we4_t05_1_lopezsanchez_rice_igarss2011.ppt
TRANSCRIPT
IEEE IGARSSVancouver, July 27, 2011
Monitoring and Retrieving Rice Phenology by means of Satellite SAR Polarimetry at X-band
Juan M. Lopez-SanchezJ. David Ballester-Berman
Signals, Systems & Telecommunications GroupUniversity of Alicante
Shane R. Cloude
AEL Consultants
IEEE IGARSSVancouver, July 27, 2011
Motivation• Remote sensing for agriculture: a tool for management and
optimization of resources
End users Demand Objective
Authorities or agencies at
national-regional-
local level
Crop-type mapping and
classification
Justification of subsidies, fraud detection,
acreages, insurance claims, etc.
Water resources
consumption
Control in regions suffering droughts or with
scarce water resources
Yield prediction Economic and market predictions, price
regulations, etc.
Farmers with extensive
fields
Timely information about
crop condition
Planning and triggering of farming practices
according to specific phenological stages
Water requirements Irrigation only when and where necessary
Final crop productivity Benefits
IEEE IGARSSVancouver, July 27, 2011
Motivation
• Motivation: examples of known demands from rice farmers in Spain– Timely information for:
• Effective germination measurements• When all plants have emerged they count their number. If low, more seeds are added
• Nitrogen fertillization stop• Once all panicles in a field have appeared, fertilization is not longer needed• Excessive fertilization may cause an increase in pests
– Detection of cultivation problems due to water salinity: areas with delayed development
• Objective: Is it possible to retrieve the current phenological stage from a single acquisition?
• Approach:– Analysis and interpretation of the polarimetric behavior of rice at different
phenological stages– If possible, proposal of a retrieval approach based on scattering properties
IEEE IGARSSVancouver, July 27, 2011
Site
• Mouth of the Guadalquivir river, Sevilla (SW Spain)
30km x 30km
IEEE IGARSSVancouver, July 27, 2011
Ground campaign
• Campaigns: 2008 and 2009
• Ground measurements over 5-8 parcels provided by the local association of rice farmers (Federación de Arroceros de Sevilla)
– Weekly (defined at field level):• Phenology: BBCH stage (0-99)
• Vegetation height
– Additional information:• Sowing and harvest dates
• Plantation density: plants/m2, panicles/m2
• Yield (kg/ha)
• Important: – A water layer is always present at ground during the campaign – Sowing is carried out by spreading seeds (from a plane) randomly over flooded
fields
IEEE IGARSSVancouver, July 27, 2011
Satellite data
2008
2009
TerraSAR-X images provided by DLR in the framework of projects LAN0021 and LAN0234
Failed orders
Available images
IEEE IGARSSVancouver, July 27, 2011
Analysis of observations• TerraSAR-X, 30 deg, 2009: Temporal evolution
HH VV HH-VV
IEEE IGARSSVancouver, July 27, 2011
Coherent acquisition of co-pol channels
Analysis of observations• TerraSAR-X HHVV dual-pol images: List of observables
– Backscattering coefficients and HH/VV ratio– Backscattering coefficients at the Pauli basis (HH+VV, HH-VV)– Correlation between HH and VV: magnitude and phase (PPD)– Correlation between 1st and 2nd Pauli channels: mag. and phase– Eigenvector decomposition (H2α): Entropy and alpha– Model-based decomposition: Random volume + polarized term (rank1)
IEEE IGARSSVancouver, July 27, 2011
Coherent acquisition of co-pol channels
Analysis of observations• TerraSAR-X HHVV dual-pol images: List of observables
– Backscattering coefficients and HH/VV ratio– Backscattering coefficients at the Pauli basis (HH+VV, HH-VV)– Correlation between HH and VV: magnitude and phase (PPD)– Correlation between 1st and 2nd Pauli channels: mag. and phase– Eigenvector decomposition (H2α): Entropy and alpha– Model-based decomposition: Random volume + polarized term (rank1)
2
2*
*2
200
0
0
HV
VVVVHH
VVHHHH
S
SSS
SSS
Single-pol (ERS, Radarsat1)
Quad-pol (ALOS-PALSAR, Radarsat-2)
Incoherent dual-pol (Envisat)
Coherent dual-pol (TerraSAR-X)
IEEE IGARSSVancouver, July 27, 2011
• Power
Analysis of observations vs phenologyHH and VV power
Wind induced roughness
Double-bounce
Vertical orientation:differential extinction
Development
Increasing randomness
Nearly random volume
Vegetative phase
Reproductive phase
Maturation
IEEE IGARSSVancouver, July 27, 2011
• Power
Analysis of observations vs phenology
HH and VV power HH / VV
Vegetative phase
Reproductive phase
Maturation Vegetative phase
Reproductive phase
Maturation
IEEE IGARSSVancouver, July 27, 2011
• Correlation between HH and VV
Analysis of observations vs phenology
Magnitude Phase (PPD)
Vegetative phase
Reproductive phase
Maturation Vegetative phase
Reproductive phase
Maturation
IEEE IGARSSVancouver, July 27, 2011
• Eigenvalue decomposition
Analysis of observations vs phenology
Entropy Alpha (dominant)
Vegetative phase
Reproductive phase
Maturation
Wind induced roughness
Double-bounce+
IEEE IGARSSVancouver, July 27, 2011
• Decomposition: Random volume + rank-1 mechanism
Analysis of observations vs phenology
Volume component Polarized component
Vegetative phase
Reproductive phase
Maturation Vegetative phase
Reproductive phase
Maturation
IEEE IGARSSVancouver, July 27, 2011
Retrieval of phenology from TSX data• Basic retrieval approach with a single acquisition (TSX)
– Four parameters• HHVV coherence and phase difference
• Entropy and alpha1
IEEE IGARSSVancouver, July 27, 2011
Retrieval of phenology from TSX data• Basic retrieval approach with a single acquisition (TSX)
– Five phenological intervals
– Decision plane
IEEE IGARSSVancouver, July 27, 2011
Retrieval of phenology from TSX data• Retrieval results (parcel F)
IEEE IGARSSVancouver, July 27, 2011
Retrieval of phenology from TSX data• Retrieval results: Comparison against ground data
– Percentage of pixels assigned to each stage within a parcel
Parcel B
Parcel C
IEEE IGARSSVancouver, July 27, 2011
Retrieval of phenology from TSX data• Retrieval results in an area without external information
IEEE IGARSSVancouver, July 27, 2011
Retrieval of phenology from TSX data• Comments on the approach
– Useful tracking of phenology:• At parcel level: BBCH agrees with the stage assigned to the
majority of pixels inside the parcels (with some exceptions)
• At (multi-looked) pixel level: parts with different development within a parcel are well identified
– But not perfect..• The algorithm is very ‘simple’: parameters and thresholds have
been selected manually (it could be optimized)
• An ambiguity between plant emergence (BBCH 18-21) and last stages (BBCH +50) is still present at some areas. Both are characterized by high entropies
IEEE IGARSSVancouver, July 27, 2011
Conclusions• Coherent dual-pol data provided by TerraSAR-X have
been useful for retrieving phenology of rice fields with a single acquisition
– Advantages when compared to other possible approaches:• 11-days revisit rate with the same sensor & mode• High spatial resolution • Retrieval with a single pass is possible (single-pol and incoherent dual-pol
are not enough)– Limitations:
• There remain some ambiguities that might be solved with full-pol data (e.g. using anisotropy), but not in operational mode with TSX
• Low coverage: TSX dual-pol swath is 15 km on ground• Some measurements are below or close to the noise level of TSX (-19 dB)
IEEE IGARSSVancouver, July 27, 2011
Future lines of research• Multi-temporal approaches (time series)
– Time coordinate provides extra information
• Multi-angular (and multi-temporal) integration– Ideal to reduce refresh time or increase spatial coverage
• Development of an operational scheme with farmers
• Pending issues:– Presence of rain– Other species within the rice fields (mixture)
• Application to rice under different farming practices:– Plantation procedures and arrangements– Dry ground at some moments