Remote Sensing of Environment 129 (2013) 197–209

Contents lists available at SciVerse ScienceDirect

Remote Sensing of Environment

journal homepage: www.elsevier.com/locate/rse

Remote sensing of suspended sediment concentration and hydrologic connectivity in

a complex wetland environment
Colleen M. Long ⁎, Tamlin M. Pavelsky
Department of Geological Sciences, University of North Carolina, Chapel Hill, NC, USA

a r t i c l e i n f o a b s t r a c t

Article history: Maintaining the ecological diversity and hydrologic connectivity of freshwater delta systems depends on
Received 8 July 2012 regular recharge of floodplains with river water, which can be difficult to observe on the ground. Rivers
Received in revised form 21 October 2012 that form deltas often carry large amounts of suspended sediment, but floodplain lakes and wetlands usually
Accepted 24 October 2012
have little sediment in suspension. Remote observation of high sediment water in lakes and wetlands
Available online 4 December 2012
therefore often indicates connectivity with the river network. In this study, we use daily 250-m MODIS
Keywords:
imagery in band 1 (620–670 nm) and band 2 (841–876 nm) to monitor suspended sediment transport
Remote sensing and, by proxy, hydrologic recharge in the Peace–Athabasca Delta, Canada. To identify an appropriate
Suspended sediment suspended sediment concentration (SSC)-reflectance model, we compare 31 published empirical equations
Wetland using a field dataset containing 147 observations of SSC and in situ spectral reflectance. Results suggest
MODIS potential for spatial transferability of such models, but success is contingent on the equation meeting certain
Peace–Athabasca Delta criteria: 1) use of a near infrared band in combination with at least one visible band, 2) development based
on SSCs similar to those in the observed region, and 3) a nonlinear form. Using a highly predictive SSC-
reflectance model (Spearman's ρ = 0.95), we develop a twelve-year time series of SSC in the westernmost
end of Lake Athabasca, observe the timing and sources of major sediment flux events, and identify a threshold
river discharge of ~ 1700 m 3/s above which SSC in Lake Athabasca is clearly associated with flow in the
Athabasca River. We also track the influx of Athabasca River water to floodplain lakes, and in three of the
lakes identify distinct discharge thresholds (1040 m3/s, 1150 m 3/s, and 1850 m3/s) which result in lake
recharge. For each of these lakes, we find a statistically significant decline in the threshold exceedence
frequency since 1970, suggesting less frequent recharge during the summer.
© 2012 Elsevier Inc. All rights reserved.

1. Introduction decreases in the amount and frequency of sediment and water input,
as well as constraints on the spatial extent of recharge, can substantially
Freshwater deltas are among the most biologically productive impact geomorphic and biogeochemical processes in deltas (Meade,
and ecologically diverse terrestrial ecosystems. To sustain their charac- 1996; Yang et al., 2003).
teristic complexity, these environments depend on regular recharge The supply of river water to delta floodplains is largely controlled
with sediment-laden river water, generally through overbank flooding by hydrologic connectivity, and an understanding of connectivity is
and flow through distributary channel networks (Lesack et al., 1998). therefore critical to assessing a delta's ecological health (Bracken &
Transport and deposition of sediment in deltas shape the landscape Croke, 2007; Pringle, 2003). Defining and quantifying connectivity,
through construction of natural levees, aggradation of wetland areas, however, remain intractable problems in many systems. Metrics
and progradation of the delta margin (Syvitski et al., 2009). Moreover, used to assess hydrologic connectivity must often be uniquely chosen
the movement of water and sediment through delta ecosystems is a based on the catchment under investigation (Ali & Roy, 2010), and
key mechanism for cycling of nutrients and contaminants (Bloesch, many techniques require extensive fieldwork (e.g. Fennessy et al.,
1995; Hernández-Ayón et al., 1993; Owens et al., 2005; Pereira et al., 2004; Jencso et al., 2009; Wolfe et al., 2007). The information these
1996). Delivery of sediment to deltas worldwide has declined in the methods provide can be limited in complex freshwater deltas where
last century due to trapping in upstream reservoirs, artificial construc- small changes in water level are highly consequential, low slopes
tion of levees, and impacts of climate change (Kummu & Varis, 2007; lead to frequent reversals in flow direction, and connections form
Syvitski, 2008; Syvitski et al., 2005; Vörösmarty et al., 2003). These through small channels or diffuse transport processes. Some previous
studies have used remote sensing to monitor connectivity and effec-
⁎ Corresponding author at: 104 South Rd., Mitchell Hall, Campus Box #3315, Chapel
tively bypass the problems associated with in situ assessments, but
Hill, NC, USA. Tel.: +1 217 300 0664. most results show general spatiotemporal variations in the degree
E-mail address: [email protected] (C.M. Long). of connectivity without fully constraining timing and magnitude of

0034-4257/$ – see front matter © 2012 Elsevier Inc. All rights reserved.
http://dx.doi.org/10.1016/j.rse.2012.10.019
198 C.M. Long, T.M. Pavelsky / Remote Sensing of Environment 129 (2013) 197–209

specific recharge events (e.g. Mouchot et al., 1991; Pavelsky & Smith, challenges of in situ data collection, and since the 1970s has been
2008; Smith & Alsdorf, 1998). used to quantify SSCs in surface waters (Ritchie et al., 1976).
Using remote sensing to monitor the presence of high-sediment The amount of sediment in water directly affects the reflectance of
river water in floodplain lakes compliments other methods of assessing solar radiation in the visible and near-infrared portions of the spectrum;
connectivity by tracking specific recharge events, even in complex delta in general, the more sediment in suspension, the higher the reflectance
systems (Pavelsky & Smith, 2009a). Because the suspended sediment (Curran & Novo, 1988; Ritchie et al., 2003). However, the exact form of
concentrations (SSCs) in delta-forming rivers generally exceed those the relationship between SSC and reflectance also depends on the min-
in floodplain lakes and wetlands, high-sediment water observed out- eralogy, color, and size of the sediments (Bowers & Binding, 2006;
side of the river often indicates connectivity to the river (Pavelsky & Bowers et al., 2007; Novo et al., 1989). These factors can be highly var-
Smith, 2009a; Wiklund et al., 2011). Sediment can therefore be used iable in natural environments, and therefore the applicability of an SSC–
as a tracer of hydrologic recharge in freshwater deltas. Because deltas reflectance relationship is generally assumed to be limited to the setting
often have highly variable flow conditions and many lakes and wet- in which the data were collected. Most studies develop unique relation-
lands, frequent and long-term in situ measurements of SSC can be ships by relating field measurements of SSC to reflectance data from sat-
prohibitively difficult to obtain. Remote sensing is not subject to the ellite imagery (for examples, Table 1). This purely empirical approach

Table 1
Compilation of published, empirically developed models relating suspended sediment concentration or turbidity to reflectance from the water surface. Maximum turbidity values
have been converted to approximate SSCs. Equations are written as they are published, where SPM = Suspended Particulate Matter, SSC = Suspended Sediment Concentration,
TSM = Total Suspended Matter, SS = Suspended Solids, and TSS = Total Suspended Solids. R is the reflectance of the water at the given wavelengths. For equations that measure
turbidity, maximum values shown in Column 5 have been converted to SSCs to facilitate intercomparison. Spearman's ρ value is the correlation coefficient between SSC values
measured in the PAD and SSC or turbidity values predicted by the model. Scatterplots of observed vs. modeled values for the six bolded equations are shown in Fig. 7.

Data products/bands Wavelengths Empirical relationship between Max SSC (mg/L) Spearman's Reference
suspended sediment (or turbidity) or Turbidity (NTU) ρ
and reflectance

SPM ¼ 29:022  e0:0335ðR1Þ

R2
Landsat TM 2 and 4 R1 = 520–600 ~2500 0.97 Doxaran et al., 2003
R2 = 760–900
SPM ¼ 26:083  e0:0366ðR2Þ
R1
Sea WiFS R1 = 545–565 ~2500 0.96 Doxaran et al., 2003
R2 = 845–885
0:0322ðR2
R1Þ
SPOT XS3 and XS1 R1 = 510–590 SPM ¼ 18:895  e ~2500 −0.96 Doxaran et al., 2003
R2 = 790–890
0:0279ðR2
R1Þ
SPOT XS3 and XS1 R1 = 510–590 SPM ¼ 27:424  e ~2500 −0.96 Doxaran et al., 2003
R2 = 790–890



R1 2
Landsat MSS 5 and 6 R1 = 600–700 lnðSSC Þ ¼ −6:2  R1
R2 þ 1:4  R2 þ 10:8 1000 0.96 Topliss et al., 1990
R2 = 700–800
SPM ¼ 12:996  eR1=0:189
R2
MODIS 1 and 2 R1 = 620–670 ~2250 0.95 Doxaran et al., 2009
R2 = 841–876
Landsat TM 1, 3, 4 R1 = 450–520 Turbidity ¼ 11:31  R1 −2:03
R3
 R2−16:42 ~12 0.87 Song et al., 2011
R2 = 630–690
R3 = 760–900
Landsat TM 4 790–900 Turbidity = 16.1 ∗ R − 12.7 ~5 0.76 Fraser, 1998
MODIS 2 841–876 ln(SSC) = (43.233 ∗ R) + 1.396 2500 0.75 Wang et al., 2008
Field 782 Turbidity = (1181 × R2) + (4062 × R) − 0.0 ~2.5 0.72 Holyer, 1978
spectrometer    
Turbidity ¼ 233:7  R2 − 1384  R2 þ
Field R1 = 652 50 0.67 Holyer, 1978
spectrometer R2 = 782 ð1120  RÞ þ ð4853  RÞ−5:08
CASI Channel 11 755.5–780.8 SSC = 529 ∗ R 2000 0.65 Wass et al., 1997
(rounded to 755–781)
AHS Advanced R1 = 819–847 ln(TSM) = 34.18 ∗ (R1 − R2) + 3.16 336 0.60 Sterckx et al., 2007
Hyperspectral Sensor R2 = 989–1019


R1 1:11
Sea WiFS R1 = 660–680 SPM ¼ 17:783  R2 ~20 0.47 D'Sa et al., 2007
R2 = 545–565
IKONOS red 632–698 Turbidity = 0.078 ∗ R − 8.7 ~1 0.44 Hellweger et al., 2007
Landsat TM 3 630–690 Turbidity = 10.0 ∗ R − 24.8 ~5 0.43 Fraser, 1998
Landsat TM 1 450–520 Turbidity = 19.0 ∗ R − 97.9 ~5 0.43 Fraser, 1998
Field spectrometer 652 Turbidity = (33.96 ∗ R2) + (5352 ∗ R) − 4.38 ~2.5 0.42 Holyer, 1978
MODIS 1 620–670 Turbidity = 1203.9 ∗ R1.087 ~1 0.43 Chen et al., 2007
Landsat TM 2 520–600 Turbidity = 6.4 ∗ R − 28.0   ~5 0.39 Fraser, 1998
CMODIS R1 = 540–560 log10 S ¼ 0:892 þ 6:2244  R1þR2 ~1000 0.36 Han et al., 2006
ðR 1=R 2Þ
R2 = 660–680
Landsat TM 3 630–690 SSC = 69.39 ∗ R − 201 1150 0.36 Islam et al., 2001



R1 2
Landsat MSS 1 and 2 R1 = 500–600 lnðSSC Þ ¼ −9:21 R1R2 þ 2:71 R2 þ 8:45 ~150 0.34 Ritchie & Cooper, 1991
R2 = 600–700



R1 2
MOS/MESSR R1 = 510–590 lnðSSC Þ ¼ −4:8 R1
R2 þ 0:9 R2 þ 10:4 1000 0.33 Topliss et al., 1990
R2 = 610–690
Landsat TM 3 630–690 log 10(S) = 0.334 + 0.098 × R 30 0.32 Keiner & Yan, 1998
Landsat MSS 5 600–700 R = 0.16 + 0.03 ∗ ln(S) 0.32 Aranuvachapun & Walling,
1988
MODIS 1 620–670 ln(SSC) = 50.171 ∗ R − 1.523 ~2500 0.31 Wang et al., 2008
MODIS 1 620–670 TSM = −1.91 + 1140.25 ∗ R 60 0.31 Miller & McKee, 2004
MODIS 1 620–670 R = 7.5 ∗ log(SSC) + 1.6 500 0.31 Chu et al., 2009
TSM ¼ 0:7581  e61:683ð 2 Þ
R1þR2
Landsat TM 2 and 3 R1 = 520–600 50 0.30 Dekker et al., 2001
R2 = 630–690
R2
Landsat TM 1 and 3 R1 = 450–520 TSS ¼ 0:0167  e12:3R1 35 −0.05 Lathrop et al., 1991
R2 = 630–690
 C.M. Long, T.M. Pavelsky / Remote Sensing of Environment 129 (2013) 197–209 199

restricts development of SSC maps from satellite imagery on regional the PAD one of the world's largest freshwater deltas. It has been named
and global scales. A limited number of previous studies have explored a United Nations Educational, Scientific, and Cultural Organization
whether models developed in one location are spatially transferable (UNESCO) World Heritage site and a Ramsar Convention Wetland of
to another (Holyer, 1978; Ritchie & Cooper, 1991; Topliss et al., 1990) International Importance because of its biological significance and role
and suggest that transferability may be possible. However, these studies as the largest alluvial-wetland habitat in the region. The delta provides
have focused on low SSCs (Holyer, 1978) and a limited range of wave- a habitat for migratory birds and land mammals including moose,
lengths (Ritchie & Cooper, 1991; Topliss et al., 1990), and they do not black bear, and wood buffalo (Prowse & Conly, 2002). The wildlife and
incorporate work from the past two decades. Development of one or vegetation in the PAD also have cultural and historical importance to
more spatially transferable models would be of great scientific and indigenous residents in the region, including the Athabascan Chipewyan
practical interest, but no recent work has compared the applicability and Mikisew Cree First Nations. A large portion of the delta (~80%) is
of existing models in a common natural setting. protected within Canada's Wood Buffalo National Park. However, the
The Peace–Athabasca Delta (PAD) (Fig. 1), located in northeastern PAD depends on the rivers that flow into it to support its aquatic and
Alberta, Canada, is an ideal setting for studying the transferability terrestrial habitats, and natural or anthropogenic changes in river flow
of SSC-reflectance models. It exhibits a wide range of SSCs, receives far upstream can have significant effects on the delta.
sediment from more than one source, and is often cloud-free during During normal flow conditions, the main source of recharge to the
the summer. Previous work has demonstrated that robust statistical delta is the north-flowing Athabasca River, and drainage is northward
relationships between SSC and remotely sensed reflectance can be through several distributaries to the Peace River and, ultimately, the
constructed in the PAD (Pavelsky & Smith, 2009a). Furthermore, recent Slave River. However, if the water level on the Peace River is higher
alterations to flow on the Peace and Athabasca Rivers associated with than that of Lake Athabasca, flow reverses along the distributary
changing climate and human impacts have raised questions about channels that connect the Peace River to the lake. Such reversals
how hydrologic recharge in the PAD may be changing (Wolfe et al., affect sediment and water delivery throughout the northern part of
2011, 2008). Remote sensing of spatial and temporal variations in SSC the delta, but because the Athabasca River consistently flows north
can help address these questions. into Lake Athabasca, the southern sector of the delta is likely to be sig-
This study is comprised of three major components. We first nificantly affected only by the strong reversals of flow on the Peace
explore the applicability of published, site-specific SSC-reflectance River associated with ice jam flooding, which have not occurred for
models to the PAD in order to understand the extent to which these more than fifteen years (Peters et al., 2006). The relationship between
relationships are spatially transferable. We then use a highly predic- flow on the Peace River and the hydrology and ecology in the Peace
tive model to distinguish the source and timing of major sediment sector of the delta have been addressed by several studies since
flux events in the western end of Lake Athabasca, which forms the the 1970s in response to the completion of the W.A.C. Bennett dam,
eastern boundary of the PAD. Finally, we use remote sensing of SSC which regulates flow from the headwaters of the Peace River
to determine discharge thresholds above which river water recharges (Farley & Cheng, 1986; Leconte et al., 2001; Peters & Prowse, 2001;
small floodplain lakes in the PAD and examine changes in the exceed- Prowse & Demuth, 1996). There is some evidence that flow regulation
ence frequency of these thresholds. has decreased the frequency of ice jam flooding on the Peace River,
which is the primary mechanism of very large-scale flooding in the
2. Study area: The Peace–Athabasca Delta PAD (Beltaos et al., 2006a, 2006b). Between 1959 and 1976, major
ice-jam flooding occurred four times in seventeen years, but between
The Peace–Athabasca Delta is a hydrologically complex and ecologi- 1976 and the present (36 years), the only major events occurred in
cally diverse freshwater delta formed by the confluence of the Peace, 1996 and 1997 (Beltaos et al., 2006a). As a result, the dominant
Athabasca, and Birch Rivers near the western end of Lake Athabasca sources of recharge in recent decades have been springtime ice jam
(Fig. 1). The three intersecting river deltas cover ~5200 km2, making floods and summertime high water events on the Athabasca River
(Pavelsky & Smith, 2008; Peters et al., 2006; Töyrä & Pietroniro,
2005).
Since the 1970s, summer discharge on the Athabasca River has
declined nearly 35% (Fig. 2) due to both natural and anthropogenic
forces. Increased evapotranspiration and diminished contributions

Fig. 2. Mean summer (May–September) discharge on the Athabasca River below Fort
Fig. 1. Landsat image of the Peace–Athabasca Delta. Inset (from Pavelsky & Smith, McMurray from 1970 to 2011. Dashed line shows statistically significant decline in dis-
2009a) shows location of the delta in Canada. charge over the time period. Since 1970, mean May–September flow has decreased ~35%.
200 C.M. Long, T.M. Pavelsky / Remote Sensing of Environment 129 (2013) 197–209

from snowpack and glaciers in the Canadian Rockies, likely related to

anthropogenic climate change, are responsible for much of this
decline (Schindler, 2001; Schindler & Donahue, 2006). Increasing
water withdrawals for industrial operations in the Alberta Oil Sands
(located on the Athabasca River about 250 km upstream from the
PAD) could add to future discharge declines (Schindler & Smol, 2006),
though winter, not summer, is the key time of year for assessing the
ecological impacts from oil sands withdrawals (Andrishak & Hicks,
2011). The effects of this decline on hydrologic recharge in the PAD
are largely unknown. In particular, the thresholds of discharge required
to deliver river water to ecologically important floodplain lakes remain
unquantified.
Recharge of floodplain lakes in the Athabasca sector of the PAD
occurs when the Athabasca River rises sufficiently to either overtop
its levees or reverse flow on the small channels that ordinarily are
directed into the river from floodplain lakes. Except during recharge
events, SSCs are higher in the Athabasca River than in floodplain
lakes. As a result, the presence of high sediment water in a floodplain
lake represents, under most circumstances, an input of river water. A
detailed time series of SSCs in these lakes would provide information
about how hydrologic recharge in the PAD is responding to the
decreases in discharge on the Athabasca River. Pavelsky and Smith
(2009a) used satellite images to capture episodic recharge of flood-
plain lakes in the Athabasca sector of the delta and related timing of
recharge to discharge on the Athabasca River. We expand on this
work by examining SSC in Lake Athabasca and determining the
threshold discharge values required for hydrologic recharge of flood-
plain lakes.

3. Methods

3.1. Collection of field data Fig. 3. Landsat image of the central PAD showing the 71 locations where field data was
collected in 2011. Outlined are the six small lakes referenced in Figs. 11 and 12. Stars
mark the locations of the virtual SSC gauges where the values in Fig. 9 were calculated.
During a field season from June 20th to July 7th, 2011, we measured
spectral reflectance and water quality in lakes, rivers, and distributary
channels in the PAD. In-situ measurements of temperature (°C), color
dissolved organic matter concentration (μg/L), turbidity (NTU), chloro- 3.2. Collection and processing of satellite imagery
phyll content (μg/L), and specific conductivity (μS/cm) were collected
using a Eureka Manta Multiprobe a total of 147 times at 71 unique loca- Previous studies have used high resolution satellite imagery to
tions visited between 1 and 4 times (Fig. 3). Multiprobe measurements track variations in SSC in wetland environments (e.g. Doxaran
were collected at two-second intervals for at least three minutes and et al., 2002; Mertes et al., 1993; Ritchie & Cooper, 1991). Because
averaged to obtain one value for each variable per site. Suspended sed- of infrequent temporal sampling, however, such imagery is unsuit-
iment concentrations were measured at each site using 275 mL water able for tracking SSC in the PAD on daily to weekly timescales. We
samples from the top ~15 cm of the water column. Water samples therefore use imagery from NASA's two Moderate-resolution Imag-
were filtered onto pre-weighed 1.2 μm Millipore cellulose filters using ing Spectroradiometer (MODIS) sensors, which each provide daily
a vacuum filtration system. After filtration, the filters were dried in an observations. Many previous studies have found that a combination
oven at 100 °C for 30 min and then reweighed on a high precision of red and near-infrared bands, such as MODIS band 1 (red; 620–
balance to determine the weight of total suspended matter. This weight 670 nm) and band 2 (near-infrared; 841–876 nm), results in robust
divided by the sample volume yielded the suspended sediment concen- quantification of SSC (e.g. Doxaran et al., 2009; Holyer, 1978; Novo
tration (mg/L). Finally, surface flow velocity was measured at most river et al., 1989). Additionally, MODIS bands 1 and 2 have a sufficiently
locations (67 total measurements) using a stopwatch, a handheld GPS, high spatial resolution (250 m) in the red and near-infrared bands to
and a small drogue (following Pavelsky & Smith, 2009b). These datasets detect moderate-sized floodplain lakes such as those found in the PAD.
augment water quality data collection in the PAD beginning in 2006 and Daily MODIS Level 2G Aqua and Terra scenes (MOD09GQ) col-
continuing in 2007, 2010, and now 2011. The 2006–2007 data are ar- lected over the PAD were downloaded using NASA's Warehouse
chived and accessible at the Oak Ridge National Laboratory Distributed Inventory Search Tool (http://wist.echo.nasa.gov) for every summer
Active Archive Center (Pavelsky & Smith, 2009b). (May–September) from 2000 through 2011. A total of 3672 images
Spectral reflectance from the water surface was measured at 1 nm were obtained. The Aqua satellite was not launched until 2002, and
intervals between 350 and1025 nm at each of the 147 field sites using therefore for 2000–2002 only one MODIS-Terra image per day is
an ASD FieldSpec® 3 Portable Spectrometer with an attached OL 731 available. All MODIS Level 2G scenes are radiometrically and geo-
Smart Detector for solar irradiance measurements. A white reference metrically corrected for variations in sun and sensor angle and atmo-
measurement was obtained from a Spectralon® white standard at spheric conditions. Scenes were projected into UTM Zone 12N/WGS
each location prior to collection of reflectance spectra to account 84, and the dataset was automatically filtered to remove scenes
for changes in lighting conditions between sites. Three reflectance where the PAD is more than 30% cloud covered based on a band 1 re-
measurements were collected at each location and then averaged to flectance threshold of 0.15 (following Chen et al., 2009). Elimination
obtain one reflectance value per wavelength per site. of cloudy images yielded 1620 total images suitable for analysis.
 C.M. Long, T.M. Pavelsky / Remote Sensing of Environment 129 (2013) 197–209 201

Most data collection points in PAD rivers are not visible in MODIS
imagery because the spatial resolution is too coarse. Among the 147
field data collection points, though, is a twenty-three point transect
in the western end of Lake Athabasca, and MODIS pixels correspond-
ing to these points are free of land contamination (Fig. 4a). We use
data from these points to directly validate MODIS-derived reflectance
and SSC against field observations. Reflectance data was extracted
from MODIS bands 1 and 2 for the 23 transect points, and comparison
of the band 2/band 1 ratio with same-day in situ reflectance at the
same wavelengths shows a high correlation (r = 0.91) and little bias
(p = 0.08 using a paired Student's t test) (Fig. 5). A ratio is used for
this comparison to account for atmospheric variations over short
spatial scales that are not addressed by the large-scale corrections
applied to the MODIS images. The high correlation suggests that in
situ reflectance can be used to test the applicability of empirical
SSC-reflectance models to the PAD, and models developed and tested
using in situ data can also be applied to MODIS imagery.

3.3. Evaluation of SSC-reflectance models

Fig. 5. Comparison between the band 2/band 1 reflectance ratio from MODIS imagery and
from field spectrometer measurements. Close fit to a 1:1 line verifies transferability of
In order to determine if an existing SSC-reflectance model can be
models between reflectance data sets.
applied in the PAD and at the same time identify which characteris-
tics result in more transferable SSC-reflectance models, we evaluate
the spatial transferability of 31 published SSC-reflectance empirical
relationships, 23 of which are independent and based on unique
field data sets. Though some of the models have one or more other
corresponding equations based on the same field data (i.e. models models in Table 1 were developed using SSCs from b 10 to 2250 mg/L,
from Doxaran et al., 2003; Fraser, 1998; Holyer, 1978; Topliss et al., but most individual studies are limited to a narrow range of concentra-
1990), they are included in the analysis because the models use dif- tions. The 147 field measurements of SSC in the PAD range from 3.9 to
ferent spectral bands and sensors. The collection of models represents 3602 mg/L, and no previous work has tested the applicability of
a range of hydrologic environments including rivers, inland lakes, es- SSC-reflectance models across such a wide range of concentrations.
tuaries, deltas, and bays. Some models predict turbidity instead of Some models predict SSCs that are highly linearly correlated with
SSC, but as in many discussions of remote sensing of SSC (e.g. measured values but are numerically dissimilar. In order to improve
Duane Nellis et al., 1998; Wass et al., 1997) we treat both metrics in- these models' predictive capacity, we perform linear regressions be-
terchangeably. We use 147 in situ spectral observations from the PAD tween measured and modeled values and divide the modeled values
to estimate SSC using each model and compare the results against in by the slope of the regression line. These scaling factors may be re-
situ SSC measurements. Because the relationships often contain out- quired in part because of physical differences in the sediment color
liers and are sometimes non-linear in form, Spearman's ρ (Spearman, and grain size among different locations. It is also possible that in
1904), a nonparametric test of correlation, is a more appropriate mea- some studies reflectance is scaled differently (e.g. as a percent, a pro-
sure of covariance in this case than is Pearson's r. Collectively, the portion, or between 0 and 255), and this is not reported in all cases.

Fig. 4. (a) False color MODIS image from the Terra platform obtained on June 27, 2011. Transect of 23 points where field measurements of SSC and reflectance were collected are
shown in white. (b) Comparison between measured SSCs and those predicted using same-day reflectance from the spectrometer, MODIS Terra, and MODIS Aqua as inputs to the
model from Doxaran et al. (2009) scaled for the PAD.
202 C.M. Long, T.M. Pavelsky / Remote Sensing of Environment 129 (2013) 197–209

3.4. Analysis of SSC in Lake Athabasca selecting the first event after June 1st when discharge increases by
at least 60% between two inflection points (for example, Fig. 6). As
Among the most highly predictive models is one developed by water rises in the Athabasca River, it recharges the floodplain lakes
Doxaran et al. (2009) specifically for MODIS data (see Section 4.1). at specific stage thresholds, but once recharge has begun the sedi-
We use this model to analyze sediment dynamics in the western ment and water dynamics become much more complex. As such, it
end of Lake Athabasca using MODIS imagery from summers 2000 to is during this first summertime high water event that we are most
2011. To monitor time series of sediment input to Lake Athabasca likely to cleanly identify a recharge threshold. After discharge reaches
from the Peace and Athabasca Rivers, we establish two virtual sedi- a maximum, the decline in river discharge and the decline in SSC in
ment gauges in the western end of Lake Athabasca: one at the termi- the lake do not happen at the same rate, so only the rising limb of
nus of the Athabasca Delta, and a second on the north shore of Lake the peak is used here. We do not consider peaks before June 1st to
Athabasca in an area that generally receives high sediment water avoid the effect of early spring floods related to ice jams, in which
only from the Peace River (Fig. 3). To create these virtual gauges, the relationships between water levels, sediment concentrations,
we quantify SSC for three horizontally adjacent pixels at each location and river discharge differ from those during the open-water season.
and then use the median value to represent the SSC. We remove The timing and duration of each event are shown in Fig. 6. In 2001,
images where any of the six gauge pixels are cloud covered (band we manually extend the analysis period by 4 days because of a
1 > 0.15). Previous work suggests that SSC-reflectance models are one-day 9% decline during the rising limb of the hydrograph.
time invariant in a localized environment so long as the source of
sediment remains the same (Dekker et al., 2001; Ritchie et al., 4. Results
2003), and we use the scaled Doxaran et al., 2009 equation to develop
time series of SSC at both virtual gauges for each summer from 2000 4.1. Spatial transferability of SSC-reflectance models
to 2011 with daily to weekly temporal resolution (on average, ~ 40%
of days per summer were suitable for inclusion). Our results confirm that it is possible to qualitatively predict SSCs
We compare the SSC time series from the two virtual gauges to daily in the PAD using models developed elsewhere. Several models pro-
river discharge measurements on the Peace and Athabasca Rivers. River duce estimates of SSC that are highly correlated with our validation
discharge data was obtained from Environment Canada for the gauge dataset (Table 1), and these models can be adapted using in situ
stations Athabasca below McMurray (Station ID: 07DA001) and Peace measurements to quantitatively predict SSCs in the PAD. Among the
at Peace Point (Station ID: 07KC001). We apply a three-day lag to the highly predictive models, the form of the best-fit relationship between
Athabasca River discharge data to account for the ~250 km distance observed and predicted SSCs varies, and it often has a power law, rather
between the gauge and the PAD. The optimal lag of three days is than a linear, form (Fig. 7 b,c,f). The model from Doxaran et al. (2009) is
based on 1) calculations with surface flow velocity measurements in the only one to produce values that are linearly correlated with the val-
the Athabasca River and the distance to the gauge, and 2) linear regres- idation dataset across the entire range of SSCs observed in the PAD
sions between 84 SSC measurements in the Athabasca River and river (Fig. 7a). This equation, with a scaling factor of 2.9 applied, accurately
discharge with 0, 1, 2, 3, and 4 day lags. Comparison of Athabasca predicts SSCs using both in situ reflectance measurements and satellite-
River discharge and sediment data suggests that the best fit relationship based reflectance from MODIS Terra and Aqua (Fig. 4b). Comparing
between the two variables is linear, and the three-day lag gives the measured SSCs to those predicted with this model results in a better
highest R2 value (0.88) in the linear regressions. The gauge for the R-squared value for the PAD (0.94) than for the Gironde Estuary (0.89),
Peace River is about 155 km away from Lake Athabasca, but because where the model was developed. Some other models produce values
flow dynamics are complicated and often reverse in this reach of the that are linearly correlated with the validation dataset at low concentra-
river, we were unable to determine a consistent lag. Instead, we use tions (i.e. b~200 mg/L), but become nonpredictive at higher concentra-
same-day discharge for the Peace River. tions (Fig. 7 d,e). For all cases in which the best-fit relationship is linear,
application of a correction factor based on the slope of the linear regres-
3.5. Analysis of floodplain lake recharge sion between the observed and modeled data is necessary to reproduce
observed SSC values.
We manually select six floodplain lakes that are consistently visi- Although some models do accurately predict SSCs in the PAD,
ble in MODIS images (Fig. 3) to determine Athabasca River discharge values from the majority of the models do not closely match the
thresholds above which river water is consistently delivered to flood- validation dataset. Some produce negative values for SSC or turbidity
plain lakes in the Athabasca sector of the PAD. We are unable to (e.g. Fraser, 1998; Hellweger et al., 2007; Islam et al., 2001; Song et al.,
quantify SSCs in these lakes with the model used in Lake Athabasca 2011) (Fig. 7f), others give values orders of magnitude different from
because the spatial resolution of MODIS results in mixed pixels con- our measurements (e.g. Aranuvachapun & Walling, 1988; Dekker
taminated by vegetation or bare soil in most of the small floodplain et al., 2001; Hellweger et al., 2007; Keiner & Yan, 1998; Lathrop
lakes. Instead, for each clear-sky MODIS image we use a binary metric et al., 1991), and many are poorly correlated with field measure-
to monitor Sediment Area Index (SAI), the proportion of lake area ments. Differences in mineralogy, sediment color, and sediment size
containing high sediment water (following Pavelsky and Smith, can likely explain some of the limited predictive power of these
2009a). To calculate SAI, we create individual masks of each lake, models in the PAD, but it is surprising that many models offer no
and remove images in which any masked pixel is contaminated by predictive capacity given the positive relationship between SSCs and
cloud. For each lake in each remaining image, we count the number of spectral reflectance that forms the basis for all of the models.
water pixels (defined as MODIS band 2 b 0.05) and the number of high Comparing the successful models to those that are ineffective sug-
sediment water pixels (defined as MODIS band 1–band 2 ≥0.01). The gests three main factors which influence the predictive ability of any
sizes of the lakes vary temporally, and analyses for each lake are limited given model and increase the potential for spatial transferability: 1)
to days when lake area exceeds 0.25 km2 (4 inundated pixels). The the use of a combination of a near infrared band with one or more
number of high-sediment water pixels divided by the total number of visible bands, 2) development based on SSCs with a maximum similar
water pixels yields the SAI. to the maximum SSC observed in the PAD, and 3) a nonlinear equa-
We compare the SAI time series for each lake to discharge on the tion form. The six best models (four of which are independent and
Athabasca River on days during the rising limb of the first major sum- all with ρ ≥ 0.95) were developed using maximum SSCs of at least
mertime hydrograph peak. These events (one per year) are found 1000 mg/L and a combination of reflectance in a near infrared and
by identifying all points of inflection on the hydrograph and then a visible band. The seventh best, though calibrated for less turbid
 C.M. Long, T.M. Pavelsky / Remote Sensing of Environment 129 (2013) 197–209 203

Fig. 6. River discharge on the Athabasca River below Ft. McMurray in 2002 as an example to show metric used for analyses. Bold, highlighted portion indicates the rising limb of the
first summertime (after June 1) hydrograph peak (defined as an increase of >60% between inflection points). Table shows timing and duration of all such events for 2000–2011.

waters, also uses a near infrared band along with visible bands. Of the 4.2. Sediment dynamics in the western end of Lake Athabasca
remaining 24 models, 16 use only a single band, 7 use a combination
of bands but do not include a near infrared band, and 1 uses a combi- To understand the dynamics of water input to Lake Athabasca by the
nation of bands not including a visible band. Finally, out of the nine Peace and Athabasca Rivers, we created daily sediment maps of the
(six independent) linear equations tested, only one is in the top ten westernmost end of the lake (Fig. 8). From the maps, we can clearly
(Fraser, 1998), suggesting that non-linear models are preferable for detect both the spatial extent and the source of sediment-laden river
modeling SSCs in the PAD. water input to the lake. We can also distinguish the source of river
Our results indicate that spatial transferability of a model is most water delivered to the lake using virtual sediment gauges near the lake's
likely to be successful if the model meets all three proposed criteria. two sources of inflow (i.e. the Peace and Athabasca Rivers) (Fig. 4).
Models that meet only one of these criteria (e.g. they predict high Comparison of SSC time series from these two locations (Fig. 9) reveals
SSCs using one band alone, or have a nonlinear form but were devel- temporally separate peaks in SSC indicating the source of river water
oped to model only low SSCs) produce weaker correlations. Strong during major flux events. The Athabasca River always flows into Lake
performance of a model in predicting SSCs in the PAD does not indicate Athabasca, but we also identify two instances, in May 2007 and July
universal transferability, but the characteristics of high-performing 2011, where peaks in SSC at the northern shore of the lake indicate in-
models identified here can be used to guide model selection in environ- flow of Peace River water.
ments where in situ measurements are unavailable to calibrate site- There is an observable relationship between the timing of signifi-
specific relationships. cant peaks in SSC in Lake Athabasca and peaks in Athabasca River

a b c
30.2
8000

500
Predicted SSC

Predicted SSC
Predicted SSC
29.8

300
4000

29.4

100

Doxaran et al., 2009 Doxaran et al., 2003 Topliss et al., 1990

0

0 2000 4000 0 2000 4000 0 2000 4000

Measured SSC Measured SSC Measured SSC
1000

d e f
5
Predicted turbidity

Predicted turbidity
Predicted turbidity

0
200
600

-5
200

-10

Holyer et al., 1978 Fraser, 1998 Song et al., 2011

0

0 2000 4000 0 2000 4000 0 2000 4000

Measured turbidity Measured turbidity Measured turbidity

Fig. 7. Subset of the 31 SSC-reflectance models used to model SSC data in the PAD. Plots show the relationship between measured values for SSC (mg/L) (a–c) or turbidity (NTU) (d–f) on
the x-axis and predicted values on the y-axis. See Table 1 for complete equations and Spearman's ρ values.
204 C.M. Long, T.M. Pavelsky / Remote Sensing of Environment 129 (2013) 197–209

discharge (Fig. 9). Parallels between fluctuations are difficult to detect

for small variations in SSC and at low discharge, but the largest peaks in
discharge do cause peaks in SSC, most notably in 2001, 2007, and 2011.
To better understand the relationship between these two variables, we
examine data from the rising limb of the first summertime hydrograph
peak, the same metric used to identify inflow thresholds for small lakes
(Fig. 6). As with small lakes, it is during these events that SSC in Lake
Athabasca is most likely to reflect river discharge instead of wind,
biological activity, or other confounding variables. Results suggest an
Athabasca River discharge threshold of ~1700 m 3/s at which Lake
Athabasca SSCs near the Athabasca Delta begin to increase dramatically
from ~200 mg/L to a maximum of 1085 mg/L at an Athabasca River
discharge of ~3200 m 3/s (Fig. 10). Our compilation of 12 years of
near-daily data shows that at discharges less than ~1700 m 3/s, SSCs
remain below ~100 mg/L and are uncorrelated with Athabasca River
discharge.
In order to determine the sensitivity of this result to the SSC virtual
gauge location, we also measured SSCs at locations closer to and further
from the margin of the Athabasca Delta. The relative values of SSC var-
ied predictably (i.e. higher closer to shore and lower out in the lake), but
Fig. 8. MODIS-derived sediment maps of the westernmost end of Lake Athabasca on (a) a
at all locations the same discharge threshold of ~1700 m 3/s was appar-
day when only the Athabasca River is flowing into the lake and (b) a day when the Peace
River is delivering high sediment water to the lake.
ent. Such a threshold is not observed between SSCs in the Athabasca
River and river discharge, however. In this case, the variables are
1500

2000 2001 2002

3000
6000
500

1000
2000
May 1 July 1 Sept 1 May 1 July 1 Sept 1 May 1 July 1 Sept 1
1500

2003 2004 2005

Suspended sediment concentration (mg/L)

3000
6000
500

1000
May 1 July 1 Sept 1 May 1 July 1 Sept 1 May 1 July 1 Sept 1 2000
1500

2006 2007 2008

3000
6000
500

1000
2000

May 1 July 1 Sept 1 May 1 July 1 Sept 1 May 1 July 1 Sept 1

1500

2009 2010 2011

3000
6000
500

1000
2000

May 1 July 1 Sept 1 May 1 July 1 Sept 1 May 1 July 1 Sept 1

SSC near the input SSC near the input Athabasca R. Peace R.
of Athabasca R. water of Peace R. water discharge discharge

Fig. 9. Annual time series of SSC in the western end of Lake Athabasca and discharge on the Peace and Athabasca Rivers.
 C.M. Long, T.M. Pavelsky / Remote Sensing of Environment 129 (2013) 197–209 205

1200
Lake Athabasca lakes occurs only when river stage is sufficiently high. To identify dis-
Athabasca River charge thresholds which consistently initiate recharge in individual
lakes, we compare Athabasca river discharge during the time periods
identified in Fig. 6 with the Sediment Area Index of six floodplain
lakes consistently visible in MODIS imagery (Fig. 11). Positive relation-
Suspended sediment
concentration (mg/L)
800

ships between Athabasca River discharge and SAI are apparent in Lakes
1, 2, and 3. For each of these lakes, there is a distinct discharge threshold
above which there is always high sediment water present and below
which there is little to no high sediment water. We know from field ob-
servations that these three lakes are hydrologically connected to the
400

Athabasca River through small distributary channels which sometimes,

but not always, flow into the lakes. Often, water flows very slowly out of
the lakes and back to the river. The threshold values likely represent the
discharges at which flow reverses on the channels and Athabasca River
water begins to enter the lakes.
0

High sediment water was detected in Lake 4 on just two days,

500 1500 2500 3500 neither of which corresponds to high discharge. This lake is likely
River discharge (m3/s) disconnected from the river system by natural levees that are not
overtopped even during high flow events. In Lakes 5 and 6, there is
Fig. 10. Athabasca River discharge vs. MODIS-derived SSC in Lake Athabasca near the some suggestion of a threshold discharge value above which SAI is
margin of the Athabasca Delta (red points). Athabasca River discharge vs. in situ mea-
always greater than 0, but the relationships for these lakes are
surements of SSC in the river (black triangles).
complicated by the presence of many instances of high SAI and low
river discharge. We know from field observations that Lakes 5 and
6, like Lakes 1, 2, and 3, are hydrologically connected to the Athabasca
linearly correlated (r= 0.94) even at low discharges (Fig. 10). Finally, River through distributary channels. The apparent discharge thresh-
we examined the equivalent dataset for the Peace River and find no olds on Lakes 5 and 6 may represent inflow of river water. In these
correlation between river discharge and SSC at the Peace River virtual lakes, however, other factors that are not as influential in Lakes 1, 2,
gauge. This is likely because input of Peace River water depends on and 3, such as resuspension of lake-bottom sediments by wind, likely
the relative levels of the river and Lake Athabasca and not only on increase SAI without influence from the river.
river discharge. The river discharge thresholds for inflow of water to the three high
river-influence lakes are 1040 m 3/s (Lake 1), 1150 m3/s (Lake 2), and
4.3. Tracking recharge of small floodplain lakes 1850 m3/s (Lake 3). Previous studies have explored hydrologic connec-
tivity in the PAD and have yielded qualitative measures of floodplain
Athabasca River water is input to Lake Athabasca under all flow con- lake connectivity (e.g. Pavelsky & Smith, 2008; Prowse & Demuth,
ditions, but summertime recharge of ecologically important floodplain 1996; Wolfe et al., 2007), but no other study has previously identified
1.0

1.0
1.0

1 2 3
0.8

0.8
0.8
0.6

0.6
0.6
Proportion of Lake with High Sediment Water
0.4

0.4
0.4
0.2

0.2
0.2

Threshold: Threshold: Threshold:

1040 m3/s 1150 m3/s 1850 m3/s
0.0

0.0
0.0

500 1500 2500 3500 500 1500 2500 3500 500 1500 2500 3500
Athasbasca River Discharge (m3/s)
1.0

1.0

1.0

4 5 6
0.8

0.8

0.8
0.6

0.6

0.6
0.4

0.4

0.4
0.2

0.2

0.2
0.0

0.0

0.0

500 1500 2500 3500 500 1500 2500 3500 500 1500 2500 3500
Athasbasca River Discharge (m3/s)

Fig. 11. Plots of Athabasca River discharge vs. Sediment Area Index, or proportion of lake water that is classified as “high sediment,” for six small lakes in the Athabasca Delta. Lake
numbers correspond to labels in Fig. 3. River discharge thresholds required to recharge Lakes 1, 2, and 3 with river water are labeled.
206 C.M. Long, T.M. Pavelsky / Remote Sensing of Environment 129 (2013) 197–209

discharge thresholds required for recharge. We use these thresholds to 5. Discussion and conclusions
identify the number of days in each year since 1970 on which Athabasca
River discharge is sufficiently high to recharge each lake (Fig. 12). Be- The first principal conclusion of this study is that models for re-
cause lake recharge often depends on relative water surface elevations motely sensing SSC developed in one location can, in some cases, be
in the river and lake, recharge likely did not occur on all days above transferred to another location to qualitatively predict SSCs without
this threshold, but a higher number of days above the discharge thresh- the need for in situ data. The first criterion we identify for making
old likely implies greater overall recharge. Since 1970, there has been a such a transfer (i.e., that a near infrared band and a visible band
statistically significant (pb 0.05) decrease in recharge frequency for used in combination are more effective than a single band used
each of the three lakes. During the first half of the study period alone or another sort of combination) is supported by previous
(1970–1990), Lakes 1 and 2 were recharged in all years, and Lake 3 work also suggesting that multispectral models are preferable for
was recharged in every year except 1981 (Table 2). In contrast, during remotely sensing SSC (e.g. Holyer, 1978; Schiebe et al., 1992; Topliss
the second half of the study period (1991–2011) discharge on the et al., 1990). Some previous studies (e.g. Doxaran et al., 2002;
Athabasca River failed to reach the threshold required to recharge Holyer, 1978; Novo et al., 1989) have more specifically suggested
Lake 2 three times, and the threshold for Lake 3 was met in just ten of that the best models use a combination of a near infrared and a red
the twenty years. band. In our study, the model that produces the best linear correlation
with our field dataset uses this combination, but five of the top seven
models use a near infrared band paired with a green or blue band.
This suggests that bands in the green and blue part of the spectrum
may be as effective as red bands for remotely sensing SSC as long as
they are paired with a near infrared band.
Our results suggest that models developed using comparatively low
SSCs have limited success predicting the higher SSCs in the PAD, while
models based on higher SSCs are more effective. This finding is corrob-
orated by past studies, which have found that different models are op-
timal for low and high SSCs (e.g. Ritchie et al., 2003) or that models
which work well at low concentrations can saturate at higher concen-
trations (e.g. Chu et al., 2009; Holyer 1978; Ritchie and Cooper, 1988;
Ritchie et al., 2003; Topliss et al., 1990). Some of the models we test
are also predictive for low SSCs but then become saturated at higher
concentrations (e.g. Doxaran et al., 2003; Song et al., 2011; Topliss et
al., 1990) (Fig. 7 d,e). Holyer (1978) found that saturation occurs
when using reflectance in the red band, but using a near infrared band
along with a red band corrects this problem. In contrast, our analysis
shows no discernible pattern in which spectral bands were used
among models that saturate. We suggest instead that the gap between
the maximum SSC from which these models were developed and the
much higher SSCs we observe in the PAD may result in saturation at
high SSCs.
Finally, we suggest that nonlinear models may be more successful
for predicting SSCs in a new location than other forms. Past studies
suggest that linear relationships are effective for remotely sensing
SSCs less than 50 mg/L, but for values greater than this, curvilinear
relationships are necessary (Ritchie et al., 2003). In the PAD, we
found that exponential relationships worked the best. This is likely
because of the large range of SSCs in the PAD, and for different
environments other nonlinear forms may also work well. In environ-
ments where SSCs are relatively low, the form of the relationship may
not be as significant when selecting a model; Ritchie and Zimba
(2006) noted that for SSCs between 0 and 50 mg/L, reflectance from
almost any visible or near-infrared wavelength is linearly related to
SSC. Nevertheless, in the PAD, where SSCs range from less than 5 to
more than 3000 mg/L, equation form is an important factor influenc-
ing success of models in predicting SSCs and it should be considered
when evaluating models for spatial transferability.
If the three primary conditions identified here are not met, then
application of an SSC-reflectance relationship beyond its area of de-
velopment may produce unreliable results. Even when these

Table 2
Number of years in which threshold discharge required for lake recharge was not
reached on the Athabasca River.

Lake 1 Lake 2 Lake 3

Fig. 12. Number of days (y-axis) when Athabasca River discharge is above a threshold in
each year (x-axis) from 1970 to 2011. Thresholds are (a) 1040 m3/s, (b) 1150 m3/s, and 1040 m3/s 1150 m3/s 1850 m3/s
(c) 1850 m3/s and are for Lakes 1, 2, and 3, respectively. Decreases in the frequency of
1970–1990 0 0 1
threshold exceedence are statistically significant (pb 0.05) in all cases. Red points indicate
1991–2011 0 3 10
years with 0 days above threshold.
 C.M. Long, T.M. Pavelsky / Remote Sensing of Environment 129 (2013) 197–209 207

requirements are met, it remains necessary to develop a constant cor- river discharge signal. Bottom-reflectance may also impact remote
rection coefficient, obtained from a linear regression between observed measurement of SSCs in these shallow lakes. Our results are in accor-
and modeled SSCs, to account for location-specific differences in factors dance with the findings of Pavelsky and Smith (2009a), who found
like sediment color and grain size. If limited availability of in situ data that suspended sediment in Lake 5 did not closely mirror discharge
prevents development of such a correction, it remains possible to accu- in the summer months, as well as with Prowse and Demuth (1996),
rately observe relative differences in SSC. For much of our work focusing whose classification shows Lake 5 as an “isolated” basin that is signif-
on spatial and temporal patterns of SSC, relative SSC measurements icantly recharged only by overbank flooding. Pavelsky and Smith
would be fully adequate. (2008) noted that Lake 5 was hydrologically connected to the river
The scaled equation from Doxaran et al. (2009) which we use to system in 2007; our observation of an apparent threshold is in accor-
map SSC in Lake Athabasca is substantially more sophisticated than dance with this finding, since in 2007 Athabasca River discharge
prior models used in the PAD (Pavelsky & Smith, 2009a), and it fully exceeded the apparent threshold required to recharge Lake 5.
addresses anomalously high predicted SSCs from that work inferred The identification of threshold discharges required to recharge lakes
to have been associated with biological activity. Our results allow us that are strongly influenced by the Athabasca River is valuable for un-
to distinguish between sediment-laden river water input from the derstanding the effects on the PAD of declines in flow on the Athabasca
Peace and Athabasca Rivers for major flux events. Input of Peace River. In the past four decades, we have observed a substantial decline
River water to Lake Athabasca is not directly controlled by discharge in the number of days on which Athabasca River discharge is sufficiently
from the river, depending instead on the relative water levels of the high to recharge the floodplain lakes studied here. In particular, re-
river and the lake (PAD-PG, 1973; Peters et al., 2006). If the lake charge from the Athabasca River to Lake 3 has changed from occurring
level is lower than the river level, the Peace River will flow into the annually to occurring irregularly. Our results also suggest that if sum-
lake regardless of river discharge. Though it is not possible to predict mer discharge on the Athabasca River continues to decline at the cur-
the input of Peace River water to Lake Athabasca using river discharge rent rate, many small floodplain lakes (e.g. Lakes 1–3 in our analysis)
alone, remote monitoring of SSC allows us to track occurrences of will no longer be recharged with Athabasca River water except during
inflow in the absence of in situ monitoring. the spring ice-jam flood period by approximately the 2040s, and possi-
In contrast to the Peace River, the Athabasca River always flows into ble as early as the 2020s in the case of Lake 3. Transformation of the
Lake Athabasca, and we expect to observe a relationship between Athabasca portion of the PAD from a frequently to an infrequently
discharge and input of river water to the lake. However, discharge and flooded environment has the potential to substantially affect delta ecol-
SSC in the lake are not significantly correlated for discharge less ogy and biological productivity (McGowan et al., 2011; Prowse & Conly,
than ~1700 m 3/s and SSC less than ~100 mg/L, which suggests that 2002; Wiklund et al., 2011). A reduction in the frequency of recharge
river discharge is not the only important control on SSC in Lake would allow willow (Salix sp.) and shrub communities to overtake
Athabasca. Wind almost certainly also influences SSC at the water sur- more productive grass- and sedge-dominated environments that cur-
face through resuspension of bottom sediment and mixing of lake and rently serve as habitats for migratory birds (Timoney, 2006; Töyrä &
river water, and at low discharges it may overwhelm the river signal. Pietroniro, 2005). Conversely, deltaic lakes that manage to persist
It is also possible that when discharge is less than 1700 m 3/s, sediment under a reduced flooding regime may see an increase in productivity
settles out of suspension within the river itself before reaching the lake. due to changes in water clarity and nutrient availability (McGowan et
At high discharge values, however, input of high sediment water al., 2011; Wiklund et al., 2011). Those lakes identified as highly river-
appears to overcome these confounding factors and SSC becomes influenced (i.e. Lakes 1, 2, and 3) are likely to face the most changes in
more strongly related to discharge. their ecology and productivity, and all of these effects could be ampli-
Even at the lowest discharges, Athabasca River water flows into Lake fied by continued climate change or increased water withdrawals on
Athabasca, but this is not the case in the small floodplain lakes in the the Athabasca River for use in the Alberta Oil Sands development
PAD. We assess the connectivity of six of these floodplain lakes to gain (Schindler & Donahue, 2006).
a more thorough understanding of recharge in the delta. Prowse and
Demuth (1996) examined hydrologic connectivity in the PAD and di- Acknowledgments
vided the delta into regions of “open,” “restricted,” and “isolated” drain-
age based on the work of Jaques (1989), and PAD-PG (1973). Based on This work was funded by student research grants from the UNC
these classifications, our high river-influence and strongly connected Department of Geological Sciences Martin Fund and the Geological
lakes (Lakes 1–3) as well as two of our low river-influence lakes (4 Society of America (Grant #9536-11). We are grateful to Robert
and 6) are in “restricted” zones. Lake 5, which we found to be more Grandjambe, Larry Benninger, Zach Miller, and the staff of Wood
connected to the river than Lake 4, is in an “isolated” zone. Wolfe et Buffalo National Park for assistance in the field. We also thank to
al. (2007) use a similar classification as Prowse and Demuth (1996) two anonymous reviewers for their useful comments.
adding also a fourth category for very shallow, rainfall-influenced
lakes. They base their classifications on δ18O-derived evaporation-
References
to-inflow ratios and examine three of our six study lakes: Lakes 2 and
3 are defined as “open,” and Lake 5 is defined as “restricted.” Our results Ali, G. A., & Roy, A. G. (2010). Shopping for hydrologically representative connectivity met-
agree more closely with the classifications of Wolfe et al. (2007) than rics in a humid temperate forested catchment. Water Resources Research, 46(12),
W12544. http://dx.doi.org/10.1029/2010wr009442.
with those of Prowse and Demuth (1996). Compared to these previous Andrishak, R., & Hicks, F. (2011). Ice effects on flow distributions within the Athabasca Delta,
methods for observing hydrologic connectivity in the PAD, the principal Canada. River Research and Applications, 27(9), 1149–1158. http://dx.doi.org/10.1002/
advantage of the method used here is that it allows us to identify spe- rra.1414.
Aranuvachapun, S., & Walling, D. E. (1988). Landsat-MSS radiance as a measure of
cific occurrences of lake recharge and thus determine discharge
suspended sediment in the Lower Yellow River (Hwang Ho). Remote Sensing of
thresholds on the Athabasca River associated with lake recharge. Environment, 25(2), 145–165. http://dx.doi.org/10.1016/0034-4257(88)90098-3.
The three lakes most proximal to Lake Athabasca are the low river Beltaos, S., Prowse, T., Bonsal, B., MacKay, R., Romolo, L., Pietroniro, A., et al. (2006).
Climatic effects on ice-jam flooding of the Peace–Athabasca Delta. Hydrological
influence lakes, and it is likely that the lake level in Lake Athabasca
Processes, 20(19), 4031–4050. http://dx.doi.org/10.1002/hyp.6418.
strongly influences the amount of water and sediment they receive, Beltaos, S., Prowse, T. D., & Carter, T. (2006). Ice regime of the lower Peace River
especially when lake levels are high (Pavelsky & Smith, 2008; Peters and ice-jam flooding of the Peace–Athabasca Delta. Hydrological Processes,
et al., 2006). The low river influence lakes are also significantly 20(19), 4009–4029. http://dx.doi.org/10.1002/hyp.6417.
Bloesch, J. (1995). Mechanisms, measurement and importance of sediment resuspension
shallower than the high river influence (Smith & Pavelsky, 2009) in lakes. Marine and Freshwater Research, 46(1), 295–304. http://dx.doi.org/10.1071/
and sediment re-suspension due to wind likely interferes with the MF9950295.
208 C.M. Long, T.M. Pavelsky / Remote Sensing of Environment 129 (2013) 197–209

Bowers, D. G., & Binding, C. E. (2006). The optical properties of mineral suspended waters. International Journal of Remote Sensing, 12(10), 2045–2063. http://dx.doi.org/
particles: A review and synthesis. Estuarine, Coastal and Shelf Science, 67(1–2), 10.1080/01431169108955235.
219–230. http://dx.doi.org/10.1016/j.ecss.2005.11.010. Leconte, R., Pietroniro, A., Peters, D. L., & Prowse, T. D. (2001). Effects of flow regulation
Bowers, D. G., Binding, C. E., & Ellis, K. M. (2007). Satellite remote sensing of the on hydrologic patterns of a large, inland delta. Regulated Rivers: Research &
geographical distribution of suspended particle size in an energetic shelf sea. Management, 17(1), 51–65. http://dx.doi.org/10.1002/1099-1646(200101/02)
Estuarine, Coastal and Shelf Science, 73(3–4), 457–466. http://dx.doi.org/10.1016/ 17:1b51::aid-rrr588>3.0.co;2-v.
j.ecss.2007.02.005. Lesack, L. F. W., Marsh, P., & Hecky, R. E. (1998). Spatial and temporal dynamics of major
Bracken, L. J., & Croke, J. (2007). The concept of hydrological connectivity and its con- solute chemistry among Mackenzie Delta lakes. Limnology and Oceanography, 43(7),
tribution to understanding runoff-dominated geomorphic systems. Hydrological 1530–1543.
Processes, 21(13), 1749–1763. http://dx.doi.org/10.1002/hyp.6313. McGowan, S., Leavitt, P. R., Hall, R. I., Wolfe, B. B., Edwards, T. W. D., Karst-Riddoch, T.,
Chen, Z., Hu, C., & Muller-Karger, F. (2007). Monitoring turbidity in Tampa Bay using et al. (2011). Interdecadal declines in flood frequency increase primary produc-
MODIS/Aqua 250-m imagery. Remote Sensing of Environment, 109(2), 207–220. tion in lakes of a northern river delta. Global Change Biology, 17(2), 1212–1224.
http://dx.doi.org/10.1016/j.rse.2006.12.019. http://dx.doi.org/10.1111/j.1365-2486.2010.02304.x.
Chen, S., Huang, W., Wang, H., & Li, D. (2009). Remote sensing assessment of sediment Meade, R. H. (1996). River-sediment inputs to major deltas. In J. D. Milliman, & B. U.
re-suspension during Hurricane Frances in Apalachicola Bay, USA. Remote Sensing Haq (Eds.), Sea-level rise and coastal subsidence: Causes, consequences, and strategies
of Environment, 113(12), 2670–2681. http://dx.doi.org/10.1016/j.rse.2009.08.005. (pp. 63–85). The Netherlands: U.S. Government.
Chu, V. W., Smith, L. C., Rennermalm, A. K., Forster, R. R., Box, J. E., & Reehy, N. (2009). Sed- Mertes, L. A. K., Smith, M. O., & Adams, J. B. (1993). Estimating suspended sediment
iment plume response to surface melting and supraglacial lake drainages on the concentrations in surface waters of the Amazon River wetlands from Landsat im-
Greenland ice sheet. Journal of Glaciology, 55(194), 1072–1082. http://dx.doi.org/ ages. Remote Sensing of Environment, 43(3), 281–301. http://dx.doi.org/10.1016/
10.3189/002214309790794904. 0034-4257(93)90071-5.
Curran, P. J., & Novo, E. M. M. (1988). The relationship between suspended sediment Miller, R. L., & McKee, B. A. (2004). Using MODIS Terra 250 m imagery to map concentra-
concentration and remotely sensed spectral radiance: A review. Journal of Coastal tions of total suspended matter in coastal waters. Remote Sensing of Environment,
Research, 4(3), 351–368. 93(1–2), 259–266. http://dx.doi.org/10.1016/j.rse.2004.07.012.
Dekker, A. G., Vos, R. J., & Peters, S. W. M. (2001). Comparison of remote sensing data, Mouchot, M. -C., Alfoldi, T., DeLisle, D., & McCullough, G. (1991). Monitoring the water
model results and in situ data for total suspended matter (TSM) in the southern bodies of the Mackenzie Delta by remote sensing methods. Arctic, 44(1), 21–28.
Frisian lakes. Science of the Total Environment, 268(1–3), 197–214. http://dx.doi.org/ Novo, E. M. M., Hansom, J. D., & Curran, P. J. (1989). The effect of sediment type on the re-
10.1016/s0048-9697(00)00679-3. lationship between reflectance and suspended sediment concentration. International
Doxaran, D., Froidefond, J. -M., & Castaing, P. (2003). Remote-sensing reflectance of Journal of Remote Sensing, 10(7), 1283–1289. http://dx.doi.org/10.1080/
turbid sediment-dominated waters. Reduction of sediment type variations and 01431168908903967.
changing illumination conditions effects by Use of reflectance ratios. Applied Optics, Owens, P. N., Batalla, R. J., Collins, A. J., Gomez, B., Hicks, D. M., Horowitz, A. J., et al. (2005).
42(15), 2623–2634. Fine-grained sediment in river systems: Environmental significance and manage-
Doxaran, D., Froidefond, J. -M., Castaing, P., & Babin, M. (2009). Dynamics of the turbid- ment issues. River Research and Applications, 21(7), 693–717. http://dx.doi.org/
ity maximum zone in a macrotidal estuary (the Gironde, France): Observations 10.1002/rra.878.
from field and MODIS satellite data. Estuarine, Coastal and Shelf Science, 81(3), Pavelsky, T. M., & Smith, L. C. (2008). Remote sensing of hydrologic recharge in the
321–332. http://dx.doi.org/10.1016/j.ecss.2008.11.013. Peace–Athabasca Delta, Canada. Geophysical Research Letters, 35(8), L08403.
Doxaran, D., Froidefond, J. -M., Lavender, S., & Castaing, P. (2002). Spectral signature http://dx.doi.org/10.1029/2008gl033268.
of highly turbid waters: Application with SPOT data to quantify suspended par- Pavelsky, T. M., & Smith, L. C. (2009a). Remote sensing of suspended sediment concentra-
ticulate matter concentrations. Remote Sensing of Environment, 81(1), 149–161. tion, flow velocity, and lake recharge in the Peace–Athabasca Delta, Canada. Water
http://dx.doi.org/10.1016/s0034-4257(01)00341-8. Resources Research, 45(11), W11417. http://dx.doi.org/10.1029/2008wr007424.
D'Sa, E. J., Miller, R. L., & McKee, B. A. (2007). Suspended particulate matter dynamics in Pavelsky, T. M., & Smith, L. C. (2009b). Surface water elevation and quality, Peace–
coastal waters from ocean color: Application to the northern Gulf of Mexico. Geophysical Athabasca Delta, Canada, 2006–2007. Data set (Available on-line [http://daac.
Research Letters, 34(23), L23611. http://dx.doi.org/10.1029/2007gl031192. ornl.gov] from Oak Ridge National Laboratory Distributed Active Archive Cen-
Duane Nellis, M., Harrington, J. A., Jr., & Wu, J. (1998). Remote sensing of temporal and spa- ter, Oak Ridge, Tennessee, U.S.A. http://dx.doi.org/10.3334/ORNLDAAC/935).
tial variations in pool size, suspended sediment, turbidity, and Secchi depth in Tuttle Peace-Athabasca Delta Project Group (1973). Technical report: A report on low water levels
Creek Reservoir, Kansas: 1993. Geomorphology, 21(3–4), 281–293. http://dx.doi.org/ in Lake Athabasca and their effect on the Peace–Athabasca Delta (pp. 176).
10.1016/s0169-555x(97)00067-6. Peace-Athabasca Delta Project Group, Ottawa, Canada.
Farley, D. W., & Cheng, H. (1986). Hydraulic impact of flow regulation on the Peace– Pereira, W. E., Domagalski, J. L., Hostettler, F. D., Brown, L. R., & Rapp, J. B. (1996). Occurrence
Athabasca delta. Canadian Water Resources Journal, 11(1), 26–42. http://dx.doi.org/ and accumulation of pesticides and organic contaminants in river sediment, water and
10.4296/cwrj1101026. clam tissues from the San Joaquin River and tributaries, California. Environmental Toxicol-
Fennessy, M. S., Jacobs, A. D., & Kentula, M. E. (2004). Review of rapid methods for ogy and Chemistry, 15(2), 172–180. http://dx.doi.org/10.1002/etc.5620150216.
assessing wetland condition. U.S. Environmental Protection Agency, Washington, Peters, D. L., & Prowse, T. D. (2001). Regulation effects on the lower Peace River, Canada.
D.C, USA. EPA/620/R-04/009. Hydrological Processes, 15(16), 3181–3194. http://dx.doi.org/10.1002/hyp.321.
Fraser, R. N. (1998). Multispectral remote sensing of turbidity among Nebraska Sand Hills Peters, D. L., Prowse, T. D., Pietroniro, A., & Leconte, R. (2006). Flood hydrology
lakes. International Journal of Remote Sensing, 19(15), 3011–3016. http://dx.doi.org/ of the Peace–Athabasca Delta, northern Canada. Hydrological Processes, 20(19),
10.1080/014311698214406. 4073–4096. http://dx.doi.org/10.1002/hyp.6420.
Han, Z., Jin, Y. Q., & Yun, C. X. (2006). Suspended sediment concentrations in the Yangtze Pringle, C. (2003). What is hydrologic connectivity and why is it ecologically important?
River estuary retrieved from the CMODIS data. International Journal of Remote Sensing, Hydrological Processes, 17(13), 2685–2689. http://dx.doi.org/10.1002/hyp.5145.
27(19), 4329–4336. http://dx.doi.org/10.1080/01431160600658164. Prowse, T. D., & Conly, F. M. (2002). A review of hydroecological results of the Northern
Hellweger, F., Miller, W., & Oshodi, K. (2007). Mapping turbidity in the Charles River, River Basins Study, Canada. Part 2. Peace–Athabasca Delta. River Research and
Boston using a high-resolution satellite. Environmental Monitoring and Assessment, Applications, 18(5), 447–460. http://dx.doi.org/10.1002/rra.682.
132(1), 311–320. http://dx.doi.org/10.1007/s10661-006-9535-8. Prowse, T. D., & Demuth, M. N. (1996). Using ice to flood the Peace–Athabasca delta, Canada.
Hernández-Ayón, J. M., Galindo-Bect, M. S., Flores-Báez, B. P., & Alvarez-Borrego, S. Regulated Rivers: Research & Management, 12(4–5), 447–457. http://dx.doi.org/10.1002/
(1993). Nutrient concentrations are high in the turbid waters of the Colorado (sici)1099-1646(199607)12:4/5b447::aid-rrr404>3.0.co;2-y.
River Delta. Estuarine, Coastal and Shelf Science, 37(6), 593–602. http://dx.doi.org/ Ritchie, J. C., Zimba, P. V., & Everitt, J. H. (2003). Remote sensing techniques to assess
10.1006/ecss.1993.1075. water quality. Photogrammetric Engineering and Remote Sensing, 69(6), 695.
Holyer, R. J. (1978). Toward universal multispectral suspended sediment algorithms. Ritchie, J. M., & Cooper, C. M. (1988). Comparison of measured suspended sediment
Remote Sensing of Environment, 7(4), 323–338. http://dx.doi.org/10.1016/0034- concentrations with suspended sediment concentrations estimated from LANDSAT
4257(78)90023-8. MSS data. International Journal of Remote Sensing, 9(3), 379–387. http://dx.doi.org/
Islam, M. R., Yamaguchi, Y., & Ogawa, K. (2001). Suspended sediment in the Ganges and 10.1080/01431168808954861.
Brahmaputra Rivers in Bangladesh: Observation from TM and AVHRR data. Hydro- Ritchie, J. C., & Cooper, C. M. (1991). An algorithm for estimating surface suspended
logical Processes, 15(3), 493–509. http://dx.doi.org/10.1002/hyp.165. sediment concentrations with Landsat MSS digital data1. JAWRA Journal of the
Jaques, D. (1989). Topographic mapping and drying trends in the Peace–Athabasca Delta, American Water Resources Association, 27(3), 373–379. http://dx.doi.org/10.1111/
Alberta. Using LANDSAT MSS imagery Ecosat Geobotanical Surveys Inc. for Parks Canada, j.1752-1688.1991.tb01436.x.
Wood Buffalo National Park, Fort Smith, Northwest Territories, Canada (pp. 33). Ritchie, J. C., Schiebe, F. R., & McHenry, R. (1976). Remote sensing of suspended sediment
Jencso, K. G., McGlynn, B. L., Gooseff, M. N., Wondzell, S. M., Bencala, K. E., & Marshall, L. in surface waters. Photogrammetric Engineering and Remote Sensing, 69(6), 695–7014.
A. (2009). Hydrologic connectivity between landscapes and streams: Transferring Ritchie, J. C., & Zimba, P. V. (2006). Estimation of suspended sediment and algae in water
reach‐ and plot‐scale understanding to the catchment scale. Water Resources Re- bodies. Encyclopedia of hydrological sciences. : John Wiley & Sons, Ltd..
search, 45(4), W04428. http://dx.doi.org/10.1029/2008wr007225. Schiebe, F. R., Harrington, J. A., & Ritchie, J. C. (1992). Remote sensing of suspended
Keiner, L. E., & Yan, X. -H. (1998). A neural network model for estimating Sea surface sediments: The Lake Chicot, Arkansas project. International Journal of Remote
chlorophyll and sediments from thematic mapper imagery. Remote Sensing of Sensing, 13(8), 1487–1509. http://dx.doi.org/10.1080/01431169208904204.
Environment, 66(2), 153–165. http://dx.doi.org/10.1016/s0034-4257(98)00054-6. Schindler, D. W. (2001). The cumulative effects of climate warming and other human
Kummu, M., & Varis, O. (2007). Sediment-related impacts due to upstream reservoir stresses on Canadian freshwaters in the new millennium. Canadian Journal of Fisher-
trapping, the Lower Mekong River. Geomorphology, 85(3–4), 275–293. http: ies and Aquatic Sciences, 58(1), 18–29.
//dx.doi.org/10.1016/j.geomorph.2006.03.024. Schindler, D. W., & Donahue, W. F. (2006). An impending water crisis in Canada's western
Lathrop, R. G., Lillesand, T. M., & Yandell, B. S. (1991). Testing the utility of simple prairie provinces. Proceedings of the National Academy of Sciences, 103(19), 7210–7216.
multi-date Thematic Mapper calibration algorithms for monitoring turbid inland http://dx.doi.org/10.1073/pnas.0601568103.
 C.M. Long, T.M. Pavelsky / Remote Sensing of Environment 129 (2013) 197–209 209

Schindler, D. W., & Smol, J. P. (2006). Cumulative effects of climate warming and other Töyrä, J., & Pietroniro, A. (2005). Towards operational monitoring of a northern wetland
human activities on freshwaters of arctic and subarctic North America. AMBIO: using geomatics-based techniques. Remote Sensing of Environment, 97(2), 174–191.
A Journal of the Human Environment, 35(4), 160–168. http://dx.doi.org/10.1579/ http://dx.doi.org/10.1016/j.rse.2005.03.012.
0044-7447(2006) 35[160:ceocwa]2.0.co;2. Vörösmarty, C. J., Meybeck, M., Fekete, B., Sharma, K., Green, P., & Syvitski, J. P. M. (2003).
Smith, L. C., & Alsdorf, D. E. (1998). Control on sediment and organic carbon deliv- Anthropogenic sediment retention: major global impact from registered river
ery to the Arctic Ocean revealed with space-borne synthetic aperture radar: impoundments. Global and Planetary Change, 39(1–2), 169–190. http://dx.doi.org/
Ob' River, Siberia. Geology, 26(5), 395–398. http://dx.doi.org/10.1130/0091- 10.1016/s0921-8181(03)00023-7.
7613(1998) 026b0395:cosaoc>2.3.co;2. Wang, F., Zhou, B., Xu, J., Song, L., & Wang, X. (2008). Application of neural network
Smith, L. C., & Pavelsky, T. M. (2009). Remote sensing of volumetric storage changes in and MODIS 250 m imagery for estimating suspended sediments concentration in
lakes. Earth Surface Processes and Landforms, 34(10), 1353–1358. http://dx.doi.org/ Hangzhou Bay, China. Environmental Geology, 56(6), 1093–1101. http://dx.doi.org/
10.1002/esp.1822. 10.1007/s00254-008-1209-0.
Song, K., Wang, Z., Blackwell, J., Zhang, B., Li, F., Zhang, Y., et al. (2011). Water quality Wass, P. D., Marks, S. D., Finch, J. W., Leeks, G. J. L., & Ingram, J. K. (1997). Monitoring
monitoring using Landsat Themate Mapper data with empirical algorithms in and preliminary interpretation of in-river turbidity and remote sensed imagery
Chagan Lake, China. Journal of Applied Remote Sensing, 5, 053506–053516. for suspended sediment transport studies in the Humber catchment. Science
Spearman, C. (1904). The proof and measurement of association between Two things. of the Total Environment, 194–195, 263–283. http://dx.doi.org/10.1016/s0048-
The American Journal of Psychology, 15(1), 72–101. 9697(96)05370-3.
Sterckx, S., Knaeps, E., Bollen, M., Trouw, K., & Houthuys, R. (2007). Retrieval of Wiklund, J. A., Hall, R. I., & Wolfe, B. B. (2011). Timescales of hydrolimnological change
suspended sediment from advanced hyperspectral sensor data in the Scheldt in floodplain lakes of the Peace–Athabasca Delta, northern Alberta, Canada.
estuary at different stages in the tidal cycle. Marine Geodesy, 30(1–2), 97–108. Ecohydrology. http://dx.doi.org/10.1002/eco.226 (n/a-n/a).
http://dx.doi.org/10.1080/01490410701296341. Wolfe, B. B., Hall, R. I., Edwards, T. W. D., Jarvis, S. R., Sinnatamby, R. N., Yi, Y., et al.
Syvitski, J. (2008). Deltas at risk. Sustainability Science, 3(1), 23–32. http://dx.doi.org/ (2011). Climate-driven shifts in quantity and seasonality of river discharge over
10.1007/s11625-008-0043-3. the past 1000 years from the hydrographic apex of North America. Geophysical
Syvitski, J. P. M., Kettner, A. J., Overeem, I., Hutton, E. W. H., Hannon, M. T., Research Letters, 35, L24402. http://dx.doi.org/10.1029/2008GL036125.
Brakenridge, G. R., et al. (2009). Sinking deltas due to human activities. Nature Wolfe, B. B., Hall, R. I., Edwards, W. T. D., Vardy, S. R., Falcone, M. D., Sjunneskog, C.,
Geoscience, 2(10), 681–686. http://dx.doi.org/10.1038/ngeo629 ( http://www. et al. (2008). Hydroecological responses of the Athabasca Delta, Canada, to
nature.com/ngeo/journal/v2/n10/suppinfo/ngeo629_S1.html) changes in river flow and climate during the 20th century. Ecohydrology, 1,
Syvitski, J. P. M., Vörösmarty, C. J., Kettner, A. J., & Green, P. (2005). Impact of humans 131–148. http://dx.doi.org/10.1002/eco.13.
on the flux of terrestrial sediment to the global coastal ocean. Science, 308(5720), Wolfe, B. B., Karst-Riddoch, T. L., Hall, R. I., Edwards, T. W. D., English, M. C., Palmini, R., et al.
376–380. http://dx.doi.org/10.1126/science.1109454. (2007). Classification of hydrological regimes of northern floodplain basins (Peace–
Timoney, K. (2006). Landscape cover change in the Peace–Athabasca Delta, Athabasca Delta, Canada) from analysis of stable isotopes (δ18O, δ2H) and water
1927–2001. Wetlands, 26(3), 765–778. http://dx.doi.org/10.1672/0277-5212(2006) chemistry. Hydrological Processes, 21(2), 151–168. http://dx.doi.org/10.1002/hyp.6229.
26[765:lccitp]2.0.co;2. Yang, S. L., Belkin, I. M., Belkina, A. I., Zhao, Q. Y., Zhu, J., & Ding, P. X. (2003). Delta re-
Topliss, B. J., Almos, C. L., & Hill, P. R. (1990). Algorithms for remote sensing of high con- sponse to decline in sediment supply from the Yangtze River: Evidence of the recent
centration, inorganic suspended sediment. International Journal of Remote Sensing, four decades and expectations for the next half-century. Estuarine, Coastal and Shelf
11(6), 947–966. http://dx.doi.org/10.1080/01431169008955069. Science, 57(4), 689–699. http://dx.doi.org/10.1016/s0272-7714(02)00409-2.