Eastern Region Flash Flood Conference
June 2 - 4, 2010
Wilkes-Barre, Pennsylvania
Flood and discharge monitoring during the 2008
Iowa flood using AMSR-E data
Authors:
Marouane
Teodosio Lacava2, Tarendra Lakhankar1, Valerio
Tramutoli4, Hosni Ghedira3, Reza Khanbilvardi1
Temimi*1,
1NOAA-CREST,
City University of New York, 160 Convent Avenue, New York,
NY, 10031, USA
2Institute of Methodologies for Environmental Analysis (IMAA) - National
Research Council (CNR), C.da Santa Loja, 85050 Tito Scalo (PZ) - Italy
3American University in Dubai, Dubai, UAE
4Department of Engineering and Physics of Environment (DIFA) - University
of Basilicata – via dell’Ateneo Lucano, 10, 85100 Potenza - Italy
June 3, 2010
temimi@ce.ccny.cuny.edu
The flood event in Iowa
http://www.flood2008.iowa.gov & http://www.iowaflood.com
Several regions particularly in Iowa
were affected by a 500 year flood
which was classified as the worst in
the history of the region. Major
damages and large inundated areas
related to these floods have been
recorded.
June 3, 2010
temimi@ce.ccny.cuny.edu
Study area delineation and objective
The objective of this
work is to demonstrate
the potential of using
passive microwave
data in monitoring
flood and discharge
conditions
June 3, 2010
temimi@ce.ccny.cuny.edu
PR used for flood monitoring in Canada
This study is an expansion of a previous work in which passive
microwave data have shown an interesting potential for flood
monitoring
The water surface fraction derived from
visible images does not include the effect of
soil wetness, and provides only an estimate of
the waterbody’s extent.
On the other hand, passive
microwave is a combination of
waterbodies, flooded area and soil
moisture responses
June 3, 2010
temimi@ce.ccny.cuny.edu
WSF(MODIS) = ow + f
WSF(AMSR-E) = ow + f + sm
Basin Wetness Index
a)
b)
WSF(MODIS)
WSF(MODIS)
WSF(AMSR-E)
WSF(AMSR-E)
c)
d)
WSF(MODIS)
WSF(AMSR-E)
WSF(MODIS)
WSF(AMSR-E)
Different configurations of inundated and wet soils as sensed by MODIS and AMSR-E; a) non flooded area and wet soil; b) large flooded area
and wet soils; c) non flooded area and large wet soil extent; d) limited wet soil extent and large flooded area
June 3, 2010
temimi@ce.ccny.cuny.edu
Definition of a new BWI
We propose, therefore, to define a Basin Wetness Index (BWI) based on
the difference between the passive microwave and visible responses.
WSF ( AMSR)  WSF ( MODIS )
BWI 
1  WSF ( MODIS )
BWI varies between 0 and 1. It represents the fraction of wetlands
within the non-flooded area.
WSF(MODIS) ??
WSF ( AMSR)  a.Q b (t )
BWI 
1  a.Q b (t )
Temimi, M., R. Leconte, F. Brissette, N. Chaouch. (2007). Flood and Soil Wetness Monitoring Over the Mackenzie River Basin Using AMSR-E
37 GHz Brightness Temperature. Journal of Hydrology. Vol. 333 (2). p. 317.
June 3, 2010
temimi@ce.ccny.cuny.edu
Leconte et al. 2001
Polarization Ratio Variation Index (PRVI)
Precipitation (in)
Observed PR
Monthly PR mean
Monthly PR + Monthly Std
0.04
PRVI 
3
0.035
PR  i
i
2.5
0.03
0.02
1.5
May
March
April
0.015
June
0.01
1
July
August
0.5
0.005
0
3/17/2008
0
4/17/2008
5/17/2008
6/17/2008
7/17/2008
8/17/2008
Date
PRVI Measures PR anomalies and minimizes surface
conditions effects
Marouane Temimi, Teodosio Lacava, Tarendra Lakhankar, Hosni Ghdira, Reza Khanbilvardi, Trumatoli, V. Flood and
discharge monitoring during the 2008 flood In Iowa using AMSR-E data. Hydrological Processes. (Under Review)
June 3, 2010
temimi@ce.ccny.cuny.edu
Precipitation (in)
Polarization Ratio (PR)
2
0.025
Where, μi and σi
are average and
standard deviation
of the PR=(TbvTbh)/(Tbv+Tbh)
respectively for a
given month i.
Average and
standard deviation
were estimated on
a monthly basis to
account for
changes in
surface conditions
(i.e soil roughness
and vegetation
density) which
might affect the
microwave signal.
PRVI vs PR
Areal average PR
Areal average PRVI
50.00
0.035
45.00
0.03
40.00
0.025
35.00
0.02
30.00
0.015
25.00
0.01
20.00
0.005
15.00
0
10.00
-0.005
Precipitation (mm)
PR and PRVI/100
Total precipitation
0.04
5.00
-0.01
3/17/08
0.00
3/31/08
4/14/08
4/28/08
5/12/08
5/26/08
6/9/08
6/23/08
7/7/08
7/21/08
8/4/08
8/18/08
Date
Calculated PR and PRVI compared to total precipitation recorded across the study area
PRVI shows a higher sensitivity to soil
moisture variation with respect to the PR
June 3, 2010
Observed total precipitation at 8 sites in May, June and July 2008
and their corresponding Thiessen Polygons.
temimi@ce.ccny.cuny.edu
PRVI vs water level
a)
b)
c)
PRVI values obtained on June 9th, 2008 in a) compared to AMSR-E soil moisture product (g/cm3) in b)
and observed water levels above flood stage as provided by the USGS (black triangle) in c)
(http://water.usgs.gov/osw/)
June 3, 2010
temimi@ce.ccny.cuny.edu
June 3, 2010
temimi@ce.ccny.cuny.edu
Observed discharge
800000
Q St Louis
Q Clinton
Q Missourri
Q Illinois
Obtained Q at St Louis
700000
600000
Discharge (cfs)
500000
400000
300000
200000
100000
0
3/3/2008
3/23/2008
4/12/2008
5/2/2008
5/22/2008
6/11/2008
7/1/2008
7/21/2008
8/10/2008
Date
June 3, 2010
temimi@ce.ccny.cuny.edu
8/30/2008
PRVI
4
250000
3
200000
2
150000
1
100000
0
50000
-1
0
-2
30-Aug-08
23-Aug-08
16-Aug-08
9-Aug-08
2-Aug-08
26-Jul-08
19-Jul-08
12-Jul-08
5-Jul-08
28-Jun-08
21-Jun-08
14-Jun-08
7-Jun-08
31-May-08
24-May-08
17-May-08
10-May-08
3-May-08
26-Apr-08
19-Apr-08
12-Apr-08
5-Apr-08
temimi@ce.ccny.cuny.edu
June 3, 2010
29-Mar-08
22-Mar-08
15-Mar-08
8-Mar-08
1-Mar-08
Date
A consistent
time lag has
been observed
300000
Discharge (cfs)
PRVI vs observed discharge downstream
5
350000
Observed discharge
PRVI
Rating Curve
 The lag term introduction
Q(t)= a PRVIb(t)
Q(t)= a PRVIb(t-d.Δt)
The lag term ``d`` will maximize the crosscorrelation function between
discharge observations and FA vectors
d<0
WSF
Q
June 3, 2010
d>0
d=0
WSF
WSF
Q
temimi@ce.ccny.cuny.edu
Q
 Use of the Kalman filter
Log (Q(t)) = log(a) + b log ( PRVI(t+ d.Δt))
Y=AX + B
Yt = Ht. At
where Yt = Y
A
At+1 = Φt At +Wt
Yt = Ht At+ Vt
(State equation)
(Observation equation)
At =
Ht =
B
X 1
With the Kalman filter, the dynamic rating curve model
continuously readjusts its parameters to satisfy the non-stationary
behavior of hydrological processes. The model is thus sufficiently
flexible and adapted to various conditions.
June 3, 2010
temimi@ce.ccny.cuny.edu
Correlation vs lag term
The best crosscorrelation value is
obtained when the
time lag is 21 days
Note that lag values
have been varying
through the summer
(before, during and
after the flood) as
discharge magnitude
and land surface
conditions vary
June 3, 2010
temimi@ce.ccny.cuny.edu
Discharge vs lag term
The time lag increases
in absolute value as the
discharge increases
A higher discharge
means larger inundated
area and therefore a
longer drainage time
The time lag is
however limited by
the time of
concentration
June 3, 2010
temimi@ce.ccny.cuny.edu
Temporal variation of the b coefficient in
the rating curve formula In hydrology, this parameter when
used in a relationship between
discharge and effective width
(Smith et al. 2008) is an indication
of the geomorphology of the river
and the characteristics of its crosssection.
Smith and Pavelsky, 2008
The coefficient ‘’b’’ converges towards a constant value. This conclusion
corroborates results by Smith et al. 2008 which demonstrated that ‘’b’’ when
estimated over several cross-sections converge towards a constant value. In this
study, the temporal variation, like the variation in space studied by Smith et al.
2008, reveals that ‘’b’’ is constant too.
June 3, 2010
temimi@ce.ccny.cuny.edu
Estimated vs Observed discharge
400000
Observed discharge
Estimated Q (no adjustement)
Estimated Q (with adjustment)
350000
300000
250000
200000
150000
100000
50000
0
8/29/08
8/22/08
8/15/08
8/8/08
8/1/08
7/25/08
7/18/08
7/11/08
7/4/08
6/27/08
6/20/08
6/13/08
6/6/08
5/30/08
Comparison between estimated and observed discharge at Saint Louis station
June 3, 2010
temimi@ce.ccny.cuny.edu
Conclusions…
•PRVI detects anomalies in soil moisture and can therefore be used
in flood and discharge monitoring
•A time lag term must be introduced in the rating curve formula
•The lag between flooded area extent and discharge downstream
seems to be sensitive to discharge magnitude and surface
conditions
•The time lag seems to be compatible with the time of concentration
of the study area
•The ‘’b’’ coefficient of the rating curve model when readjusted in
time converges towards a constant value
June 3, 2010
temimi@ce.ccny.cuny.edu