpaired. The other 89% of predictions
390
were unassessed, meaning we could not determine whether FC concentrations fell above or
391
below the NC standard after accounting for prediction uncertainty. Access to water sample data
392
on the day of prediction was strongly associated with an increased ability to assign impairment
393
status. Under a simple linear regression, any FC observations on a given day was associated with
394
a 20% reduction (p < 0.001) in unassessed stream miles.
395
Because impairment appears largely event-driven, increases in related meteorological
396
variables, particularly precipitation, produced elevated mean FC predictions. However, the high
397
temporal variability and low temporal autocorrelation range of FC concentrations limited the
398
ability of models alone to predict impairment status with high-confidence. Access to recent
399
monitoring data, however, reduced prediction uncertainty sufficiently to confidently assign
400
impairment status to larger portions of the stream network. While a recent study in a similar
401
watershed observed clear spatial patterns in FIB, they did not translate into the substantial
402
reductions in prediction error necessary to confidently assess impairment status. 59 That their
403
samples were collected monthly, without intentional storm sampling or considering temporal
404
autocorrelation, suggests spatial relationships may impact background FIB concentrations, only
ACS Paragon Plus Environment
23
Environmental Science & Technology
Page 24 of 36
405
to be drowned out by events driving brief, widespread impairment. This dynamic suggests
406
frequent, sample-based monitoring remains necessary to confidently assess the safety and legal
407
status of a water body. Implementing spatiotemporal predictive models may reduce the number
408
of concurrent sampling locations required to adequately assess water quality throughout a stream
409
network, providing the means to increase sampling frequency without increasing monitoring
410
expenses.
411
In developing geostatistical models that consider the spatiotemporal patterns and
412
environmental drivers of microbial water quality impairment, we identified clearly dominant
413
factors in each structural model component that together largely explain observed FC
414
concentrations and enable their prediction. Precipitation dominated in the model component that
415
predicts FC directly, while temporal patterns dominated spatial patterns in the component that
416
contributes indirectly to prediction by considering autocorrelation in the outcome. Rather than
417
orienting monitoring activities around specific locations of heightened concern, the dominance of
418
precipitation and temporal effects suggests a monitoring strategy that targets sampling according
419
to temporal criteria, employing frequent sampling to better capture diverse meteorological
420
conditions driving impairment events. Because we constructed the model using remotely sensed
421
and continuously reported predictors and we jointly estimated the different model component
422
parameters using likelihood-based methods, others may apply the model to geographically sparse
423
monitoring data to assess microbial water quality impairment throughout stream networks
424
without the need for additional specialized measurements or knowledge. Environmental
425
managers, regulators, and researchers can implement our model to inform management
426
decisions, target remediation efforts, and issue timely warnings for the protection of public
427
health.
ACS Paragon Plus Environment
24
Page 25 of 36
Environmental Science & Technology
428
AUTHOR INFORMATION
429
Corresponding Author
430
*Email: jill.stewart@unc.edu; phone: (919) 966-7553
431
Notes
432
The authors declare no competing financial interest.
433
ACKNOWLEDGMENTS
434
This work was supported in part by the Water Resources Research Institute of the University of
435
North Carolina, project number 70252. We are grateful to Prahlad Jat for assisting with the
436
stream network analysis and display.
437
Supporting Information Available. Profiling for maximum likelihood estimation; universal
438
kriging prediction error approximation; additive function values for flow-weighting; exploratory
439
analysis of input variables; visual interpretation of error function; mean function parameter
440
estimates; derivation of precipitation interaction effects; prediction at individual sampling sites;
441
animated maps of daily watershed predictions. This information is available free of charge via
442
the Internet at http://pubs.acs.org.
443
REFERENCES
444 445 446
(1)
Dadswell, J. V. Microbiological Quality of Coastal Waters and Its Health Effects. Int. J.
Environ. Health Res. 1993, 3 (1), 32–46, 10.1080/09603129309356762. (2)
Byappanahalli, M. N.; Nevers, M. B.; Korajkic, A.; Staley, Z. R.; Harwood, V. J.
447
Enterococci in the Environment. Microbiol. Mol. Biol. Rev. 2012, 76 (4), 685–706,
448
10.1128/MMBR.00023-12.
ACS Paragon Plus Environment
25
Environmental Science & Technology
449
(3)
Dorevitch, S.; Pratap, P.; Wroblewski, M.; Hryhorczuk, D. O.; Li, H.; Liu, L. C.; Scheff,
450
P. A. Health Risks of Limited-Contact Water Recreation. Environ. Health Perspect. 2011, 120
451
(2), 192–197, 10.1289/ehp.1103934.
452
(4)
Soller, J. A.; Schoen, M. E.; Bartrand, T.; Ravenscroft, J. E.; Ashbolt, N. J. Estimated
453
Human Health Risks from Exposure to Recreational Waters Impacted by Human and Non-
454
Human Sources of Faecal Contamination. Water Res. 2010, 44 (16), 4674–4691,
455
10.1016/j.watres.2010.06.049.
456
(5)
Marion, J. W.; Lee, J.; Lemeshow, S.; Buckley, T. J. Association of Gastrointestinal
457
Illness and Recreational Water Exposure at an Inland U.S. Beach. Water Res. 2010, 44 (16),
458
4796–4804, 10.1016/j.watres.2010.07.065.
459
(6)
Brown, J. M.; Proum, S.; Sobsey, M. D. Escherichia Coli in Household Drinking Water
460
and Diarrheal Disease Risk: Evidence from Cambodia. Water Sci. Technol. 2008, 58 (4), 757,
461
10.2166/wst.2008.439.
462
(7)
Wade, T. J.; Pai, N.; Eisenberg, J. N. S.; Colford, J. M. Do U.S. Environmental
463
Protection Agency Water Quality Guidelines for Recreational Waters Prevent Gastrointestinal
464
Illness? A Systematic Review and Meta-Analysis. Environ. Health Perspect. 2003, 111 (8),
465
1102–1109, 10.1289/ehp.6241.
466
(8)
Page 26 of 36
Colford, J. M.; Wade, T. J.; Schiff, K. C.; Wright, C. C.; Griffith, J. F.; Sandhu, S. K.;
467
Burns, S.; Sobsey, M.; Lovelace, G.; Weisberg, S. B. Water Quality Indicators and the Risk of
468
Illness at Beaches With Nonpoint Sources of Fecal Contamination. Epidemiology 2007, 18 (1),
469
27–35, 10.1097/01.ede.0000249425.32990.b9.
ACS Paragon Plus Environment
26
Page 27 of 36
470 471 472
Environmental Science & Technology
(9)
Federal Water Pollution Control Act. 33 U.S.C. §§1251-1387; 1972;
http://www.epw.senate.gov/water.pdf. (10) USEPA. National Summary of State Information. United States Environmental
473
Protection Agency http://iaspub.epa.gov/waters10/attains_nation_cy.control (accessed Sep 25,
474
2015).
475
(11) Rowny, J. G.; Stewart, J. R. Characterization of Nonpoint Source Microbial
476
Contamination in an Urbanizing Watershed Serving as a Municipal Water Supply. Water Res.
477
2012, 46 (18), 6143–6153, 10.1016/j.watres.2012.09.009.
478
(12) Mallin, M. A.; Johnson, V. L.; Ensign, S. H. Comparative Impacts of Stormwater Runoff
479
on Water Quality of an Urban, a Suburban, and a Rural Stream. Environ. Monit. Assess. 2009,
480
159 (1–4), 475–491, 10.1007/s10661-008-0644-4.
481
(13) Coulliette, A. D.; Noble, R. T. Impacts of Rainfall on the Water Quality of the Newport
482
River Estuary (Eastern North Carolina, USA). J. Water Health 2008, 06 (4), 473,
483
10.2166/wh.2008.136.
484
(14) DiDonato, G. T.; Stewart, J. R.; Sanger, D. M.; Robinson, B. J.; Thompson, B. C.;
485
Holland, A. F.; Van Dolah, R. F. Effects of Changing Land Use on the Microbial Water Quality
486
of Tidal Creeks. Mar. Pollut. Bull. 2009, 58 (1), 97–106, 10.1016/j.marpolbul.2008.08.019.
487
(15) Stumpf, C. H.; Piehler, M. F.; Thompson, S.; Noble, R. T. Loading of Fecal Indicator
488
Bacteria in North Carolina Tidal Creek Headwaters: Hydrographic Patterns and Terrestrial
489
Runoff Relationships. Water Res. 2010, 44 (16), 4704–4715, 10.1016/j.watres.2010.07.004.
ACS Paragon Plus Environment
27
Environmental Science & Technology
Page 28 of 36
490
(16) Liao, H.; Krometis, L.-A. H.; Cully Hession, W.; Benitez, R.; Sawyer, R.; Schaberg, E.;
491
von Wagoner, E.; Badgley, B. D. Storm Loads of Culturable and Molecular Fecal Indicators in
492
an Inland Urban Stream. Sci. Total Environ. 2015, 530–531, 347–356,
493
10.1016/j.scitotenv.2015.05.098.
494 495 496
(17) Leecaster, M. K.; Weisberg, S. B. Effect of Sampling Frequency on Shoreline Microbial Assessments. Mar. Pollut. Bull. 2001, 42 (11), 1150–1154. (18) Whitman, R. L.; Nevers, M. B. Escherichia Coli Sampling Reliability at a Frequently
497
Closed Chicago Beach: Monitoring and Management Implications. Environ. Sci. Technol. 2004,
498
38 (16), 4241–4246, 10.1021/es034978i.
499
(19) Whitman, R. L.; Nevers, M. B. Summer E. Coli Patterns and Responses along 23
500
Chicago Beaches. Environ. Sci. Technol. 2008, 42 (24), 9217–9224, 10.1021/es8019758.
501
(20) USEPA. Predictive Tools for Beach Notification., EPA-823-R-10-003; 2010;
502
http://water.epa.gov/scitech/swguidance/standards/criteria/health/recreation/upload/P26-Report-
503
Volume-I-Final_508.pdf.
504
(21) Nevers, M. B.; Whitman, R. L. Efficacy of Monitoring and Empirical Predictive
505
Modeling at Improving Public Health Protection at Chicago Beaches. Water Res. 2011, 45 (4),
506
1659–1668, 10.1016/j.watres.2010.12.010.
507
(22) Avila, R.; Horn, B.; Moriarty, E.; Hodson, R.; Moltchanova, E. Evaluating Statistical
508
Model Performance in Water Quality Prediction. J. Environ. Manage. 2018, 206, 910–919,
509
10.1016/j.jenvman.2017.11.049.
ACS Paragon Plus Environment
28
Page 29 of 36
Environmental Science & Technology
510
(23) Gonzalez, R. a; Conn, K. E.; Crosswell, J. R.; Noble, R. T. Application of Empirical
511
Predictive Modeling Using Conventional and Alternative Fecal Indicator Bacteria in Eastern
512
North Carolina Waters. Water Res. 2012, 46 (18), 5871–5882, 10.1016/j.watres.2012.07.050.
513
(24) Farnham, D. J.; Lall, U. Predictive Statistical Models Linking Antecedent Meteorological
514
Conditions and Waterway Bacterial Contamination in Urban Waterways. Water Res. 2015, 76,
515
143–159, 10.1016/j.watres.2015.02.040.
516
(25) Coulliette, A. D.; Money, E. S.; Serre, M. L.; Noble, R. T. Space/Time Analysis of Fecal
517
Pollution and Rainfall in an Eastern North Carolina Estuary. Environ. Sci. Technol. 2009, 43
518
(10), 3728–3735, 10.1021/es803183f.
519
(26) Nevers, M. B.; Whitman, R. L. Nowcast Modeling of Escherichia Coli Concentrations at
520
Multiple Urban Beaches of Southern Lake Michigan. Water Res. 2005, 39 (20), 5250–5260,
521
10.1016/j.watres.2005.10.012.
522
(27) Eleria, A.; Vogel, R. M. Predicting Fecal Coliform Bacteria Levels in the Charles River,
523
Massachusetts, USA. J. Am. Water Resour. Assoc. 2005, 41 (5), 1195–1209, 10.1111/j.1752-
524
1688.2005.tb03794.x.
525
(28) Safaie, A.; Wendzel, A.; Ge, Z.; Nevers, M. B.; Whitman, R. L.; Corsi, S. R.;
526
Phanikumar, M. S. Comparative Evaluation of Statistical and Mechanistic Models of Escherichia
527
Coli at Beaches in Southern Lake Michigan. Environ. Sci. Technol. 2016, 50 (5), 2442–2449,
528
10.1021/acs.est.5b05378.
ACS Paragon Plus Environment
29
Environmental Science & Technology
529
(29) Zhang, Z.; Deng, Z.; Rusch, K. A.; Walker, N. D. Modeling System for Predicting
530
Enterococci Levels at Holly Beach. Mar. Environ. Res. 2015, 109, 140–147,
531
10.1016/j.marenvres.2015.07.003.
532
(30) Money, E. S.; Carter, G. P.; Serre, M. L. Modern Space/Time Geostatistics Using River
533
Distances: Data Integration of Turbidity and E. Coli Measurements to Assess Fecal
534
Contamination Along the Raritan River in New Jersey. Environ. Sci. Technol. 2009, 43 (10),
535
3736–3742, 10.1021/es803236j.
536
Page 30 of 36
(31) Akita, Y.; Carter, G.; Serre, M. L. Spatiotemporal Nonattainment Assessment of Surface
537
Water Tetrachloroethylene in New Jersey. J. Environ. Qual. 2007, 36 (2), 508,
538
10.2134/jeq2005.0426.
539
(32) Messier, K. P.; Kane, E.; Bolich, R.; Serre, M. L. Nitrate Variability in Groundwater of
540
North Carolina Using Monitoring and Private Well Data Models. Environ. Sci. Technol. 2014, 48
541
(18), 10804–10812, 10.1021/es502725f.
542
(33) Messier, K. P.; Campbell, T.; Bradley, P. J.; Serre, M. L. Estimation of Groundwater
543
Radon in North Carolina Using Land Use Regression and Bayesian Maximum Entropy. Environ.
544
Sci. Technol. 2015, 49 (16), 9817–9825, 10.1021/acs.est.5b01503.
545
(34) Reyes, J. M.; Serre, M. L. An LUR/BME Framework to Estimate PM2.5 Explained by on
546
Road Mobile and Stationary Sources. Environ. Sci. Technol. 2014, 48 (3), 1736–1744,
547
10.1021/es4040528.
ACS Paragon Plus Environment
30
Page 31 of 36
548
Environmental Science & Technology
(35) Money, E.; Carter, G. P.; Serre, M. L. Using River Distances in the Space/Time
549
Estimation of Dissolved Oxygen along Two Impaired River Networks in New Jersey. Water Res.
550
2009, 43 (7), 1948–1958, 10.1016/j.watres.2009.01.034.
551
(36) Money, E. S.; Sackett, D. K.; Aday, D. D.; Serre, M. L. Using River Distance and
552
Existing Hydrography Data Can Improve the Geostatistical Estimation of Fish Tissue Mercury at
553
Unsampled Locations. Environ. Sci. Technol. 2011, 45 (18), 7746–7753, 10.1021/es2003827.
554
(37) Ver Hoef, J. M.; Peterson, E. E. A Moving Average Approach for Spatial Statistical
555
Models of Stream Networks. J. Am. Stat. Assoc. 2010, 105 (489), 6–18,
556
10.1198/jasa.2009.ap08248.
557 558 559
(38) Cressie, N.; Frey, J.; Harch, B.; Smith, M. Spatial Prediction on a River Network. J. Agric. Biol. Environ. Stat. 2006, 11 (2), 127–150, 10.1198/108571106X110649. (39) Jat, P.; Serre, M. L. A Novel Geostatistical Approach Combining Euclidean and Gradual-
560
Flow Covariance Models to Estimate Fecal Coliform along the Haw and Deep Rivers in North
561
Carolina. Stoch. Environ. Res. Risk Assess. 2018, 10.1007/s00477-018-1512-6.
562
(40) USACE. B. Everett Jordan Dam and Lake Master Plan Update., United States Army
563
Corps of Engineers; 2007;
564
ftp://ftp.chathamnc.org/Chatham_ConservationPlan_GIS/Plans_Policies_Ordinances/USACoE_J
565
ordan Lake Master Plan Update.pdf.
566 567
(41) NCDENR. Jordan Lake Current Allocations. North Carolina Department of Environment and Natural Resources http://www.ncwater.org/?page=313 (accessed Sep 18, 2015).
ACS Paragon Plus Environment
31
Environmental Science & Technology
568
(42) Hodges-Copple, J. An Overview of the Research Triangle Region of North Carolina.,
569
Triangle J Council of Governments; 2011; http://www.campo-nc.us/TRM/Expert-
570
Panel/Presentations/PDFs/Triangle-Overview-2011.pdf.
571
Page 32 of 36
(43) APHA; AWWA; WEF. Standard Methods for the Examination of Water and
572
Wastewater, 20th ed.; American Public Health Association, American Waterworks Association,
573
Water Environment Federation: Washington, D.C., 1998.
574 575 576
(44) USEPA. STORET Data Warehouse. United States Environmental Protection Agency http://www.epa.gov/storet/dbtop.html (accessed Sep 10, 2013). (45) NCEI. United States Climate Reference Network. National Centers for Environmental
577
Information, National Oceanic and Atmospheric Administration
578
ftp://ftp.ncdc.noaa.gov/pub/data/uscrn/products/hourly02/ (accessed May 1, 2014).
579 580
(46) NCDOT. Contour and Elevation Data. North Carolina Department of Transportation https://connect.ncdot.gov/resources/gis/pages/cont-elev_v2.aspx (accessed Jun 15, 2014).
581
(47) Peterson, E. E.; Hoef, J. M. Ver. STARS: An ArcGIS Toolset Used to Calculate the
582
Spatial Information Needed to Fit Spatial Statistical Models to Stream Network Data. J. Stat.
583
Softw. 2014, 56 (2), 1–17, 10.18637/jss.v056.i02.
584
(48) Hoef, J. M. Ver; Peterson, E. E.; Clifford, D.; Shah, R. SSN: An R Package for Spatial
585
Statistical Modeling on Stream Networks. J. Stat. Softw. 2014, 56 (3), 1–45,
586
10.18637/jss.v056.i03.
587 588
(49) Schwanghart, W.; Kuhn, N. J. TopoToolbox: A Set of Matlab Functions for Topographic Analysis. Environ. Model. Softw. 2010, 25 (6), 770–781, 10.1016/j.envsoft.2009.12.002.
ACS Paragon Plus Environment
32
Page 33 of 36
589
Environmental Science & Technology
(50) Homer, C. G.; Dewitz, J. A.; Yang, L.; Jin, S.; Danielson, P.; Xian, G.; Coulston, J.;
590
Herold, N. D.; Wickham, J. D.; Megown, K. Completion of the 2011 National Land Cover
591
Database for the Conterminous United States-Representing a Decade of Land Cover Change
592
Information. Photogramm. Eng. Remote Sensing 2015, 81 (5), 345–354.
593 594
(51) R Core Team. R: A Language and Environment for Statistical Computing, R Foundation for Statistical Computing; Vienna, Austria, 2017.
595
(52) Zimmerman, D. L.; Stein, M. Classical Geostatistical Methods. In Handbook of Spatial
596
Statistics; Gelfand, A., Diggle, P., Fuentes, M., Guttorp, P., Eds.; CRC Press: Boca Raton, FL,
597
2010; pp 29–44.
598
(53) Zimmerman, D. L. Likelihood-Based Methods. In Handbook of Spatial Statistics;
599
Gelfand, A. E., Diggle, P. J., Fuentes, M., Guttorp, P., Eds.; cr: Boca Raton, FL, 2010; pp 45–56.
600
(54) Schielzeth, H. Simple Means to Improve the Interpretability of Regression Coefficients.
601 602 603 604 605
Methods Ecol. Evol. 2010, 1 (2), 103–113, 10.1111/j.2041-210X.2010.00012.x. (55) Beale, C. M.; Lennon, J. J.; Yearsley, J. M.; Brewer, M. J.; Elston, D. A. Regression Analysis of Spatial Data. Ecol. Lett. 2010, 13 (2), 246–264, 10.1111/j.1461-0248.2009.01422.x. (56) Genton, M. G. Separable Approximations of Space-Time Covariance Matrices. Environmetrics 2007, 18 (7), 681–695, 10.1002/env.854.
606
(57) Jat, P.; Serre, M. L. Bayesian Maximum Entropy Space/Time Estimation of Surface
607
Water Chloride in Maryland Using River Distances. Environ. Pollut. 2016, 219, 1148–1155,
608
10.1016/j.envpol.2016.09.020.
ACS Paragon Plus Environment
33
Environmental Science & Technology
609
(58) NCDENR. Surface Waters and Wetland Standards, North Carolina Department of
610
Environment and Natural Resources Division of Water Quality; North Carolina, USA, 2007;
611
http://portal.ncdenr.org/c/document_library/get_file?uuid=f12e8078-b128-44cc-b55b-
612
fc5e7d876f3c&groupId=38364.
613
(59) Niell, A. J.; Tetzlaff, D.; Strachan, N. J. C.; Hough, R. L.; Avery, L. M.; Watson, H.;
614
Soulsby, C. Using Spatial-Stream-Network Models and Long-Term Data to Understand and
615
Predict Dynamics of Faecal Contamination in A... Sci. Total Environ. 2017, No. August,
616
10.1016/j.scitotenv.2017.08.151.
Page 34 of 36
617
(60) Peterson, E. E.; Ver Hoef, J. M.; Isaak, D. J.; Falke, J. a.; Fortin, M.-J.; Jordan, C. E.;
618
McNyset, K.; Monestiez, P.; Ruesch, A. S.; Sengupta, A.; et al. Modelling Dendritic Ecological
619
Networks in Space: An Integrated Network Perspective. Ecol. Lett. 2013, 16 (5), 707–719,
620
10.1111/ele.12084.
621
(61) Peterson, E. E.; Urquhart, N. S. Predicting Water Quality Impaired Stream Segments
622
Using Landscape-Scale Data and a Regional Geostatistical Model: A Case Study in Maryland.
623
Environ. Monit. Assess. 2006, 121 (1–3), 615–638, 10.1007/s10661-005-9163-8.
624
(62) Peterson, E. E.; Merton, A. A.; Theobald, D. M.; Urquhart, N. S. Patterns of Spatial
625
Autocorrelation in Stream Water Chemistry. Environ. Monit. Assess. 2006, 121 (1–3), 571–596,
626
10.1007/s10661-005-9156-7.
627
(63) Isaak, D. J.; Peterson, E. E.; Ver Hoef, J. M.; Wenger, S. J.; Falke, J. A.; Torgersen, C.
628
E.; Sowder, C.; Steel, E. A.; Fortin, M.-J.; Jordan, C. E.; et al. Applications of Spatial Statistical
629
Network Models to Stream Data. Wiley Interdiscip. Rev. Water 2014, 1 (3), 277–294,
630
10.1002/wat2.1023.
ACS Paragon Plus Environment
34
Page 35 of 36
Environmental Science & Technology
631
(64) Som, N. a.; Monestiez, P.; Ver Hoef, J. M.; Zimmerman, D. L.; Peterson, E. E. Spatial
632
Sampling on Streams: Principles for Inference on Aquatic Networks. Environmetrics 2014, 25
633
(5), 306–323, 10.1002/env.2284.
634
(65) Cho, K. H.; Cha, S. M.; Kang, J.-H.; Lee, S. W.; Park, Y.; Kim, J.-W.; Kim, J. H.
635
Meteorological Effects on the Levels of Fecal Indicator Bacteria in an Urban Stream: A
636
Modeling Approach. Water Res. 2010, 44 (7), 2189–2202, 10.1016/j.watres.2009.12.051.
637
(66) Nash, M. S.; Heggem, D. T.; Ebert, D.; Wade, T. G.; Hall, R. K. Multi-Scale Landscape
638
Factors Influencing Stream Water Quality in the State of Oregon. Environ. Monit. Assess. 2009,
639
156 (1–4), 343–360, 10.1007/s10661-008-0489-x.
640
(67) Whitman, R. L.; Przybyla-Kelly, K.; Shively, D. A.; Nevers, M. B.; Byappanahalli, M. N.
641
Sunlight, Season, Snowmelt, Storm, and Source Affect E. Coli Populations in an Artificially
642
Ponded Stream. Sci. Total Environ. 2008, 390 (2–3), 448–455, 10.1016/j.scitotenv.2007.10.014.
643
(68) Kang, J.-H.; Lee, S. W.; Cho, K. H.; Ki, S. J.; Cha, S. M.; Kim, J. H. Linking Land-Use
644
Type and Stream Water Quality Using Spatial Data of Fecal Indicator Bacteria and Heavy
645
Metals in the Yeongsan River Basin. Water Res. 2010, 44 (14), 4143–4157,
646
10.1016/j.watres.2010.05.009.
647
(69) Mallin, M. a M.; Williams, K. E. K.; Esham, E. C.; Lowe, R. P. Effect of Human
648
Development on Bacteriological Water Quality in Coastal Watersheds. Ecol. Appl. 2000, 10 (4),
649
1047–1056.
650 651
(70) Korajkic, A.; McMinn, B. R.; Shanks, O. C.; Sivaganesan, M.; Fout, G. S.; Ashbolt, N. J. Biotic Interactions and Sunlight Affect Persistence of Fecal Indicator Bacteria and Microbial
ACS Paragon Plus Environment
35
Environmental Science & Technology
652
Source Tracking Genetic Markers in the Upper Mississippi River. Appl. Environ. Microbiol.
653
2014, 80 (13), 3952–3961, 10.1128/AEM.00388-14.
Page 36 of 36
654
(71) Davies-Colley, R. J.; Bell, R. G.; Donnison, A. M. Sunlight Inactivation of Enterococci
655
and Fecal Coliforms in Sewage Effluent Diluted in Seawater. Appl. Environ. Microbiol. 1994, 60
656
(6), 2049–2058.
657
(72) Whitman, R. L.; Nevers, M. B.; Korinek, G. C.; Byappanahalli, M. N. Solar and
658
Temporal Effects on Escherichia Coli Concentration at a Lake Michigan Swimming Beach Solar
659
and Temporal Effects on Escherichia Coli Concentration at a Lake Michigan Swimming Beach
660
†. Appl. Environ. Microbiol. 2004, 70 (7), 4276, 10.1128/AEM.70.7.4276.
661
(73) Thupaki, P.; Phanikumar, M. S.; Beletsky, D.; Schwab, D. J.; Nevers, M. B.; Whitman,
662
R. L. Budget Analysis of Escherichia Coli at a Southern Lake Michigan Beach. Environ. Sci.
663
Technol. 2010, 44 (3), 1010–1016, 10.1021/es902232a.
664
(74) Nguyen, T. D.; Thupaki, P.; Anderson, E. J.; Phanikumar, M. S. Summer Circulation and
665
Exchange in the Saginaw Bay-Lake Huron System. J. Geophys. Res. Ocean. 2014, 119 (4),
666
2713–2734, 10.1002/2014JC009828.
667 668
(75) Annear, R. L.; Wells, S. A. A Comparison of Five Models for Estimating Clear-Sky Solar Radiation. Water Resour. Res. 2007, 43 (10), 10.1029/2006WR005055.
ACS Paragon Plus Environment
36