Urban Water Management

Urban Water Management
2008 Louisville, KY
Accurate Rainfall Data
How Unusual Gage Data Can
Impact Rainfall Data In
Hydrologic Models
Jim Moffitt and Ilse Gayl
Topics to Cover
• Gage-adjusted Radar Rainfall Datasets
• What we hope to see when putting
together radar and gage datasets
• Detecting problems with gage data
• Handling unusual gage data
• Gage data effects on radar rainfall
estimates
Radar Rainfall Datasets
• Our challenge is to put the right rainfall in
the right place at the right time
• Combining the strengths of two rainfall
measurement systems: gages and radar
• Radar provides excellent spatial and
temporal resolution
• Gages can provide excellent volume
measurements – the “ground truth”
Who Needs These Datasets?
• Applications of historical radar rainfall
estimates…
– Modeling & infrastructure design
– Litigation – reconstructing history
• Real-time applications too…
– Wastewater & stormwater management
– Flood warning operations
Calibrating Radar With Gages
• Provides the highest quality rainfall
estimates today – high spatial and
temporal resolution
• Can increased resolution help refine
models?
• Economic benefits of using gageadjusted radar rainfall estimates?
Rainfall is highly non-uniform
Rainfall through radar and gage eyes
Reality
Radar view
Gages view
Gage-adjusted Radar Process
Gage-adjusted Radar Process
• Correlations between gages and colocated radar pixels are used to adjust
the rest of the radar “surface.”
• An averaging process
– Averages across space and across time
– Individual gage-radar pixel pairs are not
optimized, but as a group they are.
Detecting Unusual Gage Data
• Time-series analysis comparing
radar/gage accumulations
• Monthly and event volume correlations
between gages and radar pixels
• Nearest-neighbor analysis
• Site maintenance and local knowledge
What we hope to see…
• Agreement as a group between gage and
radar accumulation plots.
• Strong correlation between gage and
radar volumes – monthly and individual
events.
Atlanta – March 2001 – 31 gages
12
10
Rainfall (inches)
8
Gages
6
Radar
4
2
0
3/2/2001
3/7/2001
3/12/2001
3/17/2001
Date
3/22/2001
3/27/2001
Single gage/radar pixel
PRCRG04
12.0
10.0
Rainfall (inches)
8.0
PRCRG04
Radar
6.0
4.0
2.0
0.0
2/28/2001 3/5/2001 3/10/2001 3/15/2001 3/20/2001 3/25/2001 3/30/2001 4/4/2001
Date
Monthly totals
14
12
y = 1.0084x
R2 = 0.9814
Radar Rainfall (inches)
10
Radar
AdjRadar
1 to 1 Line
Linear (AdjRadar)
8
6
4
2
0
0
2
4
6
8
Gage Rainfall (inches)
10
12
14
Individual event: March 19, 2001
3.0
y = 0.9887x
R2 = 0.946
Radar Rainfall (inches)
2.5
2.0
1.5
Radar
AdjRadar
1 to 1 Line
Linear (AdjRadar)
1.0
0.5
0.0
0.0
0.5
1.0
1.5
2.0
Gage Rainfall (inches)
2.5
3.0
Gage Data That Can Mislead
• Plugged gages
• Rain-turning-to-snow events
• Gages with no correlation to radar
(sprinklers, other interference)
• Timing and resolution disconnects
between radar and gage data
Plugged gage
19
Rainfall volume mixed across
two rainfall events.
St. Louis, MO, October 2004
6.000
5.000
4.000
19
Radar
AdjRadar
3.000
2.000
1.000
0.000
09-25
09-30
10-05
10-10
10-15
10-20
10-25
10-30
11-04
Rain-turning-to-snow event
9.000
8.000
7.000
Rainfall (inches)
6.000
5.000
Gages
Radar
AdjRadar
4.000
3.000
2.000
1.000
0.000
11/1
11/11
11/21
12/1
Date
12/11
12/21
12/31
Very poor correlation
Salt Lake City, UT, July 2007
7.0
6.0
5.0
Galena Gulch Gage
4.0
Rain gage
Radar
Radar
3.0
2.0
1.0
0.0
07-25
07-26
07-27
07-28
07-29
“Hmm… I haven’t see that one before…”
0.65
0.60
0.55
0.50
Rainfall (inches)
0.45
0.40
0.35
Gages
Radar
0.30
0.25
0.20
0.15
0.10
0.05
0.00
1/8
1/13
1/18
1/23
Date
1/28
Volume Outliers
14
Radar Rainfall (inches)
12
10
8
6
Radar
1 to 1 Line
4
2
0
0
2
4
6
8
Gage Rainfall (inches)
10
12
14
Nearest Neighbor Analysis
Good performance
Irrigation problem
Underreporting
Handling Strange gage Data
• Questions about gage volume are the
most serious.
• Timing is important too – if too off we
may be overlapping individual events
and/or losing volume to evaporation.
• Timing is critical in real-time
applications.
Handling Strange Gage Data
• Removing gages from analysis – entire
study period or individual events.
• For historical analyses, re-distributing
volumes if possible.
• Providing clients feedback on gage
performance to enhance future results.
• The influence a single gage can have on
broader results – right or wrong?
Gage Data Effects
• Fort Collins, CO – August 2, 2007.
• Gage 6490 appeared to be “high gage”, client thought otherwise.
• Including gage 6490 increased storm 6-hour accumulation at maximum radar
pixel by 0.47 inches.
Take home messages
• Rainfall data quality will always depend on
having high-quality data from as many wellsituated rain gages as are feasible.
• Gage maintenance is extremely important.
• Input rainfall data quality and completeness
can have the largest impact on final estimates.
• Rain gage data are used as “ground truth,”
must be accurate or the “truth” suffers, along
with everything else based upon it.