2 field visits were made over the growing season to ensure data
quality and to note variations in growth stage, harvest time, and
changes in crop type (due to recording errors and
under-seeding)
50% of the fields selected randomly for training, and remaining
50% left for validation
A decision-tree classification followed by a post
classification object-based filter
6. X- & C-Band Classification Results: Casselman 2008
Comparison of single-date RADARSAT-2 and TerraSAR-X crop
classification accuracies over Casselman, Ontario
Later season multi-frequency SAR offers good accuracy
Users Accuracies ALOS RS2 Tsx Canola Flaxseed Beans (Soy and Dry)
Corn Cereal Overall All (10) 93.4 87.4 88.4 78.3 95.1 89.4 11 Aug 9
Aug 15 Aug 94.1 73.6 88.0 83.5 93.1 89.4 All All All 95.3 91.6 92.2
83.7 92.7 91.4
11. Crop Map Generated Using All Multi-Frequency SAR Images
Carman 2009 (91.4% overall accuracy) Water Urban Shrub land Wetland
Hay-Pasture Soybean Corn Cereal Coniferous Roads Barren Legend
Broadleaf Canola Soybean Sunflower Flaxseed Fallow Potato Field
peas Beans N 0 3 6 km
12. Polarimetric SAR for Crop Classification RADARSAT-2 (June
15, July 6, August 9, September 2): Casselman 2008 RADARSAT-2 (May
29, June 22, July 16, August 9, September 2): Carman 2009 Pasture
Soy Corn Wheat Overall Cloude-Pottier 77.5 85.2 86.4 97.6 85.8
Freeman-Durden 85.1 93.7 89.9 93.3 90.9 Linear Polarizations (HH,
HV, VV, VH) 60.4 81.8 73.6 83.8 75.4 Canola Flax Beans Corn Cereals
Overall Cloude-Pottier 91.9 57.4 76.5 68.6 82.8 79.8 Freeman-Durden
94.1 81.0 80.1 69.2 88.8 83.6 Linear Polarizations (HH, HV, VV, VH)
83.1 51.3 73.7 53.5 89.4 75.5
13.
The penetration depth of the radar signal is related to the
radar frequency
Higher frequency SAR primarily captures volume scattering near
the top of the crop canopy. X-Band provides the best overall
classification accuracy.
C-band and L-Band penetrate deeper and interact with other
canopy components, and help to improve accuracies of higher biomass
crops. Too great of penetration, especially with low biomass crops,
results in interactions with the soil.
Multi-frequency SAR provides the best crop specific
accuracies
Polarimetric scattering parameters appear to outperform
multi-polarization data
Operational crop mapping and in-season crop acreage estimates
could be supported by multi-temporal, multi-frequency and
polarimetric SAR
Conclusions
14. Acknowledgements
Financial support for this project is provided by AAFCs A-base
research funding and Canadian Space Agencys (CSA) GRIP funding
The TerraSAR-X data were provided by DLR through research
project LAN0337
RADARSAT-2 data were provided by CSA
ALOS PALSAR data were provided by JAXA
Thank goes to John Fitzmaurice, Maciej Jamrozik, Colin Schut,
Jiangui Liu, Carl Puddy and Patrick Rollings for field data
collection and data processing