Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.

DSD-INT 2020 Radar rainfall estimation and nowcasting

13 views

Published on

Presentation by Ruben Imhoff, Xiaohan Li, Pieter Hazenberg, Deltares, at the Delft-FEWS International User Days 2020, during Delft Software Days - Edition 2020. Monday, 2 November 2020.

Published in: Software
  • Be the first to comment

  • Be the first to like this

DSD-INT 2020 Radar rainfall estimation and nowcasting

  1. 1. Pieter Hazenberg Ruben Imhoff Xiaohan Li 2 November 2020 Radar rainfall estimation and nowcasting
  2. 2. Meet the presenters Xiaohan Li PhD Deltares 2 Pieter Hazenberg PhD Deltares Ruben Imhoff Deltares / Wageningen UR Junior Advisor/Researcher Topics: • Radar rainfall estimation • Urban flooding applications PhD Candidate Topics: • Radar rainfall nowcasting • Hydrological predictability Senior Advisor/Researcher Topics: • Remote sensing • Hydrological extremes • Distributed modelling using HPCs
  3. 3. (Mentimeter) Let’s Chat! 3 Join the mentimeter and let’s chat! Tell us anything, who are you? Where are you from? Which organization are you from? Why are you interested in this break out session? It does not has to be radar-related ☺. What is the first thing that comes to mind when you think about the weather radar?
  4. 4. Intro 4 Estimation Post- processing Nowcasting Radar rainfall
  5. 5. From reflectivity to rainfall rate Weather Rainfall Estimation
  6. 6. 6 (Mentimeter) Do you have access to real-time radar rainfall data? • A) No • B) Yes, through our met office • C) Yes, through a third party • D) Yes, we have a radar locally • E) Yes, we have multiple sources Radarrainfallestimation
  7. 7. 7 (Mentimeter) What types of Radar data do you have access for? • A) Raw radar reflectivity • B) Processed radar reflexivity • C) Rainfall rate of high resolution (e.g. <15min) • D) Rainfall accumulation of low resolution (e.g. > 1hr) • E) Bias corrected rainfall products • D) Rainfall nowcast (deterministic / probabilistic) • F) Others: ___ Radarrainfallestimation
  8. 8. 8 (Mentimeter) How satisfied are you with the data you get? • A) Very unsatisfied • B) Somewhat unsatisfied • C) Neither satisfied nor unsatisfied • D) somewhat satisfied • E) Very satisfied Radarrainfallestimation
  9. 9. 9 (Mentimeter) If not, why? • Open question Radarrainfallestimation
  10. 10. The weather radar KNMI radar at Herwijnen, Netherlands
  11. 11. The weather radar KNMI radar at Herwijnen, Netherlands
  12. 12. Looking under the radome Turns around 360 degrees
  13. 13. Operational characteristics Radarrainfallestimation 13 PPI (plan-position indicator):
  14. 14. Operational characteristics Radarrainfallestimation 14 PPI (plan-position indicator): PPI
  15. 15. Operational characteristics Radarrainfallestimation 15 PPI (plan-position indicator): PPI Constant Altitude PPI
  16. 16. Radar polarization 16 (Beard et al., 1986) Radarrainfallestimation
  17. 17. Volumetric measurement to rainfall • Returned signal to radar or Radar reflectivity: Z • Rainfall rate R • One interested in: R = f(Z) • Classically: Z = ARb • Coefficients A and b depend on precipitation type.
  18. 18. Non-meteorological targets complicate measurements and QPE quality Weather radars do not discriminate. Even though the focus is on measuring precipitation, they also receive a return from airborne insects, birds, bats, airplanes, buildings and mountains
  19. 19. Correcting for measurement errors
  20. 20. Quantitative precipitation estimations (QPEs) Radarrainfallestimation 20 Signal processing • Noise removal • Clutter removal • Interference filtering Corrections • Beam blockage correction • VPR correction • Hydrometer classification Rainfall rate estimation • Choose variables • Choose Drop Size Distribution (DSD) Compose • Multiple radars • Multiple products
  21. 21. Impact of weather radar reflectivity correction on extreme hydrological events
  22. 22. Weather radar QPE for operational usage 2 options: 1. Apply as is and accept potential error
  23. 23. Weather radar QPE for operational usage 2 options: 1. Apply as is and accept potential error 2. Further correct using surface observations 1. Bias correction 2. Improving Z-R relations 3. Gridded interpolation (co-kriging)
  24. 24. Weather radar QPE for operational usage 2 options: 1. Apply as is and accept potential error 2. Further correct using surface observations 1. Bias correction 2. Improving Z-R relations 3. Gridded interpolation (co-kriging)
  25. 25. Derive the product for your application Radar Rainfall Post-processing
  26. 26. (Mentimeter) Which type of application would you like to use radar for? 29 Open question: Think of: Coastal, river, urban flood forecasting, real-time control, etc Radarrainfallestimation
  27. 27. (Mentimeter) Which types of weather radars are used in your data? • A) S-band • B) C-band • C) X-band • D) A combination of above • E) I don’t know Radarrainfallestimation 30
  28. 28. Weather radars around the world 31 Distribution of C-, S-, and X-band radars from the current WMO radar database plus additions (Saltikoff et al., 2016). Radarrainfallestimation
  29. 29. (Mentimeter) Which is the most important to your application? 32 A) Quantitative precipitation estimation B) Spatial patterns C) Temporal resolution D) Spatial resolution E) Real-time availability F) Others:___ G) All of above Radarrainfallestimation
  30. 30. Radar observation resolutions 33 S-band C-band X-band Frequency 2700–2900 MHz 5600-5650 MHz 9300-9500 MHz Spatial resolution 1000–4000 m 250–2000 m 100–1000 m Temporal resolution 10–15 min 5–10 min 1-5 min Maximum range 100–200 km 100–130 km 30–60 km (Thorndahl et al., 2017) Radarrainfallpost-processing
  31. 31. Fine-scale rainfall activities 34 Large scale Fine scale Particularly important for small basins, e.g. urban Radarrainfallpost-processing
  32. 32. Rainfall interpolation 35 Truth RD coarse RD interpolated (Wang, et al. 2015) Based on tracking the motion of the rainfall field, e.g. optical flow method Radarrainfallpost-processing
  33. 33. 36 Coarsen spatial resolution temporalresolution Example of rainfall interpolation Radarrainfallpost-processing
  34. 34. 37 Example of rainfall interpolation Radarrainfallpost-processing
  35. 35. Residual bias in radar observations Radarrainfallpost-processing 38 Residualbias Radom errors Systematic bias Conditional bias RG measurements Deterministic RR fields Effect of bias in radar observation will propagate into radar nowcasts (an example in Belgium)
  36. 36. Radar - RainGauge Merging 39 Category Methods Description Simple error computing-based Mean Field Bias global mean field applied across the domain Geostatistical-based Block Kriging Kriging with External Drift local mean field inferred from rain gauge/radar data Bayesian-based Bayesian Merging (Todini, 2001); Singularity-Bayesian Merging (Wang et al., 2015) co-variance of estimation errors from radar and rain gauges Merging techniques: (near real time): Radarrainfallpost-processing
  37. 37. Singularity-Sensitive Bayesian Merging 40 RG Block-Krigged 1. Block kriging Radarrainfallpost-processing
  38. 38. Singularity-Sensitive Bayesian Merging 41 RG Radar Block-Krigged Singularity Free Singularity exponent 2. Singularity analysis Radarrainfallpost-processing
  39. 39. Singularity-Sensitive Bayesian Merging 42 RG Radar Block-Krigged Singularity Free Singularity exponent Bayesian Merged 3. Kalman filter (Wang et al., 2015) Radarrainfallpost-processing
  40. 40. Singularity-Sensitive Bayesian Merging 43 RG Radar Block-Krigged Singularity Free Singularity exponent Bayesian Merged Singularity recovered Radarrainfallpost-processing 4. Recover extremes
  41. 41. 44(Plurisk, 2017) Merging Evaluation Corrected radar data is following rain gauges And is almost twice as high as the uncorrected radar data! Radarrainfallpost-processing
  42. 42. Towards short-term operational rainfall forecasts Nowcasting
  43. 43. Which sources of information do you already use for rainfall forecasting? Think of: • Numerical weather prediction models • Seasonal rainfall predictions • Nowcasting of rainfall • Satellite based information • Etc. Open question (get a cloud of words or so) Nowcasting 46
  44. 44. How many hours/days/weeks in advance should the end-user of your operational system have rainfall forecasts for timely actions? • A) <1 hour in advance • B) 1 – 6 hours in advance • C) 6 - 48 hours in advance • D) 1 week in advance • E) > 1 week in advance Nowcasting 47
  45. 45. Do you have access to (radar) rainfall nowcasts? • A) No • B) Yes, through our met office • C) Yes, through a third party • D) Yes, we run a nowcasting method locally • E) Yes, we have multiple sources Nowcasting 48
  46. 46. When can radar rainfall nowcasting be of interest to you? Nowcasting 49 After: Germann et al., J. Atmos. Sc., 2006 • A short introduction to nowcasting was given during the pitch: “Evaluation of radar rainfall nowcasting techniques for operational water management". Slides will be posted online later.
  47. 47. Plan B: Run nowcasts locally: Steps towards operational nowcasting in Delft-FEWS 1. What do you need? Where do I start? 1. Radar data Reflectivity fields (2D or 3D) Rainfall fields (2D – end product) – Easiest start! Vertically integrated liquid contents (in combination with reflectivity or rainfall fields) 2. Nowcasting algorithm Which nowcasting algorithm do you want to use? → Cross-correlation based or similar (most used), centroid tracking, analogue based, machine learning based, etc. Open source options: pySTEPS (Python - modular framework consisting of e.g. S-PROG, STEPS and ANVIL), Rainymotion (Python), TITAN (L-Rose C++ package) Nowcasting 50 Plan A: Nowcasts are issued by met office or a third party
  48. 48. Steps towards operational nowcasting in Delft-FEWS 2. Import the nowcast results in Delft-FEWS Nowcasting 51 Nowcast results from external source Nowcasts run locally, but outside Delft-FEWS Nowcasts run with a general adapter run of Delft-FEWS
  49. 49. Steps towards operational nowcasting in Delft-FEWS 2. Import the nowcast results in Delft-FEWS Nowcasting 52 Nowcast (left) useful up to approximately an hour ahead here. Note the need for bias correction in the uncorrected radar image!
  50. 50. Steps towards operational nowcasting in Delft-FEWS 3. Use the rainfall forecasts for hydrological predictions An example for the Regge catchment (water authority Vechtstromen) in the Netherlands Nowcasting 53 Imported nowcasts Preprocess data for catchment (e.g. bias correction, clip and get catchment average) Run model, here WALRUS
  51. 51. Steps towards operational nowcasting in Delft-FEWS 3. Use the rainfall forecasts for hydrological predictions Nowcasting 54 Reference Regge (Vechtstromen) S-PROG (via pySTEPS) pySTEPS probabilistic (20 ens. Members) +15 hours
  52. 52. Other rainfall estimates: What other QPE sources would you like to use for nowcasting? • Open question again. • Think of: − Satellite data − Personal weather stations − Commercial microwave links Nowcasting 55 Saltikoff et al., 2019, BAMS Global radar coverage
  53. 53. We would like to hear more from you! Wrap up
  54. 54. What makes a radar based operational system successful to your opinion? 57 Open question
  55. 55. What are the main challenges (you had/would have) to achieve so? 58 Open question
  56. 56. Contact www.deltares.nl info@deltares.nl @deltares @deltares linkedin.com/company/deltares facebook.com/deltaresNL

×