Forest Structural Classification and Above Ground Biomass Estimation for Australia

Professor Richard Lucas1

Jingyi Sun2

Centre for Ecosystem Sciences (CES) School of Biological, Earth and Environmental Sciences (BEES) University of New South Wales (UNSW) Australia Sydney, NSW, Australia 2School of Engineering 1Joint Remote Sensing Research Program (JRSRP)

GFOI R&D and GOFC-GOLD Land Cover Science Meeting, The Hague, The Netherlands 31 October – 4 November, 2016

John Armston1 Peter Scarth2

GEDI Science Team Department of Geographical Sciences 1150 Lefrak Hall, University of Maryland, College Park MD 20742, USA 3University of Queensland, St Lucia Campus, Brisbane, Qld, Australia.

Overview

• Australia’s woody vegetation – Structural classification (height and cover)

– Dominant floristics

• Defining remnant (undisturbed from direct human activity) woody vegetation – National datasets

– State-wide datasets

• TERN Biomass Library – Plot data

– Lidar data

• Identifying reference sites for undisturbed forests.

• Case study: Brigalow Belt Bioregion, Queensland

• Future sensors and updating the biomass library

• GFOI Study Site: Injune Landscape Collaborative Project

Structural Classification of Australia’s Woody Vegetation

Generated using a combination of Spaceborne optical, radar and lidar data

Structural Classification of Australia’s Woody Vegetation

Dominant Forest Types National Forest Inventory Australia

Defining Remnant Forest

• Remnant forests in Australia are defined as those that have remained undisturbed by human activities since European settlement.

• Areas of remnant vegetation mapped in Queensland through reference to historical aerial photography

• For other states, undisturbed (typically remnant) forests need to be inferred from other mapping efforts.

National Datasets: Protected Areas of Australia

Statewide Datasets

South Australia Victoria

Tasmania New South Wales and ACT

Queensland

• Remnant forests in Australia are defined as those that have remained undisturbed by human activities since European settlement.

• Areas of remnant vegetation mapped in Queensland through reference to historical aerial photography

• For other states, undisturbed (typically remnant) forests need to be inferred from other mapping efforts.

Defining Remnant Forest

TERN Biomass Library, Australia

1,073,837 hugs of 839,866 trees 1,467 tree species 15,706 plots from 16,391 observations across 12,663 sites

Australian National Biomass Library (2016).

National Airborne Lidar Validation Datasets

Summary of Biomass Library

Biomass Library Summary

Remnant Forest Plots with LIDAR

A Unique Mosaic Product for Australia

Landsat-derived persistent green, ALOS HH and HV in RGBt

Forest Growth Stage Mapping

Landsat FPC and ALOS PALSAR

L-band HH and HV (RGB)

Differentiation of early regrowth

and remnant forest

−2

0−

15

−1

0−

5

HH

(g0

; d

B)

−2

5−

20

−15

HV

(g0

; d

B)

010

30

50

70

FP

C (

%)

11

.3.1

11

.4.3

11

.4.7

11

.4.8

11

.4.9

11.4

.10

11.5

.16

11

.9.1

11

.9.5

.a

11.9

.5.b

11

.9.6

11

.11

.14

11

.12

.21

Characteristics of Remnant (undisturbed forest): Different Regional Ecosystems - Brigalow Belt Bioregion

Regional Ecosystem

Integration of ALOS PALSAR and Landsat-derived FPC

Forest Growth Stage: Brigalow Belt Bioregion

Regional Ecosystem Mapping Growth stage map

Quantifying Relative Rates of Degradation and Regeneration

Regrowth Recovery from Fire

Mature Pine and Eucalyptus

Recovering the Endangered Brigalow

Forest Ecosystems, Queensland, Australia

Regrowth classification Relevance to Vegetation Management Acts, Australia

-1500

-1000

-500

0

500

1000

1500

Pre

-19

84

19

84

19

85

19

86

19

87

19

88

19

89

19

90

19

91

19

92

19

93

19

94

19

95

19

96

19

97

19

98

19

99

20

00

20

01

20

02

20

03

20

04

20

05

20

06

20

07

20

08

20

09

20

10

20

11

20

12

20

13

20

14

20

15

20

16

20

17

20

18

20

19

20

20

20

21

20

22

20

23

20

24

20

25

WA

VIC

TAS

SA

QLD

NT

NSW

Lidar

P-band

L-band

C-band

Optical

Temporal Analysis of Biomass Library

5

10

15

Year of Data Collection

Number of satellites supporting regional to global biomass mapping

Optical C-band L-band P-band Spaceborne lidar

Landsat-7 ERS-1 SAR JERS-1 SAR BIOMASS ICESAT GLAS

Landsat-8 ERS-2 SAR ALOS PALSAR ICESAT-2

Sentinel-2 RADARSAT-1 ALOS-2 PALSAR-2 GEDI ON ISS

RADARSAT-2 SAOCOM CONAE

Sentinel-1 NISAR

Freq

ue

ncy

The Injune Landscape Collaborative Project

• RESEARCH OBJECTIVES: – To extend methods development using SAR and sensor synergy for

deforestation and degradation monitoring and retrieving estimates of AGB.

– Use time-series to better understand and quantify ecosystem response to natural and human drivers.

• OUTCOMES: – Optimised algorithm for retrieving AGB using multi-sensor data.

• Report on sensor synergy for improved AGB estimates.

– Algorithm for deforestation monitoring using time-series, multi-sensor data.

– Forest degradation mapping method using time-series, multi-sensor data.

– Report on sensor interoperability and complementarity for deforestation monitoring and degradation assessment, and landscape response natural and anthropogenic induced change.

Quantifying Tree Level Change: Injune, Qld

Tree species dynamics detected a) in the field and b) using time-series of LIDAR data

2006

2000

Datasets requested through GFOI

Forest Disturbance Monitoring: ALOS PALSAR Correlation (August – October, 2007)

Yang et al. (2016). Observation of vegetation vertical structure and disturbance using L-band InSAR over the Injune region in Australia (in preparation).

5th Aug 2007 13th Aug 2007 21st Aug 2007 21th Aug 2007 6th Sep 2007

14th Sep 2007 22nd Sep 2007 30th Sep 2007

8th Oct 2007

Hyper-temporal Landsat FPC (actual, simulated)

Hyper-temporal Landsat FPC (actual, simulated)

5th Aug 2007 13th Aug 2007 21st Aug 2007 21th Aug 2007 6th Sep 2007

14th Sep 2007 22nd Sep 2007 30th Sep 2007

5th Aug 2007

CARL framework • CARL Level 4 - Pre-operational • Table 1:

• Demonstration in a larger-scale environment. • Representative model or prototype method (near the desired performance),

which is well beyond that of level 3, is tested in a larger-scale environment. • End-to-end processing demonstrated. • Data processing methods have been (partially documented) in peer review

publications (submission for December 2016). • Methods have NOT been assessed for applicability in different forest

monitoring contexts.

• Table 2 • Prototype is available and used by different (1-2) experts, sources of

uncertainties are known and can be quantified. • Data are available for large area/national demonstrations in different tropical

country conditions • Work mostly done in research environment • Training materials/tutorials and guidance documents NOT YET developed

NOR tested in countries

• Briefly discuss the contribution of your work in the context of CARL

• Feedback on v1 of CARL framework • Assets • Limitations • Suggested modifications (will be developed further on Day 2)

CARL Framework Implementation

• Contribution to CARL • Provides national methods for:

• Quantifying vegetation height and cover • Generating open data bases of structural measures and biomass. • Discriminating and mapping relative stages of degradation and regeneration.

• CARL Framework (Version 1) Feedback • Assets

• Capacity to understand/undertake programming in development or implement programed software.

• Capacity to integrate data from different sources • Understanding of the information content of different data sources • A strong and understandable validation dataset to quantify uncertainty.

• Limitations • Historical data (e.g., ICESAT) • Relatively complex algorithms in combination.

• Suggested modifications • Methods that work well in some countries (e.g., Australia), including non-

tropical, but not applied yet to tropical countries.

CARL Framework Implementation

37th ISRSE, Tshwane (Pretoria), South Africa • As part of the 37th International Symposium on Remote Sensing of the

Environment (ISRSE) in Tshwane (Pretoria), South Africa (8-12th May, 2017), two special sessions (6 presentations each) may be of interest to you: – Session 1: Remote Sensing in Support of Ecosystem Restoration: Often, remote sensing

technologies have informed on the loss or degradation of ecosystems but there is considerable potential to also use these data to plan and monitor the restoration of previously lost or disturbed ecosystems. This session aims to highlight this potential, with particular (but not exclusive) emphasis on projects/ideas that focus on restoring significant ecosystems across large areas.

– Session 2: Interoperability for quantifying forest structure and biomass: This session aligns

with that on ecosystem restoration in that it asks how remote sensing data (optical, radar and/or lidar) have or can be used to quantify the structure (e.g., height and vertical distribution of plant material, cover) and biomass of vegetated ecosystems including and relative to the ‘undisturbed state’. Though this approach, relative states of degradation and regeneration can potentially be mapped and described across large areas, with this assisting future conservation and restoration planning.

• From these sessions, we anticipate a Special Issue of Remote Sensing for Ecosystem

Restoration for Remote Sensing in Ecology and Conservation, to be published in late 2017/early 2018.