The Multisensor Precipitation Estimator and Evaluations over the

  • Slides: 25
Download presentation
The Multisensor Precipitation Estimator and Evaluations over the Florida Peninsula Greg Quina SC DHEC,

The Multisensor Precipitation Estimator and Evaluations over the Florida Peninsula Greg Quina SC DHEC, Bureau of Air Quality Based on graduate research prepared at Florida State University

What is the Multisensor Precipitation Estimator (MPE)? • Objective merging of rain gauge and

What is the Multisensor Precipitation Estimator (MPE)? • Objective merging of rain gauge and bias-corrected radar data via optimal estimation • Hourly, 4 km resolution • Useful for providing accurate high-resolution rainfall for Flash Flood and River Flood Forecast Guidance • Implemented at RFC and some NWS offices • Final gridded precipitation estimates have less error than either the input radar or the input gauge data alone

Precipitation Sensors Rain Gauges • Accurate 8 inch diameter tipping bucket measurement • Limitations

Precipitation Sensors Rain Gauges • Accurate 8 inch diameter tipping bucket measurement • Limitations – – – High rain rates Wind and evaporative losses Electronic/mechanical issues Clogs Poor spatial resolution High maintenance cost for a meso-network

Gauge Data (Dense Network) • 622 gauges from SJRWMD, SWFWMD, and SFWMD • 3

Gauge Data (Dense Network) • 622 gauges from SJRWMD, SWFWMD, and SFWMD • 3 HRL gauges used as verification • Hourly-accumulated tipping bucket • Quality controlled

Gauge-only Products • PMOSAIC (Thiessen Polygons) – Closest available gauge • GMOSAIC (Gauge-only analysis)

Gauge-only Products • PMOSAIC (Thiessen Polygons) – Closest available gauge • GMOSAIC (Gauge-only analysis) – Optimal Estimation

Precipitation Sensors WSR-88 D Radar • Good temporal and spatial resolution (6 minute, 1

Precipitation Sensors WSR-88 D Radar • Good temporal and spatial resolution (6 minute, 1 km range x 1 degree azimuth) • Limitations – – – – Obstructions and undesired scatterers Improper beam filling and overshooting Evaporative, condensational, and wind effects below radar beam Brightband hail contamination Determining drop size distribution and appropriate Z-R relationship Truncation errors in the Precipitation Processing System (PPS) Radar calibration problems – These limitations all add up to a bias that changes from hour to hour and even over the domain of a radar!

Radar Data • Hourly Digital Precipitation Data (HDP) produced by PPS at each radar

Radar Data • Hourly Digital Precipitation Data (HDP) produced by PPS at each radar site… now called DPA • 4 km resolution • 230 km detection range from radar • Elevation angle used is based on hybrid scan data

Steps in Determining Effective Radar Coverage Area 1 Compute radar-derived precipitation climatologies for each

Steps in Determining Effective Radar Coverage Area 1 Compute radar-derived precipitation climatologies for each radar (seasonal/monthly). 2 Define max and min thresholds to place on climatology. Radar estimates are not trusted beyond these thresholds. 3 Create maps of effective radar coverage areas. – Minimize defective areas

Mosaicking Procedure Which radar to use at each grid cell > Each of the

Mosaicking Procedure Which radar to use at each grid cell > Each of the following criteria MUST be satisfied for the chosen radar for each grid cell: 1 The radar data MUST be available for the given hour, 2 The specified cell location MUST lie within the effective radar coverage area for that radar, AND 3 The height of the radar beam at the cell location MUST not exceed ANY other radar beam height that satisfies 1 and 2

Height of Lowest Unobstructed Sampling Volume Radar Coverage Map

Height of Lowest Unobstructed Sampling Volume Radar Coverage Map

Result: Mosaicked Radar Estimates (RMOSAIC)

Result: Mosaicked Radar Estimates (RMOSAIC)

Radar Bias Correction • Correct radar using “ground truth” data • Find non-zero gauge/radar

Radar Bias Correction • Correct radar using “ground truth” data • Find non-zero gauge/radar pairs that are within each specific effective radar coverage area • A radar bias correction factor is calculated by dividing the total gauge amount by the total radar amount at different time spans • Mean field bias for each radar and each hour For additional details, see Seo et al. (1999)

Remove Mean Field Bias (BMOSAIC) BMOSAIC(I, J)=BIASa(k) x RMOSAIC(I, J)

Remove Mean Field Bias (BMOSAIC) BMOSAIC(I, J)=BIASa(k) x RMOSAIC(I, J)

MMOSAIC (Final MPE Product) • Merge gauge and bias corrected radar observations • Weight

MMOSAIC (Final MPE Product) • Merge gauge and bias corrected radar observations • Weight the nearby gauges vs. radar as a function of a gauge’s distance from grid point (i, j) – Sum of all weights equal to 1 RMOSAIC MMOSAIC

Variability within a 4 x 4 km area • Much of the difference between

Variability within a 4 x 4 km area • Much of the difference between the precipitation products and point gauge observations is due to the natural spatial variations of precipitation within the 4 x 4 km cells. • We evaluate this spatial variation by making correlograms of hourly gauge data vs. gauge-to-gauge distance: – Six years of hourly gauge data (1996 -2001) – 79 gauges between 26 -27 degrees North and 80 -81 degrees West.

Hourly Correlograms 2 gauge pairs separated by an inter-gauge distance

Hourly Correlograms 2 gauge pairs separated by an inter-gauge distance

Statistical Results – 1996 through 1999 • MPE products were verified against 3 HRL

Statistical Results – 1996 through 1999 • MPE products were verified against 3 HRL gauges • All MPE products and gauge values must be valid with at least one product or gauge value recording at least 0. 01” precipitation • Hourly scatterplots (most rigorous test)… results will look better when looking at daily and monthly data. METHOD r RMSD MAD BIAS % PMOSAIC 0. 367 0. 207 0. 080 0. 060 6. 027 GMOSAIC 0. 419 0. 179 0. 076 0. 050 5. 000 RMOSAIC 0. 698 0. 136 0. 054 -0. 121 -12. 061 BMOSAIC 0. 734 0. 133 0. 052 -0. 069 -6. 888 MMOSAIC 0. 733 0. 130 0. 051 -0. 028 -2. 779

Seasonal and Precipitation Type • Correlations are generally better in stratiform type precipitation and

Seasonal and Precipitation Type • Correlations are generally better in stratiform type precipitation and cold season • Gauge-only products have poor correlations, especially in convective type and warm season • Radar biases are greatest in stratiform/cold season – 50% RMOSAIC underestimates in stratiform cases – RMOSAIC truncation errors remain in other radarinfluenced products • BMOSAIC proved “tough to beat” in warm season and convective events.

September 2001 Hydro Case Study Tropical Storm Gabrielle

September 2001 Hydro Case Study Tropical Storm Gabrielle

National Weather Service River Forecast System (NWSRFS) Interactive Forecast Program (IFP) • NWSRFS simulates

National Weather Service River Forecast System (NWSRFS) Interactive Forecast Program (IFP) • NWSRFS simulates streamflow using the Sacramento Soil Moisture Accounting Model (SACSMA) – conceptual model of the land phase of the hydrologic cycle – applied to lumped basin using 6 -hour time steps – Sixteen parameters represent basin characteristics such as percentage of impervious areas, vegetation cover, evapotranspiration, and percolation rates • NWSRFS is operational at most RFCs, and our configuration resembles that used at SERFC (i. e. , same model calibration and unit hydrographs)

Two headwaters chosen for this study • Geneva basin: large area/slow response • Wekiva

Two headwaters chosen for this study • Geneva basin: large area/slow response • Wekiva basin: small area/faster response

September 12 -16, 2001 Derived Precipitation Theissen MMOSAIC DENSE RMOSAIC SPARSE

September 12 -16, 2001 Derived Precipitation Theissen MMOSAIC DENSE RMOSAIC SPARSE

Geneva MAP and Streamflow

Geneva MAP and Streamflow

Wekiva MAP and Streamflow

Wekiva MAP and Streamflow

Check out SERFC’s MPE online at http: //www. srh. noaa. gov/serfc/qpfvsmap. shtml

Check out SERFC’s MPE online at http: //www. srh. noaa. gov/serfc/qpfvsmap. shtml