Computational Sustainability: Computational Methods for a - - PowerPoint PPT Presentation

Computational Sustainability:

Computational Methods for a Sustainable Environment, Economy, and Society Optimal Forest Fire Fuel Management and Timber Harvest In The Face Of Endogenous Spatial Risk

Cl i M t

The Next Step

Claire Montgomery Forest Economics Oregon State University C ll f F t College of Forestry

Wha What’s the the Pr PrOBLEM and how did w OBLEM and how did we g get her t here?

FIRE SUPPRESSION POLICY FIRE SUPPRESSION POLICY

William Greeley USFS chief 1920-9 “the conviction was burned into me that that fire prevention is the number 1 job

• f American foresters”

(Greeley, WB. 1951. “Forests and men” NY: Doubleday.)

“10:00 am policy” Goal – to contain every wildfire Goal to contain every wildfire by 10:00 am the day after it is reported – regardless of cost.

www.mtmultipleuse.org/images/smokey.jpg

Fi Fire i in t the we western U U.S.

NATU TURAL r RAL regime – gime – frequent (15-20 quent (15-20 year ears) low-intensity f s) low-intensity fires favor

• rs PONDEROSA PINE

s PONDEROSA PINE thic thick bar bark to to sur surviv ive low-intensity f e low-intensity fire tak take out w

• ut weak

eaker tr er trees ees --

• - "na

"natur tural al thinning“ thinning“ str strong

• nger tr

er trees esta ees establish dominance ish dominance

RESUL RESULT --

• - open stands of
• pen stands of big

big tr trees ees.

Lodg Lodgepole Pine pole Pine Mountain Pine Beetle Mountain Pine Beetle

• pioneer species

pioneer species

• seroti

tinous c nous cones nes

• “k

“k-str

• strategy” s

y” seed in eed in at g great d t density nsity

• Lar

Large ar e areas of eas of dead tr dead trees ees

• Enormous f

Enormous fuel b build ild-ups

• ups

Wh Wh ildf ildfi DO DO

ch choking out other

• ut other s

species ecies

• don’

n’t e t esta tablish d ish dominance minance

• overstoc

stocked, s stagnant s nant stands

• vulner

lnerable t le to in insect a sect and d d disease ase

Wh When en wildf ildfire res DO DO occur

• ccur
• Can be

Can be ca cata tastrophic

• Har

Hard to con to contai ain

helenair.com picasaweb.google.com

Wha What is a ca is a catastr tastroph phic f ic fire?

• Kills all

Kills all (or (or most) of most) of the the vegeta tation tion

• Kills all

Kills all (or (or most) of most) of the the vegeta tation tion

• Destr

Destroys or

• rganic ma

nic matter in tter in the soil the soil

• “R

“Red soil” ed soil” – burned urned so hot tha so hot that o t oxida xidation ion occur

• ccurs

Potential tential OBJECTIVES BJECTIVES

• f f

fire f fuel m management

• f f

fire f fuel m management

Existing analy Existing analyses es:

• maximiz

maximize minim e minimum tr travel time acr el time across a

• ss a landsca

landscape

• minimiz

minimize e expected loss fr pected loss from a

• m a fire
• maximiz

maximize e e expected net pected net pr present v esent value lue

• f
• f timber har

timber harvest est

• f
• f timber har

timber harvest est less tr less trea eatment cost on a tment cost on a landsca landscape My desir My desired objectiv d objective:

• maximiz

maximize e e expected net pected net pr present v esent value lue

• f
• f timber har

timber harvest est less tr less trea eatment and tment and suppr suppression cost ession cost less tr less trea eatment and tment and suppr suppression cost ession cost

• subject to

subject to

• wildlif

wildlife ha e habita bitat g t goal al di di f f t diti diti i i hi hi h h

• en

endi ding ng f for

• res

est t con conditi dition

• n i

in whi hich h

natural f fire re re regime i is re restored.

Potential tential Activities f ctivities for eac r each unit: h unit:

Existing Analy Existing Analyses es

• Do Nothing

Do Nothing

• Trea

eat fuels (mec t fuels (mechanical r hanical remo moval, pr l, prescribed b escribed burning) rning)

• Timber har

mber harvest est I’ I’d d lik like to add e to add:

• Modi

Modified f ed fire suppr suppress ssion ( l ( l t fi b i i d t th ) (e. e.g. l let f t fire re b burn urn i in mo moder erate we weath ther er)

Assessing Assessing CONSEQUENCES: ONSEQUENCES:

Inte Integration of ion of sim simula lation models into optimiza tion models into optimization: tion: 1) 1) Vegeta tation and tion and fuels fuels FOREST VEGET FOREST VEGETATION TION SIMUL SIMULATOR (FVS) R (FVS) with FOREST FUEL with FOREST FUELS EXTEN S EXTENTION ION (FFE) (FFE) 2) 2) Fi Fire b behavior – FL FLAMMAP (F AMMAP (Finne inney 2006) y 2006) pr predicts edicts 2) 2) Fi Fire b behavior – FL FLAMMAP (F AMMAP (Finne inney 2006) y 2006) pr predicts edicts FIRE SPREAD – FIRE SPREAD – as s a a function of: function of: vegeta tativ tive co cover and fuels r and fuels topog topography phy --

• - slope

slope, aspect aspect weather – ther – wind, fuel moistur ind, fuel moisture using minim using minimum tr m travel time el time alg algorithm rithm FIRE INTEN FIRE INTENSITY ITY-- flame length and lame length and other

• ther attrib

ttributes utes as a as a function function of:

• f:

vegeta tativ tive co cover and fuels r and fuels t h t h topog

• pography

weather ther

Trade-of de-offs and Optimiza s and Optimization ion

Eleme Elements nts of

• f the pr

the prob

• blem

lem

• ST

STOCHASTIC OCHASTIC – fir ire occur e occurrence and e ence and extent is unpr tent is unpredicta dictable le

• DYNAMIC

MIC

• DYNAMIC

MIC – optimal decisions in ptimal decisions in period period t de depend pend

• n f
• n fire occur
• ccurrence and fuel tr

ence and fuel trea eatments tments in pr in previo ious periods periods.

• SP

SPATIAL IAL

• - fuel tr

fuel trea eatment tment af affects ects fire re s spre read rat rates and, hence and, hence, fi fire r risk in adjacent units in adjacent units

• - dama

damage b by f fire in one in one unit ma unit may af y affect v ect values lues in in other units e

• ther units e.g. Grizzly cor

Grizzly corridor idors

K h K hi M M t l 2008 S

ti l d fi i k d

Emphasiz Emphasize DECISION MODEL e DECISION MODEL

Konos

• noshi

hima, ma, M M, et al. 2008. Spatial endogenous fire risk and

efficient fuel management and timber harvest. Land Economics. Specif Specifies d ies decision model a cision model as s stoc

• chastic d

hastic dynamic p namic prog

• gram

Si Si lif lifi if ifi ti f th th bl t t k it t it t t bl Si Simplif plifies es spec ecif ifica cati tion

• n of t

f the p pro robl blem em t to ma make it t it tra ractabl ble

Emphasiz Emphasize PROBLEM SPECIFICA e PROBLEM SPECIFICATION ION

Finne nney, M.

• M. 2007. A computational method for optimizing fuel

treatment locations. International Journal of Wildland Fire.

Wei Y Y et et al al 2008 An optimization model for locating fuel We Wei, Y Y., ., et et al

• al. 2008. An optimization model for locating fuel

treatments across a landscape to reduce expected fire losses. Canadian Journal of Forest Research.

Chung Chung W W et et al al 2009 OptFuels: a decision support system to Chung Chung, W W., ., et et al

• al. 2009. OptFuels: a decision support system to
• ptimize spatial and temporal fuel treatments. presented at

Symp.on Systems Analysis in Forest Resources. Sim Simplifies d ies decisi sion m

• n model

del Sim Simula lates f fire on

• n lands

landscape as r as reali ealistica call lly as poss y as possible

Konoshima, M

• noshima, M, et al. 2008

et al. 2008. Spa Spatial ial endo endogenous fi fire r risk and and ef effici cien ent fuel man fuel management and timbe and timber har

• harvest. La

Land Ec nd Economics

• nomics.

Method Method – stoc tochastic dynamic pr hastic dynamic prog

• gram
• - “cur

“curse of

• f d

dimensionality” mensionality” SO SO k kept it SIMPLE it SIMPLE 2 2 periods eriods p Sty Styliz lized land ed landscape e

• 7

7 identically sha identically shaped units ed units

• 2 initial

2 initial vegetat ation s stat ates g

• 4

4 decisions – decisions – treat, , cut, tr cut, trea eat&cut, lea t&cut, leave Stoc Stochastic w hastic weather (2) ther (2) and and ignition points (7) ignition points (7) Sim Simula lated f ted fire spread ead p initially – initially – no no wind, no wind, no slope slope added slope and wind indivi added slope and wind individu dual ally Solv Solve b e by com complete en lete enumer umeration ion y p y p

Look a Look at the r the results sults to dr to draw out g

• ut gener

neralities: lities:

Finne nney, M.

• M. 2007. A
• 7. A computa

mputational method f ional method for r op

• ptimizing f

timizing fuel trea eatment loc tment locations

• tions. Intern

International Journ ional Journal of l of Wildland F Wildland Fire. No Not dyn dynamic No Not sto stochastic Spa Spatial risk ial risk Ma Maximiz ximize minim minimum t m travel t el time of me of f fire a across

• ss la

landsca ndscape Giv Given f fire ignition ignition point point – upwind ind Gi Gi th th diti diti ili ili i i d Gi Give ven we weath ther er con conditi ditions

• ns

– prev revaili iling g wind

• sever

ere w e weather ( her (low f fuel m moistur isture) No v values a lues assigned t ssigned to c cells ls O d i O d i i i i d One ne d dec ecision

• n p

per eriod Heu Heuristic a c appr pproach – – sol

• lve iter

iteratively f y for stri r strips ps acr across lands landscape

Wei, Y i, Y., et ., et al.

• al. 2008
• 2008. An op

An optimiza timization tion model f model for loc r locating ting fuel tr trea eatments acr across a a lands landscape to to r reduce e ce expected f ected fire lo losses sses. Canadian Journal of nadian Journal of F Forest R st Resear search. No Not dyn dynamic c Sto Stochastic ic Spa Spatial l risk risk Min Minimize e expected lo cted loss p ss plus t trea eatment c tment cost st Deriv Derives “spr “sprea ead” pr d” proba

• babilitie

ilities” fr ” from ma

• m map of

p of “b “burn” pr urn” proba

• babilities

ilities Trea eatmen tments af ts affect s ect spre read proba babilities bilities p p p p Given w n weathe her c r cond ndit itions ions – prevailing ng w wind

• sever

ere w e weather ( her (low f fuel m moistur isture) Value ma lue matr trix ix – not s not spat atial p One d One decision p cision period riod In Inte teger pr prog

• gramming

Chung Chung, W , W., et ., et al.

• al. [2009]

[2009] OptF ptFuels: a ls: a decision decision suppor support sy t system stem to to

• ptimiz
• ptimize spa

spatial and tempor ial and temporal fuel tr al fuel trea eatm tmen ents ts. . pr presen esented a ted at Symposium o mposium on Systems A stems Analy alysis i is in Forest R st Resour sources ces. In Inte tertempo poral l b t b t t t d i but t no not dy t dynam namic Stoc

• chastic

hastic Spa Spatial risk ial risk Uses Uses e existing isting sim simula lation tion mode models f ls for r FUEL FUEL, , Vegeta tation tion

• choose a
• ose a 5-decade f

cade fuel t trea eatment t tment trajectory ajectory Vegeta tation tion, Fire Be Beha havior vior into he into heuristic op uristic optimiza timization f tion framewor

• rk

k

• to minim

to minimize e expected loss ected loss plus cos lus cost

• for giv

r given b budg dgets Fire risk risk is com is computed ed on landsc

• n landscape as fuel

e as fuels e s evolve giv iven tha en that NO FIRE NO FIRE OCCURS OCCURS

Wha What do I do I want to do to mo nt to do to move f forw rwar ard? d?

Actual Landsca Actual Landscape Sp Spatial Externalities: ial Externalities: fuel tr uel trea eatment on tment on fire risk risk ha habita bitat loss on t loss on ha habita bitat objectiv t objectives es (e (e.g .g. wild . wildlif life p e popula lations) tions) Dynamic Decision Pr Dynamic Decision Process

• cess

decisions in decisions in ne next xt period de period depend on pend on tr trea eatments and tments and realiza ealization tion of

• f f

fire e event in pr ent in previous ious periods periods Endog Endogeneity of neity of F Fire Suppr Suppression Cost ssion Cost Desir Desired Endin d Ending Condition Condition g to r to reac each h a a “na “natur tural sta al state” (e e” (e.g .g. . na natur tural f al fire r regime) gime) at minim minimum e m expected loss + pected loss + cost during the cost during the tr trans ansition p tion period riod Think a ink about how to

• ut how to LEARN fr

LEARN from the

• m the optimiza
• ptimization r

tion results sults

Potential Study Ar tential Study Area ea

Madras

Data currently available: ‐ Vegetation cover/forest types (LEMMA) ‐ Ecology Plot Data (IMAP) ‐ Land ownership (FS)

Culver

Land ownership (FS) Areas we would like to partner: ‐ Designing dynamic fuel models

Redmond Bend Sisters Bend