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.

The Science of Water Transport and Floods from Theory to Relevant Applications

93 views

Published on

This is the presentation given by Riccardo Mantilla at Univerisity of Illinois and about the System for forecasting floods in Iowa.

Published in: Education
  • Be the first to comment

  • Be the first to like this

The Science of Water Transport and Floods from Theory to Relevant Applications

  1. 1. Development & Applica0ons of a Fully-automa0c Calibra0on-Free Flood Forecas0ng System For the State of Iowa Ricardo Man0lla Assistant Professor Civil and Env. Eng. Dept. The University of Iowa, Iowa City, IA Research Engineer IIHR- Hydroscience and Engineering The University of Iowa, Iowa City, IA
  2. 2. Blurring the line between theore/cal and applied flood research
  3. 3. A reality check on the needs from riverine communi/es A state wide real-/me flood forecas/ng system available to the public in Iowa
  4. 4. Why were these floods so large?
  5. 5. DIVISION VI Sec. 15. NEW SECTION. 466C.1 IOWA FLOOD CENTER.   1. The state board of regents shall establish and maintain in Iowa City as a part of the state university of Iowa an Iowa flood center. In conducting the activities of this chapter, the center shall work cooperatively with the department of natural resources, the department of agriculture and land stewardship, the water resources coordinating council, and other state and federal agencies. 2. The Iowa flood center shall have all of the following purposes: a.  To develop hydrologic models for physically based flood frequency estimation and real-time forecasting of floods, including hydraulic models of flood plain inundation mapping. b.  To establish community-based programs to improve flood monitoring and prediction along Iowa's major waterways and to support ongoing flood research. c.  To share resources and expertise of the Iowa flood center. d.  To assist in the development of a workforce in the state knowledgeable regarding flood research, prediction, and mitigation strategies.
  6. 6. •  Historical observa/ons of streamflow are needed to calibrate support models for future streamflow forecasts •  There is a human forecaster in the loop •  Forecasts are issued every 6 hours at loca/ons above ac/on levels CALIBRATION VALIDATION State of the art of flood forecas/ng (Na/onal Weather Service)
  7. 7. IAHS* decade long ini/a/ve Addressing the problem of predic/on in ungauged basins (PUB) 2003 2012 *IAHS: Interna/onal Associa/on of Hydrological Sciences
  8. 8. A new modeling approach emerged from inves/ga/ng a different ques/on: What are the physical processes governing scale invariance in peak flows? Q Aθ = d ξ
  9. 9. Self-similar river networks structure drain the landscape
  10. 10. Transport on self-similar river networks
  11. 11. Typical Hillslope area: 0.05 km2 Typical Link length: 400 m 1 mile The river network imposes a geomorphology based par//on of the landscape
  12. 12. Explicit representa/on of water movement by solving flow transport equa/on at the hillslope-link scale Hillslope-Link Par//oning + Mass and Momentum conserva/ons for each control volume A full model for Iowa would par//on the state into over 3 million individual control volumes (i.e. 15 million ODEs) NSF - CMG Research: On the Quest for Power Laws in Floods: Developing Numerical and Analy/cal Tools, 2010 ($705,320) – A collabora/on with the UI Math Department.
  13. 13. What aspect of hillslope scale (local scale) dynamics are relevant to predict streamflow at larger aggregated basin scales (global scale)? Can the system be simplified while s/ll making reasonable predic/ons
  14. 14. Simulated Hydrographs at different loca/ons in the river network form a runoff event produced by a spa/ally variable* 10 minute rainfall event (axis have been renormalized using peak and /me of concentra/on) * Rainfall field is a spa/ally variable field U[20, 100] Pradeep V. Mandapaka, Witold F. Krajewski, Ricardo Man/lla, Vijay K. Gupta, Dissec/ng the effect of rainfall variability on the sta/s/cal structure of peak flows, Advances in Water Resources, Volume 32, Issue 10, October 2009, Pages 1508-1525, ISSN 0309-1708, 10.1016/j.advwatres.2009.07.005.
  15. 15. Transforma/on of a uniform 6 hour uniform intensity (6 mm/hr) rainfall event using different configura/ons* of the ODE system *configura/on refers to a different mathema/cal expression to represent flux-storage rela/ons or a different parameter set dsp dt = p(t)− qpc − qps dss dt = qps − qsc
  16. 16. Transforma/on of a uniform 6 hour uniform intensity (6 mm/hr) rainfall event using different configura/ons* of the ODE system *configura/on refers to a different mathema/cal expression to represent flux-storage rela/ons or a different parameter set qpc = k2sp qps = kI sp qsc = k3ss Linear Model
  17. 17. Transforma/on of a uniform 6 hour uniform intensity (6 mm/hr) rainfall event using different configura/ons* of the ODE system *configura/on refers to a different mathema/cal expression to represent flux-storage rela/ons or a different parameter set qpc = k2 sin2 (ω •sp ) qps = kI sp qsc = k3 exp α •ss( )−1( ) Complex Model
  18. 18. Consider five dis/nct nonlinear scenarios of runoff produc/on and compare the response to the one produced by a linear system B = 1 N Q(i) nl −Q(i) l Q(i) li ∑ Let’s define an error measure based on differences in peak flows These hydrographs have the same volume under the curve
  19. 19. Consider five dis/nct nonlinear scenarios of runoff produc/on and compare the response to the one produced by a linear system Scale: 0.1000 km2 B≅60%
  20. 20. Consider five dis/nct nonlinear scenarios of runoff produc/on and compare the response to the one produced by a linear system Scale: 1.0000 km2 B≅50%
  21. 21. Consider five dis/nct nonlinear scenarios of runoff produc/on and compare the response to the one produced by a linear system Scale: 10.000 km2 B≅45%
  22. 22. Consider five dis/nct nonlinear scenarios of runoff produc/on and compare the response to the one produced by a linear system Scale: 100.00 km2 B≅15%
  23. 23. Consider five dis/nct nonlinear scenarios of runoff produc/on and compare the response to the one produced by a linear system Scale: 1000.0 km2 B≅5%
  24. 24. Consider five dis/nct nonlinear scenarios of runoff produc/on and compare the response to the one produced by a linear system Scale: 10000. km2 B≅1%
  25. 25. What are the insights? Simple and complex models can have similar capabili/es to predict peak flows, /me to peak, hydrograph /ming. But… Which one is right? Which model provides a more accurate representa/on of flows in the landscape, for example, did most of the runoff entered the channel as subsurface or surface runoff? The answer has important implica/ons for sediment transport, water quality, and ecological models, etc. And more importantly… Why? What makes two models “equivalent”?
  26. 26. Consider the output of one of the complex model simula/ons ω = 1000, α = 1000 represen/ng water /me of arrival to channel and fit of a log-normal distribu/on with the same moments Best Fit of a log-normal to blue line dsp dt = p(t)− qpl − qpu dss dt = qpu − qsl qpl = k2 sin2 (ω •sp ) qpu = kI sp qsl = k3 exp α •ss( )−1( )
  27. 27. Hydrographs at the outlet match (5% error) at outlet Error = 1 T h1(t)−h2 (t) h1(t) " # $ % & ' 2 0 T ∫ dt Area = 16875 km2 Area = 0.47 km2
  28. 28. α = 1000 ω = 1000 ω = 700 ω = 300 ω = 100 Error = 5.7% Error = 5.9% Error = 5.8% Error = 8.3%
  29. 29. ω = 1000 ω = 700 ω = 300 ω = 100 Error = 22.0% Error = 22.3% Error = 22.8% Error = 20.9% α = 100
  30. 30. α ω γODE −γLog-Normal γ = E t −µt σt " # $ % & ' 3( ) * * + , - -
  31. 31. Difference in skewness of small scale hydrographs 25 20 15 10 5 Error at large scale hydrographs [%] 10-2 10-1 100 101 102
  32. 32. Linear Complex Lognormal Three equivalent hillslope rainfall-runoff transforma/ons! Area = 16875 km2 Area = 0.47 km2
  33. 33. •  Preserve the river network structure as it exists on the landscape. •  Aggrega/on and asenua/on of flows (in the real system) are equivalent to an averaging process in space. •  Superposi/on of signals smoothens out streamflow response at larger scales •  Unbiased errors in total volume of rainfall, loca/on and /ming of storms are averaged out by the averaging in the river network. •  Unbiased runoff par//on (runoff coefficient) •  Then, just follow the water
  34. 34. Diagnos/cs based vs. Calibra/on based model development Ques/on: What is the simplest model representa/on that can replicate the observed data Goal: Minimize model parameter sensi/vity in realis/c parameter ranges
  35. 35. HPC Cluster University of Iowa Data Center Virtual Servers IFC Central Database Server Model Forecasted Discharge River Network Topology, Rainfall Maps IFIS Underlying Cyber-Infrastructure: System Overview Rainfall Maps HTTP Data Transfer Reading from DB (TCP 5432) Wri/ng into DB (TCP 5432) Push Informa/on Pull Informa/on Model Forecasted and Real Stage and Discharge IFC Real Time Flood Forecas0ng Model Iowa Flood Informa0on System IFC Rainfall System
  36. 36. Mass Conserva/on Equa/ons Fluxes between control volumes dsp dt = RC *P t( )− qpc −ep dss dt = 1− RC( )*P t( )− qsc −es qpc = k2sp and qsc = k3sp Channel Rou/ng dqc dt = voqc λ1 Aλ1 1− λ1( )l qpc + psc − qc + qu u∈c ∑ $ % & ' ( )
  37. 37. Mass Conserva/on Equa/ons Fluxes between control volumes dsp dt = P t( )− qpl − qpc −ep dsl dt = qpl − qls −el dss dt = qls − qsc −es qpc = k2sp and qpl = klsp qls = kisl and qsc = k3ss kl = 99 1− sl Sl " # $ % & ' 3 k2 and ki = ρk2 Channel Rou/ng dqc dt = voqc λ1 Aλ1 1− λ1( )l qpc + psc − qc + qu u∈c ∑ $ % & ' ( )
  38. 38. Model (2) – Top Layer Model (1) – Constant Rc Model (2) – Top Layer Model (1) – Constant Rc
  39. 39. Model (2) – Top Layer Model (1) – Constant Rc Model (2) – Top Layer Model (1) – Constant Rc
  40. 40. Simula/on period June 24th to July 9th 2014 Soil moisture Precipita/on
  41. 41. Simula/on period June 24th to July 9th 2014 Soil moisture Flood Intensity
  42. 42. Rela/ve differences of peak flows for the May-June 2013 using different rainfall QPE products (Stage IV vs IFC product) Linear Model Non-linear Model Quintero et al. (2015) A Spa/al-Dynamical Framework for Evalua/on of Satellite Rainfall 1 Products for Flood Predic/on. Submised to Advances in Water Resources.
  43. 43. 0:0012:00 0:00 12:00 18:00 6:00 0:00 12:00 Discharge/MeanAnnualFlood Precipitation(mm/ hr) Date Sloan, Man/lla and Basu (2015) Hydrologic Impacts of Subsurface Drainage at the Field Scale: Climate, Landscape and Anthropogenic Controls, Agricultural Water Management. (In press). Sloan, Man/lla and Basu (2015) Hydrologic Impacts of Subsurface Drainage at the Watershed Scale: Climate, Landscape and Anthropogenic Controls, Agricultural Water Management. (submised).
  44. 44. •  Historical observa/ons of streamflow are needed to calibrate support models for future streamflow forecasts •  There is a human forecaster in the loop (forecasted are revised and corrected) •  Forecasts are issued every 6 hours at at most 165 loca/ons if above ac/on levels (*need rainfall forecast) CALIBRATION VALIDATION Let’s recall State of the art of flood forecast
  45. 45. •  Validated model structure is fed with real-/me rainfall es/mates (QPE) •  No human forecaster in the loop (no rainfall forecast necessary) •  Forecasts are issued every 15 minutes for all communi/es in Iowa •  So how does this work… Let’s consider the hydrograph predicted for Clear Creek at Coralville Rapid Refresh Flood Forecasts vs. Forecasted Rain Based Forecasts Time in hours Precipita/on Streamflow Minor Flood Level Old Forecast New Forecast Major Flood Level
  46. 46. Let’s see this working in real 0me Public version of IFIS: h`p://ifis.iowafloodcenter.org/ifis/main/ Research Version of IFIS: h`p://s-iihr50.iihr.uiowa.edu/ifis/sc/test1/ssc_model/
  47. 47. Thank you Acknowledgements: Sco` Small, Felipe Quintero, Walter Navarro, Tibebu Ayalew, Radek Goska, Luciana Cunha, Pradeep Mandapaka, Chi Chi Choi, Travis Baxter, Jarred Barr

×