Water Body Detection from TanDEM-X Data:   concept & first evaluation of an accurate water indication mask   A. Wendleder 1) ,  M. Breunig 1) , K. Martin 2) , B. Wessel 1) , A. Roth 1)   1) German Aerospace Center DLR  |  2) Company for Remote Sensing and Environmental Research SLU   IGARSS 2011 / Vancouver / 2011-07-28 IGARSS’11, Vancouver
Outline Introduction Definition of the TanDEM-X water indication mask  Challenges for TanDEM-X water body detection  Concept & methodology of water body detection  Test site demonstration  Evaluation of classification results  Outlook Slide
Definition of the TanDEM-X  water indication mask Global mission – global DEM – global water body mask  Water body mask primarily extracted for post-processing DEM editing  ongoing work in flattening of outpoking water bodies  correct orthorectification of remote sensing data  No production of a complete global water body inventory Slide    Kurnool Kadapa Channel / India frozen Lake Taimyr / Russia
Challenges for TanDEM-X water body detection  TanDEM-X mission with 2 global acquisition data sets in 2011 & 2012  The water body detection runs completely data-driven Processing at Raw DEM level (30*50 km ≈ 8.000*10.000 pixels) 400 up to 800 Raw DEM per day to be processed  Therefore maximum computing time of 3 minutes per product Applicable for different appearances of water bodies worldwide  (coastline, inland lake, river, tropical, arctic, arid or humide climates etc.)   Slide    tropical river & coastline in Indonesia  small inland water bodies in Minnesota / USA
Concept & Methodology (I) Input  images are amplitude & coherence image Exclusion  of desert & polar regions  SRTM WAM MODIS/Terra Land Cover Types Exclusion  of steep terrain SRTM DEM Slide
Concept & Methodology (II) Slide    Median filter  separately applied both to amplitude & coherence image Threshold method   with fix threshold values Two different thresholds to handle complexity of water appearance  1. threshold: reliable classification 2. threshold: potential classification Calculation of water body areas via  Chain Code  and elimination of water bodies < 1 hectare  Fusion  of three intermediate water body layers
Test site demonstration River Elbe, Hamburg, Germany acquired on January 27, 2011 Incidence angle 43.4° to 45.7° Slide
Evaluation of classification results (I)  Slide    Calculation of completeness and correctness reference vector layer data of digital landscape models from the Authoritative Topographic Cartographic Information System (ATKIS)
Evaluation of classification results (II)  Slide    ATKIS: Authoritative Topographic Cartographic Information System Reference Completeness Correctness Amplitude ATKIS 86.9 % 92.6 % ATKIS water bodies > 1hectare 88.1 % 92.5 % Coherence ATKIS 79.8 % 98.7 % ATKIS water bodies > 1hectare 80.9 % 98.7 %
Evaluation of classification results (III)  Water body mask derived of  amplitude  image  rich in detail susceptible to misclassifications  Water body mask derived of  coherence  image significant and robust results loss of details of small scale water bodies Maximum of a correct & complete water mask with  combination of both Slide
Outlook Accuracy assessment of the water body detection  for different climate zones robustness & global transferability of our approach  Mosaicking of different water bodies  (neighboring acquisitions resp. first & second year acquisition)  to an intermediate & final TanDEM-X water body mask product TanDEM-X DEM editing using TanDEM-X water body mask  flattening of outpoking water bodies  Slide
Slide    River Elbe,  Hamburg, Germany SAR image Water indication mask  DEM  edited DEM
Slide
Slide    Thank you for your attention!  Anna Wendleder | Markus Breunig German Remote Sensing Data Center Team SAR Topography Phone: +49 8153 28 3439 Email: Anna.Wendleder@dlr.de | Markus.Breunig@dlr.de

igarss_2011_breunig_DLR_TDM_water_mask.ppt

  • 1.
    Water Body Detectionfrom TanDEM-X Data: concept & first evaluation of an accurate water indication mask A. Wendleder 1) , M. Breunig 1) , K. Martin 2) , B. Wessel 1) , A. Roth 1) 1) German Aerospace Center DLR | 2) Company for Remote Sensing and Environmental Research SLU IGARSS 2011 / Vancouver / 2011-07-28 IGARSS’11, Vancouver
  • 2.
    Outline Introduction Definitionof the TanDEM-X water indication mask Challenges for TanDEM-X water body detection Concept & methodology of water body detection Test site demonstration Evaluation of classification results Outlook Slide
  • 3.
    Definition of theTanDEM-X water indication mask Global mission – global DEM – global water body mask Water body mask primarily extracted for post-processing DEM editing ongoing work in flattening of outpoking water bodies correct orthorectification of remote sensing data No production of a complete global water body inventory Slide Kurnool Kadapa Channel / India frozen Lake Taimyr / Russia
  • 4.
    Challenges for TanDEM-Xwater body detection TanDEM-X mission with 2 global acquisition data sets in 2011 & 2012 The water body detection runs completely data-driven Processing at Raw DEM level (30*50 km ≈ 8.000*10.000 pixels) 400 up to 800 Raw DEM per day to be processed Therefore maximum computing time of 3 minutes per product Applicable for different appearances of water bodies worldwide (coastline, inland lake, river, tropical, arctic, arid or humide climates etc.) Slide tropical river & coastline in Indonesia small inland water bodies in Minnesota / USA
  • 5.
    Concept & Methodology(I) Input images are amplitude & coherence image Exclusion of desert & polar regions SRTM WAM MODIS/Terra Land Cover Types Exclusion of steep terrain SRTM DEM Slide
  • 6.
    Concept & Methodology(II) Slide Median filter separately applied both to amplitude & coherence image Threshold method with fix threshold values Two different thresholds to handle complexity of water appearance 1. threshold: reliable classification 2. threshold: potential classification Calculation of water body areas via Chain Code and elimination of water bodies < 1 hectare Fusion of three intermediate water body layers
  • 7.
    Test site demonstrationRiver Elbe, Hamburg, Germany acquired on January 27, 2011 Incidence angle 43.4° to 45.7° Slide
  • 8.
    Evaluation of classificationresults (I) Slide Calculation of completeness and correctness reference vector layer data of digital landscape models from the Authoritative Topographic Cartographic Information System (ATKIS)
  • 9.
    Evaluation of classificationresults (II) Slide ATKIS: Authoritative Topographic Cartographic Information System Reference Completeness Correctness Amplitude ATKIS 86.9 % 92.6 % ATKIS water bodies > 1hectare 88.1 % 92.5 % Coherence ATKIS 79.8 % 98.7 % ATKIS water bodies > 1hectare 80.9 % 98.7 %
  • 10.
    Evaluation of classificationresults (III) Water body mask derived of amplitude image rich in detail susceptible to misclassifications Water body mask derived of coherence image significant and robust results loss of details of small scale water bodies Maximum of a correct & complete water mask with combination of both Slide
  • 11.
    Outlook Accuracy assessmentof the water body detection for different climate zones robustness & global transferability of our approach Mosaicking of different water bodies (neighboring acquisitions resp. first & second year acquisition) to an intermediate & final TanDEM-X water body mask product TanDEM-X DEM editing using TanDEM-X water body mask flattening of outpoking water bodies Slide
  • 12.
    Slide River Elbe, Hamburg, Germany SAR image Water indication mask DEM edited DEM
  • 13.
  • 14.
    Slide Thank you for your attention! Anna Wendleder | Markus Breunig German Remote Sensing Data Center Team SAR Topography Phone: +49 8153 28 3439 Email: Anna.Wendleder@dlr.de | Markus.Breunig@dlr.de

Editor's Notes

  • #5 Challenge…. The currently sophisticated methods like parametric and geometric active contour models and texture measures are not efficient enough to be applied for near-real-time image processing and cannot be applied fully automatically. The only possibility here to keep up with the amount of daily incoming data during the TanDEM-X mission is to process the data fully automatically and to use a less time-consuming threshold method. In this case we apply the threshold method on….
  • #6 … the amplitude image and on the coherence image as we can see here.
  • #8 As you can see all water bodies are appearing smooth and dark without any disturbances caused by wind effect, snow and ice coverage or others
  • #13 Hamburg, Nord-Ostsee-Kanal WAM erklären, wie WAM editing coherence low/amp dark filtering/Glättung
  • #14 Hamburg, Nord-Ostsee-Kanal WAM erklären, wie WAM editing coherence low/amp dark filtering/Glättung