stocks of tropical forests Ed Mitchard University of Edinburgh - - PowerPoint PPT Presentation

Mapping the changing carbon stocks of tropical forests

Ed Mitchard University of Edinburgh edward.mitchard@ed.ac.uk

Oxford Centre for Tropical Forests, 2nd November 2018

Massive uncertainty in in tropical carbon cycle

Tropical Aboveground Forest Carbon Stocks 170 – 280 Pg C (95 % C.I.)

Deforestation & degradation 0.5 – 3.1 Pg C/year Tree growth 0.9 – 2.85 Pg C/year

Carbon flux from tropics to atmosphere:

• 2.2 - +2.35 Pg C/year

Caused by uncertainty and disagreements in data from different methods (modelling, field studies, atmospheric measurements, different types of satellite data)  negative for REDD+ and climate modelling

2

Mitchard, E. T. A. 2018. The tropical forest carbon cycle and climate

• change. Nature 559 527-534

Why use satellite data to map tropical forest carbon stocks?

• Ground plots sparse &

expensive

• Cannot do enough!
• Cannot access all areas
• Need broad scale with

timely repeats

• Consistency

Optical satellite data

• About 500 satellites in orbit!
• Lots of data, mostly free
• Has been used to make maps of

forest cover and forest change (deforestation)

• Hansen et al (2013) (Global Forest

Watch) – Forest loss from 2000- 2017, based on analysis of 1 million Landsat scenes

• PRODES – Deforestation mapped

annually across Brazil

Issues with using conventional (optical) satellite data to monitor forests

• 1. Cloud cover
• 2. Only see the top of the

canopy

– biomass canopy cover

• 3. Spectral inconsistencies

The world as seen from the DSCOVR satellite May 2016

Data from MODIS L2 Cloud Mask Product (1km resolution)

• 1. Pristine
• 2. Degraded

Budongo Forest, Uganda

Greeness relates to canopy cover, not biomass

Normalised Difference Vegetation Index (NDVI) NDVI = (NIR – RED) (NIR + RED)

Wave Length (nm) Reflectance (%)

Canopy cover (m2/m2)

1 2 3 4

ASTER 2006 NDVI

0.25 0.30 0.35 0.40 0.45 0.50 0.55

Mitchard, E. T. A., et al. 2009. Measuring woody encroachment from 1982-2006 along a forest-savanna boundary in central Africa. Earth Interactions, 13, 1-29.

Inconsistent spectral signatures

• Grass, crops, secondary vegetation

and trees are all green at some point in the year

• Subtler changes occur too within

forests, relating to leaf age/phenology/flowers

• Changes too due to atmospheric

conditions, satellite platforms

Wet season Dry season

Types of remote sensing data

Optical Lidar Radar

LiDAR can image whole trees

LiDAR can map tree height, canopy size -> bio iomass

• LiDAR-estimated mean or top-of-

canopy height relate strongly to biomass within field plots, without saturation

• Can also use individual tree

recognition methods

• Depends on use of locally-derived

diameter:height relationships and wood density

Asner et al. 2010. High-resolution forest carbon stocks and emissions in the Amazon. PNAS 107: 16738-16742

Bio iomass = f( f(Dia iameter, wood density, height) LiDAR measures height only

Diameter Height Wood density

LiDAR Problems

• Data volumes large, processing

tricky & site specific

• Continuous-cover data is from

aircraft/UAV only

• Expensive
• Rare
• Non-operational
• Satellite data is dispersed

footprints

• Average figures over many trees
• Still rare – sampling tool
• Cloud cover a problem

ICESat Orbits (footprints along tracks 170 m apart, 70 m diameter)

Satellite LiDAR has been used to make pantropical biomass maps

Ground plots Spaceborne LiDAR data (ICESat GLAS) Spatial Averaging Tens of thousands of point biomass estimates Tens of thousands of point biomass estimates Full coverage

• ptical (MODIS)

and elevation (SRTM) data Spatial Modelling AGB maps Testing

Saatchi et al (2011, PNAS)

Baccini et al (2012, Nature Climate Change)

Unfortunately these biomass maps differ quite a bit

Baccini et al. (2012) Saatchi et al. (2011)

• 150 -100 -50 0 +50 +100 +150

Absolute Difference

AGB (Mg ha-1) AGB (Mg ha-1)

0 100 200 300 400 500

AGB (Mg ha-1)

0 100 200 300 400 500

Why do they differ? Do either match reality?

Wood density

Feldpausch et al. 2011 Biogeosciences Ter Steege et al. 2006 Nature

Diameter-height relationships

Field plots show a different pattern to both maps over Amazonia

413 plots

Mitchard et al. 2014. Markedly divergent estimates of Amazon forest carbon density from ground plots and satellites. Global Ecology & Biogeography

• 200
• 100

Saatchi-Krig Saatchi Baccini Kriged from plots Baccini-Krig Saatchi-Baccini

Must consider spatial variation in wood density and forest structure

≠ ≠

DBH: 30 cm H: 25 m DBH: 100 cm H: 25 m Density: 0.9 Density: 0.45

Conclusion on Optical & Lidar

• Optical data good for mapping canopy

cover (e.g. forest area, deforestation)

• But cloud cover
• But inconsistent in time
• LiDAR data good for biomass
• But height only – biases
• Data limited
• Combined Optical+LiDAR made the

first pantropical biomas maps

• But biases
• But limited to mid-2000s

 GEDI & ICESat-2 launches mean new satellite data from 2018/19

≠ ≠

There is another technology…

Optical Lidar Radar

Radar

Radar

• Side-looking
• Signal ‘scattered’

from trunks and branches

Active radar pulse Return signal (backscatter)

Optical vs. Radar

Radar interactions with forest structure

Backscatter Measurements:

฀ (H,V ) crown stem ground

(H,V) (H,V)

฀ 

฀ (H,V )  f (vol,Wd,) vol: forest volume (size) Wd : wood density (dielectric constant)  : shape and orientation of components

AGB (Mg ha-1)

ALOS HV (dB)

• 20
• 19
• 18
• 17
• 16
• 15
• 14
• 13
• 12
• 11
• 10
• 9

AGB (Mg ha-1)

ALOS HV (dB)

• 20
• 19
• 18
• 17
• 16
• 15
• 14
• 13
• 12
• 11
• 10
• 9

R2 = 0.73 p < 0.0001 n = 253

Radar backscatter is sensitive to biomass

Cameroon - MD Mozambique - NC Mozambique - NN Uganda - BFR

HV backscatter vs AGB HV backscatter vs log AGB

Mitchard, E. T. A., et al. 2009. Geophysical Research Letters

Biomass (Mg ha-1)

Biomass map – Mbam Djerem, Cameroon, 2007

2000 2006 2000 2006

Road Railway Mbam Djerem National Park Large towns

Radar wavelengths

C-band L-band P-band La Selva, Costa Rica. Radar data at different wavelengths

Wavelength

Short Long Low High

Biomass sensitivity

High Low

Maximum resolution

Radar satellites

Ban Band Wavele length Typ ypical ma maximum reso esolu lutio ion fr from

• m orbi
• rbit

Sa Satell llit ites X-band 2.5-3.75 cm ~1 m Ter erraSAR-X (20 (2007-) Tan anDEM-X (20 (2010-) COS OSMO-SkyMed (20 (2007-) C-ba band 3.75-7.5 cm ~3 m ERS-1 (1991-2000) ERS-2 (1995-2011) ASAR (2002-2012) RAD RADARSAT 1 (1995 1995-) RAD RADARSAT 2 (20 (2007-) Sen Sentin inel-1 (20 (2014-) S-ba band 7.5-15 cm ~6 m No NovaSAR (20 (2018-) L-ba band 15-30 cm ~20 m JERS-1 (1992-1998) ALOS PALSAR (2007-2011) AL ALOS-2 PALS ALSAR-2 (20 (2014-) SAOCOM (20 (2018-) NISAR (2019) P-band 70-130 cm ~50 m BIOMASS (2021)

Interferometry (tree height)

B

A1 A2

ground

H hv hg

฀ 

฀  ฀  

Phase center

ESA BIOMASS – 2021-25

Biomass Level-2 biomass t/ha 200m resolution (based on CESBIO algorithm using HV) Saatchi et al. (2011) biomass t/ha 1 km resolution Baccini et al. (2012) biomass t/ha 463 m resolution

Example data over Lope National Park, Gabon, produced using P-band aircraft campaign

Can use radar/optical in combination – e.g. mapped peatlands in Amazon and Congo

• Our research has increased known tropical peat stocks by ~35%

through mapping peatland in Congo (Dargie et al. 2017. Nature)

Conclusions - radar

• Sees through clouds
• Penetrates vegetation
• Structural information on forests
• But saturation – cannot map higher

biomass forests or changes

• No long-wavelength satellite

currently – though BIOMASS launched 2021

• Data not normally free

So what can we do to map biomass and biomass changes in tropical forest?

• Aircraft-based LiDAR works – but expensive
• Satellite-based LiDAR very patchy
• Radar data saturates

Wait for BIOMASS? – but only 2021-25, one-off mission Can we do better with current technologies?

We are entering the Sentinel era

• Sentinel 1 – C-band radar – 2 satellites
• 4 TB data per day
• 6/12/24 day repeat cycle
• Sentinel 2 – Optical 10m - 2 satellites
• 6 TB data per day
• 5 day repeat cycle
• Sentinel 3 – optical 300m (sea/land temperature and

colour) – 1 satellite

(3 from 2020)

• global coverage every 2 days
• 1.5 TB data per satellite per day
• Can we use these satellites to map biomass? Or at least

biomass change?

FODEX

Tropical Forest Degradation Experiment

42

Main problem: no suitable ground truth plots for change

Lack of forest fie ield ld plo lots wit ith la large ch change valu lues prevents alg lgorithm development and testin ing

Change in biomass in forest plot t1 to t2

% of initial biomass 0% 20 % 40% 60% 80% 100% 120 % DEGRADATION RANGE OF CHANGE IN FIELD PLOTS

43

The solution - FODEX

• Create 52 Permanent Sample Plots
• 26 in Gabon
• 26 in Peru

Before & after logging, twice during recovery:

• 1. Measure all trees
• 2. Collect Terrestrial Laser Scanner data
• 3. Collect LiDAR from Unmanned Aerial Vehicle

44

Using the fi field data

• Collect ALL useful satellite data over sites
• ~20 satellites – optical, radar & LiDAR
• Create and test algorithms for mapping biomass

change

• Single satellites – e.g. machine learning on

dense time series of optical satellite data

• Combinations – e.g. LiDAR + radar
• Classification vs continuous maps
• Create biome-scale maps
• Used by countries for REDD+
• Used by climate modellers

45

Any questions?

Thanks to :

Murray Collins (U. Edinburgh) Iain McNicol (U. Edinburgh) Simon Lewis (UCL/Leeds) many others… NASA & USGS JAXA & METI

edw edward.mitchard@ed.ac.uk @edmitchard www.g .geos.ed.a .ac.uk/homes/emitchar www.mit itchardgroup.wordpress.com