DNR Modeling meeting May 15, 2008

Philippe's first whiteboard

DNR strengths

• Large online coastal data set
• IT infrastructure

Data

• DNR Data
• NCEP/NWS predictions
• NEXRAD precipitation

Types of modeling

• "Data intensive" - make good use of DNR's database, we are data intensive?
• How can we apply our models and communicate the results to decision makers.
  • Statistics ?
    • Classic Statistics
      • Multi-linear regression (MLR) - Can use to figure out what is important, can use it for performance (is your model better, not computer performance)
    • Artificial Intelligence techniques - Can we use the power of the computer and optimization methods to get the most out of the data? Do these techniques provide an advantage?
      • Artificial Neural Net (ANN) - Our bread and butter. How much better than MLR? A very simple ANN works well [1,1]. Is the model simplicity the reason we're not seeing a big difference?
        If we want to go for a fancy ANN, we need an ensemble of ANNs^? and pick the best one? Optimize w/ a big ANN and end up w/ several versions of the same model. We would calculate predictions with all models and average the answers. Dan's dissertation.
        Training sets? Engineered data sets? "Bagging and boosting", training on extreme data sets so far hasn't yielded a better model. "Weighted bootstrap sampling" - Based on the first derivative rather than the absolute value.
        • Are we overtraining? Play w/ the new version of Matlab validation feature during the training process.
      • Random Forest (RF) - Run data through RF first to get variable importance. Given all this data, what do we do w/ it? Use it to pick what we should include in the ANN.
      • Fuzzy Logic
      • ?

If its not done by August, scrap it?

Models

• Water level/surge
• Temperature - Someone will have to fund us if they want the temperature.
  • Gridded - ramp up an existing model, break down the Corpus Christi bay into little blocks.
    • SLOSH - Drive this model w/ predictions from other models [forward data simulation]. Presently researching at NCAR.
    • ADCIRC -
    • FVCOM - National Center of Ocean Studies, a more flexible model
• Current / Packery - Linear regression between BHP and the JFK Causeway.
• Salinity
• Wave - Philippe and Dan working on CMP, due in October.
• Indicator bacteria / precipitation - Proposal w/ Joanna? Predict water quality ahead of time? Use automated precipitation for Waylon's project.
• Wind - Wind predictions from WRF-NAM are not good for our location(s). Can we improve upon them? Provide adjusted/corrected wind predictions. Sea breeze forcings?
• Thunderstorm (with NWS) - Waylon and Philippe's model.

Links related to models:

• Information on ocean models
• More information on 3-D ocean models
• Application of FVCOM to Gulf of Maine
• FVCOM Forum

Projects

• NOAA ECSC - Proposal to apply one of the gridded models and add it to the AI.
• MANERR - Mission Aransas National Estuarine ? Reserve

• Put water level stations on all platforms.

Proposals

• Offshore TCOON / surge model
• Wave prediction

When we find money, things get easier. More people-power to focus on this issue.

Add a buoy to Robin's model? Make it easier to add waves or whatever.

For the summer, keep going. [Keep building models for the existing ANNs?]

Other things

How to make the website look better? Presentation of graphs, models?

Key question: Is it possible to have a data-intensive model and still capture the extreme events?

Continue meeting during the summer? Individual progress reports in June? Goal for August is to include same group; Sergey, Angelica, Philppe, Robyn, Beate, James and Rick and Waylon?

Have another meeting in about a month?