Flood Forecasting

Flood Forecasting

Our current plans for flood forecasting involve three phases:

1. Short-term trend extrapolation
2. Crest prediction
3. Short-term forecasting

Short-term Prediction

By "short-term prediction", we mean extrapolating from current water levels in order to predict when near-term critical events like road-crossings will happen.

Linear Trend Extrapolation

• On the gage details page
  • Create extrapolated readings every 15 minutes from the last reading until last reading + 6 hrs.
  • Show extrapolated readings on chart with a different color.
  • If extrapolated readings cross a road (rising or falling) display a notice:
    • "Based on current trend, the flood water will be [higher / lower] than the road after HH:MM pm/am"

Crest Prediction

We want to be able to predict crest timing at a particular gage or area based on the timing of crests at upstream gages.

• For each gage build a model for each upstream gage that predicts the timing of the gage's crest based on the timing and height of the upstream gage
  • For example, if gage C has two upstream gages, A and B. Build two models. One that predicts the timing of the crest of C based on A and one based on B.
• Only consider gages from USGS at the falls to USGS at Duvall
• Formalize the definition of a crest
  • Something like: crest is confirmed after three down readings with no up readings.
  • How do we handle a turbulent gage?
  • How do we calculate the time of the crest? First highest point?
• Phase I: Height independent. Calculate median time based on past data.
• Phase II: Try to model hight. Maybe start with simple linear fit.
• Gage Details page:
  • Show prediction based on nearest upstream gage that has crested
  • Show confidence interval with prediction
  • Example:
    • Estimated crest at Gage C based on observed crest at Gage A: 5:10 pm (+/- 90 minutes)

Flood Forecasting

Model Based Short-term Prediction

Predict the next 4-12 hours at each gage. Most short term (4-12 hrs) flooding impacts in the lower valley will be the result of rain that has already fallen.

We should consider using the following inputs to build a model:

• Gage trend
• Upstream readings and trends
• Upstream major tributary readings and trends (Tolt, Raging, 3 forks)
• Meteorological readings from various stations in the watershed:
  • Temperature
  • Snowpack
  • Rainfall

Predict the following for each gage:

• Road crossing time
• Yellow / Red status change time
• Crest time
• Crest height