lclu practicals - home - earth onlineearth.esa.int/landtraining07/d3pa-caetano.pdf · sobreiro....
TRANSCRIPT
LCLU practicalsMário Caetano
September 5th, 2007Practicals D3PA
Sep. 5, 2007 D3PA LCLU Practicals Mário Caetano 2
Exercise1
Sep. 5, 2007 D3PA LCLU Practicals Mário Caetano 3
Areas that are in their major part constituted by waterWater10
Areas with 10% - 30% trees and with < 50% of herbaceous vegetation
Transitional-woodland forest9
Areas with 10% - 30% trees and with > 50% of herbaceous vegetationAgro-forestry8
Areas with >30% trees in which both broadleaf and needleleaf types are between 30% - 70%Mixed forest7
Areas with >30% trees in which >70% are of the needleleaf typeNeedleaf forest6
Areas with >30% trees in which >70% are of the broadleaf typeBroadleaf forest5
Areas with < 10% trees and that are not Urban / Bare soil, Sparse vegetation and Cropland
Other natural vegetation4
Areas with < 10% trees and with > 70% of herbaceous vegetationCropland3
Areas with < 10% trees and with 30% - 70% of bare soilSparse vegetation2Areas with < 10% trees and with > 70% of bare soilUrban / Bare soil1DescriptionClass nameCode
Sep. 5, 2007 D3PA LCLU Practicals Mário Caetano 4
Definition of the mapping approach
Map format = raster
Map format = vector
MMU = pixel size of input satellite data
Feature selection > Image classification > accuracy assessment
The steps required to information extraction depend on the defined mapping approach:
MMU > pixel size of input satellite data
Feature selection > Image classification > post-processing > accuracy assessment
upscaling
Spatial unit of analysis = image pixelFeature selection > Image classification > post-processing > accuracy assessment
Generalisation + Raster to vector conversionSpatial unit of analysis = object
Image segmentation > Feature selection > Image classification > post-processing > accuracy assessment
GeneralisationGenerate the objects
Sep. 5, 2007 D3PA LCLU Practicals Mário Caetano 5
Sparse herbaceous vegetation9Pine tree crown8Cork tree crown7Herbaceous vegetation6Shadow5Eucalyptus crown4Bare soil 3Shallow water2Deep water1Class nameCode
Sep. 5, 2007 D3PA LCLU Practicals Mário Caetano 6
Tree
Shrub
Herbaceous veg.
Map of landscape units
Pixel 30 m (e.g. Landsat)
Sep. 5, 2007 D3PA LCLU Practicals Mário Caetano 7
Forest
Shrubland
Pixel 30 m (e.g. Landsat)
Map of landscapeunits
Tree
Shrub
Herbaceous veg.
Sep. 5, 2007 D3PA LCLU Practicals Mário Caetano 8
Floresta
Mato
Pixel 30 m (e.g. Landsat)
Map of landscapeunits
Tree
Shrub
Herbaceous veg.
Sep. 5, 2007 D3PA LCLU Practicals Mário Caetano 9
Tree crown
Shade
IKONOS 4 m (e.g. Landsat)
Tree
Shrub
Herbaceous veg.
Sep. 5, 2007 D3PA LCLU Practicals Mário Caetano 10
Surface elementsmap
Tree
Shrub
Herbaceous veg.
Sep. 5, 2007 D3PA LCLU Practicals Mário Caetano 11
IKONOS
Surface ElementsMap
Objects
SEM + objects Landscape UnitsMap
Sep. 5, 2007 D3PA LCLU Practicals Mário Caetano 12
671346308614125451519104313712
Surface element – our decision
Surface element – your decision
Sample ID (ID)
Sep. 5, 2007 D3PA LCLU Practicals Mário Caetano 13
Segmentation
SEM+
Objects LUM
Forest = crowns + shade + herbaceous veg. + bare soil
Sep. 5, 2007 D3PA LCLU Practicals Mário Caetano 14
Water AgricultureBare
soil/urbanEucaliptus
forest
Segmentation
SEM+
Objects
LUM
Classes
Sep. 5, 2007 D3PA LCLU Practicals Mário Caetano 15
Pinus pineaforest
Quercussuber forest Mix forest
Agroforestryareas
Segmentation
SEM+
Objects
LUM
Classes
Sep. 5, 2007 D3PA LCLU Practicals Mário Caetano 16
Thematic information extraction from satellite images
12
3456
7
8 * mandatory
Geographical stratification
Image segmentation
Ancillary data integration
Post-classification processing
Definition of the mapping approach *
Feature identification and selection
Classification
Accuracy assessment
**
*
Sep. 5, 2007 D3PA LCLU Practicals Mário Caetano 17
Eucalipto
Pinheiro bravoOutras
Folhosas Pinheiro Bravo
Pinheiro Manso
Sobreiro
Eucalipto
Pinheiro Manso
Eucapiltusglobulus
Sobreiro
Sep. 5, 2007 D3PA LCLU Practicals Mário Caetano 18
0
200
400
600
800
1000
1200
1 2 3 4
Deep water Shallow water
Bare soil Eucalyptus tree crown
Shadow Herbaceous vegetation
Cork tree crown Pine tree crown
Sparce herbaceous vegetation
Spectral class means IKONOS
Sep. 5, 2007 D3PA LCLU Practicals Mário Caetano 19
In course resolution images the mixed pixels are mainly due to co-existence in the same pixel of different classes.
The problem of mixed pixels exist in coarse and fine resolution images:
The mixed pixel problem
MERIS FR
In fine resolution images the mixed pixels are mainly due to co-existence in the same pixel of different components (e.g., houses, trees).
IKONOS
Sep. 5, 2007 D3PA LCLU Practicals Mário Caetano 20
Meaningless segmentationMeaningful segmentation
Sep. 5, 2007 D3PA LCLU Practicals Mário Caetano 21
SEM+Objects
Sep. 5, 2007 D3PA LCLU Practicals Mário Caetano 22
Thematic information extraction from satellite images
12
3456
7
8 * mandatory
Geographical stratification
Image segmentation
Ancillary data integration
Post-classification processing
Definition of the mapping approach *
Feature identification and selection
Classification
Accuracy assessment
**
*
Sep. 5, 2007 D3PA LCLU Practicals Mário Caetano 23
3. Image segmentationA type of segmentation that is very common is the multi-resolution segmentation, because of its ability to deal with the range of scales within a single image.
Super-objects
Sub-objects
Sep. 5, 2007 D3PA LCLU Practicals Mário Caetano 24
Source: Jensen (1996)
Hard classificationDecision rules
0 – 30 -> Water30 - 60 -> Forest wetland
60 - 90 -> Upland forest
6. Classification
Fuzzy classification Decision rules are defined as membership functions for each class.
Membership functions allocates to each pixel a real value between 0 and 1, i.e. membership grade.
But, wow can we represent the sub-pixel information?
Sep. 5, 2007 D3PA LCLU Practicals Mário Caetano 25
Exercise2
Sep. 5, 2007 D3PA LCLU Practicals Mário Caetano 26
Thematic information extraction from satellite images
12
3456
7
8 * mandatory
Geographical stratification
Image segmentation
Ancillary data integration
Post-classification processing
Definition of the mapping approach *
Feature identification and selection
Classification
Accuracy assessment
**
*
Sep. 5, 2007 D3PA LCLU Practicals Mário Caetano 27
Water areas (lakes, rivers, ocean, etc)Water8Areas with > 80% of bare soilBarren7
Areas with shrubs, herbaceous and other vegetationOther natural vegetation6
Needleleaf tree areas with >40% crown coverNeedleleaf forest5Broadleaf tree areas with >40% crown coverBroadleaf forest4
Croplands with no vegetation (fallow lands)Agriculture without vegetation3
Croplands with vegetation (cultivated lands)Agriculture with vegetation2
Areas with a built up density >80%Urban1DescriptionClass nameCode
Sep. 5, 2007 D3PA LCLU Practicals Mário Caetano 28
The number of mixed pixels in an image varies mainly with:
Landscape fragmentation
Sensor’s spatial resolution
The mixed pixel problem6. Classification
MERIS FR pixels
Sep. 5, 2007 D3PA LCLU Practicals Mário Caetano 29
34263400239933824374231523082300327852194109
Surface element – our decision
Surface element – your decision
Sample ID (ID)
Sep. 5, 2007 D3PA LCLU Practicals Mário Caetano 30
0,000
0,050
0,100
0,150
0,200
0,250
0,300
0,350
0,400
1 2 3 4 5 6 7 8 9 10 11 12 13
Urban Agriculture with vegetation
Agriculture without vegetation Broadleaf forest
Needleleaf forest Other natural vegetation
Barren Water
Spectral class means MERIS
Sep. 5, 2007 D3PA LCLU Practicals Mário Caetano 31
9. Accuracy assessment
The most widely used method for accuracy assessment may be derived from a confusion or error matrix.
Accuracy assessment allows users to evaluate the utility of a thematic map for their intended applications.
The confusion matrix is a simple cross-tabulation of the mapped class label against the observed in the ground or reference data for a sample set.
Sep. 5, 2007 D3PA LCLU Practicals Mário Caetano 32
0,880,860,450,881,000,960,711,001,00PA2202222422425214222Total
1,00191980,911110170,74509371360,922622450,9226124141,00151530,825133124221,0022221UATotal87654321
MapR
efer
ence
Samples deterministically selected
Sep. 5, 2007 D3PA LCLU Practicals Mário Caetano 33
0,630,670,290,720,380,650,220,940,88PA0,03860158729884781139824Total1,0010610680,04521022521101170,7429012214311919460,8637432150,6184149519140,962612530,42217393261054092220,4448214110211UATotal87654321
Map
Ref
eren
ceSamples randomly selected
Sep. 5, 2007 D3PA LCLU Practicals Mário Caetano 34
Agriculture without vegetation
Urban
Sep. 5, 2007 D3PA LCLU Practicals Mário Caetano 35
Other natural vegetation
Needleleaf forest