+ - 0:00:00
Notes for current slide
Notes for next slide
These slides are viewed best by Chrome and occasionally need to be refreshed if elements did not load properly.


Press the right arrow to progress to the next slide!

1/19


Using Remote Sensing Data to Understand Fire Ignition during the 2019-2020 Australia Bushfire Season


Weihao Li, Chang L., Di Cook, Emily Dodwell

https://bit.ly/VicBushFireIgnition

ACEMS Retreat, Nov 6 2020


1/19

People and motivation

Weihao Li Emily Dodwell
Monash EBS Honours AT&T research


Motivation: Spatio-temporal visualisation and analysis of emergency call data. This is private so the bushfire data was collected because it has some similar form and structure.

2/19

🗺 Background

⚡ Lightning or 🔥Arson?

3/19

📡 Remote sensing data

Japan Aerospace Exploration Agency provides a hotspot product (reflected energy from the earth) taken from the Himawari-8 satellite.

Example code to access data provided in a G. Williamson gist post

4/19

📂 Data Sources

🔥 Historical fire origins: 2000-2019 Department of Environment, Land, Water and Planning

📡 Remote sensing data: Japan Aerospace Exploration Agency

Wind speed data: 1-day, 7-day, ..., 2-year averages from Commonwealth Scientific and Industrial Research Organisation and Automated Surface Observing System

Temperature, Rainfall and Solar exposure: 1-day, 7-day , 14-day, 28-day, ..., 720-day averages computed from Bureau of Meteorology

Fuel layer: Forest type, forest height class, forest crown cover from Australian Bureau of Agricultural and Resource Economics

Road map: Proximity to the nearest road using OpenStreetMap

Fire stations: Proximity to the nearest CFA station Department of Environment, Land, Water and Planning

Recreation sites: Proximity to the nearest camping site Department of Environment, Land, Water and Planning

5/19

💻 Data fusion

6/19

💻 Detect ignitions by clustering hotspot data

7/19

💻 Estimated ignitions

76,000 hotspots reduced to 1,000 ignition sites.

8/19

📊 Exploratory analysis of historical fire origins

Text processing of 26 causes, reduced to four major causes. Lightning and accident were the two main sources of historical bushfire ignitions, which took up 41% and 34% respectively. There were 17% bushfires caused by arson.

9/19

📊 Spatial distribution of historical fire origins

Roughly different spatial locations of ignition causes. Lightning bushfires were concentrated in the east of Victoria. Bushfires caused by arson were near Bendigo!

10/19

📊 Proximity of historical fire origins

Lightning-caused bushfires were further away from the CFA stations and roads. In contrast, bushfires caused by arson were closer to CFA stations and roads.

11/19

📐 Modelling

A random forest model outperformed other model choices to classify different causes of bushfire ignition.

80% of the data used as training set, 7497 observations, and the remaining 1872 observations was used as test set.


Model performance was compared using multi-class AUC (Hand and Till, 2001).

Model Accuracy Muti-class AUC
Multinomial logistic regression 0.53 0.74
GAM multinomial logistic regression 0.68 0.82
Random forest 0.75 0.88
XGBoost 0.74 0.88
12/19

📐 Model performance

The overall accuracy of our model was 74.95%.

  • High accuracy with lightning and accident ignitions.
  • Less accurate predictions for arson and burning off.


Lightning Accident Arson Burning_off Total
Prediction:Lightning 703 (0.9) 77 (0.12) 50 (0.15) 44 (0.32) 874
Prediction:Accident 51 (0.07) 494 (0.78) 89 (0.27) 38 (0.28) 672
Prediction:Arson 18 (0.02) 55 (0.09) 175 (0.54) 22 (0.16) 270
Prediction:Burning_off 5 (0.01) 8 (0.01) 11 (0.03) 32 (0.24) 56
Total 777 634 325 136 1872
13/19

📐 Model interpretation

Variable importance assessed using Local Interpretable Model-agnostic Explanations (lime). Proximity to the nearest road, proximity to the nearest road and average wind speed had largest influence on the prediction.

14/19

🔎 Prediction for 2019-2020 Australia bushfires

15/19

🔎 Summary of findings

  • Majority of the bushfires in 2019-2020 season were caused by lightning.
  • 138 bushfires caused by accidents which took up 14% of the total fires. Most of them were ignited in March.
  • 37 bushfires were caused by arsonists, and over half of them were in March.
  • Very few planned burns were predicted after October 2019 which suggests the correctness of our model.


Cause Oct Nov Dec Jan Feb Mar Total
Lightning 19 57 315 266 32 149 838 (0.82)
Accident 3 8 34 13 0 80 138 (0.14)
Arson 2 2 10 2 0 21 37 (0.04)
Burning_off 7 0 2 0 0 0 9 (0.01)
16/19

📜 Policy implications

  • CFA and FFMV may need to consider fire prevention solutions in remote areas of Victoria to reduce the risk of lightning-ignited bushfire. Perhaps new strategically located CFA stations might improve accessibility and speed of response.
  • CFA should be asked about the increasing number of accident-caused bushfires. Could it be a policy/coding change. Are there substantially more careless campers, or logging operations?
  • Arson is a small source of fires, and primarily a problem closer to populated areas. Possibly sensor network might be recommended.
17/19

📜 Summary

  1. Algorithm to detect bushfire ignition from hotspot data
  2. Model to predict the cause of bushfire ignition
  3. Prediction of the causes of the 2019-2020 Australia bushfires
  4. A complete and adaptable workflow for monitoring and understanding new ignitions from hotspot data
  5. Shiny app for exploration of historical fire origins, predicted causes of 2019-2020 fires and future fire risk maps
  6. All work conducted with open data and open source software.
19/19



Acknowledgements

Slides produced using Rmarkdown with xaringan styling. Monash style by the kunoichi, Dr Emi Tanaka.

Data and code for analysis is available on Weihao's GitHub repo. Shiny app code is available at Chang's GitHub repo.

Thanks for listening!


Creative Commons License
This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.

19/19


Using Remote Sensing Data to Understand Fire Ignition during the 2019-2020 Australia Bushfire Season


Weihao Li, Chang L., Di Cook, Emily Dodwell

https://bit.ly/VicBushFireIgnition

ACEMS Retreat, Nov 6 2020


1/19
Paused

Help

Keyboard shortcuts

, , Pg Up, k Go to previous slide
, , Pg Dn, Space, j Go to next slide
Home Go to first slide
End Go to last slide
Number + Return Go to specific slide
b / m / f Toggle blackout / mirrored / fullscreen mode
c Clone slideshow
p Toggle presenter mode
t Restart the presentation timer
?, h Toggle this help
Esc Back to slideshow