Multi-sensor Evolution Analysis
Dynamic phenomena may have relevant effects on environment and therefore on mankind and can leave traces on land cover changes, detectable from EO data. In order to boost automation in the identification of features with dynamic behaviour, ESA has just implemented a prototype of a Multi-temporal analysis solution, called MEA. Its aim is to permit the definition and use of Evolution Models, which describe how to detect a specific featurefrom related land cover changes over time. In order to reach this objective, it is necessary to create a common classification system and an Earth Fixed Geo-spatial reference, also applicable across different missions.
MEA was studied to help the user to:
- define time Evolution Models of features
- apply Evolution Models across 13 years of (A)ATSR data
- obtain in a few seconds the results of related spatio-temporal and semantic queries.
In addition MEA will be used to test the applicability of Evolution Models across different sensors with different resolutions:- (A)ATSR at about 1Km- AVNIR 2 at about 10 m.
As already stated above, the starting point has been the creation of a common classification system applicable across different missions. The SOIL MAPPER®was chosen as a suitable solution for optical sensors, since it provides fast, automatic classification of single pixels into classes like:
- Bare soil / Built-up
- Snow / Ice
- Water / Shadows
Therefore SOIL MAPPER® assigns also a semantic meaning to each pixel.In order to permit a multi-temporal analysis across sensors with different resolutions, MEA is based on a common grid on the Earth made of latitude / longitude aligned “Tiles??? grouping always 64*64 grid elements of variable resolutions. Currently the Tiles dimensions are selected as follows:
- ATSR: 0.25° x 0.25° (4 Tiles x degree), with pixel size ~ 434 m at Equator
- AVNIR 2: 0.23’ x 0.23’ (256 Tiles x degree), with pixel size ~ 5 m at Equator
The implementation for 13 years of (A)ATSR data, even using only data over land and with minimal cloud coverage, results in over 300 million Tiles, posing huge issues for applying the Evolution Models through searches in the space, time and semantic dimensions within a few seconds of response time.Once the images are remapped on the same Earth fixed grid, MEA permits an expert to define Evolution Models (changes in land cover classes) for periodic features or for transitions, as well as the tolerances over searched classes and time transitions. The searches are performed over the area of interest comparing changes in the single pixels, over time.
The user, when applying an Evolution Model in his time and region of interest, obtains one or more tiles over the Earth map, with pixels tagged with a colour to indicate how they match the search parameters of the Evolution Model. In the next image, for example, the transition from snow to not-snow is tagged with the following colours:
- green: perfect match
- yellow: within tolerance
- red: no such change
- black: no data