The document provides an overview of change detection techniques in remote sensing. It defines change detection as the process of identifying differences in objects or phenomena by observing them at different times using remote sensing images. The main goals of change detection are to detect land use and land cover changes over time and understand how the Earth's surface is changing. Several change detection techniques are described, including visual analysis, image differencing, image ratioing, and post-classification comparison. Practical examples of detecting changes in lakes and forests over time are also presented.
2. ObjectivesObjectives
• Introduction
• What is Change Detection?What is Change Detection?
• Pre‐processing / Requirement
• Change Detection Techniques
• Application AreasApplication Areas
• Practical Example
• Further Readings
04/07/2013 2
3. Introduction
Remote Sensing (RS) methods try to answer
four basic questions:f q
How much of What is Where?
• What: Type, Characteristic and Properties of Object.
e.g. Water, Vegetation, Land etc.g g
• How Much: determine by simple Counting,
measuring Area covered or percentage of total areameasuring Area covered or percentage of total area
coverage.
• Where: Relate locations and area covered to either a• Where: Relate locations and area covered to either a
standard map or to the actual location on the
‘ground’ where the object occursground where the object occurs.
Note: Where also refers to a moment in time04/07/2013 3
4. • What is the SHAPE and EXTENT of ... ?
(Area Boundaries Lineaments )(Area, Boundaries, Lineaments, ...)
• This extends the ‘WHERE’ to be a completely
GEOMETRIC blGEOMETRIC problem.
– Identification and Delineation of Boundaries
04/07/2013 4
5. • What is the MIX of Objects?at s t e o Objects?
Th f f th E th i d b bj t likThe surface of the Earth is covered by objects like
Soil, Water, Grass, Trees, Houses, Roads and so on.
‐ Landuse/Landcover ‐ Classification
04/07/2013 5
7. What is Change Detection?What is Change Detection?
• Change detection is the process of identifying
differences in the state of an object or phenomenon
by observing it at different times.
• It is the detection of class transition between a pair
of co‐registered imagesof co‐registered images.
• The main goal is to use remote sensing to detect
CHANGE on a landscape (landuse and landcover)
over time.
04/07/2013 7
8. • Change detection algorithms analyze multiple images of
the same scene – taken at different times – to identifyy
regions of change.
• Changes on the earth surface could be directly caused by
natural forces, by the activities of animals and human
i d dinduced.
Ti l d h d i f E h’ f• Timely and accurate change detection of Earth’s surface
features provides the foundation for a better
understanding of the relationships and interactionsunderstanding of the relationships and interactions
between human and natural phenomena in order to
better manage and use resources.
04/07/2013 8
9. • It can be performed with raw remote sensing bands
or thematic land cover maps classified from them.or thematic land cover maps classified from them.
G d Ch D t ti h h ld id th• Good Change Detection research should provide the
following:
• area changeg
• rate of change
• spatial distribution of changed types• spatial distribution of changed types
• accuracy assessment of change detection results
04/07/2013 9
10. Pre‐processing / Requirement
• Geometric Correction – Georeferencing ‐ precise
coregistration between multitemporal imagescoregistration between multitemporal images
• Radiometric Correction ‐ precise radiometric and
atmospheric calibration or normalization betweenatmospheric calibration or normalization between
multitemporal images
/ f h l• Region/Area of Interest – same geographic location
• Remote sensing system consideration – spatial,
spectral, radiometric and temporal
– whenever possible, select images acquired from the same
type of sensors, with the same spectral and spatial
resolutions, and at the same seasonal timeframe in order to
i i i t d iminimize unwanted variances.
04/07/2013 10
11. • Free of clouds in the area of analysis
• Select time periods – what is change detection
period?period?
• Select Landcover scheme – they must be classified in
accordance with the same classification schemeaccordance with the same classification scheme.
– classes must also be defined identically
Cl ifi ti h l ifi ti l ith• Classification – choose classification algorithm
• Choose change detection method
• Change detection accuracy assessment
04/07/2013 11
14. Visual AnalysisVisual Analysis
• It is the first place to start
• Visually comparing multi‐images
• Manual digitizing changes in multi‐images is often g g g g
used to both identify and classify change between
imagesg
• Elements of image interpretation combined with the
knowledge of the area of study are often usedknowledge of the area of study are often used.
04/07/2013 14
15. Drying up of Lake Faguibine ‐ Mali
1974 2006
▪ It covered area of about 590km2
▪ Water level have fluctuated widely since the beginning of 1980
▪ An extended period of reduced precipitation led to a complete drying of the
lakelake
Source: Africa: Atlas of Our Changing Environment , UNEP
04/07/2013 15
16. Declining Water Levels in Lake Chad (1972‐2007)
1972 A (12 797 k )1972 1987Area (12,797sqkm)
Area (1,563sqkm)
2007
1987 Image show that lake Chad
Lake Chad, located at the junction
of Niger, Nigeria, Chad and
Cameroon, was once the sixth
g
reduced to about one-tenth of what it
was in 1972 image.
2007 image show some improvement
but the extent of the lake is stilllargest lake in the world.
Persistent drought and increased
agriculture irrigation have reduced
the lake’s extent
but the extent of the lake is still
smaller to what it was 2-3 decades
ago.
Area (1 753sqkm)Area (1,753sqkm)
04/07/2013 16
17. Image Differencingg g
• It requires selection of corresponding bands from two
dates imageries of the same study area
• Uses software algorithm to identify and quantify the
changes between two temporal images
• The difference image is created by subtracting the
brightness values of one image from the other on a per‐g g p
pixel basis.
• Unchanged areas will have values at or nearer zero; whileg ;
areas with significant change will be progressively
positive or negative.
04/07/2013 17
19. Advantages
• It is relatively easy to understand and to implement.y y p
• This method of analysis involves only subtraction
with minimal human intervention.with minimal human intervention.
• So long as the two images have been sampled to the
same ground resolution and projected to the samesame ground resolution and projected to the same
coordinate system, the subtraction can be carried
out very quicklyout very quickly.
• The results of change detection are not subject
• to the inaccuracy inherent in classified land cover
maps.
04/07/2013 19
20. LimitationsLimitations
• this method is limited in that it fails to reveal the• this method is limited in that it fails to reveal the
nature of a detected change (e.g., the class from
which a land cover has changed).which a land cover has changed).
• identify threshold values of change and no change in• identify threshold values of change and no‐change in
the resulting images.
• direct use of raw spectral data in change analysis
makes the detected change highly susceptible tomakes the detected change highly susceptible to
radiometric variations caused by illumination
conditions and seasonality.conditions and seasonality.
04/07/2013 20
21. Image Ratioing
• Similar to Image differencing conceptually and in its• Similar to Image differencing conceptually and in its
simplicity.
• This method uses one temporal image to divide
image of another date.
• Values near to 1.0 indicate – no change
• Values greater or less than 1.0 indicate changesg g
• Usually used for vegetation studies
• All other advantages and disadvantages of image• All other advantages and disadvantages of image
differencing apply to image ratioing.
04/07/2013 21
23. Post Classification ComparisonPost Classification Comparison
• Most popular method of change detection
• In post classification comparison, each date of
rectified imagery is independently classified to fit
common landtype.
• Landcover maps are overlaid and compared pixel by
pixel basis.p e bas s
• The result is a map of landtype change
• The change map display acreage of each change• The change map display acreage of each change
class
04/07/2013 23
26. Sources of Error in Change DetectionSources of Error in Change Detection
• Errors in data – image quality
• Atmospheric errorAtmospheric error
• Mis‐registration between multiple image dates
• Seasonal variability
• Processing error
• Radiometric error – due to sensor drift or age
• Error in ClassificationError in Classification
04/07/2013 26
27. Application AreasApplication Areas
l d /l d h• landcover/landuse changes
• mapping urban growth
• rate of deforestation
• urban sprawlurban sprawl
• desertification
di t it i• disaster monitoring
• agriculture
• coastal change
• environmental impact assessmentp
04/07/2013 27
31. Landcover Type 1996
A ( k )
1996
P t (%)
2008
A ( k )
2008
P t (%)Area (sqkm) Percentage (%) Area (sqkm) Percentage (%)
Farmland 30.627 74.71 30.772 75.07
Rock Outcrop 4.6449 11.33 4.0734 9.94
Bare Soil 3.537 8.63 4.1517 10.12
Forest Reserve 2.0133 4.91 1.89 4.61
Dam 0.171 0.42 0.1053 0.26
Total 40.9932 100 40.9932 100
04/07/2013 31