1. 1 Model Calibration John M. Broemmelsiek ITS / Traffic Operations US DOT / FHWA Louisiana Division Email: john.broemmelsiek@dot.gov

2. 2 1. What is a model? 2. What is calibration? 3. Why calibrate? 4. Decision-making and calibration Agenda

3. 3 Agenda

4. 4 What is a model? What is Model? “Anything used in anyway to represent something else”

5. 5 What is a scientific model? What is Model? “A simplified abstract view of the complex reality”

6. 6 Why do we develop models? What is Model? “To predict the future”

7. 7 What is Model? Complex Reality Simplified Abstract Error Zero Error = perfect ability to predict the future. Analyst’s job: minimize error to the greatest extent possible.

8. 8 Applications What is Model?  All fields of endeavor: Engineering, Economics, Biology, Geology, Physics, Psychology have models that must be calibrated.  Transportation  Travel Demand models  Emissions models  Capacity models  Safety models  …

9. 9 1. What is a model? 2. What is calibration? 3. Why calibrate? 4. Decision-making and calibration Agenda

10. 10 What is Model? Complex Reality Simplified Abstract Error Zero Error = perfect ability to predict the future. Analyst’s job: minimize error to the greatest extent possible.

11. 11 Reducing model error is a process What is “calibration”? Verification Calibration Validation

12. 12 Reducing model error is a process What is “calibration”? Verification Calibration Validation Has the model been built correctly? Is the model statistically significant? Can we even determine significance? Do we know what the model is, such that it can be verified?

13. 13 Reducing model error is a process What is “calibration”? Verification Calibration Validation Process by which the analyst selects model parameters that cause the model to best reproduce real world conditions for a specific application.

14. 14 Reducing model error is a process What is “calibration”? Verification Calibration Validation Process to determine that a model is an accurate representation of the real world.

15. 15 What is “calibration”? Complex Reality Simplified Abstract Error Verification Calibration and Validation

16. 16 Reducing model error is a process What is “calibration”? Verification Calibration Validation Calibration and Validation are iterative processes!

17. 17 Calibration activities vary by the model used and the user’s tolerance for error What is calibration?  selection and confirmation of field data  application of a numerical constant  statistical comparison of model to field data  visual inspection

18. DATA COLLECTION

19. DATA COLLECTION- QUEUE OBSERVATIONS

20. CALIBRATION – TRAVEL TIME

21. CALIBRATION – DELAY

22. 22 1. What is a model? 2. What is calibration? 3. Why calibrate? 4. Decision-making and calibration Agenda

23. 23 Why calibrate?  No single model can be expected to be equally accurate for all possible conditions  No single model can include the whole universe of variables  Models are developed with a subset of limited, real data  Models have default values for variables, i.e. models assume that users have varied amounts of data All models have error that needs to be minimized.

24. 24 What is “calibration”? Complex Reality Simplified Abstract Error All models have error! The real issue: What is that error?

25. 25 If all models have error, how do I know: Why calibrate?  if the error has been sufficiently minimized?  if the results reflect reality?  if the prediction of the model is valid? Judgment

26. 26 Judgment required Why “calibration”? Verification Calibration Validation Has the base model I am using been sufficiently verified? Has the base model I have modified been sufficiently calibrated? Has the calibrated model been sufficiently validated?

27. 27 1. What is a model? 2. What is calibration? 3. Why calibrate? 4. Decision-making and calibration Agenda

28. 28 Six different software programs found that calibration differences of 13 percent in the predicted freeway speeds for existing conditions increased to differences of 69 percent in the forecasted freeway speeds for future conditions Decision-making and Model Calibration

29. 29 Known limitations of popular traffic analysis models Decision-making and Model Calibration  Roads with driveways  Over-saturated conditions  Roads that have more than zero crashes  Roads that have two-way left-turn lanes  Tight diamond interchanges  Roads with bicycles  Roads with on-street parking  Roads with commercial vehicle loading  Roads that function in inclement weather  Roads with a roadside environment that may impact drivers in any way

30. 30 Known limitations of popular traffic analysis models Decision-making and Model Calibration Sensitivity to changes in parameters is often unknown and could result is very large changes in model output!

31. 31 Decision-making and Model Calibration  When attempting to model the “real world”, an un-calibrated model is usually meaningless  How a model is calibrated should be agreed upon before work is started.  Calibration and validation should be extensively documented  Properly calibrating a poor model doesn’t make the results acceptable  Poorly calibrating a superior model doesn’t make the results acceptable

32. 32 Decision-making and Model Calibration  Properly calibrating a superior model doesn’t make the results acceptable  Embrace, manage and disclose uncertainty  Endeavor to convey uncertainty to decision makers  Engineering principles and judgment  Accuracy versus Precision

33. 33 What is a model? What is “calibration”? “Anything used in anyway to represent something else”

34. 34 http://ops.fhwa.dot.gov/trafficanalysistools

35. 35 Model Calibration John M. Broemmelsiek ITS / Traffic Operations US DOT / FHWA Louisiana Division Email: john.broemmelsiek@dot.gov