Modern Change Detection Flashcards
two reasons that time-series based change detection methods are superior to older alternatives
- can provide highly accurate change identification with good characterization of the timing of the change
- can continuously monitor time series and flag changes more or less as they happen on Earth
Explain how the opening of the Landsat archive set the stage for the development of modern change detection techniques
we used to think of data as coming in whole images, and we desired scenes or images that were cloud free, or mostly cloud free over our areas of interest, we now can consider data for individual pixels
one of the most significant advances in finding the buried treasure in the Landsat archive
development of the FMASK algorithm for automatic cloud and cloud shadow detection
FMASK algorithm
- combination of optical bands and thermal imagery from Landsat’s sensors to identify clouds as bright and cold
- uses information about the sun angle and the likely cloud height to estimate where the cloud’s shadow should be on the ground
Once we have a reliable way of automatically masking out clouds and their shadows in large image archives, we can begin to develop the pixel-based approaches that utilize all these vast amounts of data
The most revolutionary approach to doing this is called Continuous Change Detection and Classification, or CCDC
key principle behind CCDC and all related approaches
model the time series of spectral reflectance, or an index like NDVI, using statistical models
overview of how the CCDC algorithm works
- create statistical model using several years of data during a time period where no change has occurred
- run the model with other observations to find when change occurred
- perform a land cover classification to identify when and what has changed
- After enough data have become available after the change, we can fit a new model for that pixel and continue monitoring for subsequent changes in the same manner
