1. Mining Changing Patterns in Satellite Image Time Series Thales Sehn Korting http://www.dpi.inpe.br/~tkorting/

2. Background Remote Sensing Data Mining Image Processing Multitemporal Images Bio Computer Engineer (FURG) MsC in Applied Computing (INPE) PhD candidate in Remote Sensing (INPE) Advisors  Leila Fonseca  Gilberto Câmara  Angela Schwering

3. The Earth is constantly changing.

4. Science challenges Understand patterns of change in local and global scale What is the impact of human-induced land cover change? How are ocean, atmosphere and land processes coupled? Where are changes taking place? How much change is happening? Who is being impacted by the change? (Kumar, 2001)(Câmara, 2008)

5. time Warning 1 Can we detect deforestation before it is too late? Warning 2 Loss > 90% Loss > 50% Warning 3 (Câmara, 2009)

6. (DEGRAD, 2009)

7. (Heas, 2005) Changes in different time-scales

8. (Goodchild, 2007) Changes in geographical objects

9. (Goodchild, 2007) We focus on stationary objects.

10. How to model changing patterns in land use/cover?

11. SITS 06/2008 07/2008 08/2008 SITS – Satellite Image Time Series

12. SITS Detect changes What? When? Where? t1 road construction t2 deforestation t3 deforestation

13. ≠ Temporal resolutions NDVI Variations in image attributes define temporal signatures.

14. Signature for deforestation (Freitas, 2008) Similar signatures define changing patterns.

15. (Kumar, 2001) What attributes that best describe changing patterns? Image objects Pixels Cells Regions

16. Algebra TransformationClassification Visual Interpretation

17. Hypothesis Classification methods based on data mining are efficient to identify temporal signatures.

18. Data Mining Traditional techniques are unsuitable due Large-scale High dimensional Heterogeneous Complex Statistics AI, M. Learning, P. Recognition Data Mining data data bases

19. Data Minining Challenges Take advantage of spatial and temporal correlation Volume of information 250m x 250m grid → about 10 billion for the globe Earth Science data sets are constantly increasing High Dimensionality Long time series are common in Earth Science

20. Decision trees to classify changes Independence of number of attributes amplitude of attributes Easy to understand the result

21. Objective Provide a technological framework to identify land use/cover changing patterns.

22. GeoDMA – Geographical Data Mining Analyst

23. ifgi Project

24. SITS classification scheme

25. Extended GeoDMA framework Timeline Visualization Mining

26. Mining Changing Patterns in Satellite Image Time Series Thales Sehn Korting http://www.dpi.inpe.br/~tkorting/