Lecture 14 Flashcards
What does change detection involve
Identifying differences in the state of an object/feature or phenomenon
* Quantitatively analysing the temporal dynamics of the phenomenon or object/feature
Provide examples of phenomena that can lead to surface change
- urbanisation
- deforestation
- changing agricultural land use patterns
- Desertification
-Natural disaster - Geomorphic processes
- plant community changes
What is change detection
The process of identifying the differences in the state of an object or phenomenon by observing it at one or more different times.
what does change detection assume?
We want to identify the change between two images that is due to actual change in surface (e.g. landcover, temperature) and not due to external factors (e.g. ‘noise’)
▪ Need to account for changes in reflectance due to differences in the imagery due to (e.g.) :
▪ vegetation phenology
▪ atmospheric conditions
▪ solar angle (time of day/year)
▪ sensor characteristics
What is important to consider when mapping change detection?
It is important to use data with the same spectral and spatial resolution
Key things to consider :
▪ Imagery from the same time of the day
▪ Solar angle differences (e.g. shadow)
▪ Images from same time of year - minimizes the influence of seasonal Sun-angle and plant phenological differences
▪ Agricultural crops - planted at approximately the same time of year – delays in planting date between fields can cause detection error.
What are three types of change detection approaches?
- image differencing
- Temporal trajectory analysis
Post-classification comparison
What is image differencing?
When imagery from one date is subtracted from imagery of another.
- the “imagery” can be reflectance wavebands or ban ratios (e.g. NDVI)
- image differencing produces positive and negative values
What should be considered when using image differencing for image values of the same time of year?
- vegetation should be at a similar phenological stage
- Differences due to external factors (e.g.)
- drought
- vegetation loss/gain
- change in vegetation type
What would we expect the value to be/ how is this interpreted
Image differencing is easy to apply and interpret
- due to external factors - the stable features may not always have 0 value as you’d expect.
- therefore a threshold value to differentiate change from no-change should be decided.
What is image timeseries analysis often applied to?
▪ Wildfires
- deforestation
-regeneration
- phenological events
How does timeseries analysis work
Daily to annual temporal image
frequency
▪ need to consider the speed
at which the surface
change occurs and
dissipates
▪ More advanced methods used to
identify inflection points in the
timeseries of data
▪ i.e. date of change
What is post classification analysis
▪ Involves independently classifying images from two dates
▪ An algorithm is used to compare the classified images to produce a map indicating areas of change
▪ Provides qualitative information on the nature of the change
▪ Change of land cover X to land cover Y
What does post classification analysis require
Accurate classifications
▪ Errors will propagate in change detection output
▪ Misses changes within a class
What are the advantages of image differencing?
*Simple method
*Results are easy to interpret
Disadvantages of image differencing
*Inadequate qualitative
information regarding the
nature of the change
*Selection of change
threshold values can be
problematic
