1 of 11

 GROUND FUEL DISCOVERY

Jack Gihl

Cristian Washburn

Ram Prabu

Haofei Tian

1

5/11/2022

TFRSAC Spring 2022

2 of 11

Fire Fuels Discovery Package Use Overview

2

5/11/2022

Image adapted courtesy: https://www.newport.com/n/lidar

3 of 11

Technical Results 

3

5/11/2022

ML Algorithm

Category

Specifications

Canopy Fuel Algorithm

Training

-Number of Samples in data set > 1000:700

-Training error optimized to prevent underfit (high bias) and overfit (high variance)

Testing/Validation

-Regression Accuracy: 85%: ~70%

-Classification Accuracy: 95%: ~75%

Timing

-Maximum calculation time = 9 sec : ~60 sec (with minimum hardware requirements)

Ground Fuel Algorithm

Training

-Number of Samples in data set > 1000: 760

-Training error optimized to prevent underfit (high bias) and overfit (high variance)

Testing/Validation

-Regression Accuracy: 80%: >88%

Timing

-Maximum calculation time = 10 sec: 120 sec (with minimum hardware requirements)

Goal: Achieved

4 of 11

Fire Fuels Discovery Package: How we are training overview

4

5/11/2022

Image adapted courtesy: https://www.newport.com/n/lidar

Lidar Data

Canopy Plot Data

Ground Plot Data

Train CFA

Train GFA

DBH, CrHt, CrRad, Canopy Density

Duff, Litter, Ground Fuels (time)

5 of 11

Prototype Software

5

5/11/2022

3D view of LIDAR

Selecting input LIDAR file

Selecting Ground fuel parameters

Selecting Canopy fuel parameters

Selecting 3D or heat map

Output for CFA (Heat map or 3D view) 

Output for GFA (Heat map) 

6 of 11

Prototype - Canopy Fuel Parameters

6

5/11/2022

Fire Fuel Detector Prototype Software

Input LIDAR File

Canopy Height

Canopy Base 

Height

7 of 11

Prototype - Canopy Fuel Parameters

7

5/11/2022

Fire Fuel Predictor Prototype Software

LIDAR input

Small Branches

Dead/Live

Species

Crown Type

Fuel Weight

8 of 11

Prototype - Ground Fuel Parameters

8

5/11/2022

Fire Fuel Predictor Prototype Software

LIDAR input

f1h

f10h

f100h

duff

litter

9 of 11

Future Plans

  • Continued work with Andy Hudak and Ben Bright (USFS) 
  • In IP and copyright process for our software
  • Reach out to other forest services in other countries
  • Pursue SBIR grants to continue work
  • Move from Beta to full product
    • Leverage FIA data
    • Improve Accuracy 
    • Broaden ecological scope for algorithms
    • Smoother interface
    • Compatibility for transfer learning

​

​

9

5/11/2022

10 of 11

Demonstration

​

  • Video Link of the demo:  Demo Video

10

5/11/2022

11 of 11

Wrap Up

Thank you for your Participation

​

11

5/11/2022