## How Uncertain are Numerical Models?Neill Cooper Overheads of a talk given to the NSCA Dispersion Model User Group on 11th May, 2000 Background Issues - Models are approximate
- Data is approximate
- Validation data is sparse
- The atmosphere is turbulent
- Plume behaviour is unpredictable
- Modelling of single episodes is important
Sources of uncertainty include - Input Data Errors
- Insufficient data
- Meteorology
- Emission data (Talk 2)
- Atmospheric Turbulence
- Model Simplifications
- Future weather
- Spatial and Temporal Averaging
Categories of uncertainty - Parameter Uncertainty
- Conceptual Model Uncertainty
- Scenario (Future) Uncertainty
- Variability (Turbulence)
Parameter Uncertainty - Tests with a simple ‘R91’ type model
- Undertake a Monte Carlo study
Uncertainty Study: input values - Source
*Q*: 1.0 ± 20% - 10m Wind speed
*u*: 5 ± 2 ms^{-1} - Boundary layer height
*A*: 750± 250m - Roughness length
*z0*: 0.1 to 0.4m - Stability category : B C or D
- Effective source height
*z*: 25 to 50 m - Receptor height
*H*: 10m ± 5m - Number of runs : 100
Uncertainty Study: results - Concentration 100m downwind varies by a factor of 17.
- Variation 1 and 10 km downwind about a factor of 5
*Important parameters:*- 100 m: source height, receptor height
- 1 km: wind speed
- 10 km: wind speed; boundary layer height
Conceptual Model Uncertainty - Most models use a Gaussian cross-section for plumes.
- Model intercomparisons show large differences.
Scenario (Future) Uncertainty -
How often, and for how long, will the site have specific meteorology (e.g. calm conditions) next year? The year after?
Variability - Turbulence -
The same release in the same meteorology will give different maximum surface concentrations.
Quantifying uncertainty - Parameter Uncertainty - gives a factor of 2 on the concentration from a single plume
- Conceptual Model Uncertainty - ?
- Scenario (Future) Uncertainty - ?
- Variability (Turbulence) - factor of 2
How accurate are models? - QQ plots show excellent agreement
- Scatter plots show poorer agreement
- Validation of ADMS and AERMOD give average differences of about a factor of 2 for single, short term releases.
Conclusions For a single release, the uncertainty in numerical model results is about a factor of 2 or 3. © 2000 Neill S Cooper
Classification of the uncertainty in numerical models can be found
here. |