Page 1
> ISRSE 37 > Kersten Clauss > 2017-05-081
Mapping Paddy Rice in Asiaa multi-sensor, time-series approach
Kersten Clauss1, Marco Ottinger1, Wolfgang Wagner2, Claudia Kuenzer3
1 Department of Remote Sensing, Institute of Geography and Geology, University of Wuerzburg2 Department of Geodesy and Geoinformation, Vienna University of Technology3 German Remote Sensing Data Center (DFD), Earth Observation Center (EOC),
German Aerospace Center (DLR)
[email protected]
Page 2
Motivation
• Food Security
• >40% of calorie intake in South Asian
countries
• single most important food crop in
Asia
• stable demand as food crop even with
dietary change
• Trade
• production volume, market price, food
security interact
• rice is a globally traded commodity
• Livelihoods
• ~75% of the worlds farms are in Asia,
of which 80% are smaller than 2 ha
• high environmental risk (drought,
flood, salinization)
> ISRSE 37 > Kersten Clauss > 2017-05-082
Global Rice Science
Partnership 2013
Page 3
Rice Production 2014
> ISRSE 37 > Kersten Clauss > 2017-05-083
0 >208 mio. tonnes
rice productionFAOSTAT
Page 4
Motivation
• Food Security
• >40% of calorie intake in South Asian
countries
• single most important food crop in
Asia
• stable demand as food crop even with
dietary change
• Trade
• production volume, market price, food
security interact
• rice is a globally traded commodity
• Livelihoods
• ~75% of the worlds farms are in Asia,
of which 80% are smaller than 2 ha
• high environmental risk (drought,
flood, salinization)
> ISRSE 37 > Kersten Clauss > 2017-05-084
AMIS
FAOSTAT
Page 5
Motivation
• Food Security
• >40% of calorie intake in South Asian
countries
• single most important food crop in
Asia
• stable demand as food crop even with
dietary change
• Trade
• production volume, market price, food
security interact
• rice is a globally traded commodity
• Livelihoods
• ~75% of the worlds farms are in Asia,
of which 80% are smaller than 2 ha
• high environmental risk (drought,
flood, salinization)
> ISRSE 37 > Kersten Clauss > 2017-05-085
DeltAdapt project
Page 6
Methodology
• combine results from time-series
of different sensors to reduce
data size
• ability to transfer to different rice
regions
• MODIS:
• high temporal resolution
• global coverage
• moderate data size
• free, open access
• Sentinel-1:
• high spatial resolution
• unaffected by cloud cover
• sensitive to surface water
• free, open access
> ISRSE 37 > Kersten Clauss > 2017-05-086
Page 7
Rice Area Detection from Time-Series
• rice fields are commonly
flooded prior to
transplanting/seeding
• water level is maintained
throughout the growing cycle
• emergence, horizontal and
vertical growth influences
spectral and microwave
response
• flooding and phenological
growing stages of rice create
a distinct temporal footprint
time-series data
> ISRSE 37 > Kersten Clauss > 2017-05-087
Page 8
Rice Area Detection from Time-Series
• rice fields are commonly
flooded prior to
transplanting/seeding
• water level is maintained
throughout the growing cycle
• emergence, horizontal and
vertical growth influences
spectral and microwave
response
• flooding and phenological
growing stages of rice create
a distinct temporal footprint
time-series data
> ISRSE 37 > Kersten Clauss > 2017-05-088
Page 9
Rice Area Detection from Time-Series
• rice fields are commonly
flooded prior to
transplanting/seeding
• water level is maintained
throughout the growing cycle
• emergence, horizontal and
vertical growth influences
spectral and microwave
response
• flooding and phenological
growing stages of rice create
a distinct temporal footprint
time-series data
> ISRSE 37 > Kersten Clauss > 2017-05-089
Page 10
Rice Area Detection from Time-Series
• rice fields are commonly
flooded prior to
transplanting/seeding
• water level is maintained
throughout the growing cycle
• emergence, horizontal and
vertical growth influences
spectral and microwave
response
• flooding and phenological
growing stages of rice create
a distinct temporal footprint
time-series data
> ISRSE 37 > Kersten Clauss > 2017-05-0810
Page 11
Clauss, Yan, Kuenzer (2016)
doi:10.3390/rs8050434
Rice Production in China
> ISRSE 37 > Kersten Clauss > 2017-05-0811
Page 12
Feature Calculation
> ISRSE 37 > Kersten Clauss > 2017-05-0812
LSWI 75th percentile
EVI - LSWI inversions
EVI 90th percentile
Page 13
Feature Calculation
> ISRSE 37 > Kersten Clauss > 2017-05-0813
LSWI 75th percentile
EVI - LSWI inversions
EVI 90th percentile
Feature Bands
10th percentile EVI, LSWI, Blue, Red, NIR, MIR
25th percentile EVI, LSWI, Blue, Red, NIR, MIR
50th percentile EVI, LSWI, Blue, Red, NIR, MIR
75th percentile EVI, LSWI, Blue, Red, NIR, MIR
90th percentile EVI, LSWI, Blue, Red, NIR, MIR
amplitude NDVI, EVI, LSWI, Blue, Red, NIR, MIR
75th - 25th percentile EVI, LSWI, Blue, Red, NIR, MIR
90th - 10th percentile EVI, LSWI, Blue, Red, NIR, MIR
local maxima > 0.8 EVI, LSWI
local maxima > 0.7 EVI, LSWI
local maxima > 0.6 EVI, LSWI
local minima < 0.3 EVI, LSWI
local minima < 0.2 EVI, LSWI
local minima < 0.1 EVI, LSWI
EVI LSWI inversions EVI, LSWI
Page 14
Rice Area in China derived from MODIS with OCSVM
> ISRSE 37 > Kersten Clauss > 2017-05-0814
2002
2005
2010
2014
Clauss, Yan, Kuenzer (2016)
doi:10.3390/rs8050434
Page 15
2014
2010
2005
Rice Area Change - Beimin Lake
Clauss, Yan, Kuenzer (2016)
doi:10.3390/rs8050434
2002
> ISRSE 37 > Kersten Clauss > 2017-05-0815
Page 16
Rice Area Change - Heilongjiang
2014
2010
2005
2002
> ISRSE 37 > Kersten Clauss > 2017-05-0816
Clauss, Yan, Kuenzer (2016)
doi:10.3390/rs8050434
Page 17
Accuracy Assessment
> ISRSE 37 > Kersten Clauss > 2017-05-0817
classified area compared to statistical yearbook data
Overall
Accuracy
User‘s
Accuracy
Producer‘s
Accuracy
rice no rice rice no rice
0.90 0.90 0.89 0.89 0.79
Clauss, Yan, Kuenzer (2016)
doi:10.3390/rs8050434
Page 18
Sentinel-1 Coverage 2015
Clauss, Ottinger, Kuenzer (2017)
in review
Sentinel-1A IW DV coverage 2015
> ISRSE 37 > Kersten Clauss > 2017-05-0818
Page 19
Sentinel-1 Study Sites
Clauss, Ottinger, Kuenzer (2017)
in review
Sentinel-1A IW DV coverage 2015
A
B
CFED
> ISRSE 37 > Kersten Clauss > 2017-05-0819
Page 20
Sentinel-1 Study Sites
Clauss, Ottinger, Kuenzer (2017)
accepted with revisions
Sentinel-1A IW DV coverage 2015
A
B
CFED
Giao Thuy, Vietnam Soc Trang, Vietnam Poyang Lake, China
Sacramento, USA Ebro Delta, Spain Isla Mayor, Spain
> ISRSE 37 > Kersten Clauss > 2017-05-0820
Page 21
Clauss, Ottinger, Kuenzer (2017)
in review
Methodology
> ISRSE 37 > Kersten Clauss > 2017-05-0821
Page 22
Clauss, Ottinger, Kuenzer (2017)
in review
Pre-Processing
> ISRSE 37 > Kersten Clauss > 2017-05-0822
Page 23
Clauss, Ottinger, Kuenzer (2017)
in review
Segmentation
> ISRSE 37 > Kersten Clauss > 2017-05-0823
Page 26
Clauss, Ottinger, Kuenzer (2017)
accepted with revisions
Time Series per Object
> ISRSE 37 > Kersten Clauss > 2017-05-0826
Page 28
from time series per pixel to
time series per object
Page 29
Classification
> ISRSE 37 > Kersten Clauss > 2017-05-0829
Page 30
Classification
> ISRSE 37 > Kersten Clauss > 2017-05-0830
Page 31
Giao Thuy, Vietnam Soc Trang, Vietnam
Classified Rice Areas
Clauss, Ottinger, Kuenzer (2017)
in review
> ISRSE 37 > Kersten Clauss > 2017-05-0831
Page 32
Poyang Lake, China Sacramento, USA
Classified Rice Areas
Clauss, Ottinger, Kuenzer (2017)
in review
> ISRSE 37 > Kersten Clauss > 2017-05-0832
Page 33
Ebro Delta, Spain Isla Mayor, Spain
Classified Rice Areas
Clauss, Ottinger, Kuenzer (2017)
in review
> ISRSE 37 > Kersten Clauss > 2017-05-0833
Page 34
validation points, Soc Trang study site
Study Site Overall
Accuracy
Producer’s
Accuracy
User’s
Accuracy
rice no rice rice no rice
Giao Thuy 0.87 0.88 0.85 0.85 0.88
Soc Trang 0.85 0.82 0.88 0.89 0.81
Poyang Lake 0.81 0.78 0.85 0.87 0.75
California 0.78 0.73 0.86 0.89 0.67
Ebro Delta 0.87 0.90 0.83 0.82 0.91
Isla Mayor 0.82 0.78 0.88 0.90 0.74
Clauss, Ottinger, Kuenzer (2017)
in review
Accuracy Assessment
> ISRSE 37 > Kersten Clauss > 2017-05-0834
Page 35
Multi-Sensor Methodology
> ISRSE 37 > Kersten Clauss > 2017-05-0835
Page 36
Moderate Resolution Rice Area (MODIS)
Mekong Delta,
Vietnam
rice 2015
> ISRSE 37 > Kersten Clauss > 2017-05-0836
Page 37
High Resolution Rice Area and Seasonality (Sentinel-1)
Mekong Delta,
Vietnam
> ISRSE 37 > Kersten Clauss > 2017-05-0837
Page 38
• large area rice mapping is possible with remote sensing time-series
• frequent cloud cover in rice growing regions requires high revisit time of multi-
spectral sensors
• SAR time-series enable high resolution rice mapping in cloud prone regions
• require frequent coverage over large areas
• seasonality extraction depends on temporal density of time-series
• combined approach reduces data and can aid towards current, high resolution
rice area mapping at large scale
• large scale mapping is limited by calibration/validation, not EO data
Conclusions and Outlook
> ISRSE 37 > Kersten Clauss > 2017-05-0838