Traditional Saharan underground canalization mapping from RS data

1
Traditional Saharan undergroundcanalization
mapping from RS data

• M.Quartulli, N.Pace, P.Laureano, M.Burgi

ESA-EUSC 2005 Image Information Mining -
Applications
2
Riade

• It is a project cofinanced by MIUR (the Italian
  Ministry of Education, University and Research)
  within the National Program of Research,
  Technological Development and High Level Training
  2000-2006.?The project objective is the
  development of an integrated and technologically
  innovative system for monitoring desertification
  processes localized in the southern areas of
  Italy, able to contribute at determining the
  cause - effect relationship in the observed
  phenomena, and to promote interventions towards
  the territory safeguard.??The partnership
  composition is as follows?
• Advanced Computer Systems A.C.S. S.p.A.?
• ENEA (Institute for the New Technologies, Energy
  and Environment)?
• Desertification Research Group (NRD) of the
  University of Sassari.

3
Foggaras

• Foggara/ Qanat/ Falaj/ Khottara/ Madjirat
  underground water drainage tunnels in arid and
  semi-arid regions
• 3000/6000 km total extension, half of which
  currently in use

4
Foggara canalizations

• 4/8 km long, 1/2 m diameter
• ventilation/maintenance wells 8 to 10m apart

5
Optical phenomenology
6
Optical superv. class. at 10m res.
7
C-band phenomenology at 10m
8
C-band SAR phenomenology
9
C-band SAR classification
10
C-band SAR classification
11
Foggara canalizations

• 4/8 km long, 1/2 m diameter
• ventilation/maintenance wells 8 to 10m apart

12
Stochastic geometry approach
A hierarchy of models is composed as in D ? T ?
S ? L ? G with prior probabilities pT(Tt),
pS(Ss), pG(Gg) describing models for the
elements and likelihoods pT,D(TtDd),
pS,T(SsTt), pG,S(GgSs) linking them.
13
Level 0 image data
14
1level points
15
2nd level segments
16
3rd level segment clusters
17
4th level linear segment clusters
18
5th level connected foggara
19
Inference algorithm

• Optimization by complete enumeration made
  possibile by limited problem size
• 4 min PC time, interpreted language

20
Conclusions

• IIM for resource management
• Very structured phenomenon over natural
  background easy discrimination
• HR L-band SAR
• Graph-based models and reconstruction algorithms