Combined use of Electromagnetic Scattering Models, Fuzzy Logic and Mathematical Morphology for Flood Mapping from COSMO-SkyMED data L. Pulvirenti 1 , M. Chini 2 , N. Pierdicca 1 , L. Guerriero 3 (1) Sapienza, University of Rome (2) Istituto Nazionale di Geofisica e Vulcanologia (3) Tor Vergata University of Rome
Several overflows occurred in Italy in the recent years
ASI is presently funding some investigations about the use of Earth Observation data for civil protection from floods (e.g., OPERA ).
Potential of SAR for flood monitoring
The synoptic view , the good spatial resolution , the capabilities to operate in almost all-weather conditions and both during daytime and nighttime are the key features of radar sensors.
Possible advantages of using X-band COSMO-SkyMED images :
Very high spatial resolution (especially in the spotlight configuration) -> An accurate flood boundary delineation can be expected.
Short revisit time (constellation of 4 satellites) -> A Multi-temporal analysis can be performed.
SAR data interpretation often not straightforward, especially in the presence of vegetation ( challenging problem ).
Need to rely on electromagnetic scattering models , simulating 0 under flooded conditions, to correctly interpret the SAR observations.
Fuzzy logic suitable for representing the set of flooded pixels in SAR images for which the definition of a criterion of membership is a difficult task.
Spatial details of high resolution images generally smaller than the dimensions of the targets -> large within-class variances (also because of the speckle noise)
Need to segment imagery for dealing with homogeneous areas.
Mathematical morphology allows identifying objects with different spatial extension (when used in a multi-scale manner).
Steps of the designed algorithm and case studies
Segmentation of a multi-temporal series of CSK observations of a flood.
Computation of the average 0 for each segment.
Application to the segmented images of the fuzzy-logic-based approach
allowing us to account for different scattering mechanisms.
Methodology tested on two case studies (OPERA team activated) :
the overflow of the Tanaro River, close to the city of Alessandria (Northern Italy) in April 2009 ( 4 CSK images used ).
the flood occurred in Tuscany (central Italy) in December 2009 near the Massaciuccoli lake ( 5 CSK images used ).
In both cases a portion of flooded area was vegetated.
The opening (erosion followed by dilation) and closing morphological operators are applied with structuring elements ( se ) of different sizes.
Structures in a SAR image may have a high response for a specific selected se size and a lower response for other sizes .
The morphological profile is built for each image of the available multi-temporal series.
A K-means clustering is applied to the multi-temporal profile.
The final segmentation (extraction of contiguous objects belonging to the same class) is performed.
The fuzzy sets
Degree of membership to a fuzzy set defined through the standard S and Z functions.
Default values of the fuzzy thresholds based on the outputs of the EM scattering model developed at the Tor Vergata University of Rome. It assumes:
Bare soil : IEM
Vegetation : homogeneous half space overlaid by a layer filled with discrete dielectric scatterers representing stems and leaves.
Flooded conditions : simulated substituting the soil with a semi-infinite layer having the of water and a negligible roughness.
S function 0 1 0 1 Z function
Fuzzy set of flooded bare soils
Flooded bare soils generally much smoother than the surrounding dry land, thus acting as specular reflectors, giving low 0 .
Flooded 0 1 X Band, HH pol 0 HH SMC [%]
Set of vegetated flooded areas
Protruding vegetation may produce large 0 .
Reflections between water surface and upright vegetation may enhance backscattering -> flooded vegetation may show a bright radar return in a SAR image.
0 1 wheat plant height=70 cm 0 HH SMC [%]
Multitemporal analysis Tanaro river overflow . RGB color composite of the CSK observations of the flood Red : April, 29 2009 Green : April, 30 2009 Blue : May, 1 2009 NDVI map ( AVNIR-2 image acquired on April 23, 2009 ) bare vege t ated vegetated Mean ( 0 ) [dB] April, 29 April, 30 May, 1 May, 16 dry
Dependence of 0 on water level
Radar return predicted by the EM model (small leaves) versus the water level ( h w ) .
Large double buonce effect Small double buonce effect h w > h 60 cm plant 25cm plant 75 cm plant 0 [dB] h w [cm]
Other fuzzy rules (multitemporal analysis)
Non-flooded objects at time t are generally non-flooded at time t+ 1
Flooded objects that at time t have small 0 may be flooded at time t+ 1 if 0 ( t+ 1 ) considerably larger than 0 ( t ) (decrease of h w )
Flooded objects that at time t have large 0 may be flooded at time t+ 1 if 0 ( t+ 1 ) < 0 ( t ) (decrease of h w )
Non-flooded objects surronded by flooded ones placed at higher altitude are probably flooded (DEM-based correction)
60 cm plant 25cm plant 75 cm plant 0 [dB] h w [cm] 0 1
The Tanaro overflow
Occurred near the town of Alessandria (Northern Italy) on April 27-28, 2009.
~ 6000 people were evacuated for precaution.
Some agricultural fields were inundated. These fields were either bare or covered by wheat at different stage of growth (early – intermediate) .
Segmentation results Tanaro flood: original images RGB color composite (3500x5000 pixels) Red : April 29, 2009 Green : April 30, 2009 Blue : May 1, 2009 Tanaro flood: segmented image ~ 8000 objects
Flood evolution map (Apr. 29- May 1) Cyan :flooded Blue: water bodies
Flood map of April 29, 2009
The Tuscany flood
Occurred near the Massaciuccoli lake (Central Italy) on December 25-26, 2009.