Difference between revisions of "Using Gaussian Stochastic Processes (GaSP) for Hazard Mapping"
| Line 5: | Line 5: | ||
'''Approach:''' Develop and evaluate efficient statistical surrogates for computationally complex computer models | '''Approach:''' Develop and evaluate efficient statistical surrogates for computationally complex computer models | ||
| − | '''Summary:''' Hazard mapping is an essential tool used to estimate the risk faced by residents living in areas susceptible to natural disasters. However, accurate computer models are often computationally complex and can take hours or even days to compute. This limits the utility of computer models during storms, where conditions are frequently changing. An alternative is to create a statistical surrogate using Gaussian stochastic process (GaSP) models that can be computed more efficiently than a computer model. In this project, we hope to apply this method to model | + | '''Summary:''' Hazard mapping is an essential tool used to estimate the risk faced by residents living in areas susceptible to natural disasters. However, accurate computer models are often computationally complex and can take hours or even days to compute. This limits the utility of computer models during storms, where conditions are frequently changing. An alternative is to create a statistical surrogate using Gaussian stochastic process (GaSP) models that can be computed more efficiently than a computer model. In this project, we hope to apply this method to model landslides, using data gathered from previously conducted simulations. |
| Line 39: | Line 39: | ||
=== Week 5 (6/29 – 7/3) === | === Week 5 (6/29 – 7/3) === | ||
* '''Mini-presentation on July 2nd''' | * '''Mini-presentation on July 2nd''' | ||
| − | * Work on applying surrogate to | + | * Work on applying surrogate to landslide model (♦) |
* Identify challenges in doing so | * Identify challenges in doing so | ||
Revision as of 22:32, 13 July 2015
Student Researchers: John Bihn, Tao Cui, and Dakota Sullivan
Mentor: Elaine Spiller
Approach: Develop and evaluate efficient statistical surrogates for computationally complex computer models
Summary: Hazard mapping is an essential tool used to estimate the risk faced by residents living in areas susceptible to natural disasters. However, accurate computer models are often computationally complex and can take hours or even days to compute. This limits the utility of computer models during storms, where conditions are frequently changing. An alternative is to create a statistical surrogate using Gaussian stochastic process (GaSP) models that can be computed more efficiently than a computer model. In this project, we hope to apply this method to model landslides, using data gathered from previously conducted simulations.
Contents
Milestones
Week 1 (6/1 - 6/5)
- Read about statistical surrogates
- Plot test function
- Find optimization routine in Python
- Become comfortable with LaTeX
Week 2 (6/8 - 6/12)
- Complete reading on storm surges
- Technical paper from Dr. Spiller
- Non-technical, independent research
- Use optimization routine to find parameters to fit
Week 3 (6/15 - 6/19)
- Play around with reading data files from computer experiment
- Hopefully storm surge data – maybe move to Week 4
- Implement optimization of model posterior (∴)
Week 4 (6/22 – 6/26)
- Continue on (∴)
- Play around with reading data files and plotting outputs
- Reading about Monte Carlo calculations
- Regroup and discuss plan for July
Week 5 (6/29 – 7/3)
- Mini-presentation on July 2nd
- Work on applying surrogate to landslide model (♦)
- Identify challenges in doing so
Week 6 (7/6 – 7/10)
- Continue to work on (♦)
- Use surrogate in probability calculation (∗)
- Discuss possible scenario model
Week 7 (7/13 – 7/17)
- Work on (♦) and (∗)
Week 8 (7/20 – 7/24)
- Work on (♦) and (∗)
Week 9 (7/27 – 7/30)
- Posters due July 29th
Week 10 (8/3 – 8/7)
- Finish presentations and papers
