scientific python in disaster simulation and ... - SciPy India 2017

“SCIENTIFIC PYTHON IN DISASTER SIMULATION AND VISUALIZATION : MODELING AND SIMULATION OF WILDLAND FIRE SPREAD USING CAN” GUGAN SELVARAJ

Objectives • Present the frame work of Cellular Automaton Network (CAN) to simulate the forest fire spread • Study the reliability of the proposed framework

12/11/2016

Ⓒ EERC @ IIIT-H

2

Need for the Study Forest Fire Affected Forests

Safe Forests

Others

9% 10%

Severe Fires 6.17 %

81%

State of Forests in India

Causes

(Mostly)

Impacts

Need Forest Map of India 12/11/2016

Ⓒ EERC @ IIIT-H

3

Fire Behavior in Forests Generally, Fire behavior depends on three components such as Air, Heat and Fuel characteristics. But in case of forest environment, in addition to those it also depends on the topography and weather characteristics. Dynamic Factor

Static Factor

Temperature, Wind velocity, Precipitation, Hu midity, Atmospheric Stability etc.

Elevation, Aspect, Steep ness, Position, Shape etc.

Affects the rate of spread significantly

Individually affects the behavior

Most Important Factor Fuel load, Size and Shape, Density, Chemical comp., Moisture Content, Porosity, Type of Fuel etc. Individually affects the behavior 12/11/2016

Ⓒ EERC @ IIIT-H

COMPLEX BEHAVIOR 4

Fire Spread Principle The spread of fire in forests also depends on complex interaction between the heat transfer mechanisms (conduction, convection or radiation) leading to formation of surface or crown fires.

Crown Fire Convection & Radiation

Spatially, its spreads in all the directions radially from the fire point.

Surface Fire Conduction & Radiation 12/11/2016

Ⓒ EERC @ IIIT-H

5

Forest Fire Simulation using CA Cellular Automata

Space time models that represents a computational paradigm for complex phenomena that evolve basis of local interactions. The efficiency of simulation depends on type of grid, neighborhood, transition rule and global states.

Process Flow of CA Simulation

Fire Spread Phenomena

1. Choice of CA Grid and Possible States. 2. Formulate Neighborhood & Transition Rule based on the phenomena. 3. Formulate the iterative process for simulation. 12/11/2016

Ⓒ EERC @ IIIT-H

CA Neighborhood & Transition Rule

6

Fire Spread Rule • •

•

•

•

Model Description Grid : Rectangular Global States : Combustible, InCombustible, Burnt / Burning Neighborhood : Moore In case of CFSR the fire spreads in a square pattern but in case of IFSR the fire spreads in a approx. circular pattern as given by Cunbin et.al. using Rothermel model and Huygens principle. Thus IFSR is found to be more reliable. 12/11/2016

Conventional Fire Spread Rule (CFSR)

Homogenous Forest Model

Fire Spread Trend (Analytical) Ⓒ EERC @ IIIT-H

Improved Fire Spread Rule (IFSR)

Fire Spread Trend (CFSR)

Fire Spread Trend (IFSR) 7

Cellular Automata Network (CAN) Model Fire Simulation Model Elevation Slope Aspect Fuel Model Canopy Cover Canopy Height Crown Base Ht. Crown Density Fuel Type

Model Description • •

•

Grid : Rectangular Global States : Combustible, InCombustible, Burnt/B-urning Neighborhood : IFSR CAN Layers considered For study

• •

Extended version of conventional CA approach Provides flexible and reliable framework to simulate multivariate phenomenon

12/11/2016

Ⓒ EERC @ IIIT-H

• • • •

Slope Elevation Fuel Type Fuel Density

8

Fire Simulation using CAN Model A Case Study : Mizoram Mizoram Location Map

Land Cover Forest Cover Terrain Char. Climate Forest Fire Vulnerability

21,078 sq.km 91% Steep Slopes Wet, High Rainfall, equable temp., High Relative Humidity Very High

CAN Layers Considered For study • • • •

Elevation Slope Fuel Type Fuel Density et.al. 2008)

12/11/2016

: Obtained from CARTOSAT DEM V3 : Computed from CARTOSAT DEM V3 : Hypothetically Generated : Computed using FCD Model using IRS LISS III Imagery 2012 (Azazi

Ⓒ EERC @ IIIT-H

9

CAN Layers Elevation

Slope Morphometry

Fuel Type

Forest Density

12/11/2016

Ⓒ EERC @ IIIT-H

10

Forest Fire Simulation using CAN : Results Fire Spread Trend a

Fire Spread Trend Cases

b

a – Conventional CA Model with Fuel Type b –CAN Model with Fuel Type + Slope c

d c –CAN Model with Fuel Type + FCD

d –CAN Model with Fuel Type + Slope + FCD 12/11/2016

Ⓒ EERC @ IIIT-H

11

Forest Fire Simulation using CAN : Results CA Vs CAN Model a

a

d CA Model d

CAN Model (FT + FS + FCD)

12/11/2016

Ⓒ EERC @ IIIT-H

12

Forest Fire Simulation using CAN : Results CCA Vs CAN Model CAN Models b

b

FT + FS

c FT + FCD c d d FT + FS + FCD

12/11/2016

Ⓒ EERC @ IIIT-H

13

Conclusions • CA models are found to be more suitable to study the behavior of fire in the forest environment. • CAN models are more preferable over the conventional CA approach for their accuracy because of their multivariate considerations. • CAN models with more precise layers are found to be more reliable and accurate.

12/11/2016

Ⓒ EERC @ IIIT-H

14

Packages Used • • • • • • • •

Numpy – For matrix manioulations Matplotlib – For the graphs, contours Math – for mathematical expressions Image – Image processing Sys, Shutil – for kernel or terminal excecutions GDAL – Satelite image processing FFMpeg – Videos …. and more

12/11/2016

Ⓒ EERC @ IIIT-H

15

Future Scope • Inclusion of – Influence of wind – Crown fire spread behavior – Random spot fire behavior

• Incorporating more number of influential factors such as canopy height, aspect, etc. in CAN model • Application of CAN model to study the, – – – –

Post effects of forest fires on the stability of forest slopes Statistics of loss of fauna and flora Behavior of fire spread Etc.

• Can be reverse engineered ,to find the origin of the fire

12/11/2016

Ⓒ EERC @ IIIT-H

16

Thank You For your kind attention

“Research is creating a new knowledge”