1
Abstract
2
We describe an approach for estimating daily flows and loads from small coastal streams
3
in central coastal California. Estimations are derived from monthly measured flow and
4
concentration data, coupled with daily flow and dilution models. Daily flow estimation makes
5
use of average annual stream flow estimates from the National Hydrography Dataset Plus
6
Estimated Runoff Model (EROM). From locally available stream gage data, daily ratios are
7
developed between daily flow averages and the EROM average annual flow estimates for each
8
gage location. Where multiple gages are used, the ratios are averaged. These ratios are then
9
applied to the EROM average annual stream flow estimates at nearby stream sampling locations
10
to estimate daily flows. Measured monthly flows are used as benchmarks to assess performance.
11
Loading estimates derived from linear interpolation of monthly flow and concentration
12
data are likely to have a high degree of error because of intervening storm events which may
13
dramatically effect loading and may cause dilution effects. In some cases high flow may reduce
14
concentrations due to dilution and in other cases, pollutants may increase in concentration with
15
flow because of overland transport. We characterize this relationship at the scale of individual
16
watersheds and describe a method for improving the accuracy of loading estimates by
17
application of watershed specific relationships between pollutant concentration and flow.
18
19
Introduction
20
The California central coast is characterized by relatively small coastal watersheds that in
21
some cases transport relatively high loads of pollutants to the Pacific Ocean. Few of these
22
watersheds contain flow gages and very few contain any water quality instrumentation, but
23
monthly water quality and flow data is available from the Central Coast Ambient Monitoring
1
24
Program (CCAMP) and related efforts. An approach was needed to better characterize flows and
25
loads leaving these watersheds; there is a continuing need to understand pollutant loading to the
26
ocean in support of epidemiological studies of marine mammal disease (M.A. Miller et al., 2002;
27
M.A. Miller et al., 2005; W.A. Miller et al., 2006; Stoddard et al, 2008), studies of the role of
28
anthropogenic nutrient inputs in plankton bloom initiation (Lane, Kudela etc), and other
29
purposes. Our approach is adaptable to other areas where local gage data and stream flow and
30
water quality measurements are available.
31
CCAMP is the California Central Coast Water Board’s regionally-scaled water quality
32
monitoring program. It operates along a 300-mile stretch of California coastline from southern
33
San Mateo County through northern Ventura County. CCAMP’s “Coastal Confluences”
34
monitoring program provides a long-term dataset of monthly stream flow and water quality
35
measurements at the ocean discharge point of thirty-three streams and rivers. Associated
36
watersheds range in size from X to Y square kilometers.
37
The Central Coast Daily Flow Estimation (CCDFE) approach described here utilizes
38
annual average flow estimates available from the Estimated Runoff Model (EROM) within the
39
National Hydrography Dataset Plus Version 2 (NHDPlus V2) (Horizon Systems, 2005) and
40
improves upon these estimates for more locally-scaled purposes, using locally available stream
41
gage and flow measurement data.
42
NHD+ V2 is blah blah
The EROM estimates average annual flows for stream reaches within the NHDPlus V2
43
geospatial framework. Estimation of monthly flow averages is planned for inclusion in future
44
revisions. Runoff, estimated for each NHD catchment, is derived from a “flow balance model”
45
(Wolock and McCabe, 1999), which incorporates precipitation, potential evapotranspiration,
46
evapotranspiration, and soil moisture storage (McCay, et al., 2012). Incremental flows are
2
47
calculated by routing and accumulating catchment levels flows using NHDPlus flow lines. This
48
estimate of “natural” runoff is further refined when necessary by accounting for instream losses
49
due to natural hydrologic processes, an important correction for flow estimates in the arid West.
50
The estimate is also corrected for a negative bias in base flow estimates through a log-log
51
regression-based adjustment using “reference” gage data from largely unaltered river and stream
52
systems (McCay, et al., 2012). The model can account for flow additions, removals and transfers
53
where data is available. Flow estimates are adjusted where on-stream flow gage data is
54
available, pro-rating for differences in drainage area. Only gages that meet specific screening
55
criteria (accuracy of reported drainage area, period of record, number of years of record) are used
56
(McCay, et al., 2012).
57
Unmodified, the NHDPlus EROM model provides annual average flow estimates for
58
medium resolution (100 K) hydrographic stream reaches defined by the National Hydrography
59
Dataset. CCDFE uses the EROM annual flow estimates as the basis for more spatially and
60
temporally explicit estimates of flow, adjusted with locally available flow gages and verified
61
with measured stream flows. CCDFE describes flow at CCAMP Coastal Confluence monitoring
62
sites at daily resolution. and reconciles the following issues inherent in the NHDPlus EROM
63
model because of its national scale: (1) CCDFE allows use of flow data from more locally
64
available stream gages that may have a shorter period of record or otherwise not be employed in
65
the EROM calculations , (2) CCDFE allows for flow adjustments using gages from nearby
66
watersheds with similar hydrologic, climatic, and land use characteristics , (3) CCDFE makes
67
use of monthly measured flow data at coastal confluence sites to optimize choice of gages and to
68
assess “fit” of the model.
69
Methods
3
CCAMP enhancements to the NHDPlus EROM model include selection of one to three
70
71
USGS gages that more directly represent localized flow conditions at the site of interest. Gages
72
are also selected to ensure that flow data throughout the time period of interest is available from
73
at least one gage at any given time. In some cases a gage may reside on the same stream system.
74
These gages may not be represented in the existing EROM model because of its required period
75
of record and other criteria. Ratios are developed between gaged daily flow measurements and
76
the EROM annual mean flow at each gage location. If more than one gage is used these ratios
77
are averaged. Mean daily flow ratios are then multiplied by the NHD+-derived annual mean
78
daily flow at the discharge location of interest to estimate flow at that location and point in time.
79
Gage choice is optimized by evaluating performance against CCAMP measured monthly stream
80
flows, collected since 2005.. Measured flows provide benchmark data for flow model validation.
81
Results
82
Linear correlations between model results and measured stream flows for most Coastal
83
Confluences of interest are high (Table 1), with r-square values typically exceeding 0.8. Not all
84
watersheds are modeled as successfully, depending on gage location, anthropogenic
85
disturbances, and watershed behavior. In particular, the Old Salinas River site, has flow
86
influence from an adjacent watershed via a tide gate during higher flow events, and is heavily
87
influenced by irrigation runoff and tile drain water. Field measurements are at times confounded
88
by tidal influence. The error associated with flow estimates at this site is likely to be higher than
89
others.
90
91
92
4
93
Error associated with linear interpolation was examined in more detail using high
94
frequency nitrate concentration data from an in-situ ultraviolet spectrophotometer sensor,
95
deployed by Monterey Bay Aquarium Research Institute through its Land Ocean
96
Biogeochemical Observatory (LOBO) network. The L03 sensor is located at the lower
97
end of the Old Salinas River in Moss Landing Harbor. This instrument collects nitrate
98
readings continuously at an hourly interval and has been in operation since 1994. Its
99
location in a tidal area presents additional sources of variability that would not be
100
encountered were the sensor located in a non-tidal riverine environment. We adapted this
101
data for use by selecting daily measurements collected at salinity low points. We
102
extracted monthly interval measurements from this dataset and used the subset to create a
103
daily linear interpolation of nitrate concentrations. Linear regression between the
104
measured and interpolated daily concentrations produced a significant relationship (y =
105
0.7228x +4.7881; P=0.00, r2 = 0.53). When averaged daily interpolated concentrations
106
were compared to averaged measured concentrations for the period of record, the
107
interpolated values underestimated average concentrations by 3%.
108
109
Monthly flow and macronutrient data were collected by CCAMP in accordance
110
with California State Board’s Surface Water Ambient Monitoring Program Quality
111
Assurance Program Plan www.waterboards.ca.gov/water_issues/programs/swamp/).
112
Depth-integrated samples are collected into 1 L plastic bottles from the center of the
113
stream flow or thalweg and immediately placed in cold ice chests (4 C) for transport to
114
the analyzing laboratory. Flow measurements used for validation of the CCAMP stream
5
115
flow estimation model are collected monthly along a ten-point cross section using a
116
Marsh-McBirney conductive probe flow meter and setting rod.
117
118
119
McKay, L., Bondelid, T., Dewald, T., et al, “NHDPlus Version 2: User Guide”, 2012
120
121
Wolock, David M. and Gregory J. McCabe. 1999. Estimates of Runoff Using Water-Balance
122
and Atmospheric General Circulation Models. Journal of the American Water Resources
123
Association 35(6):1341 – 1350, December 1999.
124
125
126
6
127
128
Figure From usepa 2003
129
130
Notes below:
131
132
Estimating Loads with available data
133
134
Given constraints of the available dataset:
135
Analyte concentrations were measured monthly
136
Flows were measured monthly when conditions permitted such measurements
137
7
138
We pursued an approach guided by tenets set forth in Richards,1997
139
"
140
1. Find a way to estimate "missing" concentrations: i.e. concentrations to go with the flows
141
observed at
142
times when chemical samples were not taken.
143
2. Abandon most of the flow data and calculate the load using the concentration data and just
144
those
145
flows which were observed at the same time the samples were taken.
146
3. Do something in between - find some way to use the more detailed knowledge of flow to
147
adjust the
148
load estimated from matched pairs of concentration and flow.
149
The second approach is usually totally unsatisfactory because the frequency of chemical
150
observations is
151
inadequate to lead to a reliable load estimate when simple summation is used. Thus almost all of
152
the
153
load estimation approaches which have been shown to give good results are variants of
154
approaches 1 or
155
3. Total load vs. unit load
156
If all of the flow and concentration observations are available for all n intervals in formula (3) or
157
(4) above,
158
the summation is an easy task. The summation is the total load for the time period of interest;
159
each
8
160
individual product, ciqiDt or ciqiti, could be called the unit load. If the total load is an annual
161
load, the unit
162
load might be the daily load. If the total load is a weekly load, the unit load might be the hourly
163
load.
164
This distinction is important, because we tend to focus on the total load as our goal. The
165
considerable
166
complexities of the various methods for calculating loads, however, are almost always related to
167
trying to
168
accurately characterize the unit loads. Once this is done, the total load is easily obtained.
169
"
170
171
We addressed flow measurement issues through use of the ccamp flow model (Worcester,
172
Paradies, initally presented at the 6th National Water Quality Monitoring Conference, Atlantic
173
City NJ 2008)
174
Daily flows were estimated using the ccamp flow model
175
176
We addressed issues associated with flow dependent concentration variability in the following
177
fashion:
178
179
We initially explored using linear interpolation to impute concentration values for days between
180
measured concentrations.(as in Chescheir etal and other studies)
181
Exploratory data analysis revealed that site specific relationships existed between concentrations
182
and flows. (as in Birgand etal, Stelzer etal,Zamyadi etal and other studies)
9
183
In some instances concentrations were reduced as flows increased.(as in Birgand etal, Stelzer
184
etal,Zamyadi etal and other studies)
185
In some instances concentrations were increased as flows increased.(as in Birgand etal, Stelzer
186
etal,Zamyadi etal and other studies)
187
188
We were concerned about the subjectivity associated with graphical identification of
189
concentration and dilution effects (e.g. the K. Worcester pers comm and the USGS 'worked
190
record' methodology).
191
We explored various methods of detecting the presence of these relationships.(we found no
192
literature which provided an automated scanning methodology for detecting possible
193
concentration or dilution effects)
194
We adopted a hybrid approach for scanning site - analyte combinations to identify possible
195
concentration or dilution effects.
196
The scanning algorithm employs flow interval, ratio estimation, and regression (as described in
197
EPA 841-B-03-004, July 2003 and Richards,1997).
198
199
We calculated average concentrations at each site in each of 30 flow interval-regime bins to
200
produce a table of of limits for each flow interval regime
201
We successively calculated the regression based coefficient of determination (R squared) for
202
average concentration in each flow regime for each site analyte combination
203
204
10
205
in stepwise fashion begining with all flow regimes and successively eliminating low flow regime
206
values one bin at a time until an R squared value exceeds .5 (or not)
207
208
The slope of the regression line is used to distinguish between concentration and dilution effects.
209
Notes:
210
The use of regression here is used only to detect site-analyte situations where concentration -
211
dilution adjustments are needed.
212
The results of the regression are not used in the process of 'load estimation'.
213
214
We explored various methods of improving load estimates which address detected site specific
215
relationships.
216
We adopted a hybrid approach based on a framework set forth in EPA 841-B-03-004, July 2003
217
and Richards, P 1997.
218
The hybrid approach includes:
219
A 'worked record' approach (as described in EPA 841-B-03-004, July 2003 and Richards, P
220
1997) modified to utilize the flow regime algorithm described above to mitigate the subjectivity
221
normally associated with the 'worked record' approach with respect to dilution and concentration
222
effects.
223
A 'flow interval' approach (as described in EPA 841-B-03-004, July 2003 and Richards, P 1997)
224
which utilizes the flow interval regimes described above to impute daily flux, but not to calculate
225
loads for more than a single day.
11
226
A 'ratio estimator' approach (as described in EPA 841-B-03-004, July 2003 and Richards, P
227
1997) which utilizes the flow interval regimes described above to impute daily flux, but not to
228
calculate loads for more than a single day.
229
230
The hybrid algorithm initally utilizes linear interpolation to establish a trial value for
231
concentration for each day which is missing a concentration measurement.
232
In the case of dilution effects; When a trial value exceeds the average concentration in a given
233
flow regime, the average concentration for the flow regime is used as the imputed value.
234
In the case of concentration effects; When a trial value is less than the average concentration in a
235
given flow regime, the average concentration for the flow regime is used as the imputed value.
12