Sediment in the Trinity River Basin, Texas - Center for Research in ...

lyricalwillingMécanique

22 févr. 2014 (il y a 3 années et 8 mois)

54 vue(s)



Virginia Smith

CE 397



Spring 2009

Introduction

Data Resources

Hydrostatistical Analysis of Mainstream Gages

Hydrostatistical Analysis of Catchment Gages

Conclusions

References

Appendix

Introduction

This project is an analy
sis of the sediment budget for the Trinity River Basin.
Sediment is any
particulate matter that can be transported by fluid flow.
In any river basin s
ediment plays an important
role
by impacting geomorphology

(impacting the channel geometry and
the flood

plain)
, transporting
nurturance
, building deltas, and
aiding the construction of

coastlines.
S
ediment can also cause
impacts
on

water resource management
;

such as
,

altering the capacity of reservoirs, causing a decrease in
water quality, and limiting photosynthesis through high turbidity.
Shifts in the balance of the sediment
budget can have far reaching affects.
[1]

The sediment budget
o
f any river basin can be de
scribed by the
change in sediment,

S
, as it
is
related to the sediment output,
O,

minus the sediment input,
I.
[2]

This
relationship is

described by the following equation:








It is likely that in highly populated basin anthropogenic effects impact

the sediment budget.
Many
variables have been shown to influence sediment generation.
Crop land can generate up to thirty
times

more sediment than background natural regions.

[3]



Dams also impact the transport of sediment
in the river system. A study

investigating the 600 largest reservoirs in the world found that the average
reservoir was 80% efficient at trapping sediment. Therefore, while human influence may cause
additional sediment to be generated, dams (also due to human influence) may limit th
e amount of
sediment being transported through the river system.

[4]


Other influential factors may be the
topography, the natural land cover, the geology, the vegetation density, and the climate.

[2]

In t
his report

only
sediment in the Trinity River Basi
n

was investigated
. The basin is shown in figure 1.
The Trinity River Basin is located in northeast Texas. The Trinity River flows 710 miles through Texas.
The river’s basin spans 17,969 square miles.
The basin extends over three

climate areas (as def
ined by
the National Oceanic and Atmospheric Administration), ten geologic regions and many types of land
cover.
Hill slopes vary in the basin, but general the

upstream area of the basin is hilly,
and

the basin
becomes flatter as the river flows towards t
he coast. Over time urban area and crop land has
increased.

[5]



Figure 1: This map shows the location of the Trinity River Basin in Texas
.
[6]

Over the past 25 years the Trinity River Basin has experienced an intense increase in population.
The
Trinity River Basin is the most heavily p
opulated river basin in Texas. The population density of the
basin is
due

largely

to the population centers Dallas, Fort Worth and Houston.
The population of the
basin was approximately 3.2 million people in 1980.

In 2006 the basin population was estimated to be
more than 13 million people.

[5]

The population is projec
ted to continue growing, as is

the water
demand. According to the Texas Water Development Board (TWDB), 80 to 90 percent of the water
supply in th
e Trinity River Basin comes from surface water. There are currently 28 water supply
reservoirs within the basin, with thirteen more planned. The current reservoirs contain more than
5,000 acre
-
feet
of water storage.

[7]

The reservoir capacity and water

demand
of the
basin

is shown in
the graphs below
, in figure 2
. The demand in the upper Trinity River Basin exceeds the current supply.
Overtime, the demand will continue to rise and additional resources will been needed to meet the
demand.

[5]


Figure
2
: The graphs above show water demand in the Trinity Basin. The graph on the represents the upper basin and the
graph on the left represents the lower basin. The red lines show the predicted demand. The blue lines show resources as
they currently

exist. The cyan lines show the amount of water resources available in the future.

[5]

High water demands, notable anthropogenic impacts and physical diversity of the basin make the
Trinity River Basin an interesting study for investigation. Data was col
lected for sediment, flow and a
variety of predictor variables for this study

area
. T
his data was analyzed statistically to gain a deeper
understanding of the sediment budget of the
basin
. This report will describe the analysis by reviewing
the data sour
ces and collection, the gages on the mainstream and the gages on the smaller
catchments.

Data Resources

Suspended s
ediment in the Trinity River has been
recorded

by both the

TWDB and the US Geological
Survey (USGS).
The TWDB data was collected from TWDB r
eports

online
.

[8]

The data is reported as
total monthly values.
The USGS data was collected from the USGS Sediment Database.
The USGS
sediment and flow values were stored as mean daily data. Pivot tables were used to group the

USGS

values into months
as to

facilitate a comparison with the TWDB data.

[9]


Figure 3

shows the s
ediment
gages of the Trinity River basin. The record of the
se gages is from 1965 to 1986. The gages maintained
by the TWDB col
lect
ed

suspended sediment data t
hrough the Texas Method. The USGS gages collect
ed

suspended sediment data through the USGS Depth Integrated Method.

The TWDB reports that the
methods may give different

individual

measurements, particularly when loads contain greater amounts
of fine sand
. However, TWDB reports that regardless of these facts, the sampling methods maintain a
high correlation when tested over many diverse type
s of water bodies.
[
8
]


Figure
3
: USGS and TWDB gages within the Trinity River Basin. The USGS gages are labeled
yellow. The TWDB gages are
colored orange

Because the Trinity River sediment is sandy the sediment sampling methods were tested again for this
investigation.

[10]

Gage 08062700 was sampled by both TWDB and USGS from 1966 to 1968, producing
two suspended

sediment datasets for the same location and period of time. The homogeneity of the
data was te
sted using the Wilcoxon Sum

Rank Test in Statistical Analysis Software (SAS). The Wilcoxon
Test was
chosen

to evaluate the datasets because
it is a non
-
paramet
ric test.

[11]

The datasets were

equal in size
, but not normally distributed
.
The results of the te
st are shown

in the appendix
.
As can be
seen the test showed

that the
datasets

do not appear to be diverse.

All of the sediment gages, TWDB
and USGS, are located by USGS stream flow gage stations.
Additional s
tream

flow

data was

obtained
from the National Water Information System (NWIS).

[12]

D
ata

was collected

from a variety of sources
.

This data
was used for

basemap information and
potential sediment predictor variables.
A large portion of the data was downloaded from the Texas
Natural Resource Information System (TNRIS).

[13]

All the data provided by TNRIS was GIS shapefiles

and raster images
.
The TRNIS
issued
files were used to create the

basema
p for the project. The
basemap
included the streams and rivers, drainage basins, river basin,

reservoirs,

counties and cities.
Other data downloaded from

TNRIS included the hill slopes

and the DEMS. The DEMs we
re used to
delineate catchments for each non
-
mainstream gage

in GIS
.


By delineating catchments draining to the
individual gaging stations, shapefiles could be made of the catchments pertaining to each
non
-
mainstream
gaging station.

The ca
tchment files wer
e then used to solve for variables that describe the catchments; such as land
cover, climate, and storm depth.
Catchment shapefiles were paired with

the hill slopes, land cover,
natural regions and counties
shapefiles to clip out data
for the

individual c
atchments.
The land cover
data came from the National Land Cover Dataset (1992).

[14]

The natural regions data was downloaded
from Texas Parks and Wildlife.

[15]

This data was then attached to erosion rates provided by the TWDB.

[16]

The county
shap
efile

was used to make estimates of types of data described by county. Data
described by county included storm depth
,

climate and vegetation data.


Figure
4
:

The images above are examples of using GIS to extract descriptive variables for specific catchments.
The image
top

left shows the hill slope rasters for each of the catchments.
The

image
to the

top right

represents the catchments by
natural region. The

image

on the bottom left

shows land

cover of the catchments. And, the image to th
e

bottom

right

shows the catchments according to the counties they contain.


The storm depth data is a product of USGS report 1725.

The storm depth describes the rainfall d
epth
during a storm.

These values were calculated through a multi
-
step process.
First, seven interevent
minimum times were calculated for storm events. Then, storm depth and duration were calculated for
every station in a given county. Next
,

L
-

moment
statistics were done to assess the storm depth for a
state or region. For storm depth, a weighted
-
average storm depth L
-
moment mean depth was found
.
This was
solved

for each county in Texas, Oklahoma and New Mexico. For the purpose of this
investigation

the values were averaged. The
county values were then attached to the shapefile of
counties and a weighted average storm depth was produced for each catchment.

It should be noted
that this data was redistributed twice, degrading its spatial integrity.

[
17]



The climate and vegetation data are products of the North American Regional Reanalysis (NARR).

The
NARR data was collected from the NOAA/National Operational Model Archive Distribution System
(NOMADS).
Precipitation
,
precipitation

intensity and
vegetat
ion
density were collected for the T
rinity
region. It is important to note that this data is modeled, not
observation

records.
The components
used to create the NARR are the National Center for Environmental Prediction (NCEP) Eta Model
and its Dat
a Assimilation System, a version of the NOAA land surface model and many additional
data sets.
The data was collected for a fifteen year period, from 1979 to 1994.
The

NARR variables
collected where converted from
NetCDF

to
raster

files. The rasters were

then averaged using the
cell
statistics tool in
spatial analysis
; which
produced a mean for all of the rasters by cell
. Then, the average
value rasters where converted to features and spatially joined to counties. Finally¸ a weighted average
of each NAR
R variable for each catchment could be estimated. It should be noted that the multiple
redistributions

in this procedure most likely altered the data.

[18]

After all the data had been processed and a basemap was complied the sediment budget of the basin
c
ould be analyzed.
It should be noted that the
variable
d
ata was normalized based on the area of the
catchment.
S
tatistical comparison
s

were

made between the catchment sediment concentration and
the variables describing each catchment.


Hydrostatistical An
alysis of Mainstream Gages

For the sediment analysis the gages were divided into two groups: mainstream gages and branch
catchment gages. (For the remainder of this report the branch catchment gages will be referred to as
catchment gages.)
First
,

the ma
instream gages were analyzed. There were four suspended sediment
gage sites along the main channel of the Trinity River. Three of the gages were maintained by
the
TWDB, and one by the USGS. Although the sampling methods were different, the differences a
re
assumed negligible based on the previous Wilcoxon Test.
(Statistical Analysis Software (SAS) was used
to compute the Wilcoxon Test. The output can be seen in the appendix.)
The record for the gages
ranges from 1965 to 1982; however, not all of the ga
ges have the same length of record.

Below are the sediment rating curves. The x
-
axis represents discharge and the y
-
axis represents
sediment load.
The points represent monthly estimated values.
The black line is the linear trend in the
data. Sedimen
t
concentration can be described as the ratio of sediment load to stream flow.

[19]


Therefore, the slope of the trend line (m=sediment load /discharge) is an approximation of
concentration. The concentration
was shown to be

greatest just south of the Da
llas/Fort Worth area.
The
sediment concentration decreased

and then i
ncreased

again at the coast.


Figure
5
: The image below shows the sediment rating curves for the mainstream sediment gages in the Trinity River Basin.
The y
-
axis represents the
sediment load and the x
-
axis represents the flow. The according to the equation C=L/Q, the slope
of the trend should correlate with the concentration. A greater slope represents a higher concentration and a smaller slope
implies a lower concentration of
sediment.

The top three gages rep
resent a period of record after dams where
emplaced

upstream. The last gage,
08066500 represents a period before and after a dam was
emplaced

upstream. To view the effects of
the dam on the sediment rating curve record wa
s divided into a pre
-
dam and post
-
dam.
In t
he image
below
, figure 6,

the
original

sediment curve compared to the divided record sediment rating curve.

The
slope before the dam
was

greater than the slope after the dam. This insinuates that concentration
d
ecreased after the reservoir was impounded.

This makes sense because the dam acted as a barrier to
sediment transport.


Figure
6
: The graphs above show the effects of an upstream dam on the sediment rating curve. The graph on the left
shows the
sediment rating curve for one gage over a period of time. The graph on the right represents the sediment rating
curves for the same gage; however the period of record has been broken into two sections, pre
-
dam and post
-
dam. As can
be seen in the graph th
e slope of the pre
-
dam data is much greater than that of the post
-
dam data. This implies that the
concentration of sediment in the water decreased after the dam was emplaced.

Although the data
there was not adequate to
compare the pre
-
dam sediment rating
curve to the post
-
dam sediment rating curve

for all the gages
,

the flow regime pre
-

and post
-
dam
could

be compared.
T
o
gain a deeper understanding of how the dams may
have impacted

these gages comparisons

of

the flow
regimes were
made
. Below
, in figure 7
,

are the daily mean flow records for each of the gages. This
data was downloaded from the
NWIS
. The
graphs on the left show

the mean daily flow values through
time. The blue lines represent the mean daily flow and the red lines represent the constructi
on of
dams. The
graphs

on the right s
how

the flow duration curves before and after the dams influenced the
upstream.

The pre
-
dam and post
-
dam records were separated based on the location of the red lines in
the flow record.

The red line represents the p
re
-
dam flow record and the blue line represents the post
-
dam record.


Figure
7
: Graphs above represent the flow regimes
that

occurred at the mainstream sediment gages of the Trinity River.
The graphs on the left show the flow values as a time series. The red lines on these graphs represent the emplacement of
the dams. The graphs on the right represent the flow duration curve
s pre
-

and post
-
dam for each of the mainstream gages.

All four gages appear to have higher flow duration curves after the emplacement of upstream dams
.

When the means for the pre
-
dam and post
-
dam records
were compared
it showed that the mean daily
discha
rge did increase in all four cases. However, the increase was not as great as it appears in the
graphs. In all four graphs the pre
-
dam rose above the post
-
dam in the highest percentiles. However,
the increase in mean daily flows seemed counter intuitive
. There are several
expla
n
ations for this
occurrence
. First, the last quarter of the twentieth century was wetter than the rest of the century.
Secondly, the values are the mean daily values. This could have caused a damper of the pre
-
dam
extreme flows
.

Another influence of this
flow shift

might have been due to runoff from water pumped
from ground water systems.

[20]


Hydrostatistical Analysis of Catchment Gages

The second stage of the analysis was to investigate the catchment gages.
To gain a deepe
r
understanding of the variables influencing sediment concentrations the concentration was analyzed in
comparison to a variety of variables
describing

the catchments. Unlike the mainstream gages the
catchments drained a small area. The mainstream
gages d
rained large areas and were subject to a
wide variety of potential influences. The smaller area
s

of the branch catchments made it easier to
define all the potential
variables influencing
sediment

generation
. Also, none of the catchments have
major dams u
pstream impacting their flow.

The catchments can be seen
i
n the
map in
figure

8

below.

The catchments are shown in the top image,
colored beige. Beside the image is a graph of the flow duration curves for the basin. The curves
correlate to the catchme
nt by area
, as can be seen with the arrows
. The highest curves correspond to
the

catchments with the

greatest area
s

and the smallest curves belong to the catchments with the
smallest area
s
.

However, these do not correlate to concentration. The lower ima
ge shows the
relationship between concentration and flow duration curves at each gage.
The red dots represent the
lowest concentration and the green dots represent the highest concentration. The scale is given in the
legend on the right.
There does not
appear to
be a clear trend between the flow duration

curves and
the concentration
.


Figure 8: The images above show the flow duration curves in relation to the catchment size and the gage concentration.
There is a strong correlation between catchment
size and flow, but not between concentrations.


Next
,

sediment rating curves were made for each of the catchment gages. The sediment load is on the
y
-
axis and the flow is shown on the x
-
axis. As mentioned
previously
, the slope of these curves
correspond
s

to concentration
. The equation of sediment concentration is as follows
: sediment
concentration =sediment load ÷ flow.

[19]

This matches the results depicted in the image. Where the
slope is greatest the concentration
wa
s the greatest (gage 8051500).

As th
e concentration values
decreased

the slope of the graphs decrease
d
. However, there appears to be no clear geographic

trend

in the sediment data.


Figure
9
: This image shows all of the sediment rating curves for the Trinity River Basin. The gages w
ith higher
concentration are green (and have sediment rating curves with higher slopes). The gages will lower sediment
concentrations are red (and have sediment rating curves with a small slope).

To gain insight into the spatial variation of the concentra
tion a
multiple
regression analysis was done.
The regression used nine predictor variables described in the data section: hill slope, average storm
depth, erosion of natural region, vegetation density, precipitation, precipitation intensity, urban land
c
over, agriculture land cover, and forested land cover.

The goal of creating a regression model was to
relate th
ese variables to the catchments’ concentration
in order to see the sediment production
influences. Identifying catchment traits that correspond

to higher sediment concentration would
explain the variation in sediment concentration at the catchment gages across the basin.

Before modeling the basin through multiple regression the correlation between the descriptive
variables had to be identified.

Multicollinearity occurs when two or more predictor variables within the
regression model are highly correlated.

When

multicollinearity

occurs the variables are biased, but

the
overall model results will still be accurately predicted
. T
he t
-
statistics
for the correlated variables will
not be. This hinders the ability to decide which variables to include in the model and which variables to
dismiss. The variation inflation factor (VIF) shows the degree of
bias

present for a variable. In this
study, vif
levels greater than ten were assumed to

indicate the

bias
.

[21]


To identify variables that may be highly correlated
statistical te
st of correlation where preformed using
SAS

(the statistical significance can be seen in the
SAS output in the appendix)
.
The correlation
between all nine v
ariables was found through Kendall’s t
au, Spearman’s rho and Pearson’s r

te
sts
.

The
first correlation te
st preformed was Pearson’s r. This is the most common test; however, it is a
parametric

te
st and requires normal dis
tribution. While the variables appeared to have a relatively
normal distribution
,

there were only eight observations for each variable. However, the values with
the highest correlation also had the highest

statistical significance

(see the

SAS correlatio
n output in
the a
ppendix)
. An important difference between Kendall’s tau and Spearman’s rho is that the influence
outliers have on the results. Kendall’s tau allows for some resistance to outliers. Kendall’s tau also
tends to have lower values.

The cor
relations identified in all three methods were relatively similar. The
non
-
parametric te
st gave very similar t
ests.

[11
]


Based on the results of the correlation te
st there appeared to be correlation between precipitation and
precipitation intensity. The
re also seemed
some correlation between types of land cover and the types
of land cover, precipitation and vegetation and erosion and urban area, and agriculture area and
precipitation. These correlations were taking into account when performing the regre
ssion.

The regression was done in

first done

SAS
, and then redone in
excel. The SAS output can be seen in the
appendix.

Many regression models were tested to find the best combination of variables to describe
the trends in the data. First, all the varia
bles were ran in a regression model with the variance inflation
factor tool. This result showed that there was a bias present in all of the variables to some degree. In
fact, the test reported that the agriculture land and forest land variables were line
ar combinations of
the other variables. However, these variables were identified as such because of the order in which the
variables were being analyzed. In the next run of the model the variables with the highest correlation
were removed. The first var
iables to be removed were precipitation intensity and forest land.

Many
variables and combinations of the regression model were created subsequently.

The variables used in
the tria
l models were selected based on the t
-
stat value (ltl>2), the p
-
value (p<=
0.05) and the vif value
(vif<10). The final result showed tha
t the best possible result for wa
s as follows:


Conc. = Hill slope*(5.16)

t=5.16

R
2
=0.792


p
-
value = 0.0013



F = 26.65

When the same tests were
performed

in excel the same results were found.

Hill slope is a known
key variable to sediment generation.

It is not surprising that it was found to be

influential to the
generation of sediment. It is one of the variables in the soil loss equation.
E
rodibility
, precipitation
and crop land also play a role in the soil loss equation.

However, in the variations of the
regression model ran these variables were found to be statistically insignificant to describing
sediment concentration. One possible reason for
this is all of the observations were averages
representing a long period of time.
Tak
ing the averages may have dampen
ed the concentration
variable relationship.
If the data had been analyzed using short time steps
a

greater relationship
between the varia
bles and the sediment concentration might have been established.

One
approach to doing this would be to create regression models for each individual catchment, so that
the variables could be analyzed through time.

[2
2
]

Conclusions

The primary products of
this investigation were sediment

rating

curves
,

flow duration curves and a
regression model
for the sediment concentration of the catchments in the basin.

All of these products
were
enlightening
to the sediment budget of Trinity River Basin. First, in st
udying the sediment
concentration of the mainstream gages a discontinuity in the sediment concentration trend from
upstream to the coast was identified. Also, a drop in sediment concentration was shown after the
emplacement of the dam upstream of gage 080
66500. The drastic change in concentration before and
after the dam implied an influence from the dam itself on the concentration. The flow duration curves
also appeared to have been influenced by the dams. After the dams were emplaced the flow rose in
all
four cases. Upon further investigation
,

it was evident that
o
the
r

influences
also were likely to have
played a role in the shift of the flow duration curves.

Next, the catchment gages were analyzed. Flow duration curves showed a strong correlation
to the
drainage area. However, there did not appear to the relationship between concentration, drainage
area or geographic location. The concentration appeared high and low in both the upper and lower
basin. An additional investigation was done to inspe
ct the variables that may have influenced sediment
generation in the catchments (resulting in the varied concentration levels at the gages). This was done
through a regression model. However, after accounting for multicollinearity and statistical signifi
cance
the only
variable that appeared influential was hill slope.

This study posed several interesting questions; such as, why exactly did the flow duration curves
increase after dams were impounded upstream, why did more variables matter in the linear reg
ression
model, and how did the averaging and redistribution of data influence the final results. While this
investigation gave some insight into the sediment budget of the Trinity River Basin,
there are still
several issues not fully comprehended.
To und
erstand this system fully a deeper investigation into
these issues is required.


References

1.

Gray, John R. "Fluvial Sediment."
Office of Surface Water
. USGS. 6 May 2009
<http://water.usgs.gov/osw/techniques/sediment.html>.


2.

Dietrich, William E.; Dunne, Thom
as; Humphrey, Neil F.; Reid, Leslie M.


1982.

Construction of
sediment budgets for drainage basins.


In: Sediment Budgets and Routing in Forested Drainage Basins:
Proceedings of the Symposium; 31 May
-

1 June 1982; Corvallis, Oregon. Gen. Tech. Rep. PNW
-
141.
Portland, Oregon: Pacific Northwest Forest and Range Experiment Station, Forest Service, U.S.
Department of Agriculture; 1982: 5
-
23.


3.

Wilkinson, B. & McElroy, B., 2007, The Impact of Humans on Continental Erosion and Sedimentation,
GSA Bulletin
, v.119
, p.140
-
156.


4.

Vorosmarty, C.; Meybeck, M.; Fekete, B.; Sharma, K.; Green, P.; Syvitski, J., 2003, Global Fluvial
Sediment Retention by Registered Dam Systems, EUG Joint Assembly, Program and Abstracts.


5.

Trinity River Basin Master Plan. Trinity River
Authority of Texas. 2007. 1
-
48


6.

"The Trinity River Basin."
Wikipedia
. 29 Feb. 2009. 30 Apr. 2009
<http://en.wikipedia.org/wiki/Trinity_River_(Texas)>.


7.

Water for Texas. Texas Water Development Board. 2007. Document number GP8
-
1


8.

Dougherty, John P.
Suspended Sediment Load of Texas Streams, Compilation Report, October 1971
-

September 1975
. Rep. no. 233. 1979.


9.

Gray, John R. "Suspended
-
Sediment Database Daily Values of Suspended." 8 May 2009. USGS. 8 May
2009 <http://co.water.usgs.gov/sediment/>.


10.

Phi
llips, Johnathan D., Michael C. Slattery, and Zachary A. Musselman. "Channel adjustments of the
lower Trinity River, Texas, downstream of Livingston Dam."
Earth Surface Processes and Landforms

30
(2005): 1419
-
439.


11.

Helsel, D. R., and R. M. Hirsch.
Statistical Methods in Water Resources
. Vol. 4. USGS.


12.

USGS Water Data for the Nation."
National Water Information System: Web Interface
. 08 May 2009.
USGS. 8 May 2009 <http://waterdata.usgs.gov/nwis>.


13.

"Texas Natural Resources Information System."
Www.tnr
is.org
. 08 May 2009. Texas Water
Development Board. Mar. 2009


14.

."National Land Cover Dataset 1992 (NLCD 1992)."
The USGS Land Cover Institute (LCI)
. Mar. 2007.
USGS. Mar. 2009 <http://landcover.usgs.gov/natllandcover.php>.


15.

"GIS Lab Data Downloads." 20 Jan
. 2009. Texas Parks and Wildlife. Mar. 2009
<http://www.tpwd.state.tx.us/landwater/land/maps/gis/data_downloads/>.


16.

Greiner, John H.
Erosion and Sedimentation by Water in Texas, Average Annual Rates Estimated in
1979
. Rep. no. 268. Austin, TX: Texas
Department of Water Resources, 1982.


17.

Asquith, William H., Meghan C. Roussel, Theodore G. Cleveland, Xing Fany, and David B. Thompson.
Statistical characteristics of storm interevent time, depth, and duration for eastern New Mexico,
Oklahoma, and Texas
. Te
ch. no. 1725. B ed. Vol. 200. Ser. 1725. Texas Water Science Center, 2006.


18.

Messinger et al. "North American Regional Reanalysis." Bulletin of the American Meteorological Society
(2005): 1
-
42
.


19.

Jahani, A. "Calculating the suspended sediment load of the Dez River."
Erosion and Sediment Transport
Monitoring Programmes in River Basins (Proceedings of the Oslo Symposium)

210 (1992): 219
-
24.


20.


Quantifying downstream impacts of impoundment on flow regime and channel planform, lower
Trinity River”, Texas,
Slattery, M.C.; Phillips, J.D.; Wellmeyer, J.L.

(
Geomorphology
,
2008
-
11
-
18
).


21.

"Multicollinearity."
Wikipedia
. 07 May 2009.

07 May 2009
<http://en.wikipedia.org/wiki/Multicollinearity>.


22.

Stone, Robert P., and Don Hilborn. "Universal Soil Loss Equation (USLE)."
Ministry of Agriculture, Food
and Rural Affairs
. 15 Dec. 2008. Ontario Government. 07 May 2009.





Appendix

Wilcoxon Test

Correlation Tests

Regression

Wilcoxon Test:


Correlation Tests:





Regression:

All Variables: agriculture and forest are linear
combinations of the other variables.


Agriculture and Forest were removed




Other variables were removed based on t
-
stat, correlation and variance inflation


Finally, nothing was left except hill slope