Validation of NWS Stage IV Multi- Sensor Precipitation Estimates …exh5102/Assets/... · 2008. 10....
29
Validation of NWS Stage IV Multi- Sensor Precipitation Estimates (MPE) in Mid-Atlantic Region Validation of NWS Stage IV Multi Validation of NWS Stage IV Multi - - Sensor Precipitation Estimates (MPE) Sensor Precipitation Estimates (MPE) in Mid in Mid - - Atlantic Region Atlantic Region Boone F. Larson 1 , Ali Tokay 2 , Emad Habib 1 , Brian R. Nelson 3 1 Department of Civil Engineering University of Louisiana at Lafayette - Lafayette, LA 2 JCET-University of Maryland Baltimore County NASA Goddard Space Flight Center, Greenbelt, MD 3 NOAA National Climatic Data Center 151 Patton Ave, Asheville, NC 28801 USA Boone F. Larson Boone F. Larson 1 1 , Ali Tokay , Ali Tokay 2 2 , Emad Habib , Emad Habib 1 1 , Brian R. Nelson , Brian R. Nelson 3 3 1 1 Department of Civil Engineering Department of Civil Engineering University of Louisiana at Lafayette University of Louisiana at Lafayette - - Lafayette Lafayette , LA , LA 2 2 J J CET CET - - University of Maryland Baltimore County University of Maryland Baltimore County NASA Goddard Space Flight Center, Greenbelt, MD NASA Goddard Space Flight Center, Greenbelt, MD 3 3 NOAA National Climatic Data Center NOAA National Climatic Data Center 151 Patton Ave, Asheville, NC 28801 USA 151 Patton Ave, Asheville, NC 28801 USA
Transcript of Validation of NWS Stage IV Multi- Sensor Precipitation Estimates …exh5102/Assets/... · 2008. 10....
Validation of NWS Stage IV Multi-Sensor Precipitation Estimates (MPE)
in Mid-Atlantic Region
Validation of NWS Stage IV MultiValidation of NWS Stage IV Multi--Sensor Precipitation Estimates (MPE) Sensor Precipitation Estimates (MPE)
in Midin Mid--Atlantic RegionAtlantic Region
Boone F. Larson1, Ali Tokay2, Emad Habib1, Brian R. Nelson3
1Department of Civil EngineeringUniversity of Louisiana at Lafayette - Lafayette, LA
2JCET-University of Maryland Baltimore CountyNASA Goddard Space Flight Center, Greenbelt, MD
3NOAA National Climatic Data Center151 Patton Ave, Asheville, NC 28801 USA
Boone F. LarsonBoone F. Larson11, Ali Tokay, Ali Tokay22, Emad Habib, Emad Habib11, Brian R. Nelson, Brian R. Nelson33
11Department of Civil EngineeringDepartment of Civil EngineeringUniversity of Louisiana at Lafayette University of Louisiana at Lafayette -- LafayetteLafayette, LA, LA
22JJCETCET--University of Maryland Baltimore CountyUniversity of Maryland Baltimore CountyNASA Goddard Space Flight Center, Greenbelt, MDNASA Goddard Space Flight Center, Greenbelt, MD
33NOAA National Climatic Data CenterNOAA National Climatic Data Center151 Patton Ave, Asheville, NC 28801 USA151 Patton Ave, Asheville, NC 28801 USA
ObjectivesObjectivesObjectives
• Validation of MPE with independent rain gauge observations. Three-fourths of the gauges are at the coast.
• Pave the way for further studies regarding comparison of NEXRAD MPE with satellite rainfall products including NASA’s Tropical Rainfall Measuring Mission (TRMM) Multi-sensor Precipitation Analysis (TMPA)
•• Validation of MPE with independent rain gauge Validation of MPE with independent rain gauge observations. Threeobservations. Three--fourthsfourths of the gauges are at of the gauges are at the coast. the coast.
•• Pave the way for further studies regarding Pave the way for further studies regarding comparison of NEXRAD MPE with satellite rainfall comparison of NEXRAD MPE with satellite rainfall products including NASAproducts including NASA’’s Tropical Rainfall s Tropical Rainfall Measuring Mission (TRMM) MultiMeasuring Mission (TRMM) Multi--sensor sensor Precipitation Analysis (TMPA) Precipitation Analysis (TMPA)
Study Site and DataStudy Site and DataStudy Site and Data
• Study site: Eastern Virginia, Delmarva, and northeast North Carolina.
• Gauge Network consists of 28 gauge sites maintained by TRMM satellite validation office.
• Each site contains dual or triple gauges.
• Each gauge site contained approximately 1-2 years of data. All gauge data spans from mid ‘04 - early ’07
• Precipitation occurs mostly by tropical cyclones, convective thunderstorms, and frontal systems
•• Study site: Eastern Virginia, Delmarva, and northeast Study site: Eastern Virginia, Delmarva, and northeast North Carolina.North Carolina.
•• Gauge Network consists of 28 gauge sites maintained by Gauge Network consists of 28 gauge sites maintained by TRMM satellite validation office.TRMM satellite validation office.
•• Each site contains dual or triple gauges.Each site contains dual or triple gauges.
•• Each gauge site contained approximately 1Each gauge site contained approximately 1--2 years of 2 years of data. All gauge data spans from mid data. All gauge data spans from mid ‘‘04 04 -- early early ’’0707
•• Precipitation occurs mostly by tropical cyclones, Precipitation occurs mostly by tropical cyclones, convective thunderstorms, and frontal systemsconvective thunderstorms, and frontal systems
Rain Gauge Network MapRain Gauge Network MapRain Gauge Network Map
Each gauge site is Each gauge site is within its own MPE within its own MPE pixel with exception of pixel with exception of Wallops Island. At the Wallops Island. At the island, 2 MPE pixels island, 2 MPE pixels each contain 3 gauges each contain 3 gauges while another MPE while another MPE pixel contains 1 gauge.pixel contains 1 gauge.
Rain Gauge Data PeriodsRain Gauge Rain Gauge Data PeriodsData Periods
Rain GaugesRain GaugesRain Gauges
• Met One tipping bucket
• Resolution = 0.01 inches
• Maintenance- once a month
• Shortcomings:
– Snow melt tips
– Pivotal point failure
– Battery failure
•• Met One tipping bucketMet One tipping bucket
•• Resolution = 0.01 inchesResolution = 0.01 inches
•• MaintenanceMaintenance-- once a monthonce a month
•• Shortcomings:Shortcomings:
–– Snow melt tipsSnow melt tips
–– Pivotal point failurePivotal point failure
–– Battery failureBattery failure
Petersburg, VA
Multi-sensor Precipitation Estimates (MPE)MultiMulti--sensor Precipitation Estimates (MPE)sensor Precipitation Estimates (MPE)
• MPE ~ merging operational radar, automated gauge, as well as geo-stationary satellite rainfall following manual quality control and bias adjustment.
• Stage IV ~ Mosaicking of MPE products on over a 4km resolution grid (HRAP) over the ConUS.
• Hourly MPE obtained through NWS National Center for Environmental Prediction (NCEP) Stage IV archives.
• MPE products are used by River Forecast Centers (RFC) in hydrologic model prediction.
•• MPE ~ merging operational radar, automated gauge, as MPE ~ merging operational radar, automated gauge, as well as geowell as geo--stationary satellite rainfall following manual stationary satellite rainfall following manual quality control and bias adjustment.quality control and bias adjustment.
•• Stage IV ~ Mosaicking of MPE products on over a 4km Stage IV ~ Mosaicking of MPE products on over a 4km resolution grid (HRAP) over the ConUS.resolution grid (HRAP) over the ConUS.
•• Hourly MPE obtained through NWS National Center for Hourly MPE obtained through NWS National Center for Environmental Prediction (NCEP) Stage IV archives.Environmental Prediction (NCEP) Stage IV archives.
•• MPE products are used by River Forecast Centers MPE products are used by River Forecast Centers (RFC) in hydrologic model prediction. (RFC) in hydrologic model prediction.
Methodology (Rainfall Statistics)Methodology (Rainfall Statistics)Methodology (Rainfall Statistics)
=r
Pearson Correlation CoefficientPearson Correlation Coefficient
Weighted Absolute BiasWeighted Absolute Bias
∑=
−=
N
i iG
iGiMPEi
lll
biasweighted1
_ω
Nll
G
iGi ×><=
)(ω, where the weighting function is:
( ) ( ) ( )( ) 2/1,2 MPEGMPEGDiff llCovlVarlVarStd ×−+=
Standard Deviation of the DifferenceStandard Deviation of the Difference
( )( ) ( )MPEG
MPEG
lVarlVarllCovr
×=
,
33--HourlyHourly
66--HourlyHourlyHourlyHourly
10-1 100 101 10210-1
100
101
102
Gauge Rainfall (mm)
MP
E R
aina
ll (m
m)
Gauge Station 29 (0mm Threshold)
Timestep: 3-HOURLYWeighted Abs Bias =0.45Weighted Bias =-0.07Std Difference = 4.09R = 0.78
10-1 100 101 10210-1
100
101
102
Gauge Rainfall (mm)
MP
E R
ainf
all (
mm
)
Gauge Station 29 (0mm Threshold)
Timescale: 6-HOURLYWeighted Abs. Bias =0.39Weighted Bias = -0.07Std Difference = 4.57R = 0.86
10-1 100 101 10210-1
100
101
102
Gauge Rainfall (mm)
MP
E R
ainf
all (
mm
)Gauge Station 29 (0mm Threshold)
Timestep: HOURLYWeighted Abs. Bias = 0.57Weighted Bias = -0.07Std. Difference = 2.80R = 0.65
MonthlyMonthlyDailyDaily
10-1 100 101 10210-1
100
101
102
Gauge Rainfall (mm)
MP
E R
ainf
all (
mm
)
Gauge Station 29 (0mm Threshold)
Timestep: DAILYWeighted Abs. Bias = 0.33Weighted Bias = -0.07Std Difference = 6.53R = 0.91
100 101 102100
101
102
Gauge Rainfall (mm)
MP
E R
ainf
all (
mm
)
Gauge Station 29 (0mm Threshold)
Timestep: MONTHLYWeighted Abs Bias = 0.18Weighted Bias = -0.06Std Difference = 21.12R = 0.91
0.980.970.67Correlation Coefficient (non-zeros)
0.980.970.67Correlation Coefficient (All MPE vs. gauge)
0.190.441.31Normalized RMSE w.r.t gauge average
0.010.01-0.17Normalized Bias w.r.t gauge average
53.9113.373.37Standard Deviation (gauge) (mm)
57.9913.943.08Standard Deviation (MPE) (mm)
62.017.162.01Mean (gauge) (mm)
62.667.231.67Mean (MPE) (mm)
MonthlyDailyHourly
20060.970.970.65Correlation Coefficient (non-zeros)
0.970.970.65Correlation Coefficient (All MPE vs. gauge)
0.130.381.34Normalized RMSE w.r.t gauge average
-0.01-0.01-0.17Normalized Bias w.r.t gauge average
36.9612.543.27Standard Deviation (gauge) (mm)
36.6712.662.66Standard Deviation (MPE) (mm)
69.577.801.90Mean (gauge) (mm)
68.827.711.58Mean (MPE) (mm)
MonthlyDailyHourly
2005Hourly Statistics of all 28 Gauges
10-1 100 101 10210-1
100
101
102
Gauge Rainfall (mm)
MP
E R
ainf
all (
mm
)
Gauge Station 28 (1mm Threshold)
Gauges Averaged: G28Weighted Abs Bias = 0.45Weighted Bias = -0.16Std Difference = 2.88R = 0.73
Gauges Averaged: G30Weighted Abs Bias = 0.65Weighted Bias = -0.41Std Difference = 4.32R = 0.43
Gauges Averaged: G29Weighted Abs Bias = 0.51Weighted Bias = -0.11Std Difference = 3.94R = 0.57
10-1 100 101 10210-1
100
101
102
Gauge Rainfall (mm)
MP
E R
ainf
all (
mm
)
Gauge Station 32 (1mm Threshold)
Timestep: HOURLYAverage of Gauge: G32Weighted Abs. Bias = 0.59Weighted Bias = -0.29Std Difference = 4.12R = 0.48
Timestep: HOURLYAverage of Gauge: G33Weighted Abs. Bias = 0.60Weighted Bias = -0.30Std Difference = 4.66R = 0.51
Timestep: HOURLYAverage of Gauge: G34Weighted Abs. Bias = 0.56Weighted Bias = -0.23Std Difference = 3.95R = 0.52
Wallops Island Hourly Gauge AnalysisWallops Island Hourly Gauge Analysis
Pixel One: Single GaugePixel One: Single Gauge(G28, G29, G30)(G28, G29, G30)
Pixel Two: Single GaugePixel Two: Single Gauge(G32, G33, G34)(G32, G33, G34)
10-1 100 101 10210-1
100
101
102
Gauge Rainfall (mm)
MP
E R
ainf
all (
mm
)
Avg. of Gauge Stations G28,G29 (1mm Threshold)
Timestep: HOURLYAvg. of Gauges: G28,G29Weighted Abs. Bias = 0.53Weighted Bias = -0.05Std Difference = 3.24R = 0.63
Timestep: HOURLYAvg. of Gauges: G28,G30Weighted Abs. Bias = 0.60Weighted Bias = -0.19Std Difference = 3.40R = 0.57
Timestep: HOURLYAvg. of Gauges: G29,G30Weighted Abs. Bias = 0.59Weighted Bias = -0.20Std Difference = 3.60R = 0.59
10-1 100 101 10210-1
100
101
102
Gauge Rainfall (mm)
MP
E R
ainf
all (
mm
)
Avg. of Gauge Stations G32,G33 (1mm Threshold)
Timestep: HOURLYAvg. of Gauges: G32,G33Weighted Abs. Bias = 0.64Weighted Bias = -0.20Std. Difference = 4.15R = 0.48
Timestep: HOURLYAvg. of Gauges: G32,G34Weighted Abs. Bias = 0.63Weighted Bias = -0.21Std. Difference = 4.43R = 0.40
Timestep: HOURLYAvg. of Gauges: G33,G34Weighted Abs. Bias = 0.60Weighted Bias = -0.19Std. Difference = 3.95R = 0.53
Hourly Gauge AnalysisHourly Gauge Analysis
Pixel One: 2Pixel One: 2--Gauge AverageGauge Average(G28(G28--G29, G28G29, G28--G30, G29G30, G29--G30)G30)
Pixel Two: 2Pixel Two: 2--Gauge AverageGauge Average(G32(G32--G33, G32G33, G32--G34, G33G34, G33--G34)G34)
10-1 100 101 10210-1
100
101
102
Gauge Rainfall (mm)
MP
E R
ainf
all (
mm
)
Avg. of Gauge Stations G28,G29,G30 (1mm Threshold)
Timestep: HOURLYAvg. Gauges G28,G29,G30Weighted Abs. Bias = 0.58Weighted Bias = -0.10Std. Difference = 3.25R = 0.59
10-1 100 101 10210-1
100
101
102
Gauge Rainfall (mm)
MP
E R
ainf
all (
mm
)
Avg. of Gauge Stations G32,G33,G34 (1mm Threshold)
Timestep: HOURLYAvg. Gauges G32,G33,G34Weighted Abs. Bias = 0.64Weighted Bias = -0.16Std. Difference = 4.17R = 0.46
Pixel One: 3Pixel One: 3--Gauge AverageGauge Average(G28(G28--G29G29--G30)G30)
Pixel Two: 3Pixel Two: 3--Gauge AverageGauge Average(G32(G32--G33G33--G34)G34)
Hourly Gauge AnalysisHourly Gauge Analysis
Internal MPE AnalysisInternal MPE AnalysisInternal MPE Analysis
11.600.466.650.59M.C. / B.R.8.080.784.600.83M.C. / B.L.3.490.952.270.95M.C. / B.C.5.000.953.160.94M.C. / T.R.
12.860.427.140.57M.C. / T.L.4.750.942.950.94M.C. / T.C.7.750.794.450.83M.C. / M.R.
11.180.536.140.66M.C. / M.L.
Wt. Abs. BiasCorrelationWt. Abs. BiasCorrelationPixels Compared
Gauge Threshold: 10 mmGauge Threshold: 5 mm
3.950.733.490.78M.C. / B.R.2.890.872.550.89M.C. / B.L.1.490.961.340.97M.C. / B.C.2.310.932.060.93M.C. / T.R.4.080.733.610.77M.C. / T.L.1.860.951.660.95M.C. / T.C.2.810.872.470.89M.C. / M.R.3.490.793.060.83M.C. / M.L.
Wt. Abs. BiasCorrelationWt. Abs. BiasCorrelationPixels Compared
Gauge Threshold: 1 mmGauge Threshold: 0 mmGauge 29
Correlation MapsCorrelation MapsCorrelation Maps
HourlyHourly
33--HourlyHourly
66--HourlyHourly
DailyDaily
MonthlyMonthly
Weighted AbsoluteBias Maps
Weighted AbsoluteWeighted AbsoluteBias MapsBias Maps
HourlyHourly
33--HourlyHourly
66--HourlyHourly
DailyDaily
MonthlyMonthly
ConclusionsConclusionsConclusions
• Overall, MPE showed promising results within this region when using gauge as a reference.
• Averaging gauges within an MPE pixel at Wallops Island did not seem to improve the statistics.
• The internal MPE analysis reveal similar statistics between center pixel and surrounding pixels while the largest differences occurred mostly within the corner pixels. Also, decreases in correlation (increases in bias) were evident as the threshold was raised.
• Spatially representing correlation and weighted bias showed areas of consistent high correlation and low bias. However, low correlation and high bias were observed in the southern Delmarva region.
•• Overall, MPE showed promising results within this region Overall, MPE showed promising results within this region when using gauge as a reference.when using gauge as a reference.
•• Averaging gauges within an MPE pixel at Wallops Island Averaging gauges within an MPE pixel at Wallops Island did not seem to improve the statistics.did not seem to improve the statistics.
•• The internal MPE analysis reveal similar statistics The internal MPE analysis reveal similar statistics between center pixel and surrounding pixels while the between center pixel and surrounding pixels while the largest differences occurred mostly within the corner largest differences occurred mostly within the corner pixels. Also, decreases in correlation (increases in bias) pixels. Also, decreases in correlation (increases in bias) were evident as the threshold was raised.were evident as the threshold was raised.
•• Spatially representing correlation and weighted bias Spatially representing correlation and weighted bias showed areas of consistent high correlation and low bias. showed areas of consistent high correlation and low bias. However, low correlation and high bias were observed in However, low correlation and high bias were observed in the southern Delmarva region.the southern Delmarva region.
AcknowledgementsAcknowledgementsAcknowledgements
• Funding provided by NASA’s Graduate Student Summer Program (GSSP)
• University of Louisiana at Lafayette
•• Funding provided by NASAFunding provided by NASA’’s Graduate Student s Graduate Student Summer Program (GSSP)Summer Program (GSSP)
•• University of Louisiana at LafayetteUniversity of Louisiana at Lafayette
