Colorado Decision Support System for Prediction of Wildland Fire … · 2016-05-26 · 5/26/2016 1...
Transcript of Colorado Decision Support System for Prediction of Wildland Fire … · 2016-05-26 · 5/26/2016 1...
5/26/2016
1
Colorado Decision Support System for Prediction of Wildland Fire Weather, Fire Behavior, and Aircraft Hazards
0
CO-FPS Stakeholder Meeting #523 May 2016
NCAR, Boulder, CO
5/26/2016
2
Mesa Lab
NCAR-Wyoming Supercomputer Center
Aviation FacilityCenter Green
Foothills Lab
Quick View of a Test Case
TEST CASE
5/10/2016
Starts at 20:00 UTC2 PM LT
15 hour forecastWx Source: HRRR
5/26/2016
3
Progress Since April 2016
• Staging of CO‐FPS input datasets
• Development of the data ingest subsystem (HRRR, NAM, LANDFIRE, MMA, CO‐WIMS)
• Implementing data security measures
• Testing of data flows
• Working with Intterra on the development of the Interface Control Document (ICD)
• Updating the Functional Requirements Specification (FRS) based on user feedback
5
5/26/2016
4
Progress Since April 2016 (cont.)
• Testing a variety of CAWFE® model and computer configurations to optimize performance
• Configuring CAWFE® to run daily over the mountains and plains to evaluate run‐time performance (MPI vs. OpenMP, etc.) and robustness
• Created a test wildfire ignition capability to exercise CAWFE®
Progress Since April 2016 (cont.)
• Developed product viewing capability outside of CO‐WIMS for testing and evaluation purposes
• Prepared a Draft Training Plan and coordinated plan with Intterra
5/26/2016
5
Due Soon
• Functional Requirements Specification (FRS)
• Design Specification
• Performance Verification Metrics and Methodology Report
• Stakeholder Meeting Minutes (May meeting)
• Computer augmentation specification
Additional CO‐FPS DevelopmentInformation
Jim Cowie (NCAR)
5/26/2016
6
CO‐FPS DevelopmentBuilding up Real‐Time Capability
• High Resolution Rapid Refresh (HRRR)– 3km Numerical Weather Prediction model– From the National Weather Service– Supplies boundary conditions for the CAWFE® model– And provides CO‐scale weather products– Updated hourly– 15 hour forecast today, but expanding to 18 hrs this summer
• North American Model (NAM)– Also from NWS– Provides forecast information past 18 hours
CO‐FPS Development
Building up Real‐Time Capability
• Datasets still to acquire:
– AWOS, ASOS, RAWS, etc.
– MMA data (via CO‐WIMS)
– VIIRS data
• Used by CO‐FPS and/or
• Used for verification
5/26/2016
7
CO‐FPS Development
Process Automation
• HRRR & NAM automatically ingested• Working on automated processing scripts for product generation
• CAWFE® model triggering– Prototype script, currently run by hand– Initiates a CAWFE® run– Uses defined ignition point– Weather, terrain and fuel (LANDFIRE)– Eventually connected to CO‐WIMS
CO‐FPS Development
Interfacing To CO‐WIMS
• Working with Intterra
– Bi‐weekly telecons
• Interface Control Document
– Initial version delivered
– Describes how and what CO‐WIMS and CO‐FPS will communicate
– Secure communications
– Web services
5/26/2016
8
CO‐FPS Development
Interfacing To CO‐WIMS
• Providing web service URLs for testing
– For CO‐scale weather
– For fire simulations
• Working on interface services next
– Handling CO‐WIMS requests
– Tracking fire simulations
CO‐FPS Development
CAWFE® Model Development
• Optimizing model performance (compiler and parallelized configurations)
• Real‐time simulation stability testing– Different weather
– Different terrain
– Different fuels
• Case studies
5/26/2016
9
CO‐FPS Development
• Product symbology
• Product units
• Products will be set up as web services
• Dedicated product server to be set up
Starting CAWFE® Model Simulation
5/26/2016
10
Triggering a Fire Simulation
• Bill Petzke will demonstrate the manual triggering of a fire simulation, but not through CO‐WIMS yet
• Location: Hall Ranch, Boulder County, west of Lyons
• Weather: 10 May 2016 at 20:00 UTC (2 PM LT)
• Fuel: LANDFIRE
• Output will be demonstrated later in the meeting
Starting CAWFE® Model Simulation
Required Inputs (today it will be done manually)
• Location (LAT, LON)• Time (UTC)• Date• Fire boundary (min default = 100 m)• Fire flaming front boundary
In future, the ignition process will be done through CO‐WIMS
5/26/2016
11
Hall Ranch Fire Simulation Site Initialization Parameters
• Initialization Date/Time
– 10 May 2016 at 20:00 UTC
• Coordinates
– 40.2140979, ‐105.2923467
• Fire boundary radius at ignition
– 100 m
• Length of CAWFE® simulation
– 15 hoursHall Ranch
Hall Ranch Fire Simulation Site Initialization Parameters
Generalized weather at 20:00 UTC initialization (from the HRRR Model)
• Temperature: 64 F
• Gust: 28 mph
• Wind Speed: 16 mph
• Wind Direction (coming from): 248 degrees
• Relative Humidity: 28 %
5/26/2016
12
Triggering a Fire Simulation
Hall Ranch
Lyons
Hall Ranch Fire Simulation Site Terrain
5/26/2016
13
Hall Ranch Fire Simulation Site Ground View North
Hall Ranch Fire Simulation Site Ground View North West
5/26/2016
14
General Weather Situation on 10 May 2016
AIR TEMPERATURE
Starts at 20:00 UTC2 PM LT
15 hour forecastSource: HRRR
ScenarioTemps in the mid 60s cooling down towards evening
General Weather Situation on 10 May 2016
RELATIVE HUMIDITY
Starts at 20:00 UTC2 PM LT
15 hour forecastSource: HRRR
ScenarioRH around 28% rising to 70% towards evening
5/26/2016
15
General Weather Situation on 10 May 2016
WIND GUST
Starts at 20:00 UTC2 PM LT
15 hour forecastSource: HRRR
ScenarioGusts around 28 mph falling to 15 mph towards evening
General Weather Situation on 10 May 2016
WIND DIRECTION & SPEED
Starts at 20:00 UTC2 PM LT
15 hour forecastSource: HRRR
ScenarioWind direction was initially from the southwest near Hall Ranch but veered to west and northwest and weakened quickly in the evening.
5/26/2016
16
Will View Hall Ranch Wildland Fire Simulation Towards End of Meeting
Updates on CAWFE® ModelConfiguration and Development
Branko Kosović, Pedro Muñoz Jimenez, Domingo Muñoz-Esparza
5/26/2016
17
Coupled Wildland Fire Spread and Weather Model
Surface fire spread model parameterized using Rothermel(1972) model
LANDFIRE – fuel data with Anderson (1982) fuel model Fuel burnout calibrated based on Albini et al. (1995) Optimizing model performance for the operational
implementation Implementing crown fire model by Finney (1998, FARSITE) Fire and burn area perimeter modeled using an improved
higher order level set method Assimilating MMA burn area perimeter and fire line data Explored using Scott and Burgan fuel model, however, fuel
burnout parameters for some fuel categories are not available
CAWFE® Uses LANDFIRE Fuel Data
Fuel parameters required by CAWFE® are: Surface fuel load [kg/m2] Fuel load decrease weighting parameter [s] Surface area to volume ratio [1/m] Fuel depth [m] Fuel moisture content of extinction Canopy fuel load [kg/m2] Canopy fuel burnout time [s]
5/26/2016
18
Fuel Burnout
Developments of WRF‐CAWFE® fire model
Basics of the spread model
• Fire spread model is based on Rothermel 1972
• Rate of spread (ROS) calculated at each grid point and used to propagate fire line forward in time
• This mathematically done using the ‘level‐set method’
5/26/2016
19
Developments of WRF‐CAWFE® fire model
Parallel computing
• Code rewritten to take maximum advantage of parallel computing
• Redundant calculations removed
• Increase in speed of the model ‐> the model runs faster
CPU#1 CPU#2
CPU#3 CPU#4
Developments of WRF‐CAWFE® fire modelSolution convergence (I)
• We need to discretize the level‐set equation
• How much do we need to refine the grid to reach convergence on the fire propagation/perimeter?– “Idealized fire”: uniform wind (U = 5 ms‐1), flat terrain, uniform fuel (short
grass), no feedback to the atmosphere (uncoupled mode) [initial fire line of 1 km length]
t=1min t=30min
5/26/2016
20
Developments of WRF‐CAWFE® fire modelSolution convergence (II)
• We have implemented high‐order (=more accurate) solutions for the fire propagation model
• Using higher‐order numerical schemes ‐> half the resolution is needed, 4 times faster
Developments of WRF‐CAWFE® fire modelSolution accuracy
• New implementation provides a more robust solution
dx=50m
dx=25m
dx=12.5m
Wind speed [ms‐1] Burnt area
5/26/2016
21
Developments of WRF‐CAWFE® fire modelSummary of the fire model developments
• New implementation takes maximum advantage of parallel computing and removes unnecessary calculations
• Resolution required by the new implementation is half ‐> faster
• Solution is more robust
Current work
• Refining canopy parameterization in WRF‐CAWFE®
• Implementing Rothermel 1991, Finney 1998, Scott & Reinhardt 2001
• Use higher‐fidelity models to calibrate & improve canopy parameterization in WRF‐CAWFE®
Forecasting System Performance Time required to perform a 1‐h simulation
The data preparation to run a simulation requires 3‐4 min
5/26/2016
22
High Park, CO, Fire
A wildfire in the mountains. Fire started Monday, June 9th,
2012, West of Fort Collins, CO, Larimer County.
Southwesterly winds The hot and dry conditions (fire
danger extreme) led to a rapid intensification of the fire.
87,000 acres burned – third largest wildfire in Colorado history by burned area.
259 homes burned. The fire was fully contained by
June 30th
Caused by lightningIgnition Point
17mi
High Park, CO, Fire
Ignition Point
17mi
5/26/2016
23
High Park, CO, Fire
d1
d2
d1 – 127 km x 127 km (1km grid cell)
d2 – 14 km x 14 km (111m grid cell)
High Park, CO, Fire
d1 – 127 km x 127 km (1km grid cell)
d2 – 14 km x 14 km (111m grid cell)
Contours – CAWFE® simulation
d2
d1
5/26/2016
24
High Park, CO, FireJune 9th, 20:20 UTC June 10th, 4:10 UTC
Last Chance, CO, Fire
Grassland fire Fire started Monday, June 25, 2012,
south of Last Chance, CO, Washington County
Strong southerly winds 45,000 acres burned – second
largest wildfire in Colorado in 11 structures burned, including four
houses The fire was fully contained by
Tuesday evening, June 26 Cause of fire ‐sparks an automobile
wheel following a tire blowout ( Wikipedia)
Fuel:Red – no fuelBlue – grass
Last Chance
Woodrow
Ignition Point
17mi
5/26/2016
25
Last Chance, CO, Fire Grassland fire Fire started Monday, June 25, 2012,
south of Last Chance, CO, Washington County
Strong southerly winds 45,0000 acres burned – second
largest wildfire in Colorado in 11 structures burned, including four
houses The fire was fully contained by
Tuesday evening, June 26 Cause of fire ‐sparks an automobile
wheel following a tire blowout ( Wikipedia)
Last Chance, CO, Fire
5/26/2016
26
Last Chance, CO, Fire
d1 – 127 km x 127 km (1km grid cell)
d2 – 28 km x 28 km (111m grid cell)
Squares – VIIRS data Contours – CAWFE®
simulation
d2
d1
Last Chance, CO, Fire
5/26/2016
27
Limited Demonstration of CO‐FPSProducts on GIS Displays
Jennifer BoehnertKevin Sampson
Molly Hausmann (Intterra)
User Questions from Functional Requirements Specification
• Displayed time units?
• Rate of spread units?
• Heat release units?
• Smoke concentration units? micro‐gram/m3
• Wind speed units?
• Turbulence units? Light, moderate, severe?
• More aviation thresholds and units later…
5/26/2016
28
Viewing the CAWFE® Model SimulationTriggered at Start of the Meeting
Hall Ranch CAWFE® Model AnimationTriggered at Start of the Meeting
5/26/2016
29
General Questions
Reference Slides
5/26/2016
30
Initial Operating Capability CO‐FPS Products
18 hour predictions (at user defined increments) of:
• Fire extent• Rate of spread• Heat release• Smoke concentration• Significant fire phenomena
• Turbulence intensity• Downdraft and updraft regions• Wind shear regions
• Wind speed, direction, gustiness• Surface air temperature• Surface relative humidity
Fire behavior product groupWill be calculated on 100 m fire scale
grids when triggered
Aviation hazard product groupWill be calculated on 3 km (state scale)Could also be provided on 1 km, and fire
scale grids (100 m)
Fire weather product groupWill be calculated on 3 km (state scale)Could also be provided on 1 km, and fire
scale grids (100 m)