Total Forest Carbon Using UAV Data

1
Total Forest Carbon Estimation Using 3D Data from
UAV Imagery
By Jwan M. Aldoski Geospatial Information
Science Research Center (GISRC), Faculty of
Engineering, Universiti Putra Malaysia, 43400
UPM Serdang, Selangor Darul Ehsan, Malaysia.
2
Outline

• Introduction
• Methods for Forest Carbon Measuring
• Study Area
• Data Used
• Satellite Data
• UAV Data
• What is the UAV?
• How UAS Works?
• Purposes for UAS and Benefits?
• Data Used and Process
• Methodology
• Applied Techniques and Analysis Methods
• Research Expected Results
• Research Timetable

3
Introduction

• Carbon is one of the most common elements on
  earth, and is found in all living organisms
• Carbon is the basis of most molecules found in
  vegetation Carbohydrates, Sugars, Fats,
  Proteins, Alcohols, DNA, Chlorophyll
• Big problem now its high level as co2 in
  atmosphere

4
Introduction

• Forest trees take co2 from atmosphere and stored
  in five pools within and around vegetation
• Above-ground Biomass stems, bark, leaves, etc
  of tree and non tree plants
• Below-ground Biomass roots of all sizes of tree
  and non tree plants
• Dead Wood
• Litter
• Soil Organic Carbon (SOC)

5
Introduction
Accurate forest biomass and carbon measurement
are necessary for

• managing forest resources, informing climate
  change modeling studies, and meeting national and
  international reporting requirements for
  greenhouse gas inventories IPCC and REDD .
• also necessary at the sub-national level for
  purposes such as completing the Malaysia Forest
  Service Climate Change Scorecard that
  necessitates annual estimates of carbon stocks
  and fluxes for each National Forest , and for
  quantifying changes in forest biomass on regional
  scales in response to disturbance.

6
(No Transcript)
7
Forest Pools and Carbon Measurement Methods
Pool Methods Suitability
Above-ground tree biomass Plot Very suitable and cost-effective, commonly adopted and familiar. Plot selection is key to the method
Above-ground tree biomass Plot-less, transect Good but not suitable in dense vegetation
8
Forest Pools and Carbon Measurement Methods
Pool Methods Suitability
Above-ground tree biomass Harvest Expensive, time consuming, not appropriate all the time. Used to develop allometric equations
9
Forest Pools and Carbon Measurement Methods
Pool Methods Suitability
Above-ground Tree Biomass Modeling or Allometric Method Suitable for projections, requires basic input parameters from field measurements
10
Forest Pools and Carbon Measurement Methods
Pool Methods Suitability
Above-ground Tree Biomass Carbon flux measurements Expensive and needs skilled human resources
Eddy covariance instrumentation
11
Forest Pools and Carbon Measurement Methods
Pool Methods Suitability
Above-ground Tree Biomass Remote sensing Needs field measurements for calibration. Data are usually at large spatial scales, needs expertise to be used and can be expensive
12
Forest Pools and Carbon Measurement Methods
Pool Methods Suitability
Below-ground Tree Biomass Root extraction and mass measurement Expensive and not suitable at large scales
Below-ground Tree Biomass Root to shoot ratio This study Most commonly used Requires AGB measurement
Below-ground Tree Biomass Biomass equations Requires input data e.g. height, diameter, girth
13
Combing Remote Sensing Data Ground Data
Remote Sensing Forest Carbon Measurement Method

• Two primary methods
• 1) Stratify And Multiply
• assigns a biomass value, or a range of biomass
  values, to areas of land distinguished by
  characteristics such as vegetation type or land
  use.
• Limitation
• uses ground-based measurements to determine
  biomass values
• the ambiguities present in land area
  classification
• the wide range of variability in aboveground
  biomass within a given land cover type
• Most country (such as Malaysia) under redd
  program using this methods (Lu et al.,2017)

14
Remote Sensing Forest Carbon Measurement Method

• 2) Direct Mapping Approach
• It employs a set of spatially continuous
  variables to predict biomass values or carbon at
  unobserved locations.
• The direct mapping approach takes advantage of a
  variety of geospatial variables, such as,
  spectral , vegetation indices, backscattered
  energy, climate and topography, and other
  information from remote sensing platforms like
  Optical Data, Radar and LiDAR
• Advantage
• Resulting map are more accurate across the
  landscape
• Update changes are easier

15
Remote Sensing Forest Carbon Measurement Method
Pool Methods Suitability
Above-ground Tree Biomass Below-ground Tree Biomass Remote sensing data ground data Needs plots measurements for calibration. Data are usually at large to small spatial scales, needs expertise to be used and can be expensive for large area.
16
Most of country under redd program using these
RS Imagery
Remote Sensing Forest Carbon Measurement Method
RS images types used Multispectral images,
Radar images and Lidar images
Spatial resolution Sensor
Coarse MODIS
gt250 m LANDSAT MSS
Medium Landsat ETM 7
20-250m ASTER
30m LANDSAT
RADARSAT
Fine IKONOS
lt20m SPOT-5
Lidar
17
Challenges Remote Sensing Forest Carbon
Measurement Method

• RS needs to be calibrated with field measurements
• Some satellite imagery is very expensive
• RS data requires technical expertise to be
  interpreted
• Clear and practical methodologies are needed not
  only in field measurements, but also in the
  application of remote sensing
• New technology and methodologies (e.g. Lidar
  technique data acquisition, radar data) could
  contribute further to improve precision and
  accuracy of assessment, if their costs could be
  brought down.
• However , with availability of two freely
  satellite base data Landsat 8 and sentinel open
  a new technology for forest biomass and carbon
  estimation it needs to be applied in tropical
  forest.

18
UAV Forest Carbon Measurement Method

• It also called Drones, Remotely Piloted Vehicle
  (RPV) by the Federal Aviation Administration
  (FAA) adopted by United States Department of
  Defense (DOD) Civil Aviation Authority(UK).
• The Forest Biomass and Carbon estimation involves
  extensive area and getting reliable ground
  information is critical. Forest managements rely
  on ground staffs to report on field conditions.
  Most of the times, you need a holistic view to
  see what's out there.
• Powered, aerial vehicles
• No human operator on board
• Can fly autonomously or be piloted remotely
• Can be expendable or recoverable
• Can carry weapons or surveillance equipment)

19
UAV Forest Carbon Measurement Method

• UAV Types
• Fixed wing
• Rotary Wing

20
UAV Components
1) Vehicle or platform itself
Predator
Nano Hummingbird
Puma AE
Solar Eagle
Honeywell T-Hawk
21
2) UAV Payload
22
3) UAV Support Equipment

• Such as control station, data links, telemetry,
  communications and navigation and related
  equipment necessary to operate the UAV.

23
UAS Works

• Collects Data 
• Processes it into images
• Sends images to centers for furfure analysis

24
UAV Advantages

• Safety, No pilot to be shot down, Can fly into
  hurricanes or at low altitudes over the ocean
• Little damage when they crash due to their light
  weight
• It can be made and built in a time of 3-4 days.
• All components are locally available.
• Flight need not be scheduled. It can be based on
  the weather conditions and preferences of the
  farmer.
• Availability of data and imagery immediately
  after the flight.

Disadvantages

• Significant experience required to fly the UAV.
• Easily destructible.

25
Study Objectives

• Main objective of this is to develop TFCS model
  based on 3D data generated from high resolution
  UAV imagery.
• The specific objectives are
• To develop TFCS forest stand from UAV data
• To develop the tree height models
• To develop TFCS and TFB models based on the tree
  heights and UAV data
• To compare the accuracy of the UAV data and
  satellite data (LANDSAT-8 OLI Sensor and Sentinel
  -2 Data) for biomass estimates.
• To evaluate the impacts of produced DTMs based on
  different approaches and parameters on TFCS and
  TFB models

26
Study Area Kelantan sate, Malaysia
27
Kelantan Forest Type
Forest Type in Kelantan code
Virgin Inland Forest lowland Hill Forest 1
Virgin Inland Forest High Hill Forest 2
Logging Forest lowland Hill Forest (1-10 years ) 3
Logging Forest High Hill Forest (1-10 years ) 4
Logging Forest lowland Hill Forest (11-20 years ) 5
Logging Forest High Hill Forest (11-20 years ) 6
Logging Forest lowland Hill forest (21-30 years ) 7
Logging Forest High Hill Forest (21-30 years ) 8
Logging Forest lowland Hill Forest (gt30 years ) 9
Logging Forest High Hill Forest (gt30 years ) 10
Non-Reserved Inland Forest Forest 14
Protection Forest lowland Hill forest 16
Protection Forest High Hill forest 17
Protection Forest Mountain Forest 18

28
Methodology
UAV and Remote sensing (RS) techniques
Field measuring techniques for estimating Total
Forest Carbon Stocks

Total Forest Biomass Or Carbon Stocks
29
Methodology
30
Research Data Used and Process

• Remote Sensing Data
• a. Landsat 8
• 1. OIL sensor L1T product
• 2. TIRS Sensor L1T product
• b. sentinel 2
• c. ASTER-GDEM
• 2. Meteorological Data3.TIRS Data
• 4. NFI Data
• 3. UAV Data

Sentinel 2 (Example 30/30/2016)
Landsat-8 (Example 22/4/2013)
31
Lansat8 pr-prosesing for OLI Sensor L1T product
Satellite Data Process
Radiometric correction

• 1. Converts to Spectral Radiance
• using the radiance scaling factors
• L?    MLQcal AL
• where
• L?  Spectral radiance (W/(m2 sr
  µm))ML  Radiance multiplicative scaling factor
  for the band.AL  Radiance additive scaling
  factor for the band.
• Qcal                L1 pixel value in DN
• 2. OLI Top of Atmosphere Reflectance
• Equation converts Level-1 DN values to TOA
  reflectance
• ??'    M?Qcal A?
• where ??'  TOA Planetary Spectral Reflectance,
  without correction for solar angle. 
  (Unitless)M?  Reflectance multiplicative
  scaling factor for the band.A?  Reflectance
  additive scaling factor for the band.Qcal  L1
  pixel value in DN
•  

32
Geometric correction
Satellite Data Process
Lansat8 pr-prosesing for OLI Sensor L1T product
Images were geometrically corrected using 23
ground control points (GCPs) of major features
(e.g. roads and buildings ) and digital elevation
models (DEM) to attain improved geodetic accuracy
and a geometrically rectified product free from
distortions (NASA, 2013), The first order
polynomial function was used and a
nearest-neighbour resampling protocol was applied
to correct for systematic shifts occurring in a
few cases between neighboring images. The total
transformation root mean square error (RMSE) of
less than a pixel was attained.
33
Satellite Data Process
Lansat8 pr-prosesing for OLI Sensor L1T product
Atmospheric correction and re-projected

• Lansat8 images were Atmospheric ally using the
  MODTRAN based on the Fast Line-of-sight
  Atmospheric Analysis of Spectral Hypercube
  (FLAASH) radiative transfer algorithm (Matthew et
  al., 2000 Perkins et al., 2005), topographic
  correction using ccorrection method, Then the
  Landsat images re-projected to the Universal
  Transverse Mercator (UTM) coordinate system with
  datum WGS 1984 and zone 47 north using the
  nearest neighbor resampling method. The final
  image mosaic of the Kelantan state

34
ASTER Global Digital Elevation Data

• ASTER Global Digital Elevation Data (ASTER-GDEM)
  with 30 m. a fill-sink process as pre-processing
  was applied ASTER-GDEM data in a GIS environment
  ,Then topographical variables in this study,
  altitude information (elevation) (m) ,aspect (in
  azimuth degrees) slope (in percentage), land
  curvature (concave, convex or ?at) ,a measure of
  potential relative radiation, and Insolation (W
  h/m2), were derived at 30 m spatial resolution
  using surface analysis tools in a GIS
  environment)

35
Meteorological Data

• Inverse Distance Weighting (IDW) and Kriging
  methods will use for mapping precipitation
  rainfall (mm datasets) was derived from
  Meteorological Data Malaysia acquired from
  Meteorology Department. There are total of 73
  weather stations that can provide annual
  meteorological records in Kelantan.

Precipitation Map
Weather stations, Kelantan state, Malaysia
Station No Station Code Station Name State District Latitude Longitude
1 4614001 Brook Kelantan Gua Musang 04 40 35 101 29 05
2 4614002 Lojing Kelantan Gua Musang 04 36 00 101 24 00
3 4717001 Blau Kelantan Gua Musang 04 46 00 101 45 25
4 4720026 Ldg. Mentara Kelantan Gua Musang 04 45 20 102 01 00
5 4721001 Upper Chiku Kelantan Gua Musang 04 45 55 102 10 25
6 4726001 Gunung Gagau Kelantan Gua Musang 04 45 25 102 39 20
....... ....... ....... ...... ..... ......
72 6121067 Stn. Keretapi Tumpat Kelantan Tumpat 06 11 55 102 10 10
73 6122064 Stor JPS Kota Bharu Kelantan Kota Bharu 06 06 30 102 15 25
36
TIRS Data

• 1.Converts to Spectral Radiance
• using the radiance scaling factors L?   
  MLQcal AL
• where L?  Spectral radiance (W/(m2 sr
  µm))ML  Radiance multiplicative scaling factor
  for the band.AL  Radiance additive scaling
  factor for the band.
• Qcal    L1 pixel value in DN
• 2. Atmosphere Brightness Temperature (LST) is
• The conversion formula is as follows
  TK2/In (K1/ L? 1)
• where T    TOA Brightness Temperature, in
  Kelvin.L?    Spectral radiance (Watts/(m2 sr
  µm))K1     Thermal conversion constant for the
  band
• K2    Thermal conversion constant for
  the band
• 3. Atmosphere Brightness Temperature , in
  Kelvin convert to Celsius LST T-273

37
NFI Data
1. Forest mapping/stratification 2. Number of
stems per ha (N) 3. Basal area per hectare
(m2) 4. Volume per ha (V) and 5. Dry biomass
(tones per ha) 6. carbon (tones per ha)
38
Plot types Temporary, Permanent data for Five
pools
NFI Data
Calculate number of plots needed
precision in inventory 5 of the mean at 95 CI
Locating of NFI Plots
39
UAV Data
40
UAV Pre- Processing Data
41
Forest Biomass and Carbon Techniques
42
Variables Measurement
Spectral and Vegetation Indices
code Abbrev. Name Formula
V1 CIgreen Chlorophyll Index Green NIR /Green-1
V2 NDVI Normalized Difference Vegetation Index NIR - Red /NIR Red
V3 GNDVI Green Normalized Difference Vegetation Index NIR -Green/ NIR Green
V4 GSAVI Green Soil Adjusted Vegetation Index NIR -Green/ NIR Green L(1L)
V5 (SAVI) Soil Adjusted Vegetation Index (Red -Green)(1L)/( Red Green L) L 0.5
V6 GRNDVI Green-Red NDVI NIR-(Green Red) / NIR (Green Red)
V7 (NDII) Normalized difference infrared index Red - NIR / Red NIR
V8 RI Normalized Difference Red Green Index Red-Green / Red Green
V9 NGRDI Normalized green red difference index Green-Red/Green Red
V10 EVI2 Enhanced Vegetation Index 2 2.5 NIR Red/ NIR 2.4Red1
V11 WDRVI Wide Dynamic Range Vegetation Index 0.1 NIR -Red/0.1 NIR Red
V12 Norm G Norm G Green/ NIR Red Green
V13 Norm NIR Norm NIR NIR / NIR Red Green
V14 Norm R Norm R Red/ NIR Red Green
V15 TNDVI Transformed NDVI v-(NDVI)0,5
V16 MSRNir/Red Modified Simple Ratio NIR/RED (NIRRED)1/ v (NIRRED)1
43
UAV- Variables Measurement
Spectral Band Ratios, Spectral Band Differencing
D1 Difference Green Red Index Green - Red
D2 Difference Green NIR Index Green NIR
GDVI Difference NIR/Green Index NIR -Green
RDVI Red Difference Vegetation Index NIR-Red
DVI Difference Vegetation Index Red -Green
D6 Difference Red NIR Index Red NIR
G/RSR Simple Ratio Green Red Green / Red
G/ NIR SR Simple Ratio Green Near- Infrared Green / NIR
GRVI Green Ratio Vegetation Index NIR / Green
RRVI RED Ratio Vegetation-Index NIR / Red
R/G SR Red/Green Ratio Vegetation-Index Red / Green
R/NIR SR Red/NIR Ratio Vegetation-Index Red / NIR
Tasseled Cap Indices Bright Index (BI) Green
Vegetation Index(GVI) Wetness Index(WI)
44
Image Texture variables
Topographical Variables
Variables Measurement
Slope
Aspect
Meteorological Variables
Atmosphere Brightness Temperature
Elevation
Precipitation Rainfall Variables
45
Data Analysis

• Statistical Analysis will do by
• ArcGIS , ENVI software's and SPSS statistic
  analysis
• Statistical Analysis Methods ( Simple and
  multilinear regression methods, Stochastic
  gradient boosting (SGB) algorithm, and Radom
  Forest algorithm

Experimental Phases
Experimental phases Different variables groupings
Exp.1 UAV- UAV Spectral imagery (Most important Variables selected)
Exp.2 UAV- Spectral Bands Env. variables
Exp.3 UAV- Spectral Bands Env. Variables Topography Variables
Exp.4 UAV- Vegetation Indices (VIs) (Most important Variables selected)
Exp.5 UAV- Vegetation Indices (VIs) Env. variables
Exp.6 UAV- Vegetation Indices (VIs) Env. variables Topography variables
46
Results and Expected Outcomes

• In novation in modelling off accurate estimation
  of Total Forest Biomass and Total Forest Carbon
  Stock which will have a significant impact on
  monitoring costs. The results of Total Forest
  Biomass and Total Forest Carbon Stock from
  deforestation and degradation associated with
  transition between land use and land cover types
  which is require by the relevant authorities for
  any planning and development. UAV data
  integration with remote sensing data enable large
  area mapping and monitoring of forest cover and
  change at regular intervals, providing
  information on where and how changes are taking
  place at bi-annual or even annual time scales.

47
Research Time Table
48
Research Budget
49
References

• Beuchle, R., et al. 2011. A Satellite Dataset
  for Tropical Forest Change Assessment.
  International Journal of Remote Sensing 32
  70097031. doi10.1080/01431161.2011.611186.
• Bodart, C., et al. 2011. Pre-processing of a
  Sample of Multi-scene and Multi-date Landsat
  Imagery used to Monitor Forest Cover Changes over
  the Tropics. ISPRS Journal of Photogrammetry and
  Remote Sensing 66 555563.
• Eva, H. D., et al. 2012. Forest Cover Changes in
  Tropical South and Central America from 1990 to
  2005 and Related Carbon Emissions and Removals.
  Remote Sensing 4 (5) 13691391.
  doi10.3390/rs4051369.
• FAO (Food and Agriculture Organization). 2010.
  FAO Forest Resources Assessment of 2010 (FRA).
  Rome FAO. http//www.fao.org/docrep/013/i1757e/i1
  757e.pdf.
• FAO. 2015. FAO Forest Resources Assessment of
  2015 (FRA). Rome FAO. http//www.fao.org/forestry
  /fra/fra2015/en/.
• GFOI (Global Forest Observations Initiative).
  2014. Integrating Remote-sensing and Ground-based
  Observations for Estimation of Emissions and
  Removals of Greenhouse Gases in Forests Methods
  and Guidance from the Global Forest Observations
  Initiative. (Often GFOI MGD.) Geneva,
  Switzerland Group on Earth Observations, version
  1.0. http//www.gfoi.org/methods-guidance/.
• GOFC-GOLD (Global Observation of Forest Cover and
  Land Dynamics). 2014. A Sourcebook of Methods and
  Procedures for Monitoring and Reporting
  Anthropogenic Greenhouse Gas Emissions and
  Removals Associated with Deforestation, Gains
  and Losses of Carbon Stocks in Forests
  Remaining Forests, and Forestation. (Often
  GOFC-GOLD Sourcebook.) Netherland GOFC-GOLD Land
  Cover Project Office, Wageningen University.
  http//www.gofcgold.wur.nl/redd/index.php.
• Hansen, M. C., et al. 2013. High-resolution
  Global Maps of 21st-century Forest Cover Change.
  Science 342 850853.
• Herold, M. 2009. An Assessment of National Forest
  Monitoring Capabilities in Tropical Non-annex I
  Countries Recommendations for Capacity Building.
  Report for The Princes Rainforests Project and
  The Government of Norway. Friedrich-Schiller-Unive
  rsität Jena and GOFC-GOLD. http//princes.3cdn.net
  /8453c17981d0ae3cc8_q0m6vsqxd.pdf
• IPCC, 2003. 2003 Good Practice Guidance for Land
  Use, Land-Use Change and Forestry, Prepared by
  the National Greenhouse Gas Inventories
  Programme, Penman, J., Gytarsky, M., Hiraishi,
  T., Krug, T., Kruger, D., Pipatti, R., Buendia,
  L., Miwa, K., Ngara, T., Tanabe, K., Wagner, F.
  (eds.). Published IGES, Japan.
  http//www.ipcc-nggip.iges.or.jp/public/gpglulucf/
  gpglulucf.html (Often referred to as IPCC GPG)
• Raši, R., et al. 2011. An Automated Approach for
  Segmenting and Classifying a Large Sample of
  Multi-date Landsat-type Imagery for pan-Tropical
  Forest Monitoring. Remote Sensing of Environment
  115 36593669.

50
Questions?

• Thank You