Calibrating a Complex Environmental Model to Historic Data

1
San Francisco, October 25-26, 2007. Goldsim User
Conference

• Calibrating a Complex Environmental Model to
  Historic Data

Presented By Alan Keizur, P.E. Golder Associates
2
Agenda

• Background and brief overview of example
  application
• Discuss calibration approach and case study
• Issues and considerations
• Question and answer period

3
Golder Associates

• International engineering and earth sciences
  consulting firm with over 5,000 employees

4
(No Transcript)
5
Tailings Pond
6
Fresh Water Reservoir
7
Open Channel with Staff Gage
8
SAG Mill
9
Model Overview

• Model is used to evaluate the potential impact of
  operational changes (e.g., increase production)
  or different climate profiles
• Forecasts are developed based on specific
  scenarios
• Dam raise projections
• Forecast water levels in storage reservoirs under
  different conditions
• Exceed discharge limits
• Anticipate riparian flow violations
• Forecasts typically 3 months to 5 years
• Primarily deterministic simulations (most
  uncertain variables are addressed via scenarios)
• The model does support Monte Carlo analysis

10
Model Overview (1)
11
Model Overview (2)
Legend here
12
Example Model Logic
13
Water Balance Concept
Discharge from Gates or Valves
Direct Precipitation
Pond (Storage)
Spillway
Runoff
Pumping
Transfers from Other Facilities
Seepage
Evaporation
14
Calibration Approach

• How do we know how well the model works?
• One approach is to begin model simulations some
  time in the past and compare with measured data.
• Enter actual values for key model inputs (time
  histories)
• Adjust calibration variables as necessary to
  improve fit (within reasonable ranges)

15
Model Calibration

• A calibration dashboard was developed to compare
  model projection against measured data for
  backward-looking simulation
• Calibration points were defined at several key
  locations
• Water levels in reservoirs
• Flow rates in diversion ditches and receiving
  waters
• Primary calibration variable is runoff
• GoldSim optimization capability was utilized

16
Calibration Dashboard
17
Calibration Run Reservoir Water Level
18
Calibration Run Stream Flow
19
Prerequisites

• A reliable record of measured data at key
  locations (generally at least one year,
  preferably longer)
• Ability to minimize the number of unknown
  variables (ideally limit to one)
• Reasonable conceptual model for the behavior of
  the unknown variable (with appropriate bounds)
• e.g., empirical equation with one or more
  coefficients

20
Use of Optimization Capability

• Optimization was used as a starting point to
  determine what combinations of input variables
  perform well
• Credit goes to Golders Brisbane office for the
  idea
• Steps
• Extract individual submodels into standalone
  files for computational efficiency
• New submodel feature may eliminate need to do
  this
• Run optimization on key input variables (within
  reasonable bounds), minimizing difference between
  predicted and actual
• Determine if resulting combinations make physical
  sense consistent with what is known about the
  site
• Make manual adjustments as necessary

21
Optimization Settings Dialog
22
Model Maintenance Thoughts

• Model calibration is updated periodically to
  incorporate newly-collected data
• Enables longer comparison period to be utilized
• Increases confidence in model forecasts
• Becomes complicated if the system has changed
  over time
• Helps to identify additional data collection
  needs
• Helps to identify errors in conceptual model or
  input data
• if good match cannot be made without selecting
  unreasonable values for calibration variables
• We are developing a spreadsheet to store QAd
  site data for use as data source for GoldSim
  model

23
Example Disconnect
24

• Questions and Answers
• Muchas Gracias!