GLC mapping in semi-arid regions: a case study in West Africa Jean-François Pekel and Pierre Defourny Department of Environmental Sciences and Land Use Planning - GEOMATICS UCL Université Catholique de Louvain BELGIUM In close collaboration with E. Bartholomé (JRC-SAI) Supported by the Joint Research Center European Commision
20
Embed
GLC mapping in semi-arid regions: a case study in West Africa Jean-François Pekel and Pierre Defourny Department of Environmental Sciences and Land Use.
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.
Transcript
GLC mapping in semi-arid regions:
a case study in West Africa
Jean-François Pekel and Pierre DefournyDepartment of Environmental Sciences and Land Use Planning - GEOMATICS
UCL Université Catholique de Louvain BELGIUM
In close collaboration with E. Bartholomé (JRC-SAI)
Supported by the Joint Research Center European Commision
Department of Environmental Sciences and Land Use Planning - UCL GLC 2000, 28 & 29 mars 2001
S10 NDVI data are S10 NDVI data are temporally consistenttemporally consistentthanks to the BRDF reduction effects and thanks to the BRDF reduction effects and the low impact of clouds in the MVC compositethe low impact of clouds in the MVC composite
S1 multispectral data are S1 multispectral data are spatially consistentspatially consistent
• Regional seasonality allows to assume that theRegional seasonality allows to assume that the beginning of the dry season provides the mostbeginning of the dry season provides the most spatially constrasted imagespatially constrasted image
Overall approachOverall approach
Unsupervised classificationsUnsupervised classifications (ISODATA) (ISODATA) based onbased on
+ the + the best cloud free S1best cloud free S1 image(s) image(s) 21/10/99 complemented by 22/10/9921/10/99 complemented by 22/10/99
+ + 5 phenological variables5 phenological variables either computed either computed - - by pixelby pixel
- - by classeby classe
3 methods3 methods• S1S1 interactive classes merging and labeling interactive classes merging and labeling
• S1 + pixel-based phenological variablesS1 + pixel-based phenological variables interactive classes merging and labeling interactive classes merging and labeling
• S1 S1 classe-based phenological variablesclasse-based phenological variables interactive classes merging and labeling interactive classes merging and labeling
Department of Environmental Sciences and Land Use Planning - UCL GLC 2000, 28 & 29 mars 2001
5 phenological variables from S10 NDVI time series5 phenological variables from S10 NDVI time series
Time
NDVI
NDVI min = average of 5 lowest NDVI valuesNDVI min = average of 5 lowest NDVI values
NDVI max = average of 3 highest NDVI valuesNDVI max = average of 3 highest NDVI values
NDVI range = NDVI max - NDVI min NDVI range = NDVI max - NDVI min
Veg. Duration = Veg. Duration = Date Veg. End - Date Veg. Start Date Veg. End - Date Veg. Start
Start Date = first date > Start Date = first date > NDVI min + 0.33 NDVI rangeNDVI min + 0.33 NDVI range End Date = last date > End Date = last date > NDVI min + 0.33 NDVI rangeNDVI min + 0.33 NDVI range
Best S1 unsupervised classificationBest S1 unsupervised classification• interactive merging into meaningful interactive merging into meaningful classesclasses• classes labelingclasses labeling
Department of Environmental Sciences and Land Use Planning - UCL GLC 2000, 28 & 29 mars 2001
R, PIR, MIR
( 21/10/1999)
ISODATAMerging
Labeling
Maps and high resolution images
LC map 16 classes
30 classes
Department of Environmental Sciences and Land Use Planning - UCL GLC 2000, 28 & 29 mars 2001
Best S1 classification Best S1 classification 16 classes LC map 16 classes LC map
S1 classes merged automaticallyS1 classes merged automatically • according to per class average phenological according to per class average phenological variables variables computed on the year-long S10 time seriescomputed on the year-long S10 time series
Department of Environmental Sciences and Land Use Planning - UCL GLC 2000, 28 & 29 mars 2001
70
NDVI
Decades
Class n°
Decade
NDVI
30 classes
30 masks
LC map 16 classes
4 ph.var.
Cl.nbr Min Range Dur.End
ISODATA
ISODATA
Labeling
Phenological variables
Mean profile per class
Department of Environmental Sciences and Land Use Planning - UCL GLC 2000, 28 & 29 mars 2001
Best results Best results (expert, computing)(expert, computing)
More robust resultsMore robust results(computing) (computing) water !water !
Most efficient resultsMost efficient results(expert)(expert)
Department of Environmental Sciences and Land Use Planning - UCL GLC 2000, 28 & 29 mars 2001
Tree densityTree density
Crop areaCrop area
LCCS use - definition of 16 classes fromLCCS use - definition of 16 classes from steppe to open forest ! ! !steppe to open forest ! ! !
Department of Environmental Sciences and Land Use Planning - UCL GLC 2000, 28 & 29 mars 2001
Evolution temporelle de l'activité photosynthétique
0
0,05
0,1
0,15
0,2
0,25
0,3
0,35
0,4
0,45
0,5
0,55
0,6
0,65
0,7
0,75
0,8
Temps
« steppe arbustive »
Time
NDVI Temporal evolution of NDVI Temporal evolution of NDVI
Complementary information to classe name/codeComplementary information to classe name/code average NDVI profile per classeaverage NDVI profile per classe
Department of Environmental Sciences and Land Use Planning - UCL GLC 2000, 28 & 29 mars 2001
3 additionnal issues :3 additionnal issues :
• year to year variationyear to year variation
• agriculture density retrievalagriculture density retrieval
• method use for very large areamethod use for very large area
Department of Environmental Sciences and Land Use Planning - UCL GLC 2000, 28 & 29 mars 2001
Detection feasibility of the agriculture Detection feasibility of the agriculture densitydensity
Department of Environmental Sciences and Land Use Planning - UCL GLC 2000, 28 & 29 mars 2001
Procedure to apply the method to very large areaProcedure to apply the method to very large area Indpt unsup. classification on moving windowIndpt unsup. classification on moving window
Department of Environmental Sciences and Land Use Planning - UCL GLC 2000, 28 & 29 mars 2001
Procedure to apply the method to very large areaProcedure to apply the method to very large area Biomes stratification based on NDVI rangeBiomes stratification based on NDVI range
Department of Environmental Sciences and Land Use Planning - UCL GLC 2000, 28 & 29 mars 2001
Perspectives for GLC2000 data setPerspectives for GLC2000 data set
great information contentgreat information content
information extraction quite feasibleinformation extraction quite feasible
New (?) issues because of the information quality :New (?) issues because of the information quality :
classes classes mergingmerging becomes a more becomes a more difficultdifficult exercise exercise
classes classes labelinglabeling becomes a becomes a tufftuff exercise exercise
need for labeling assistance through actual need for labeling assistance through actual matchingmatching of classification output and reference informationof classification output and reference information
need for 1-km² ‘ field ’ observation methodneed for 1-km² ‘ field ’ observation method