Do-Hyung Kim , Raghuram Narashiman, Joseph O. Sexton, Chengquan Huang, John R. Townshend

1. A METHODOLOGY TO SELECT PHENOLOGICALLY SUITABLE LANDSAT SCENES FOR FOREST CHANGE DETECTIONIGARSS 2011 , Jul, 27, 2011 Do-Hyung Kim, Raghuram Narashiman, Joseph O. Sexton, Chengquan Huang, John R. Townshend Global Land Cover Facility, University of Maryland - College Park

2. CONTENT • 1. Background • 2. DATA • 3. METHOD • 4. RESULT • 5. DISCUSSION • 6. REFERENCE

3. Background • Influence of phenology on forest change detection • example of path 116/ row 32 (Korea) • Profile based techniques by time series data can resolve the issue of influence of phenology on change detection performance (Coppin et. al., 2004) Original Scene Oct 7 2006 Change detection False forest change by seasonality Sep 2 1999 Change detection Aug 28 2006 Aug 28 2006 Replacement Scene

4. Background • Global Land Survey • Global, orthorectified, typically cloud-free Landsat imagery centered on the years 1975, 1990, 2000 and 2005 with a preference for leaf-on conditions(Gutman, 2008). • LARGE AREA SCENE SELECTION INTERFACE (LASSI) • Global Land Survey 2005 is a dataset which is selected using such an automated method, LARGE AREA SCENE SELECTION INTERFACE (LASSI) (Franks, 2002). • An automated scene selection method which specialized for forest cover change detection is needed • Seasonality is not the only one parameter for LASSI. • GLS is not only for forest cover change analysis.

5. DATA • MODIS data • MOD13C1 : 5km NDVI dataset for the years 2000-2009 (Hueteet. al., 2002) • Land cover data • MOD12C1 : 5km Land Cover dataset (Friedl el. al., 2002) consists of the IGBP classification system from which the % forest, % evergreen, % deciduous and % crop layers were extracted. • Landsat METADATA • Metadata of globally available Landsat scenes dating back from the 1970s to present(http://landsat.usgs.gov/consumer.php )

6. METHOD • DATA process S = pixels > 40% deciduous & number of samples > 15 from MOD13C1 When I = composite (1<= i <=23) and j = year (2000 <= j <= 2009) Median value of the above samples for each ith composite at jth year NDVIij= Median(S) 10 year norm, NORM at each i composite NORMi= Median (NDVIij)

7. METHOD • Filtering NDVIij value which is greater or smaller than NORMi+- σ(NDVIij) is replaced by NORMi NDVI Composites

8. METHOD • Peak growing season selection • SOP and EOP

9. METHOD • Scene selection – web based app SOP, EOP User input Search Conditions – Date/Month/Year, Quality, Cloud, Path/Row SOP, EOP for each WRS2 tiles Perform Search Data Base update tool Data Base Landsat 7 ETM+ (SLC-on)Landsat 7 ETM+ (SLC-off)Landsat 1-5 TMLandsat 4-5 MSSLandsat 1-3 MSS Metadata UNZIP Search results Shown as Table

10. Deciduous forest Path/Rows Number of Path/row Deciduous 1546 7836 GLS

11. RESULT • SOP, EOP • Temporal consistency • Trend compared to latitude and biome • GLS replacement scene

12. SOP of 10 year norm

13. EOP of 10 year norm

14. SOP variation from 1999 to 2007 Variation (date)

15. EOP variation from 1999 to 2007 Variation (date)

16. Start of Peak by latitude and by Biomes Temperate Broad Leaf Tropical Dry Broad Leaf

17. End of Peak by latitude

18. GLS 2000 scenes need to be replaced Number of Scenes 424 1546 7836

19. GLS 2005 scenes need to be replaced Numbers of Scenes 435 1546 7836

20. Replacement scene selection • Browse through available scene list • Pick the best image based on visual observation • Criteria: Minimal cloud cover and within phenology bounds

21. P17 R28: Canada Peak Season Range: 5/25/2002 – 9/30/2002 GLS Replacement scene Replacement scene date: 8/24/2001 GLS2000 date: 5/15/2002

22. P22R49 (Guatemala)Peak Season Range: 6/10/1999 – 11/17/1999 GLS image: 12/4/1999 Replacement image: 8/6/1999 GLS date is just out of date range. Replacement scene has clouds. This is an example of replacement scene not being a better choice.

23. Replacement Scenes GLS 2000 284 Replacement / 424 scenes need to be replaced GLS 2005 252 Replacement /435 scenes need to be replaced

24. DISCUSSION • 1. Snow effect • 2. Scale issue • 3. Selection of path/row with seasonality • 4. Threshold selection • 5. Validation against ground measurement

25. Acknowledgement • This work has been carried out as part of the Global Forest Cover Change project, funded by the NASA MEaSUREs program (NNH06ZDA001N-MEASURES)

26. Reference • A. Huete, et al., “Overview of the radiometric and biophysical performance of the MODIS vegetation indices,” Remote Sensing of Environment, vol. 83, no.1-2, pp. 195-213, Nov., 2002. • Friedl, M.A., et al., “The MODIS land cover product: multi-attribute mapping of global vegetation and land cover properties from time series MODIS data,” Proceedings of the International Geoscience and Remote Sensing Symposium (IGARSS), vol. 4, pp. 3199-3201, 2002 • Coppin et. al., "Digital change detection methods in ecosystem monitoring: a review, “ IN T. J. REMOTE SENSING, 10 MAY, 2004, VOL. 25, NO. 9, 1565–1596 • U.S. Geological Survey (2010, Dec. 30), Landsat Bulk Metadata Service. Available: http://landsat.usgs.gov/consumer.php • Gutman, G., Byrnes, R., Masek, J., Covington, S., Justice, C., Franks, S., and R. Headley, Towards monitoring land cover and land-use changes at a global scale: The Global Land Survey 2005, Photogrammetric Engineering and Remote Sensing, 74, 6-10, 2008. • Franks, S., Masek, J.G., Headley, R.M.K., Gasch, J., and Arvidson, T., Large Area Scene Selection Interface (LASSI). Methodology for selecting Landsat imagery for the Global Land Survey 2005, in press Photogrammetric Engineering and Remote Sensing.