Increasing Remotely Sensed Snow Cover through Composite AMSR- E/MODIS Product

vetinnocentΛογισμικό & κατασκευή λογ/κού

7 Νοε 2013 (πριν από 3 χρόνια και 10 μήνες)

150 εμφανίσεις

Increasing
Remotely Sensed Snow Cover through

Composite AMSR
-
E/MODIS Product


By

Charley Follett


An Undergraduate Thesis

Submitted in Partial Fulfillment for the Requirements of

Bachelor of Arts

in

Geography and Earth Science








Carthage College



Kenosha, WI




April
, 2010






2



Increasing MODIS Snow Cover

through

Composite

AMSR
-
E/MODIS Product


Charley Follett



Abstract

Accurate mapping of snow cover in mountainous terrain is extremely important, as snow
-
melt is a major source of water storage and runoff. Therefore, remote sensing of snow cover has
wide applications, especially in mountainous terrain where recording groun
d station data is
difficult. Both MODIS and AMSR
-
E satellite instruments can be used to study snow cover,
with
the possibility of a composite AMSR
-
E and MODIS snow cover product increasing accuracy of
remote sensing based snow cover
. As each instrument com
plements the other, as AMSR
-
E can
detect through clouds whereas MODIS cannot, and the resolution of MODIS is superior to
AMSR
-
E.
The focus is

in the study area of the Colorado Headwaters region

during the month of
January, 2006.















3


Table of
Contents

List of Figures
………………………………………………………………………
……..
4


Lists of Tables

…………………………………………………………………………...
4



Literature Review
………………………………………………………………….....
.
.......5



Introduction.
........................
......................................................................
......…
..
...
.
.5


Background…………………….
.
………………………………………………
.….5





Hypothesis
……

…………………………………………………………………
..
……...
13


Method
s
.
……
…….
……………………………
……………………………………….…..
14



Study Period
.............
......……………
……
…………
……………………………...
14


Study Area…………………………..

……….
……
…………………………….....
14


Acquisition
………………………………………………………
…………………
15

Processing……………………………...
………………………………...
…………
.17


Results
……………………
……
……………………………………………………

.........
20






Discussion
…………………………………………………………………………
….........
25



Future Research……………………
……………………………
………………....
25


Acknowledgements
................................................................................
..............................
2
7



Works

Cited
.......................
...........................................................................
.......
...............
..
2
8
















4



List of Figures




Figure 1: Map of study area, shown by red outline







14



Figure 2: AMSR
-
E and MODIS overlay




18



Figure 3: 2006 January 1st composite AMSR
-
E and MODIS map.





19


Figure 4: January 8th MODIS snow cover









20


Figure 5: January 8th composite snow cover









20


Figure 6: Chart of accuracy between MODIS and Composite product

21


Figure 7: January 5th snow cover.









22



List of Tables



Table 1: MODIS integer values, from National Snow and Ice Data
Center




16

Table 2: Percentage of remotely sensed snow and land cover





21

Table 3: Percentages of snow, land, and cloud cover






22


Table 4: Paired T test of means.










23



5





Introduction


Remote sensing
of snow cover has wide application
s
, especially in mountainous terrain
whe
n

ground station data is sparse. Remote sensing of snow cover allows for reliable, continuous
study without the need for coverage to be interpolated through techniques such as isohye
tal
mapping.

Both

the

Moderate Resolution Imaging Spectroradiomete
r
(
MODIS
)

and
Advanced
Microwave Scanning Radiometer
-

Earth Observing
System (
AMSR
-
E
)

satellite instruments can
be used to

remotely sense

snow cover
. C
omposite snow cover products that comb
ine the AMSR
-
E and MODIS instruments have been
developed

by researchers

and

shown to increase accuracy
over using either instrument alone
.

The purp
ose of this thesis was

to compare a

composite
AMSR
-
E and MODIS product with
the
cloud removal
algorithm
against

MODIS
snow cover
product

to

see if accuracy is improved

as a result
.




Background



Accurate mapping of snow cover in mountainous terrain is extremely important, as it
contributes to modeling of snow
-
melt, which is a major source of water storage and runoff in
many watersheds such as the Colorado River. Data relating to water

storage change is relevant to
agricultural interests, reservoir operators, and other agencies and corporations that manage water
resources.
The high albedo reflectance of snow also makes it an important factor in climate
change studies and energy budget e
quations

(Foster et al. 2005)
.
Streamflow is

statistically
linked to snow cover

(Yang et al. 2007)
,
thus

snow cover is a vital factor to be investigated when
6


researching areas where s
now cover is extensive and hydrologically significant
.

The Colorado
Headw
aters meets
this criterion
, as snowmelt in the sub
-
basin is the leading contributor to the
Colorado River

(Short, 2010)
.
Remote sensing is an excellent
source that has been used
to map
snow cover, and the introduction of new satellite instruments and techn
iques increase our ability
to map sno
w cover with each passing year.
Through

developing a composite

snow cover
mapping method
, which uses both

MODIS and AMSR
-
E snow cover
products
,

this proposal
aimed to

increase

accuracy

of mapped

snow cover and provide valuable
information

to
water
managers a
nd other users of the data
.


Remote sensing provides valuable tools

for looking at snow
-
covered area

(SCA
) and

other
hydrological

factors, such as snow water
equivalent (
SWE).

SCA

refers to t
he geographical
extent of snow

cover
, while
SWE
is an estimate of how
much water would be obtained from a
melting snowpack.
Remote sensing offers continual, uniform, and reliable data over large areas,
and can
be used to
complement
other sources of data
s
uch as ground
observation networks
.

Ground based stations can provide continuous and reliable information about snow cover and
other factors

for a specific location
, but in mountainous

terrain,
these values (ex. SWE) may vary
wi
dely due to the complex
terrain
.

In addition, other difficulties arise from
inconsistencies

between gro
und stations, and accessibility issues in maintaining a network of ground stations

in
mountainous areas
. Remote sensing fills the gaps left by ground station data and can prov
ide
among other things percent snow cover over a mountainous region, which is
often
not possible
using ground station data.
Snowpack Telemetry

(SNOTEL) stations record long
-
term data
(i.e.
SWE, temperature, precipitation)
which can be statistically analyze
d against remote sensi
ng
measurements for validation. The SNOTEL
data
network
has been in operation from the 1970s,

7


and
provides data

at high elevations and in rugged terrain, where validation of remote sensing
data is most useful.


Besides academia,
SCA e
stimates obtained from

remote sensing
are
used extensively in
the
private and government sectors. Organizations such as the National Oceanic and
Atmospheric Administration provide real
-
time snow cover and snow water equivalen
cy

maps
produced
using
remote
ly

sens
ed data

to the public. These maps are interactive, and allow casual
users to
explore
regions

with only a few clicks. In the private sector, news corporations use these
maps for weather forecasts and visualizations.
Communities can use remote sensing
snow
products to make decisions regarding water resources and agricultural activity
, such as irrigation.


Remote sensing instruments collect data about the earth by measuring electromagnetic
radiation e
mitted or reflected by the earth’s surface
.

Remote sen
sing instruments must collect
data from “windows” in the electromagnetic spectrum where the Earth’
s atmosphere does not
interfere

(Campbell, 1996)
.
This limits remote sensing to using the visible, infrared, and
microwave radiation
portions
of the electroma
gnetic spectrum.

Remote sensing is either passive,
or active. Passive remote sensing refers to detect natural radiation emitted or reflected by the
earth’s surface, whereas active remote sensing refers to instruments sending their own energy to
the earth a
nd detecting how it changes when it is reflected back to the remote sensing instrument.

Passive remote sensing in the visible and infrared spectrums measures the reflectance of solar
radiation by the earth’s surface. The wavelength spectrums

have differen
t properties and
applications; visible radiation can penetrate water, infrared radiation is extremely useful for land
-
use and vegetation differentiation, and microwave radiation is not attenuated by
the
passage
through the atmosphere

and can be used at nig
ht because there is no dependence on reflected
solar
radiation

(Campbell, 1996)
.
A m
u
lti
-
spectral scanner

use
s

both the visible and infrared
8


spectrum to measure several

bands, or

ranges o
f radiation wavelength.

Everything has
its

own
spectral signature; a field of corn has different reflectivity when it is planted, when it is growing,
and when it is harvested.
Multi
-
spectral scanners are sensitive enough to detect such differences
in reflectivity, which adds to the type of analysi
s that can be done.

For instance
, bands

can be
combined into a single image, or selected to emphasize differences in reflectivity due to
wavelength.


A multi
-
spectral instrument, t
he Moderate Resolution Imaging Spectroradiometer

(MODIS)
sensor
on the Terra satellite provides
global
sn
ow and ice products at a 500 meter
resolution on a daily basis, with
each swath measuring
233
0 kilometer
s
.

With a high radiometric
sensitivity across 36 spectral bands, MODIS also has a high temporal cycle of 1
-
2 days globally.
The MODIS instrument uses both visible and infrared spectral bands.

This resolution is excellent
at both regional and more local scales
.

Applications of MODIS data range from measuring
surface temperature, to measuring global vegetation and

snow cover.

Snow cover has high
reflectance in the visible spectrum and low reflectance in the short
-
wave infrared spectrum.

High
reflection in the vi
sible spectrum is due to albedo, or the measure of how strongly something
reflects light from light sources. If an object or surface is very white, albedo is high, if it is
darker, albedo will be low, thus snow’s high reflectance in the visible spectrum. R
eflectance in
the infrared spectrum is a function of energy emitted.


The
algorithm used to creat
e

the
MODIS snow cover products
contains several
enhancements designed to increase the accuracy of the SCA

estimates
. To prevent snow cover in
forests from bei
ng misread,
the
Normalized Difference Vegetation Index

(NDVI)
(insert citation)
is calculated to help determine differences between snow
-
covered
areas
and snow
-
free forest
s

(Yang et al. 2007)
. Other criteria, related to
the
reflection
of snow
in the different
spectral
bands
9


ar
e used to allow for the discrimina
tion of snow from water
and clouds

(Yang et al. 2007)
.

Comparisons of MODIS data

and
SNOTEL station

observations of SWE

show high statistical
agreement when cloud
-
cover
does not interfere

with the viewing of snow cover from the MODIS
instrument
(Klein, 2003
).


MODIS has a nominal resolution of 500 meters, so sometimes a
MODIS pixel will show that there is snow cover when a SNOTEL station disagrees, because
the
pixel is averaged across 500
meters whereas the station detects snow in
its

immediate area.

Because the
MODIS
instrument
cannot
detect
snow cover
under
clouds

it

is
ineffective for the
mapping of SCA
under
these
conditions. Other sensors exist
that can

detect
SCA

through cloud
cover,
for example those that utilize the
microwave

portion of the spectrum (i.e. AMSR
-
E)
.

The A
dvanced Microwave Scanning Radiometer
-
Earth Observing
System (
A
MSR
-
E
)
instrument

on the Aqua satellite provides passive microwave measurements
o
f snow cover
at a
global scale. The sensitivity of the sensor allows it to
penetrate cloud
-
cover

and differentiate
between snow, water, ice, and
other type of
land cover. AMSR
-
E goes a step further
than other
SCA

products
and can
be used to estimate
SWE as

well. However, the spatial resolution of
AMSR
-
E
products is 25 kilometers, which is very coarse when compared to other
.

Even so, the
AMSR
-
E spatial resolution is a major improvement over previous sensors,
such as
the Special
Sensor Microwave/
Imager (
SSM/
I) instrument. The SSM/I instrument used four frequencies;
AMSR
-
E uses six
, and therefore has a wider range of the microwave spectrum to use for
analysis.

Microwave instruments have
coarser
resolution compared to visible and infrared
spectrum instruments
due to physical limitations. Microwave
radiation has a longer wavelength
than visible and infrared radiation
.

An unrealistically sized antenna would be required to obtain
better spatial resolution, and is not possible with current technology. For vast areas such as the
ocean a low resolution is not a serious issue, but in heterogeneous terrain such as mountainous

10


areas, it may present problems. Therefore, AMSR
-
E snow cover would be well suited for a
supplementary role, but the resolution is too poor and insurmountable to be the primary
instrument used for studying snow cover.

Sanjay et al
. 2008

found that
that the
re was good agreement between
MODIS
derived
SCA data

and

ground station data in mountainous terrain and
concluded

that the MODIS data is
can be used as
an effective estimator of
SCA

in the Himalayas region

(Sanjay et al. 2008)
.
However,
the rugged terrain
in the area contributed to a
mountain
shadowing effect (caused by
the low angle of solar illumination in complex terrain
, not to be confused with a mountain rain
shadow effect
) leading to

a less accurate reading of snow
-
cover.

Visible spectrum

cannot be
u
sed during
night
time
hours,

so shadow
ing will

decrease their effectiveness as well.
The study
also compared SCA estimates derived from other sensors as well and found that MODIS
outperformed the other sensors at detecting snow in areas where shadowing was
a problem

(2008)
.
In a study using MODIS
SCA data to examine the relationship between s
now cover in
the Tibetan plateau and the East Asian Summer Monsoon,
Che Tao

and Li
Xi found

that
the
MODIS snow cover product
was
useful and accurate when compar
ed against ground station data

(Che et al. 2004)
.

In
the paper titled,
“Snow depth estimation over north
-
western Indian Himalayas using
AMSR
-
E”, Indrani Das and R. N. Sarwade found that AMSR
-
E was suitable for use in the
Himalayas, although snow depth was

off by
an average error of
20.34 cm due to the coarse
spatial
resolution of AMSR
-
E

(Sarwade et al. 2003)
.

However, the errors were not related to the
question of whether snow was detected

by the sensor
, but were instead related to the snow depth
estimate
s

derived by the sensor
.

A few studies have used
AMSR
-
E
solely

to derive estimates of
SCA, but found that the coarse spatial resolution of the data hindered its use even though it had
11


the ability to penetrate cloud
-
cover

(Comiso, 2003)
.
As a result
, there
have been
numerous
studies that
have
use
d

combined products of AMSR
-
E and
other
high resolution sensors such as
MODIS to
map
snow cover.

MODIS and AMSR
-
E c
omposite products have been developed
to take into account

the
advantages of each sensor to provide a more complete
SCA product
.

This concept is not new and
has been applied
in the past
using
MODIS and SSM/I data
. This combination yielded more
accurate estimates of SCA

than the use of each product alone.
AMSR
-
E
,
however, is

a much
more powerful microwave sensor than SSM/I and has produced even stronger results

(Tait et al.
2000)
.


AMSR
-
E
has the ability to penetrate
cloud
-
cover,

and also
can penetrate the atmosphere
without losing accuracy.

As a supplement to
the high resolution MODIS snow product, AMSR
-
E has been found to
significantly
increase
the
accuracy
of SCA
estimate

(Wang et al. 2008)
.
U
sing a combined MODIS and AMSR
-
E product suppresses cloud cover
effects while

still
providing the user with a
relatively fine spatial resolution

(2008)
.

When v
alidated against
ground stations,
the
combined MODIS and ASMR
-
E
product
show
s

high agreement

with surfa
c
e
observations
,
but the accuracy drops when the snow
cover
is shallow or the snow distribution is
scat
tered spatially

(2008)
.

The mountain
-
shadow effect encountered

in previous studies that used
MODIS to map
SCA might

also be decreased
by using
AMSR
-
E

data
, which operates the same
regardless of
the
brightness.


A combined MODIS and ASMR
-
E
SCA
product
has been found to be
more accurate than
when
either product
is used
alone, but
the integration of
cloud cover removal
may increase the
accuracy even further
.
Relatively few studies have utilized cloud removal algorithms in relation
to MODIS data, but those

that have report excellent results

(Zhengming 2008)
.

Cloud removal
techniques are new and relatively unexplored, and a delicate balance between
predicting the
12


presence of snow under clouds and
maintaining accuracy must be
preserved

(
Bardossy 2009)
.

Using

a combined MODIS and AMSR
-
E product enhanced with an additional cloud cover
removal
technique
could make for an extremely powerful and accurate snow
cover
product in
mountainous regions.

The goal of this proposal is to see if accuracy is increased and us
e tools and techniques
that are replicable without esoteric features. Future research may include refinement of
techniques used and porting of scripts to other GIS and remote sensing programs, such as
IDRISI, ERDAS, or GRASS.












13


Hypothesis

The
composite AMSR
-
E and MODIS snow cover product was designed t
o test the
following hypothesis:

Null Hypothesis
:

The snow and land cover detected with the composite snow cover product and
the standard MODIS snow cover product is not significantly different.

A
lternate Hypothesis

1
:

The composite AMSR
-
E and MODIS product has significantly higher
snow and land cover detected than the standard MODIS snow cover product.













14



Methods:

Study Period:


MODIS and AMSR
-
E snow cover products are both available as far back as 2003. This
proposal will focus on
the winter regime, specifically, January 2006
. MODIS snow cover
products are available daily, 8
-
day composite, or monthly, and AMSR
-
E is available dail
y, 5
-
day
composite, or monthly. Therefore, the composite products will be built using daily snow cover
products of both instruments. Because the concern of this proposal is
a proof of concept
, and not
determining long term trends in the region, the study p
eriod is
January 1
st

through January 8
th
,
2006.


Study Area:


The Colorado River provides
water resources for seven states and Mexico, and
understanding changes in the water availability is essential to people and interested within the
region
.

75 percent o
f the total water within the Colorado River watershed originates as snow in
the Rocky Mountains

(Short 2010)
.




15










Figure 1: Map of study area, shown by red outline


Acquisition

Data from both the AMSR
-
E and MODIS instruments will be obtained from the Earth
Observing System Data and Information System (EOSDIS), using python scripting to
automatically download data and subset it to the study area.
Although t
he MODIS instrument is
l
ocated on the Terra satellite
,

and the AMSR
-
E instrument is located on the Aqua satellite
, snow
products from both instruments

can be subset to the study area

extent

while being acquired.

This
is done using a vector outline of the study area, and then proc
essing the imported HDF files to
have pixels completely within the vector selected and subset.

16




Both AMSR
-
E and MODIS data utilized are level 3 in terms of processing, which
ensures that necessary adjustments have already been made to the datasets. MODI
S derived
snow coverage contains cells with a range of values that indicate either snow, clouds, land,
water, or no data(indicating an error).
In comparison, AMSR
-
E SWE cell values from 1 to 240
indicate millimeters of snow water equivalent. Therefore, any

value between 1 and 240 indicates
presence of snow and is considered an indicator of snow cover, and 0 indicates a lack of snow
cover. Values above 240mm are used to indicate parameters besides SWE, such as detection of a
lake or other large body of water
, and should be disregarded. Figure 2 shows what each value in
a MODIS HDF
snow cover
file corresponds to.












17


Daily MODIS Snow
Cover Coded
Integer Values

Sample
Value

Explanation

0

Data
missing

1

No decision

11

Night

25

No snow

37

Lake

39

Ocean

50

Cloud

100

Lake ice

200

Snow

254

Detector
saturated

255

Fill




Table 1
: MODIS integer values
, from National Snow and Ice Data Center

Processing

Using python scripting within ArcGIS, the following steps will be automated.
The
ArcGIS product suite allows for extensive scripting capabilities and includes a “Model
-
Builder”
tool that makes it easy for users without programming experience to automate tasks within
ArcGIS.
One script
was

devoted to automatically creating the compo
site product.
After
processing an appropriate amount of combined AMSR
-
E and MODIS snow

cover products, the
resulting product was compared to MODIS snow cover product without AMSR
-
E.

18


The script
involve
d

several steps of processing. First, b
oth MODIS and AMS
R
-
E data
must be projected in the same coordinate system. To interpolate MODIS images, the HDF file
was

converted into GRID format at 500m resol
ution. Then, the cell values were

classified, using
the metadata and documentation, so that the MODIS image is c
lassified accord
ing to snow,
cloud, and no snow.

Next, the AMSR
-
E data was

converted into grid format. Using the Mask tool, AMSR
-
E
pixels

(which are

much larger t
han MODIS pixels due to the coarser

resolution) that are
completely within MODIS detected clo
ud cover
are

selected. If an AMSR
-
E pixel is not
completely within MODIS detected cloud cover, it will not be selected

or used in the composite
product
.

Originally during the design of scripting and the project, analysis was intended to include
the use of

a cloud removal algorithm, and so the replacement of MODIS cloud cover with
AMSR
-
E was designed to allow for that, so that the tolerance for non
-
cloud cover pixels was
very low. Any MODIS pixels that are not cloud cover stop AMSR
-
E from replacing nearby
c
loud cover pixels.

If, however, an AMSR
-
E pixel is over an area that is completely cloud cover, AMSR
-
E
replaces that pixel.
Because AMSR
-
E reads through clouds, the encapsulated pixel will represent
either snow or land cover
. It is possible for AMSR
-
E swaths to have errors, but no AMSR
-
E
swaths in the study had errors within the study area. When a
pixel

of A
MSR
-
E
is
within

MODIS

cloud cover
, the AMSR
-
E pixel is

resampled

using Nearest Neighbor analysis to match the 500m
resol
ution of MODIS
. Raster operations use

the values of

selected AMSR
-
E to replace the

19


MODIS cloud cover

detecte
d.

Afterwards,
using raster operations the values are

combined into a
single composite image. The result is a composite MODIS and AMSR
-
E product
.

Fo
llowing the creation of composite MODIS and AMSR
-
E

product, statistical testing was

done
to see if there is a significant replacement of cloud cover with AMSR
-
E sensed land or
snow cover.



Figure
2
: AMSR
-
E and MODIS overlay. For the pixels

in red, cloud cover was detected. The outline of AMSR
-
E
pixels can be seen as well. The bottom pixels are mostly completely within the cloud cover and would be replaced
with AMSR
-
E.







20








Results

Visualization


Below are shown several maps of the
composite imagery produced, as well as maps
showing the non
-
composite imagery.


Figure
3
: 2006 January 1st composite AMSR
-
E and MODIS map.



The next two images compare MODIS snow cover product and a composite snow cover
product over the study area. Visua
lly, it appears as if a dramatic increase in snow cover has
occurred, but the table that follows shows that statistically, the gains are negligible.

21



Figure
4
: January 8th MODIS snow cover



Figure

5
: January 8th composite snow cover

22



Below
in figure 6 is a

chart measuring the snow and land cover remotely sensed by MODIS, as
well as the snow and land cover found with the composite product.
The most noticeable increase is with
January 1
st
, with
a

10% aggregate increase in snow and land cover s
ensed, as shown by Table 2.


Figure

6
: Chart of accuracy between MODIS and Composite product



Table 2
: Percentage of remotely sensed snow and land cover

0%
20%
40%
60%
80%
100%
120%
1
2
3
4
5
6
7
8
Percentage of remotely sensed snow and land

January 1st
-

January 8th

MODIS
Composite
MODIS
Composite
1-Jan
29.08%
39.39%
2-Jan
63.66%
63.66%
3-Jan
97.22%
97.88%
4-Jan
53.12%
53.12%
5-Jan
95.24%
95.24%
6-Jan
83.22%
83.32%
7-Jan
77.67%
78.08%
8-Jan
65.23%
66.16%
23



Table 3
: Percentages of snow, land, and cloud cover between MODIS and composite imagery.


The above table measures the day to day changes in snow cover and land cover by the MODIS
sensor as well as the composite product.
There were several days (January 2
nd
, 4
th
, and 5
th
) that AMSR
-
E
could not be used to replace cloud cover. On January 2
nd
, for

instance, AMSR
-
E coverage did not include
the study area. On January 5
th
,

shown in Figure

7
, no pixels were replaced due to lack of cloud cover.



Figure

7
: January 5th snow cover. No pixels could be replaced.


1-Jan
2-Jan
3-Jan
4-Jan
5-Jan
6-Jan
7-Jan
8-Jan
MODIS
Snow
19.61%
1.20%
4.40%
2.80%
8.68%
3.68%
5.39%
2.68%
Land
9.47%
62.46%
92.82%
50.33%
86.56%
79.54%
72.28%
62.55%
Cloud
70.92%
36.34%
2.78%
46.88%
4.76%
16.78%
22.33%
34.77%
Composite
Snow
22.36%
1.20%
4.61%
2.80%
8.68%
3.74%
5.59%
3.51%
Land
17.02%
62.46%
93.27%
50.33%
86.56%
79.58%
72.49%
62.65%
Cloud
60.61%
36.34%
2.12%
46.88%
4.76%
16.68%
21.92%
33.84%
Replacement?
Yes
No
Yes
No
No
Yes
Yes
Yes
24



Paired T for Composite versus MODIS


N Mean StDev SE Mean

Composite 8 0.7210 0.2035 0.0719

MODIS 8 0.7056 0.2278 0.0805

Difference 8 0.0155 0.0355 0.0126

95% lower bound for mean difference:
-
0.0083

T
-
Test of mean difference = 0 (vs > 0):


T
-
Value = 1.23 P
-
Value = 0.129





Table 4
: Paired T test of means.



Table 4

shows the statistical testing that was done at a significance level of .05

using a paired t
test in MiniTab
, suggesting that there was not a significant increase in accuracy
through the creation of
the composite product.













25



Discussion


Originally this research was intended to answer two independent hypotheses, with the first
regarding the efficacy of a composite product compared to a standard MODIS snow cover

product, and
the second intended to be a comparison of said composite product against a snow cover product using a
cloud removal algorithm. The second hypothesis proved too difficult to achieve with the tools available
with multiple errors prohibiting a s
tatistical analysis of results, leading to a shift towards testing the
composite product alone

and maintaining the experimental design and hypothesis setup at the beginning
of the experiment.

The results
showed that the gains of using the composite produc
t
were not significant and suggest
a flaw in methodology

and approach
, which reflect the challenges of this project
.
Originally, the
processing

of the composite product was meant to make it easier for a cloud removal algorithm to be
applied, which was outs
ide the scope of this undergraduate thesis. This
might have affected the efficacy
of the composite product, as a composite product with a lower tolerance for replacing MODIS cloud
cover could have led to a significant level of increased accuracy.




Accept
ing the null hypothesis does not mean that increasing the accuracy of MODIS with
AMSR
-
E cannot be done, only that the approach and methodology of this study would need to be re
-
examined. For instance, the focus was on first 8 days in January of 2006, and A
MSR
-
E could not be used
during several of those days to replace MODIS pixels, because of the coverage of AMSR
-
E missing the
study area, and for the other days, the lack of cloud cover for AMSR
-
E to replace. In table 3, it is evident
that January 3rd and 5
t
h

have negligible cloud cover. However, January 1
st

had an almost 10% decrease in
cloud cover due to AMSR
-
E, and would likely have been even higher if the methodology had been altered
to be more tolerant of non
-
cloud cover pixels. If the experiment had bee
n set up to test the efficacy of a
composite product when there is a certain amount of cloud cover with an alternate hypothesis suggesting
26


that a composite product significantly increases snow cover of MODIS during periods of high cloud
cover, the design m
ight have shown such a compos
ite product to be significantly effective.


The results of this study do support the conclusion that as a daily product, a composite AMSR
-
E
and MODIS snow cover product would likely not be a significant improvement.


Future R
esearch

More research with a more robust methodology using the lessons learned from this project would
be needed in order to test a composite AMSR
-
E and MODIS product with a different approach. Flaws
with the processing steps were re
alized early on, and al
ternate methods and tools could be explored. In
addition, the results of January 8
th

did not seem accurate during verification, suggesting that a bug in the
scripting process may have skewed results.

A follow up project could be an experimental design tha
t does not attempt to implement a cloud
removal algorithm and instead attempts to maximize the potential of a AMSR
-
E and MODIS composite
product.









27







Acknowledgements







Brian Harshburger, for giving me

professional and personal

direction and guidance
through
out the last two semesters.


Joy Mast, for being my Senior Seminar advisor, and reminding me of deadlines in her
forceful but extremely polite way.


Wenjie Sun, for being my academic advisor and helping me with remote sensing

as well
as helpin
g me attend the AAG conference.












28


Works Cited

Armstrong, E, Chin, M, LaFontaine, F, et al.

An Enhanced MODIS/AMSR
-
E SST composite
product.
” Last accessed December 6.

NASA Earth Science Office.
http://weather.msfc.nasa.gov/sport/conference/pdfs/GHRSST_workshop_paper_Jedlovec_v3.pdf

Balk, B, Elder, K.

Combining Binary Decision Tree and
Geostatistical Methods to Estimate
Snow Distribution in a Mountain Watershed.


Water Resources
. 2000; 36: 13
-
26.

Campbell, James B.

1996.
Introduction to Remote Sensing
. New York: Guilford.

Che, T, Xin, L, Feng, G.

Estimation of Snow Water Equivalent in t
he Tibetan Plateau Using
passive Microwave Remote Sensing Data(SSM/I).


Journal of Glaciology and Geocryology
.
2004.

Comiso, J.C; Cavalieri, D,J, Markust, T.

S
ea ice concentration, ice temperature, and snow depth
using AMSR
-
E data.


Geoscience and Remote
Sensing.

2003
; 41(2): 243
-
252.

Das, I, Sarwade, R. N.

Snow Depth Estimation of north
-
western Indian Himalaya using AMSR
-
E.


International

Journal of Remote Sensing.

2003; 29: 4237


4248.

Dressler, K, Leavesley, G.

Evaluation of gridded snow water equiva
lent and satellite snow
cover products for mountain basins in a hydrologic model.


Hydrological Processes.

2006;
20(4); 673
-
688.


Foster, J, Sun, Chaojiao, Walker, J, Kelly, R, Chang, A, Dong, Jiarui, Powell, H.

Quantifying
the uncertainty in passive
microwave snow water equivalent observations.


Remote Sensing of
Environment.

2005; 94: 187
-
203.

Gafurov, A,

Bardossy,

A.

Cloud
removal

methodology from
MODIS

snow cover product.


Hydrology and Earth System Sciences.

2009
;
13: 1261


1373.

Hall, D, Riggs, G.

Accuracy assessment of the MODIS snow products.


Hydrological
Processes.

2007; 21: 1534
-
1547.


Harshburger, B, Humes, K, Walden, V, Blandford, T, Moore, B, Dezzani, R.

Spatial
Interpolation of Snow Water Equivalency Using Surface Obser
vations and Remotely Sensed
Images of Snow
-
Covered Area.


Hydrological Pr
ocesses.

2010; 24: 1285


1295.


Klein, A. G, Barnett, Ann, C.

Validation of daily MODIS snow cover maps of the Upper Rio
Grande River Basin for the 2000
-
2001 snow year.


Remote Sens
ing of Environment.

2003; 86:
162
-
176.

29


Lee, S., Klein, A. G., Over, T. M.

A comparison of MODIS and NOHRSC snowcoverproducts
for simulating streamflow using the Snowmelt Runoff Model.


Hydrological Processes.

2005;
19(15): 2951
-
2972.


Maurer, E, Rhoads, J, Dubayah, R, Lettenmaier, D.

Evaluation of the snow
-
covered area data
product from MODIS.


Hydrological Processes.

2003; 17(1): 59
-
71.

National Snow and Ice Data Center. MOD10A2 and MYD10A2 Local Snow Cover Attributes,
Version 5[Inter
net]. Available from:


http://nsidc.org/data/docs/daac/mod10_modis_snow/version_5/mod10a2_local_attributes.html


Sanjay, J, Ajanta, G, Saraf, A.

Accu
racy assessment of MODIS, NOAA and IRS data in snow
cover

mapping under Himalayan conditions.


International Journal of Remote Sensing.

2008;
29(20).


Short, David.

Effect of TRMM Orbit Boost on Ra
dar Reflectivity Distributions.


American
Meteorological
Society
.

2
010
; 27(7).


Tait,

A.,B,

Hall,

D.

K
,

Armstrong

R.L
.


Utilizing multiple datasets for snow
-
cover mapping.


International

Journal

of

Remote

Sensing
.

2000;
22
(
17
):
3275
-
3284
.



Wan, Zhengming.

New refinements and validation of the MODIS Land
-
Surface
Temperature/Emissivity products.


Remote sensing of Environment.

2008; 112: 59
-
74.

Wang, X, Xie, Hongjie, Liang, T.

Evaluat
ion of MODIS snow cover and clo
u
d

mask and its
application in Northern Xinjiang, China.


Remote Sensing of Environment.

2008; 112(4): 1497
-
1513.

Xiaobing Z, Hongjie X, Jan H.

Statistical evaluation of remotely sen
sed snow
-
cover products
with constraints from streamflow.


Remote Sensing of Environment.

2005; 94: 214
-
231.


Yang, D., Robinson, D., Zhao
, Y, Armstrong, R, Brodzik, M.J.

Streamflow response to seasonal
snow cover extent chang
es in large Siberian
watersheds.


Journal of Geophysical Research.

2007; 112.