Multi-temporal LIDAR data for forestry an approach … › 2018 › presentations › Multi...Forest...

11 April 2018 Locate Conference

Forest yield predictions from multi-temporal LiDAR data

UniSA

Stefan Peters, Jixue Liu, David Bruce, Jiuyong Li

ForestrySA

Jim O’Hehir, Mary-Anne Larkin, Anthony Hay

Multi-temporal LIDAR data for forestry –

an approach to investigate timber yield changes

1



Intro

Study

area

Data-

processing

Modeling

Conclusion

• Use of existing ALS datasets

• Model and predict the ‘change’

Why investigating Multi-Temporal LIDAR data ?

2



Intro

Study

area

Data-

processing

Modeling

Conclusion

𝑃𝑖𝑛𝑒 𝑡𝑟𝑒𝑒 𝑊𝑜𝑜𝑑 𝑉𝑜𝑙𝑢𝑚𝑒 = 𝐷𝐵𝐻

200

2∗ 3.142 ∗ 𝑡𝑟𝑒𝑒_ℎ𝑒𝑖𝑔ℎ𝑡 ∗ 0.35

Determining Timber Yield

3

nikon.com

Measuring circumference DBH

𝐷𝐵𝐻 = 𝐷𝑖𝑎𝑚𝑒𝑡𝑒𝑟 𝑎𝑡 𝑏𝑟𝑒𝑎𝑠𝑡 ℎ𝑒𝑖𝑔ℎ𝑡 (𝑎𝑡 1.37𝑚)

Measuring tree height



Intro

Study

area

Data-

processing

Modeling

Conclusion

1) Forest plots:

ground truthing

wood volume derived from

measured DBH and tree heights

2) ALS over the entire area

metrics

3) Cut plots out of entire ALS

metrics

4) Build model

WoodVolume = f plot_LiDARmetrics

5) Impute plot-level Wood Volume to a

stand level target grid

Wood Volume from LiDAR metrics

4

ALS

Vol (m3 / ha)

Legend

Y time1

Yield_p_Ha

50 - 55

56 - 60

61 - 65

66 - 69

70 - 72

73 - 75

76 - 80

81 - 85

86 - 95

Y time2

Yield_time

0 - 50

50 - 100

100 - 150

150 - 200

200 - 250

250 - 300

350 - 400

400 - 450

450 - 500

Plot LiDAR metrics > 100 descriptors of the cloud

Point density, return intensity, height

(std dev, skewness, avg, percentiles…)



Intro

Study

area

Data-

processing

Modeling

Conclusion

5



Intro

Study

area

Data-

processing

Modeling

Conclusion

• Single tree type: Radiata Pine

• Planted in 1997

• ~ evenly distributed

Forest features

6



Intro

Study

area

Data-

processing

Modeling

Conclusion

Discrete multi-return Airborne Lidar Scanning (ALS)

7

Flight / data characteristics 2012 ALS flight 2015 ALS flight

Data supplier De Bruin Spatial Technologies AAM

Date of acquisition 15.01.2012 01.04.2015

Data format LAS 1.2 LAZ 1.4

Flying altitude (m ASL) 800 1300

Scan Angle (°) 15 28

Scan overlap (%) 25 15

Mean footprint diameter (cm) 19 25

Mean point density 8.9 8

Point density range inside inventory plots (m-2)

3.4 - 9.2 4 - 8

Horizontal / Vertical accuracy 0.5 m / 0.25 m RMS (1 Sigma) 0.5 m / 0.25 m RMS (1 Sigma)

Datum and projection GDA 1994 / MGA Zone 54 GDA 1994 / MGA Zone 54

ALS metadata



Intro

Study

area

Data-

processing

Modeling

Conclusion

LiDAR point cloud (2015) – clipped plots

8



Intro

Study

area

Data-

processing

Modeling

Conclusion

LIDAR point cloud (LIDAR.las)

LIDAR.laz

Forest boundaries

(.shp)

Initial input data

plot boundaries (.shp)

Ground truth plot tree data: heights, DBH, age etc. (.xls)

forest.laz

convert to

clip

remove outlier

normalize

CHM (.img)

Spike free CHM

Spatial alignment

temporal alignment with

LIDAR flight date

tree wood volume

calculation

LIDAR.laz

Tree / plot wood volume (and

further change information, such as thinning or fire)

buffer 2m

PlotTreeBuffer Polygons.shp

thiessen

PlotTree Thiessen.shp

clip

clip

PlotTreeCrowns.shp

Zonal Statistics - Build Raster

Attribute Table - Raster to Polygon

PlotTreeCrowns withHeight.shp

for each

temporal

LIDAR data

set

clip

ArcGIS model Python script

TreeCrowns.laz

lascanopy

Tree Crown Forest Metrics development of LIDAR derived

wood volume MODEL (e.g. kNN imputation) create

Fishnet

Grid Cells .shp

apply MODEL (imputation)

Forest / Stand Wood Volumes

Lastools batch scripts

aggregate cells

lascanopy

Grid Forest Metrics

LiDAR and

GIS data

processing

flowchart

plot tree stem xy positions

On screen digitizing or TS/GNSS or ITD

input

Method/ tool

output

ArcGIS

Lastools

Grid Cells with predicted

Wood Volumes 9



Intro

Study

area

Data-

processing

Modeling

Conclusion

Plot data

Timber yield modelling: Cost effective update of timber yield estimates

10

time 2 to be

updated

time 1

ALS 1

Vol (m3 / ha)

Legend

Y time1

Yield_p_Ha

50 - 55

56 - 60

61 - 65

66 - 69

70 - 72

73 - 75

76 - 80

81 - 85

86 - 95

Y time2

Yield_time

0 - 50

50 - 100

100 - 150

150 - 200

200 - 250

250 - 300

350 - 400

400 - 450

450 - 500

ALS 2

Plot data



Intro

Study

area

Data-

processing

Modeling

Conclusion

time 2 to be updated

ALS 2015

…first test at our study area

11

time 1

ALS 2012

Timber yield modelling: II) Cost effective update of timber yield estimates



Intro

Study

area

Data-

processing

Modeling

Conclusion

Approach A:

12

* K-Nearest Neighbors prediction model

(k=2, 6-fold cross validation)

plot lidar data

time 1

Approach B:

I grid wood volume time 1

plot lidar data

time 2

II grid level wood volume growth rate (time 1 - 2)

grid wood volume time 2

grid wood volume time 2

measured plot volume

time 2

measured plot volume

time 1

plot lidar data

time 1

measured plot volume increments (time 1 - 2)

kNN* classification

grid level lidar data

time 1 Apply

kNN* classification

grid level lidar data

time 1 Apply

kNN* classification

Apply



Intro

Study

area

Data-

processing

Modeling

Conclusion

Approach A and B grid prediction: relative difference: 1%

Models A and B applied to predict grid-level wood volumes

13

Approach A Approach B

GridID Pred Vol 2015 GridID Pred Vol

2012 Pred Vol Rate Calc Vol 2015

1 24.23 25 17.805 0.379597 24.5637

2 22.46 26 17.805 0.381722 24.6016

3 22.46 27 17.58 0.381722 24.2907

… … … … … …

263 24.12 263 16.395 0.381722 22.6533

264 24.12 264 15.7 0.423105 22.3427

Vol total:

RMSE (plots):

5751.41

1.51

Vol total:

RMSE (plots):

5783.8

1.70

all wood

volumes in m3



Intro

Study

area

Data-

processing

Modeling

Conclusion

Resulting Grid map compared with Site Quality map (2007)

14

Site Quality map (Manual assessment) of wood volume at age 10 (2007)

2015 LiDAR based volume prediction



Intro

Study

area

Data-

processing

Conclusion

Modeling

Proof of concepts:

- Multi-temporal LiDAR data added values

- Model, estimate, predict change of timber yield

- ALS + UAV-LiDAR for cost efficient update of yield estimates

…potentially applied to other applications

(agriculture, vegetation monitoring…)

Conclusion

15



Intro

Study

area

Data-

processing

Conclusion

Modeling

• Enhance wood volume change model and future prediction

– Larger test area

– 3+ multitemporal LiDAR scans

– Higher point density (30 ppm2)

• Investigate change in more detail

– Change of basal area, biomass, …

– Thinning, Harvested trees, gap dynamics, damaged crowns

Outlook

16



Further information:

17

[email protected]

Many thanks

Multi-temporal LIDAR data for forestry an approach … › 2018 › presentations › Multi...Forest...

Documents

Transcript of Multi-temporal LIDAR data for forestry an approach … › 2018 › presentations › Multi...Forest...