Presentation is loading. Please wait.

Presentation is loading. Please wait.

Detection of Heavy Precipitation OHD’s Research and Development in Radar and Multisensor Applications David Kitzmiller Hydrologic Science and Modeling.

Similar presentations


Presentation on theme: "Detection of Heavy Precipitation OHD’s Research and Development in Radar and Multisensor Applications David Kitzmiller Hydrologic Science and Modeling."— Presentation transcript:

1 Detection of Heavy Precipitation OHD’s Research and Development in Radar and Multisensor Applications David Kitzmiller Hydrologic Science and Modeling Branch Hydrology Laboratory Office of Hydrologic Development February 1, 2006

2 Topics  Potential for improving detection of heavy rainfall through use of radar mosaics  Implications of spatial resolution impacts on radar-rain gauge correlation  Probabilistic radar rainfall estimates and potential applications in Flash Flood Monitoring and Prediction  Quantitative nowcasts of rainfall

3 Hydrometeorology Group  Feng Ding  Richard Fulton  Shucai Guan  David Kitzmiller  Chandra Kondragunta  Dennis Miller  Kiran Shrestha

4 Collaborating Partners  University of Iowa Department of Civil and Environmental Engineering (Krajewski, Ciach, Villarini)  National Severe Storms Laboratory WISH WISH RRAD RRAD  Princeton University Department of Civil and Environmental Engineering (Jim Smith)  RS Information Systems, Inc. (McLean VA)

5 Acronyms…  DHR: Digital Hybrid Reflectivity (also precipitation accumulations from DHR)  DPA: Digital Precipitation Array (one-hour radar rainfall accumulations)  FFG: Flash Flood Guidance (amount of rainfall required to cause small streams to flood)  FFMP: Flash Flood Monitoring and Prediction System (Part of SCAN in AWIPS)  Z-R: Reflectivity-to-rainrate model used to estimate rainfall from radar measurements  Bias: (  Gauge Rainfall) / (  Radar Rainfall) for many collocated individual 1-h estimates

6 Typical One-Hour Rainfall Product:

7 Radar-to-Raingauge Comparisons Are Often Discouraging…

8 Range Influence on Radar Detection of Rainfall  Detection efficiency decreases with range  Radar detects hydrometeor distribution that is different from that at surface  Horizontal advection of precipitation affects apparent radar-gauge agreement  Pointing accuracy of radar  Location errors for gauges Most of these errors are magnified at longer ranges at longer ranges

9 Height FieldRadar Coverage Field AWIPS Multisensor Precipitation Estimator Mosaicking Technique

10 Detection of Heavy Rain 4-km Mosaic vs. 1-km Single-Site Estimates  Created a set of matching estimates: Rain gauge Rain gauge Single-radar (based on Digital Hybrid Reflectivity products used in FFMP) Single-radar (based on Digital Hybrid Reflectivity products used in FFMP) Multisensor Precipitation Estimator mosaic Multisensor Precipitation Estimator mosaic  Sites: KPBZ (Pittsburgh) KPBZ (Pittsburgh) KLWX (Sterling) KLWX (Sterling) KFCX (Blacksburg) KFCX (Blacksburg)  2004 Warm season, 7 rain events  Result: mosaic field has higher correlation with gauge estimates than does single-radar

11 Probability of Detection of 1-h rainfall ≥12.5 mm by single radar (DHR) vs. mosaic

12 False alarms for ≥12.5 mm 1-h rainfall by single radar (DHR) vs. mosaic

13 Correlation Between Radar Estimates and 1-h, 12.5-mm Rainfall Events

14 Effects of Spatial Smoothing on Radar Rainfall Estimates  We found that some degree of spatial smoothing generally improves radar-rain gauge correlations  There are several potential sources of radar/gauge location disagreement: Pointing accuracy of radar Pointing accuracy of radar Horizontal advection of raindrops Horizontal advection of raindrops

15 Raindrops detected at 60 nm range are about 1 nm above ground (assuming level surface) Terminal velocity of large drops ~ 20 knots Travel time to ground ~ 3 minutes If mean wind is 20 knots, raindrops can travel 1 nm horizontally, or two WSR-88D range gates Homogeneity of rainfall field mitigates advection effects, but advection might play large role in poor radar/gauge correlations in light, spotty rain

16 Effects of Spatial Smoothing of Radar Estimates on Radar/Gauge Correlation

17 Effects of Spatial Smoothing of Radar Estimates on Radar/Gauge RMS Error

18 Effects of Spatial Smoothing On Areal Rainfall Estimates  Quality of areal rainfall estimates based on 1km x 1  DHR products and 4km x 4km DPA assessed by comparing with collections of rain gauges in ARS Oklahoma micronet  Quality of DHR-based and DPA-based areal estimates is nearly identical  The issue warrants further investigation for confirmation  What horizontal resolution is optimum for FFMP?

19 Probabilistic Relationships Between Radar and Rain Gauge Estimates  Most common forms of bias correction are based on long-term collections of 1-h radar/gauge paired observations  Actual bias between radar estimates based on Z-R and rain gauges depends on magnitude of the rainfall rate  A common method of selecting radar rainfall alert thresholds (fraction of critical ground truth value) is not statistically reliable

20 Flash Flood Guidance  An estimate of the rainfall required to cause small headwater streams to reach bankfull  Commonly expressed as 1-h, 3-h, 6-h amounts  Routinely produced by River Forecast Centers based on soil type, antecedent rainfall

21 1-Hour FFG from MARFC, OHRFC

22 Common Operational Strategy  Take action when radar rainfall estimate is 80% of FFG value Closer examination of basin rainfall history Closer examination of basin rainfall history Call for spotter reports Call for spotter reports  However, threat of actual rainfall exceeding FFG is strongly dependent on the radar estimate itself

23 Probability of Gauge Rainfall ≥ 120% of Radar Estimate Data from KTLX, KINX, KSRX, 2004-2005 warm seasons

24 Probability of Gauge Rainfall ≥ 120% of Radar Estimate  Probability of exceeding a given gauge/radar ratio decreases with radar rainrate  For a radar estimate of 0.4 inch, there is a 45% chance that rainfall will exceed 0.5 inch  For a radar estimate of 1.5 inches, there is only a 15% chance that rainfall will exceed 1.8 inches

25 Probabilistic Relationships Between Radar and Rain Gauge Estimates  Work carried out at University of Iowa (Krajewski, Ciach, Villarini) shows that radar rainfall errors can be modeled with a set of power-law functions  Results confirmed on a larger data sample by OHD

26 After correcting radar estimates for overall long-term bias:  Radar underestimates lighter amounts and overestimates higher amounts  A simple power law relates expected rainfall to initial Z-R estimate

27 1-h Radar Rainfall Estimate, bias corrected, mm KTLX, 1996-2003 From Krajewski and Ciach, 2005 After Correcting Radar Estimates For Long-Term Bias, a Magnitude-Dependent Bias Remains…

28 From Krajewski and Ciach, 2005 1-h Radar Rainfall Estimate, bias corrected, mm KTLX, 1996-2003 Standard Deviation of The Radar Estimate Error (Spread of Estimates) Can Also Be Modeled As A Power-Law Function:

29 After correcting radar estimates for overall long-term bias:  Rainrate-dependent bias is approximated by a power-law curve  Standard deviation of multiplicative error is also a power-law curve  Distribution of multiplicative errors for any given radar estimate is approximately normal  Formulation of probability of rainfall exceeding a critical value:

30 B is long-term gauge/radar bias a,b are parameters of bias power law; c,d,e are parameters of standard deviation power law; RR is initial radar estimate; THRES is FFG or other critical rain amount Formulation of Probability of Rainfall (RR) Exceeding a Critical Value THRES

31 Application of Probabilistic QPE:  Power-law parameters are determined from extended gauge/radar sample  Parameters have seasonal and site dependence  Probability equation could be incorporated as new option in FFMP

32 Multisensor Precipitation Nowcaster (MPN)  Extrapolative forecast model for 0-1 hour rainfall amounts  Based on mosaicked radar and/or radar/gauge rainrate field  Produces forecasts of 1-h, 3-h, 6-h rainfall amounts ending 1 hour in the future  Advantages over operational SCAN 0-1 hour extrapolative algorithm: Radar mosaic input, rather than single radar Radar mosaic input, rather than single radar Incorporates real-time radar/gauge bias information Incorporates real-time radar/gauge bias information

33 Scores For MPN, Persistence Forecasts: Detection of > 15mm, 1-Hour Rainfall, 4-km Resolution 20 Flash Flood Cases

34 Potential Applications of MPN  Direct application in FFMP  Precipitation output serves as input to hydrologic models (distributed, Site Specific)  Reed et al. reporting results using MPN output in distributed hydrologic modeling at AMS Hydrology Conference this week

35 Observed Hydrograph Model Hydrograph with QPF Modeled Hydrograph, Assuming 0 QPF Modeled Hydrograph, Assuming rain persistence

36 Summary  In many parts of the U.S. the radar network is dense enough that mosaics of rainfall estimates could provide significantly better detection of rainfall than single-radar estimates  Use of prior knowledge of error statistics, through a probabilistic model, could simplify warning decisions.  Quantitative nowcasts of rainfall can improve forecasting of flood events through application in either FFG or physical hydrologic models  The optimum degree of spatial smoothing for radar rainfall estimates needs further investigation

37 Questions Raised:  How to integrate multisensor mosaics into flash flood operations? Enhancement of Multisensor Precipitation Estimator in AWIPS Enhancement of Multisensor Precipitation Estimator in AWIPS Ongoing investigation of Q2 algorithm originated at NSSL Ongoing investigation of Q2 algorithm originated at NSSL  Using probabilistic formulation of radar/rain gauge error in FFMP  Determining optimum spatial resolution of radar rainfall estimates

38 OSIP/HOSIP Status: Providing radar mosaic estimates for flash flood detection (MDL, NSSL, OHD) Providing radar mosaic estimates for flash flood detection (MDL, NSSL, OHD) OSIP Stage 1OSIP Stage 1 Multisensor Precipitation Nowcaster: Multisensor Precipitation Nowcaster: HOSIP Stage 2HOSIP Stage 2 Radar Probabilistic Precipitation Estimates: Radar Probabilistic Precipitation Estimates: HOSIP Stage 2HOSIP Stage 2 Distributed hydrologic models for flash flooding Distributed hydrologic models for flash flooding HOSIP Stage 3HOSIP Stage 3

39 Related OHD Projects Distributed hydrologic modeling for flash flooding (Hydrology Group) Distributed hydrologic modeling for flash flooding (Hydrology Group) Precipitation estimates from Terminal Doppler Radar (NEXRAD Software Project) Precipitation estimates from Terminal Doppler Radar (NEXRAD Software Project) Evaluation of NSSL dual-pol precip algorithm Evaluation of NSSL dual-pol precip algorithm Automated merging of rain gauge / radar / satellite estimates in AWIPS (Hydrometeorology Group) Automated merging of rain gauge / radar / satellite estimates in AWIPS (Hydrometeorology Group) Automated QC for rain gauge data (Hydrometeorology Group) Automated QC for rain gauge data (Hydrometeorology Group) Baltimore Flash Flood Forecast Project (Princeton University, USGS, Hydrometeorolgy Group) Baltimore Flash Flood Forecast Project (Princeton University, USGS, Hydrometeorolgy Group)

40 Questions?

41 3-Hour FFG from MARFC, OHRFC

42

43 Expected 1-Hour Rain Gauge Value As Function of Radar Estimate Radar estimates adjusted for bias (gauge/radar): KTLX bias: 0.7 KLWX bias: 1.07 KFFC bias: 0.93 May-Sep 2004


Download ppt "Detection of Heavy Precipitation OHD’s Research and Development in Radar and Multisensor Applications David Kitzmiller Hydrologic Science and Modeling."

Similar presentations


Ads by Google