Capturing some convective aspects of wildfire in a empirical ......Capturing some convective aspects...

Capturing some convective aspects of wildfire in a empirical model – PHOENIX RapidFire

Kevin Tolhurst & Derek ChongDept Forest and Ecosystem ScienceUniversity of Melbourne / Bushfire CRC

Coupled Atmosphere-Bushfire Modelling WorkshopUniversity of Melbourne, 16 – 18 May 2012

Bunyip Ridge Fire, photo: Derek Chong

13 minutes later

Wind ReductionFactors

INPUTS

Fire History

Topography

Assets & Values

Road Proximity

Weather

Suppression Resources

Fire Grid Creation

DATA MANAGEMENTGeospatial Database

Fire BehaviourBehaviour Models

(McArthur MkV, CSIRO grass)

Spotting / Embers(Indraught, Ember transport, Ignition)

Slope Correction(Wind-Slope interaction)

Wind Field Models(Wind Wizard, Wind Ninja)

Road /River/Break Impact(Fuel-f ree linear features)

Fire Perimeter Propagation

Point Spread Modeling(Huygen’s, Buildup )

Self-Extinction(Moisture, Fuel, Fire Intensity)

Reprojection on Map(Surface to Plan conversion)

Suppression Model(Resources, Fuels, Topo, Fire,Road)

MODELS

OUTPUTS(to Fire Grid)

FIRE:

ORIGINEXTENT

INTENSITYFREQUENCY

FLAME HEIGHTSIZE (AREA)

TIME TO IMPACTSPOTTING

Solar Radiation Model(Fuel Moisture in space & time)

Fuel Types

Fuel Disruptions

FIRE

Fuel Accumulation(Veg Type, Time Since Fire)

FIRE GRID

Fuel Moisture(Time, Date, Cloud, Cover, Latitude)

Convection / Heat Centres(Heat output, Extent)

Spot Fires(Ember density, Ignition probability,

Buildup, Coalescence)

Asset Impact

Asset Type & Location(Value, Number)

Vulnerability(Intensity, Flame Ht, Freq, Embers)

LOSSVALUE

ASSETS

PHOENIX RapidFire

A deterministic, dynamic, continuous, empirical fire characterization model

– Deterministic – combines mechanistic and empirical elements (not stochastic or physical)

– Dynamic – inputs conditional on base fire behaviour conditions (not steady-state).

– Continuous – fire spread is a continuous process calculated as perimeter point vectors (not discrete event or transition model, not CA).

– Empirical – dynamics in model have been tuned to observed fire behaviour patterns over a range of wildfire situations

“Modelling is an abstraction of a complex reality in the simplest way that is adequate for the purpose.” (Mulligan, M. and Wainwright, J. (2004). Modelling and model building. )

Design

Input Data from 30m source to PHOENIX grid

Cann River 1983

Effect of spotting on rate of spread

Ember distribution

Fig. 5. Ground-level distribution of N0 fallen firebrands for a fire intensity of 30 MWm−1 and a wind speed of 11.17 ms−1for (a) νc = 0 and Ntot = 5052, (b) νc = 0.39 and Ntot = 7338, the left-hand scale corresponding to “short-distance” firebrands,the right-hand scale to “long-distance” firebrands, and (c) only “char oxidation process” and Ntot = 8337. Black-filled barscorrespond to firebrands that land at a temperature higher than 500 K.

N. Sardoy et al. / Combustion and Flame 154 (2008) 478–488

Convective Centres

Red – extent of spotting

Pink/blue –ember density

Brown/yellow –flame height

Grey/black –convective centre

Green/grey –fuel loads

Convection

• Heat centres = most intense 25% of segments (10 vertices)

• Coalescence if “pull” strong enough

• Unstable conditions (constant)

• No convective inertia

Comparison of Phoenix convection model (bubbles)with Laverton weather radar data looking side on.

Smoke as reflectance > 33 dBZ (pink rays)12:37 pm

1:06 pm ≈ 550 ha

2:21 pm ≈ 2000 ha

Bunyip Ridge fire (12:20 am start)

Comparison of Phoenix convection model (bubbles)with Laverton weather radar data looking side on.

Smoke as reflectance > 33 dBZ (pink rays)

12:37 pm

1:40 pm

2:21 pm

Kilmore fire (11:45 am ignition)

12:37 pm

1:40 pm

2:21 pm

6:10 pm

6:44 pm

7:36 pm

Kilmore post wind change front on

Study Data (houses)

Principal Component Analysis

House loss and convective strength

“Flame height”

“Convective Density”

50/50 Thresholds

Interactive Factors

Operational Implementation

• Fire shape metrics

– Size, shape, orientation– Performance evaluation– Real-time model reparameterization

Planned Enhancements(Shape metrics)

Raw gridded weather forecast(65,000 vs 41,000 ha)

3.6km Wx grid, 5mins(100,000 ha)

3.6km Wx grid, 5mins (1 hr offset)(86,000 ha)

3.6km Wx grid, 5mins (1 hr offset, +5 degrees)(82,000 ha)

3.6km Wx grid, 5mins (1 hr offset, +10 degrees)(70,000 ha)

Summary

• Some aspects of convection can be caught in a 2D model if a number of assumptions are made.

• Convective processes are driving the spotting process and house loss estimation.

• PHOENIX RapidFire is being used operationally for fire spread prediction in Victoria through an automated process taking 1 to 2 minutes.

• PHOENIX RapidFire is being used for strategic planning and risk assessment (house loss, water, biodiversity).

• Many underpinning assumptions of the convective model need testing.

• Input data quality is the biggest single area for improvement