Regionalized and application-specific compositing - a review of requirements, opportunities and challenges Patrick Griffiths & Patrick Hostert Geography Department, Humboldt University Berlin Joanne White & Mike Wulder Canadian Forest Service patrick.griffiths@geo.hu-berlin.de http://www.hu-geomatics.de Phone: +49 30 2093-6894 May 22 nd 2014 ESA, ESRIN S2 Science 2014
Pixel-based compositing Image compositing now allows to create cloud free, radiometrically consistent image datasets over large areas from the Landsat archive: Cloud reduction no longer main objective Surface reflectance allows integration of L4, L5, L7 (, L8) Global WELD, Roy et al. 2014 Compositing strategies are shaped by specific information needs, the regional context and data constraints: Annual deforestation vs. agricultural change mapping Land use, phenology & ecosystem dynamics Climate, cloud cover & data availability 2
Varying regional data availability Total number of scenes acquired by L5 in 1987 Total number of scenes acquired by L7 in 2010 3
Best-pixel compositing Parametric scoring approach to determine best-pixel observation: Griffiths, P., van der Linden, S., Kuemmerle, T., & Hostert, P. (2013). A Pixel-Based Landsat Compositing Algorithm for Large Area Land Cover Mapping. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 6, 2088-2101. White, J., Wulder, M., Hobart, G.W., Luther, J.E., Hermosilla, T., Griffiths, P., Coops, N.C., Hall, R.J., Hostert, P., Dyk, A., & Guindon, L. (in review). Pixel-based image compositing for large-area dense time series applications and science. Canadian Journal of Remote Sensing. Determination of best-pixel considering different parameters: Acquisition year, day-of-year balance annual & seasonal consistency Distance-to-clouds, thermal temperature, atmospheric opacity, sensor, local solar illumination angle, etc. optimize radiometric consistency 4
Best-pixel compositing Different compositing flavors for best observations: Annual best-observation composite 5
Best-pixel compositing Different compositing flavors for best observations: Multi-year best-observation composite 6
Best-pixel compositing Different compositing flavors for best observations: Proxy-value best-observation composite 7
Best-pixel compositing Utilizing good observations: Spectral-temporal variability metrics 8
Pampas, Argentina Land cover & crop type mapping Requirement: two seasonal observations within 2009 25 footprints, 572 images No topography, few clouds > low data constraints 9
Pampas, Argentina Approach: Spring (DOY 60) & Fall (DOY 258) composites for 2009 Variability metrics Extensive field-based training data 10
Pampas, Argentina Seasonal consistency: Spring composite Fall composite 11
Pampas, Argentina Land cover map 12
Overall accuracy 81% for 10 classes High class specific accuracies for corn (92%), soybean (86%), others 13
Canada: National Terrestrial Ecosystem Monitoring System (NTEMS) To support national programs and reporting obligations (e.g., NFI, Carbon Accounting) Very large area, 1224 footprints (terrestrial) >605,000 scenes in Canadian archive Requirement: annual land cover, land cover change, forest structure Shorter growing season in north but substantial 80% across track overlap 2010 14
Canada: NTEMS National multiyear composite (2009, 2010, 2011) 15
Areas with no observations Annual composite 2003 August 1 ± 30 days 16
Proxy value composite 2003 Areas with persistent no data are assigned a synthetic value, which is determined using a trajectory of available values for the pixel. 17
Carpathians, Eastern Europe Reconstruction of land change histories Different requirements: forest vs. agricultural dynamics ~5,000 images, 1984-2012, 32 footprints Strong topography, frequent cloud cover > moderate constraints 18
Spring composite Fall composite 19
Carpathians, Eastern Europe 20
Carpathians, Eastern Europe 21
Andes, Ecuador Best-observation composite, target year 2000 22
Andes, Ecuador Data scarcity! 17 footprints, 1538 images, 1984-2012 Cloud free pixel observations (median=40) 23
Andes, Ecuador Annual consistency 24
Summary Applications & information needs drive compositing decisions: Need to balance annual & seasonal consistency Maintain radiometric consistency Preserve synopticity Global gradients in data availability: Importance of USGS archive consolidation efforts Need for global acquisition strategies Need to explore strategies for data scarce regions: Data fusion approaches hold great potential Multi-mission & multi-sensor approaches 25
Summary L8 and L7 acquisitions, June 1 st January 31 st, 2014 26
Implications for Sentinel-2 Great potential for complementary S2-L8 data use: Merged surface reflectance product Need for rigorous cross-calibration Need for comparable L1 processing Need for joint application-focused research User requirements for S2: Free data Systematic global terrestrial acquisition and archive Standard analysis-ready product Easy data access On the long run: Bring the algorithms to the data.. Seamless temporal-spatial data queries via virtual constellations 27
Thank you! This research is partly funded by the German Federal Ministry for Economic Affairs and Energy (BMWi) through the SenseCarbon project (FKZ 50EE1254) We acknowledge support through the German Aeronautics and Space Research Centre (LDR) and Humboldt-University Berlin This research contributed to the Global Land Project and the Landsat Science Team www.hu-geomatics.de/projects/sensecarbon Research in support of the National Terrestrial Ecosystem Monitoring System (NTEMS): Timely and detailed national cross-sector monitoring for Canada project was jointly funded by the Canadian Space Agency (CSA) Government Related Initiatives Program (GRIP) and the Canadian Forest Service (CFS) of Natural Resources Canada. We acknowledge contributions of Geordie Hobart, Txomin Hermosilla, Nicholas Coops patrick.griffiths@geo.hu-berlin.de http://www.hu-geomatics.de Phone: +49 30 2093-6894 28
References: Griffiths, P., Mueller, D., Kuemmerle, T., & Hostert, P. (2013). Agricultural land change in the Carpathian ecoregion after the breakdown of socialism and expansion of the European Union. Environmental Research Letters. Griffiths, P., van der Linden, S., Kuemmerle, T., & Hostert, P. (2013). A Pixel-Based Landsat Compositing Algorithm for Large Area Land Cover Mapping. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 6, 2088-2101. Griffiths, P., Kuemmerle, T., Baumann, M., Radeloff, V.C., Abrudan, I.V., Lieskovsky, J., Munteanu, C., Ostapowicz, K., & Hostert, P. (2013). Forest disturbances, forest recovery, and changes in forest types across the Carpathian ecoregion from 1985 to 2010 based on Landsat image composites. Remote Sensing of Environment. 35