Presented by Solichin Manuri, Senior Advisor at Diameter Consulting, Bogor, Indonesia, at "Online Webinar 2: Biophysical Attributes and Peatland Fires", on 14 October 2020
In this session the speaker shared information on mapping fire (extent and occurrences) in tropical peatlands including in Indonesia. This session also shared insights on the existing methods that can be used for fire mapping and comparisons. This session also emphasized that spatial explicit criteria for fire should be developed depending on the method and data used.
2. Background
• In the past four decades, severe wildfires have become common
and repetitive events in the tropical ecosystem of Indonesia
(Goldammer and Seibert, 1990).
• These fires were the result of unsustainable practices of forest and
land management, which coincide with extreme drought attributed
to the El-Niño Southern Oscillation and the Indian Ocean Dipole
Mode (Goldammer etal, 1990; Cochrane etal, 2003).
• Drained peatland during dry season become susceptible to fires.
During prolonged dry season, peat fires contribute huge amount of
GHG emissions (Siegert etal, 2001; Lohbeger etal, 2017).
• The impact of peat fires in Indonesia is limitedly explored, or with
relatively high uncertainty, in particular regarding the size of the
burn areas.
3. Challenges in fire mapping in the tropics
Mapping of burn areas in tropical region is challenging, not
only because of cloud persistence but also due to smoke or
haze occurrence during the fires and short windows of
opportunity. NASA, MODIS
4. 1982/83 1997/98 2015 20192000
Lennertz and
Panzer, 1984
Siegert and
Hoffm
an, 2000
MODIS Burn Area MCD 45
MODIS Burn Area MCD 64
2006
MoEF
GFED4.0
Lohberger,
etal, 2017 LAPAN
2016
Copernicus
Sub NationalGlobal National
Fire mapping in the tropics and Indonesia
5. Fire mapping methods in tropics and Indonesia
Burned Year
and location
Scope, resolution,
continuity
Method
Lennertz and
Panzer, 1984
1982/83; East Kalimantan,
discontinued
Visual classification of Landsat MSS images
Siegert and
Hoffman, 2000
1997/98; East Kalimantan; 25
m; discontinued
Object based image analysis of ERS2 SAR data; PCA
method
GFED4.0 1995 –
present;
Global; 27.8 m Digital classification using MODIS MCD64A1, VIIRS, ATRS,
TRMM
MODIS Burn Area
MCD 45
MODIS Burn Area
MCD 64
2000 –
present;
Global; 500 m;
continued
Thresholds of burn index was applied to define burn
areas and filtered with active fire data
Copernicus Burn
Area
2014-present Global; 500 m The annual BA derived from SPOT Vegetation using
supervised classification
Lohberger, etal,
2017
2015; Indonesia; 10 m Object based image analysis of Sentinel 1 backscatters
layers and temporal change metrics, training data from
ground and drone survey
MoEF, 2016 2006 - present Indonesia; ?;
continued
Visual classification of Landsat ETM, OLI images, with
aid of hotspot clusters and verified with ground data or
high resolution images
LAPAN 2015, 2016,
2019,
Indonesia;30-m; most
likely continue during
extreme fire years;
Change detection using dNBR of Landsat ETM/OLI
images and VH polarization of Sentinel 1, hotspot>30%
confidence level, filtering water body
6. Comparison of 2015 Fires Size
1,8
2,5
4,6
2,6
2,385
0
0,5
1
1,5
2
2,5
3
3,5
4
4,5
5
GFED4.0 MODIS MCD 64 Lohberger etal, 2017 MoEF, 2016 LAPAN, 2016
2015 Burn Area Comparison (in mill ha)
Peat Mineral soil All
7. Summary of Used Methods
Component
Sensor types Active (optical) and passive (radar) sensors; or hybrid
Spatial resolution 10 m (Sentinel), 30 m (Landsat), 500m (MODIS), 1km (SPOT
Vegetation
Temporal repetition Monthly to annually
Number of image data
or sensors
Single to multiple data sources, including additional GIS
layers, such as water body, active fires/hotspots.
Classification Visual classification: Visually inspect and delineate burn
areas; require prior knowledge or ground thruthing data on
burn area
Pixel based classification: apply threshold for change index,
which defined by analysing training samples
Object based image analysis: based on probabilities of
objects fall into burn category, derived from fuzzy logic
threshold value and training samples
8. Sensors + -
Passive • Easy to process
• Long history of data
acquisition
• Hindered by cloud, smoke
and haze
Active • Ability to penetrate cloud • Relief displacement effects
Resolution + -
Medium • Better accuracy
• Sufficient for national wall to
wall mapping
• Lower temporal resolution
Low • High temporal resolution, thus
able to get cloud and haze free
image in shorter period
• Lower spatial resolution and
accuracy
Classification + -
Manual • Full control of operators • Time consuming
• Required skilled operators
Automatic • Fast and consistent results • Rely on good training
samples
Pros and Cons
9. Satellite sensors used for burned area mapping
Satellite (Sensor) Operator
Operational Dates Temporal
Resolution
Spatial Resolution
Launch Date End Operation
Landsat 1-3 (MSS) NASA/USGS July 23, 1972 September 7, 1983 18 days 375-750 m
Landsat 4-5 (TM) NASA/USGS July 16, 1982 June 5, 2013 16 days 30-120 m
Landsat 7 (ETM+) NASA/USGS October 5, 1993 Planned lifespan for 5-21 years or 2022 16 days 15/30-60 m
Landsat 8 (OLI/TIRS)NASA/USGS
February 11,
2013
Planned lifespan for 5 years, but still
operating
16 days OLI: 15/30 m TIRS: 100 m
SPOT 1–7 (HRV) CNES
February 22,
1986
2024* 26 days 2.5 to 20 m
SPOT 4-5 (VGT) CNES March 24, 1998 July 1, 2013 1-2 days 1000 m
NOAA-7-20
(AVHRR)
NOAA
October 19,
1978
Still operating 1-2 days 1100 m
JPSS (VIIRS) NOAA
October 28,
2011
2031 1-2 days 375-750 m
Aqua (MODIS) NASA May 4, 2002 Still operating 1-2 days 250-1000 m
Terra (MODIS) NASA
December 18,
1999
Still operating 1-2 days 250-1000 m
ENVISAT (MERIS) ESA March 1, 2002 May 9, 2012 2-3 days 300-1200 m
ERS-2 ESA 1995 2011 35 days 25 m
PROBA V ESA May 7, 2013
Missions duration 2-5 years, but will be end
in 2020
1-2 days 300 m
Sentinel 1A ESA April 3, 2014 7-12 years or until 2021-2026 6 days 5-20 m
Sentinel 2A ESA June 23, 2015 7.25 years or until 2022 5 days 10-20-60 m
Sentinel 1B ESA April 25, 2016 7-12 years or until 2023-2028 6 days 5-20 m
Sentinel 3A ESA
February 16,
2016
7.5 years or until 2023 1-2 days 500 SLSTR m
Sentinel 2B ESA March 7, 2017 7.25 years or until 2024 5 days 10-20-60 m
Sentinel 3B ESA April 25, 2018 7 years or until 2025 1-2 days 300 OLCI mChuevieco, et al (2019)
10. 2018 7 years or until 2025
2024*
Satellite Lifespan
Sentinel 1B
2016
Landsat 1-3
(MSS)
1972 1983
Landsat 4-5 (TM)
1982 2013
SPOT 4-5 (VGT)
1998 2013
EVISAT (MERIS)
2002 2012
ERS-2
1995 2011
1978
NOAA-7-19 (AVHRR)
Still Operating
SPOT 1-7 (HRV)
1986
Landsat 7 (ETM+)
1993 Planned lifespan for 5-21 years or 2022
Terra (MODIS)
1999 Still Operating
JPPSS (VIIRS)
2011 2031
Landsat 8 (OLI/TIRS)
2013 Planned lifespan for 5 year, but still operating
Aqua (MODIS)
2002 Still Operating
PROBA V
2013 Missions duration 2-5 years, but will be end in 2020
Sentinel 1A
2014 7-12 years or until 2021-2026
Sentinel 2A
2015 7.25 years or until 2022
7-12 years or until 2023-2028
Sentinel 3A
2016 7.5 years or until 2023
Sentinel 2B
2017 7.25 years or until 2024
Sentinel 3A
11. Giglio etal, 2018
Siegert and Hoffman (2000) conducted an
accuracy assessment using aerial survey for the
25-m resolution ERS SAR BA. While Giglio et al
(2018) carried out uncertainty analysis for the
500 m MODIS BA using higher resolution
images, i.e. 30-m Landsat
Accuracy Assessment
Siegert and Hoffman, 2000
12. Burn Peat Depth Mapping
• Accurate information on burn peat depth is crucial for estimating emissions from
peat fires.
• Yet limited studies related to burn peat depth mapping in Indonesia, involving
manual measurement and lidar application
• Manual measurement poses potential biases due to the difficult accessibility of
remote areas and time consuming
• On the other hand, airborne lidar has been used for various height-related studies
with high precision (sub meter).
Balhorn etal, 2009
Study using Lidar
Data
Burnt Peat
Depth (m)
SE (m)
Konecny et al, 2016 0.13 0.16
Simpson et al, 2016 0.23 0.19
Ballhorn et al, 2009 0.33 0.18
13. • Several initiatives on fire mapping are available globally and
nationally using different satellite images and different
classification method.
• Spatially explicit criteria of fire mapping should be developed
specifically depending on the method and data used, combined
with knowledge from ground or aerial surveys.
• Validation of the burn area using ground thruthing data or high-
resolution images is required to understand the uncertainty of
the data
• Selection of method will rely on the need of fire mapping, i.e.
mapping coverage, reporting interval, technical capacity and
funding availability.
• It is important to select the remote sensing data based on the
spatial resolution, technical and funding capacity as well as the
lifespan and the continuity of the satellite program.
• Apart from mapping the size of fires, it is also important to
estimate the depth of peat fires
Main messages