October 11-15, 2010 Hamburg, Germany
Emad Habib1(*), Alemseged T. Haile1, Mohamed Elsaadani1, Mohamed ElShamy2, Robert Kuligowski3, & Yudong Tian4 Civil Engineering, University of Louisiana at Lafayette, LA, USA; 2Nile Forecasting Center, Cairo, Egypt; 3NOAA/NESDIS/CSTAR; 4NASA/GSFC * Corresponding Author,
[email protected], ph.: 337-482-6513
• This study reports on validation of two high-resolution satellite rainfall products: TRMM 3B42 3-hourly 0.25° product, and CMORPH 8-km 30-minute product. • The two products are assessed in terms of their ability to re-produce spatial variability in monthly and diurnal rainfall cycles across the full Nile Basin domain in Africa. • The CMORPH product is also used to drive a basin-wide hydrological forecasting model (Nile Forecasting System) and compared to simulations based on IR-only rainfall product operationally used by the Nile Forecasting Center in Egypt. • Finally, the CMORPH product is evaluated at its finest resolution using a dense independent rain gauge network in south Louisiana in the US, focusing on hydrologically-relevant validation attributes.
Diurnal cycle of rainfall in Lake Victoria basin, source of White Nile 1.25
Rainfall (mm h -1)
1.25
8
9
10
11
Rainfall (mm)
Rainfall (mm)
Malakal, Sudan
1
2
3
4
5
6 7 Month
Rainfall (mm h -1)
8
9
10
11
12
200
0
0.5
0
0
0
3
4
5
6 7 Month
8
9
10
11
12
100
6
9 12 15 Local time (h)
8
9
10
11
12
Gambela, Ethiopia
Juba, Sudan
500
400 300 200
0
1
2
3
4
5
6 7 Month
200
8
9
11
12
Monthly Rainfall Mean
600
100
10
IR-only
Dongala (IR)
1.5
Jinja, Uganda 500 1
2
3
4
5
6 7 Month
8
9
10
11
12
Rainfall (mm)
0
Monthly Rainfall Mean
600 Kitega, Burundi
400
0 60 4 x 10
120
180
240
300
400 300
2
3
4
5
6 7 Month
Monthly Rainfall Mean
600 Kisumu, Kenya
200
1
8
9
10
11
12
Discharge (m 3 s-1)
Rainfall (mm)
100
2
60
1.5
20
Dongala
IR+Gauge (IR+Gauge)
100 360
0
3
3.5
80
0 4 60 x 10
120
180
240
300
2.5
100 360
20
2.5 40
2 Dongala
1.5
1 80
0
60
120 180 240 Julian Day
300
40
CMORPH Dongala +Gauge
1.5
60
CMORPH (CMORPH)
1
2
60
(CMORPH+Gauge) 80
0.5
100 360
0
0
0
60
120 180 240 Julian Day
300
2
3
4
5
6 7 Month
8
9
10
11
12
Rainfall (mm)
1
18
21
24
Khartoum IR-only (IR)
10000
15000
Observed Simulated
20 10000
40
40
IR+Gauge Khartoum (IR+Gauge)
60 5000
60
5000 80
0
60
120
180
240
300
80
100 360
0
0
0
60
120
180
240
300
15000
Khartoum (CMORPH)
CMORPH 5000
12000
0
10000
20
10000
20
10000
40
CMORPH Khartoum (CMORPH+Gauge) +Gauge
60
60
5000 80
Observed Simulated
8000 6000
0
60
120 180 240 Julian Day
400
300
0
0
60
120 180 240 Julian Day
300
40
80
2000 0
60
120
180
240
8000
40
6000 60
IR-only
4000
Diem IR+Gauge (IR+Gauge)
4000
20
60 80
2000
100 0 360 0 12000 0
300
10000
60
120
180
240
300
10000
60
CMORPH Diem 6000 +Gauge (CMORPH+Gauge)
80
2000
40
6000
Diem CMORPH (CMORPH)
4000 2000 0
60
120 180 240 Julian Day
40 60
4000
0
100 360
300
100 360 0 20
8000
8000
0
80
100 360
Diem (IR)
12000
100 360
20
40
0
0
0
20 10000
12000 0
20
0
0
9 12 15 Local time (h)
100 360
500
100
6
Streamflow simulations performed using the Nile Forecasting System (NFS) model for 2007. Both the IR-based and CMORPH rainfall were adjusted using the same monthly weights fixed by NFS.
0
3
20
40 60
0.5
0
500
0
3
0
Observed Simulated
1
15000
200
40
80
0
300
x 10
3
1
15000
Discharge (m 3 s-1)
300
0
24
2.5
0.5
100
3.5
20
2
0
400
21
2.5
3.5
Discharge (m 3 s-1)
Rainfall (mm)
Monthly Rainfall Mean
18
0
3
0
500
600
Rainfall (mm)
Monthly Rainfall Mean
600 6 7 Month
CMORPH TRMM-3B42 Gauge
4
x 10
0.5
5
24
Evaluation of satellite rainfall for streamflow simulations in the Nile Basin 3.5
3
21
0.5
0.25
Nekemte, Ethiopia
2
18
0.75
1
Monthly Rainfall Mean
1
9 12 15 Local time (h)
24
1.5
4
100
4
21
CMORPH TRMM-3B42 Gauge
300
300
3
6
1
400
200
2
3
1.25
2
400
1
18
12
Monthly Rainfall Mean
0
0
2.5
500
500
9 12 15 Local time (h)
300
600
600
6
Discharge (m 3 s-1)
6 7 Month
3
Discharge (m 3 s-1)
5
0
MAP (mm day -1)
4
0.5
0
MAP (mm day -1)
3
0.75
0.25
MAP (mm day -1)
2
CMORPH TRMM-3B42 Gauge
MAP (mm day -1)
1
1
400
0
0
24
0.25
0
100
100
21
0.5
Gondar, Ethiopia
200
200
18
MAP (mm day -1)
300
9 12 15 Local time (h)
CMORPH TRMM-3B42 Gauge
MAP (mm day -1)
400
6
0.75
Discharge (m 3 s-1)
Rainfall (mm)
Khartoum, Sudan
3
1
Monthly Rainfall Mean
500
0
Rainfall (mm h -1)
0
Rainfall (mm h -1)
500
0.5
1.25
600
Rainfall (mm)
Monthly Rainfall Mean
600
0.75
0.25
Seasonal cycle of rainfall across the Nile basin: comparison against long-term rainfall climatology Box plots are based on GHCN-monthly rain gauge (1950-present). Synmbols on the left of each box refer to TRMM-3B42 (19982008) ; symbols on right refer to CMORPH (2003-2008).
Analysis period: April 1998-2008 (TRMM-3B42), April 2003-2008 (CMORPH), and April 1969–1976 (gauge).
CMORPH TRMM-3B42 Gauge
1
Rainfall (mm h -1)
1Dept.
80
0
60
120 180 240 Julian Day
300
100 360
100 360
300
Evaluation at fine time-space resolution using dense gauge network and MPE radar-based product in Louisiana, US
200 100 0
1
2
3
4
5
6 7 Month
8
9
10
11
12
800
Analysis period: (Aug 2004 – Dec. 2006). The 4x4 km2 MPE and the gauge data were re-sampled to the 8x8 km2 CMORPH resolution
Based on historical Gauge (Camberlin, 2009) Rain gauges
Produced using CMORPH
600
Rainfall depth (mm)
Diurnal cycle along a latitudinal cross section (15o N)
21.83 % 17.16 %
Analysis of Bias
400 6.24 %
200
3.09 %
0.04 %
1.97 %
0 -0.69 %
-200
-3.09 %-1.93 % -1.93 % Hit Bias
-400
Missed Rain False Rain
-600
Total Bias
-21.10% -800
CMORPH_Rs
-21.43 %
MPE_Rs
100
CMORPH_MPE
4
90
Rainfall (mm hr -1)
1
0.5
0
0
3
6
9 12 15 Local time (h)
18
21
1.5
3
6
9 12 15 Local time (h)
18
21
0.5
0
0
3
6
9 12 15 Local time (h)
18
21
24
Rainfall (mm hr -1)
Rainfall (mm hr -1)
CMORPH TRMM-3B42 Rain Gauge
1.5
1
0.5
0
0
0
3
6
9 12 15 Local time (h)
3
6
9 12 15 Local time (h)
18
21
24
18
21
24
CMORPH TRMM-3B42 Rain Gauge
1.5
1
0.5
0
0.5
Falsely Detected Rain Depth (% of total rain)
CMORPH on MPE MPE on Rs
20 15
10
10
5 0
1
2
3 -1
Threshold (mm h )
5 0
0
1
2
3 -1
Threshold (mm h )
CMORPH and an operational IR-only product resulted in relatively poor performance of the NFS rainfall-runoff model. Significant improvements in model performance were gained by adjusting satellite-rainfall inputs using gauge data. In Blue Nile basin, the relative volume error of simulated flow decreased from -40% when using CMORPH instead of IR-estimates as a model input. However, merging either product with real-time gauge data resulted in nearly equivalent performance.
1
Rainfall (mm hr -1)
Rainfall (mm hr -1)
1.5
3
Over Lake Victoria basin, both products successfully identified regions that have morning rain rate maxima and those that receive their maximum in late-evening to mid-night local times. However, products significantly underestimated the rain rate at the middle of the lake.
2
CMORPH TRMM-3B42 Rain Gauge
2 CMORPH on Rs
25
15
20
1
CMORPH performed better than TRMM-3B42 in terms of capturing diurnal cycle of rain rate over Lake Tana basin (source of Blue Nile). The major limitation is over the mountains and the Lake shore. TRMM-3B42 extremely underestimated rain rates in this basin.
24
2
16
0
1
0 0
12
0
3
Both products capture the overall patterns of rainfall climatology over the Nile Basin. Over basin areas north of the equator, monthly rainfall amounts are mostly overestimated by CMORPH and underestimated by TRMM-3B42. The agreement is better around the equator.
0.5
0
8
2
20
Rainfall threshold (mm h-1)
2
1
CMORPH MPE Rs
-2
10
1
25
0
CMORPH TRMM-3B42 Rain Gauge
2
1.5
1
0
4
2
24
CMORPH TRMM-3B42 Rain Gauge
50
2
Conclusions Rainfall (mm hr -1)
Analysis Period: JuneAug, 2007
10
Missed Rain Depth (% of total rain)
Probability of exceedance
Diurnal cycle of rainfall in Lake Tana basin, source of Upper Blue Nile 1.5
60
Detection Analysis
30
Probability of exceedance
-1
CMORPH TRMM-3B42 Rain Gauge
70
40
Produced using TRMM-3B42
2
3 POFD (%)
POD (%)
80
0
3
6
9 12 15 Local time (h)
18
21
24
•Evaluation of CMORPH at fine space-time scales showed that: CMORPH has very negligible total bias (0.04%) but its missed-rain and false-rain biases are significantly large (-21.10 and 21.83 %). CMORPH has poor rainfall detection capability in the study area particularly for light rain rates (< 2 mm h-1). Correlation between CMORPH estimates and gauge data increased from 0.4 at hourly scale to 0.6 at a daily scale. Similarly, the relative RMSE decreased from 840 % of mean rain rate at hourly scale to ~200% at 2-days time scale. Using radar based MPE as reference results in favorable accuracy of CMORPH than its actual accuracy in terms of POFD and falsely detected rain depth. It results in less favorable accuracy in terms of total bias and hit bias
Acknowledgment Funding provided by the National Science Foundation (NSF) through US-Egypt Cooperative Research Program (Award Number 0914618)