This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
CzechAgri, Workshop on Control and Management of Agricultural Land in IACS, Baveno, 23.5.2016
Optical and SAR comparison
� Optical based classification – internal validation sample
� SAR based classification – internal validation sample
CzechAgri, Workshop on Control and Management of Agricultural Land in IACS, Baveno, 23.5.2016
11
No. crops vs. single LPIS polygon
� Single crop declared on single LPIS polygon for more than 80% of all polygons
� LPIS polygons with multiple crops need to be detected automatically (and need to be differentiated from LPIS polygons where more crops are incorrectly classified)
� Two step approach� Statistical analysis� Segmentation within LPIS blocks
� Initial test� Statistical analysis based on application of minimum parcel size and
individual crop ratios calculation� Applied for validation sample: 1481 LPIS polygons declared with multiple
crops, more than 90% identified using statistical approach� But still number of misidentified polygons to be corrected� Object based approach to be developed and tested
CzechAgri, Workshop on Control and Management of Agricultural Land in IACS, Baveno, 23.5.2016
Multiple crops on single LPIS polygon I
Mar 2015 Apr – Jun 2015 Aug – Sep 2015
� Multiple crops grown on single LPIS polygon
confirmed
� „Easy“ to detect examplewinter rapeseed
winter cereals
spring cereals
sugar beet
potatoes
maize
fodder crops
other annual crops
CzechAgri, Workshop on Control and Management of Agricultural Land in IACS, Baveno, 23.5.2016
12
Multiple crops on single LPIS polygon II
Mar 2015 Apr – Jun 2015 Aug – Sep 2015
� Misclassification - multiple crops grown on
single LPIS polygon not confirmed
� „Difficult“ to detect examplewinter rapeseed
winter cereals
spring cereals
sugar beet
potatoes
maize
fodder crops
other annual crops
CzechAgri, Workshop on Control and Management of Agricultural Land in IACS, Baveno, 23.5.2016
Multiple crops on single LPIS polygon III
Mar 2015 Apr – Jun 2015 Aug – Sep 2015
� Multiple crops vs. crop anomaly
� „Difficult“ to detect example
winter rapeseed
winter cereals
spring cereals
sugar beet
potatoes
maize
fodder crops
other annual crops
CzechAgri, Workshop on Control and Management of Agricultural Land in IACS, Baveno, 23.5.2016
13
Classification & Validation 2015/2016
CzechAgri, Workshop on Control and Management of Agricultural Land in IACS, Baveno, 23.5.2016
� Sentinel-2 only - composite based on two acquisitions (March 17/27)
� Regional sample product - eastern part of Central Bohemia
Winter crop classification 2015/2016
� Winter rapeseed, winter cereals, fodder
crops, no vegetation
� Mono-temporal classification, same
approach as for 2015 crop type map
� Training dataset based on visual
interpretation
� Validation dataset collected during field
campaign
CzechAgri, Workshop on Control and Management of Agricultural Land in IACS, Baveno, 23.5.2016
14
� Sentinel-2 composite, SVM classification, aggregation into LPIS database
Classification
winter rapeseed
winter cereals
fodder crops
no vegetation
CzechAgri, Workshop on Control and Management of Agricultural Land in IACS, Baveno, 23.5.2016
Validation
� Internal validation - done by Gisat, sample size: 137 LPIS polygons� Validation data collected during dedicated filed campaign
CzechAgri, Workshop on Control and Management of Agricultural Land in IACS, Baveno, 23.5.2016
15
� Winter rapeseed misdetection - different pheno-phase, will be removed
using subsequent image acquisition
Early classification
CzechAgri, Workshop on Control and Management of Agricultural Land in IACS, Baveno, 23.5.2016
Conclusions & Next steps
CzechAgri, Workshop on Control and Management of Agricultural Land in IACS, Baveno, 23.5.2016
16
� Accuracy of Sentinel-1 based classification
� LPIS needs to be available
� Mostly automated processing (no manual post classification improvement applied)
� Huge amount of data to be processed (1.3 TB for single crop growing season – will increase for 2016)
� Early winter crop detection possible already in March with high accuracy (using optical imagery)
Main findings
CzechAgri, Workshop on Control and Management of Agricultural Land in IACS, Baveno, 23.5.2016
� Crop type classification 2016� Full Sentinel-2 time series� Individual crop accuracy vs. month of production � Simulation of iterative delivery during crop growing season
� Further analysis of 2015 and 2016 results (inter-annual)� Crop rotation� Crop area statistics� Crop diversification� …
� Automation� Detection of multiple crops on single LPIS polygon� Integration of optical and SAR based classification
Next steps
CzechAgri, Workshop on Control and Management of Agricultural Land in IACS, Baveno, 23.5.2016
17
Acknowledgment
� ESA: CzechAgri initiative, funding through Sentinel-2 for Agriculture
project
� UCL: Project management, support to 2015/2016 crop classification
� DG-JRC: “Towards Future Copernicus Services Components for
Agriculture”
� SZIF: IACS data provision, external validation, consultations
� MoA: LPIS data provision
CzechAgri, Workshop on Control and Management of Agricultural Land in IACS, Baveno, 23.5.2016
Sentinel-1: Backscatter analysis
� Crop-backscatter signatures during crop growing season
18
Sentinel-1: Separability analysis
� Crop-backscatter signatures
� Statistical Tests: Generalized linear models (GLMs), Analysis of Variance (ANOVA) and Tukey’s HSD to
identify which crops or crop groups have significantly different range of backscatter each month.
� Crop-backscatter signatures
� Statistical Tests: Generalized linear models (GLMs), Analysis of Variance (ANOVA) and Tukey’s HSD to
identify which crops or crop groups have significantly different range of backscatter each month