1 / 37

1 Cartography and GIS Research Group-Department of Geography 2 Department of Hydrology and Hydraulic Engineering

32 nd EARSEL Symposium 2012 “Advances in Geosciences” 21-24 May 2012 - Mykonos , Greece. Session on “ Imaging Spectroscopy”. Use of Land-Cover Fractions Obtained from Multiple Endmember Unmixing of Chris/Proba Imagery for Distributed Runoff Estimation.

bian
Download Presentation

1 Cartography and GIS Research Group-Department of Geography 2 Department of Hydrology and Hydraulic Engineering

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. 32nd EARSEL Symposium 2012 “Advances in Geosciences” 21-24 May 2012 - Mykonos, Greece Session on “Imaging Spectroscopy” Use of Land-Cover Fractions Obtained from Multiple Endmember Unmixing of Chris/Proba Imagery for Distributed Runoff Estimation Luca Demarchi1, Eva M.Ampe2, Frank Canters1, Jonathan Cheung-WaiChan1,3, Juliette Dujardin2, ImtiazBashir2, OkkeBatelaan2 1Cartography and GIS Research Group-Department of Geography 2Department of Hydrology and Hydraulic Engineering 3Department of Electronics and Informatics VrijeUniversiteitBrussel, Belgium

  2. Introduction • Use of remote sensing and GIS technology in hydrological modeling has strongly increased in the last decades: • Mapping spatial variability of several parameters for deriving runoff estimation Land-cover types • Hyperspectral have opened up new possibilities for land-cover mapping • Recent work has focused on the potential of hyperspectral CHRIS/Proba data for estimating sub-pixel land-cover fractions in urban areas[1]. Multiple Endmember Spectral Mixture Analysis (MESMA) [1] Demarchi L., Canters F., Chan J.C.-W and Van de Voorde, T. (2012). Multiple endmember unmixing of CHRIS/Proba imagery for mapping of impervious surfaces in urban and suburban environments. IEEE Transactions on Geosciences and Remote Sensing, DOI: 10.1109/TGRS.2011.2181853.

  3. Objectives • Integrate the results of MESMA in the Wetspass model: • Spatially distributed hydrological model for estimating the main water balance components: evapotranspiration, surface runoff and groundwater recharge • Compare the effects of different land-cover input scenarios on the spatial distribution of runoff. • Study area: • Woluwe catchment • Brussels Capital Region, east of city center • High heterogeneity and dense urban morphology • Hyperspectral CHRIS/Proba: • Spectral range: 410 – 1050 nm • MODE3: 18 spectral bands, 18m spatial resolution • (August 2009)

  4. Overview • Introduction • Methodology: • Unmixing CHRIS/Proba data with MESMA • Wetspass hydological modeling • Improving Wetspass with MESMA results: scenarios definition • Results and discussion • Conclusions

  5. Mapping land-cover types with multiple endmember unmixing (MESMA) • Sub-pixel land-cover mapping with medium-resolution multispectral imagery (Landsat, SPOT,...) in urban areas: • Spectral similarity of impervious surfaces and other non-artificial land-cover types (bare soil, dark vegetated areas,...) • Spectral heterogeneity of impervious surfaces  difficult to define representative endmembers for unmixing • MESMA: effective method on increasing land-over mapping accuracy when heterogeneity of land-over classes is very high. • Brightness normalization: technique proposed by Wu [2], reduces within-class spectral heterogeneity and emphasizes the shape information [2] Wu C. (2003). Normalized spectral mixture analysis for monitoring urban composition using ETM+ imagery. Remote Sensing of Environment, vol. 93, no. 4, pp. 480-492

  6. Brightness normalization Brightness normalization • r’b is the normalized reflectance for band b, • rb is the original reflectance for band b, • μ is the average reflectance for the pixel, • M is the total number of bands. • Before • After

  7. Brightness normalization Brightness normalization • r’b is the normalized reflectance for band b, • rb is the original reflectance for band b, • μ is the average reflectance for the pixel, • M is the total number of bands. • Before • After

  8. Mapping land-cover types with multiple endmember unmixing (MESMA) • Linear spectral unmixing • Unique endmembers are used for the entire scene: • Differences in land-cover composition and spectral variations within the same land-cover type are not taken into account • rib is the reflectance of endmember i for a specific band b, • fi is the fraction of endmember i, • N is the total number of endmembers, • ebis the residual for band b. • Multiple endmember unmixing • Per-pixel basis approach • Each pixel is unmixed with multiple models: • all combinations of endmembers, using 2, 3 or 4 EMs: only one retained! • Different spatial distribution of models using 2-, 3- or 4- land-cover classes for unmixing  more efficient fraction estimation

  9. Spatial distribution of models selected by MESMA

  10. Sub-pixel validation • Stratified validation set: • Fractions derived for each land-cover class from 0.25m ortho-photos • 30 pixels for all combinations of 2-, 3- and 4- land-cover classes • 30 pure pixels for each of the four land-cover classes (grey sealed surfaces, red sealed surfaces, vegetation and bare soil) • 12 combinations=360 ground truth pixels • Amplitude • Overall • Systematic

  11. Land-cover fractions Impervious surfaces

  12. Land-cover fractions Vegetation

  13. Land-cover fractions Bare soil

  14. Overview • Introduction • Methodology: • Unmixing CHRIS/Proba data with MESMA • Wetspass hydological modeling • Improving Wetspass with MESMA results: scenarios definition • Results and discussion • Conclusions

  15. Wetspass: a spatially distributed hydrological model for runoff estimation • Stands for Water and Energy Transfer between Soil, Plants and Atmosphere under quasi-Steady State [3]. • Physically based model able to simulate long-term average spatial patterns of groundwater recharge, surface runoff and evapotranspiration • It is able to handle the spatial distribution of several inputs such as soil types, land-use types, slope, groundwater depth and long-term average climatic data • Fully integrated in a geographical information system as a raster model [3] Batelaan, O. and De Smedt, F. (2001). WetSpass: a flexible, GIS based, distributed recharge methodology for regional groundwater modeling. Impact of Human Activity on Groundwater Dynamics, (IAHS Publ. No. 269). pp. 11-17.

  16. Wetspass: a spatially distributed hydrological model for runoff estimation • Water balance computation at cell level : • For each raster cell, the balance is split into independent water balances • By summing up each water balances, weighed by the corresponding fraction component, the total water balance at raster level can be obtained

  17. Overview • Introduction • Methodology: • Unmixing CHRIS/Proba data with MESMA • Wetspass hydological modeling • Improving Wetspass with MESMA results: scenarios definition • Results and discussion • Conclusions

  18. Improving Wetspass estimations using land-cover fractions from MESMA • Wetspass defines default fractions for each land-use type.

  19. Improving Wetspass estimations using land-cover fractions from MESMA • Wetspass defines default fractions for each land-use type. • Sub-pixel estimates obtained from CHRIS/Proba imagery are used in this study to improve runoff mapping within Wetspass. • Scenario 1: Semi-distributed (based on default Wetspass parameters) • av, ab, aiand aw are fixed a priori for each land-use class • Scenario 2: Semi-distributed (based on MESMA derived parameters) • av, ab, aiand aw are fixed a priori for each land-use class • Mean land-cover fractions are calculated from MESMA results for each land-use type • Scenario 3: Fully-distributed (pixel-based derived from MESMA) • av, ab, aiand aw are obtained at pixel-level from the MESMA results • In each scenario, average estimation and standard deviation of runoff are calculated for each land-use type

  20. Overview • Introduction • Methodology: • Unmixing CHRIS/Proba data with MESMA • Wetspass hydological modeling • Improving Wetspass with MESMA results: scenarios definition • Results and discussion • Conclusions

  21. Improving Wetspass estimations using land-cover fractions from MESMA • In scenario 2 new average land-cover fractions were calculated based on MESMA results - 10%

  22. Improving Wetspass estimations using land-cover fractions from MESMA • In scenario 2 new average land-cover fractions were calculated based on MESMA results + 10%

  23. Results of Runoff estimates • For each scenario, average runoff and standard deviation values have been calculated for each land-use type Urban land-use classes - 10%

  24. Results of Runoff estimates Urban land-use classes + 10%

  25. Results of Runoff estimates High standard deviations

  26. Scenario 1 and 2: similar values of runoff, spatial pattern very similar and clearly linked to the pattern of land-use

  27. Scenario 1 and 2: similar values of runoff, spatial pattern very similar and clearly linked to the pattern of land-use

  28. Scenario 1 and 2: similar values of runoff, spatial pattern very similar and clearly linked to the pattern of land-use • Variations within each land-use type are limited, confirming the small standard deviation obtained

  29. Scenario 3: strong local variation • In each pixel the Runoff is derived from its land-cover class composition and not from the land-use type • High local variability= high standard deviation. and therefore more realistic hydrological parameters estimates

  30. Runoff scenario 1

  31. Land-use map

  32. Runoff scenario 3

  33. Impervious surfaces from MESMA

  34. Overview • Introduction • Methodology: • Unmixing CHRIS/Proba data with MESMA • Wetspass hydological modeling • Improving Wetspass with MESMA results: scenarios definition • Results and discussion • Conclusions

  35. Conclusions • For most urban land-use classes, land-cover fraction values derived from RS are different from Wetspass default parameters: • Impervious surfaces level is systematically overestimated in Wetspass • Smaller runoff values are produced when RS data are used • Strong link between runoff and imperviou-sness level within each land-use type • Sub-pixel estimates derived from MESMA directly used at cell level: • Local variation of land-cover composition fully taken into account: high local variation of runoff within each land-use type • Benefits of using RS for obtaining more detailed information on the spatial pattern of runoff. • Combining MESMA-per-pixel basis unmixing approach-with Wetspass-a spatially distributed modeling-allows to generate fully distributed and more realistic hydrological estimates. • Limitations of CHRIS/Proba in urban areas are pointed out: • Spectral similarity of some land-cover types may negatively affect the quality of runoff in some locations. • Hyperspectral data with higher spectral resolution and wider spectral range may enhance this distinction and therefore runoff estimation

  36. Conclusions Thank you

More Related