OPERATIONAL SAR DATA PROCESSING IN GIS ENVIRONMENTS FOR RAPID DISASTER MAPPING

ABSTRACT: Having access to SAR data can be highly important and critical especially for disaster mapping. Updating a GIS with contemporary information from SAR data allows to deliver a reliable set of geospatial information to advance civilian operations, e.g. search and rescue missions. Therefore, we present in this paper the operational processing of SAR data within a GIS environment for rapid disaster mapping. This is exemplified by the November 2010 flash flood in the Veneto region, Italy. A series of COSMO-SkyMed acquisitions was processed in ArcGIS using a single-sensor, multi-mode, multi-temporal approach. The relevant processing steps were combined using the ArcGIS ModelBuilder to create a new model for rapid disaster mapping in ArcGIS, which can be accessed both via a desktop and a server environment.


Introduction
Having access to SAR data can be highly important and critical especially for disaster mapping.Updating a GIS with contemporary information from SAR data allows to deliver a reliable set of geospatial information to advance civilian operations.
The operational processing of SAR data within a GIS environment for rapid disaster mapping is exemplified by the November 2010 flash flood in the Veneto region, Italy.A series of COSMO-SkyMed acquisitions was processed in ArcGIS® using a single-sensor, multi-mode, multi-temporal approach.The relevant processing steps were combined using the ArcGIS ModelBuilder to create a new model for rapid disaster mapping in ArcGIS, which can be accessed both via a desktop and a server environment.
Case Study: Flood Detection with High-Resolution COSMO-SkyMed Data The Bacchiglione River burst its banks on Nov. 2 nd 2010 after two days of heavy rainfall throughout the northern Italian region (see Figure 1).The community of Bovolenta, 22 km SSE of Padova, was covered by several meters of water (see Figure 2).(see Figure 1).The community of Bovolenta, 22 km SSE of Padova, was covered by several meters of water (see Figure 2).
The extent of this flooding is documented by a series of COSMO-SkyMed acquisitions.For the present case study we focused on a COSMO-SkyMed-3 image acquired at Nov. 5 th 2010, i.e. 3 days after the flash flood.As a reference we used a COSMO-SkyMed-2 image, acquired at May 9 th 2011.Both images were acquired in StripMap mode and ordered as Detected Ground Multi-look (DGM) product (Level 1B) with a GSD of 2.5 m and HH polarization (see Figure 6).

>
Step 1: The single COSMO-SkyMed images are filtered with a Gamma Distribution-Entropy Maximum a Posteriori (DE-MAP) filter to reduce the speckle effect (see Figure 7).

>
Step 2: The filtered images are geocoded using a reference DEM (SRTM3v4).Geocoding is performed by rigorously applying the Range and Doppler equations.This step includes the radiometric calibration, which is carried out by following the radar equation law (see Figure 7).

>
Step 3: A subsequent change detection analysis generates the final map showing the extent of the flood (see Figure 8).This customized process combines band arithmetic, segmentation, morphology filtering, and raster-to-vector conversion.

Implementation of the Workflow in ArcGIS
These 3 processing steps are combined using the ArcGIS ModelBuilder to create a new model for rapid disaster mapping.
Filtering (Step 1) and geocoding (Step 2) of a single COSMO-SkyMed image are integrated in a sub-model (see Figure 3), which is called twice in the main model, both for the "before" and the "after" image (see Figure 4).The two process sequences converge in the final change detection analysis (Step 3).
The underlying algorithms are provided by three different sources: Step 2) is a standard functionality from the ENVI SARscape Toolbox for ArcGIS.

>
This toolbox is extended by the filter tool (Step 1), which is called from the ENVI SARscape.

>
The change detection analysis (Step 3) is based on ENVI processing routines and scripted with IDL (Interactive Data Language).
The three tools are implemented in ArcGIS using a Python script file.This script file retrieves the parameters from the ArcGIS user interface and invokes the ENVI/IDL-based entry point procedure stored in an IDL .savfile, consisting of the precompiled IDL code (see Figure 5).

Conclusions
We developed an operational workflow for change detection analysis with SAR data, provided as an additional ArcGIS toolbox.Thus disaster zones, e.g. after severe flooding, can be automatically identified and mapped to support local task forces.
The benefits of the resulting processing chain include: > Execution by the responsible operators without SAR expert knowledge in a familiar GIS environment.

>
The open architecture platform IDL allows to customize single processing steps, or to create additional tools from the full complement of ENVI functionality.These image analysis processes can be integrated into any other ArcGIS models to update a GIS with contemporary information from remotely sensed data.

>
The ArcGIS model presented here can be dissolved from its desktop environment and published to users across the ArcGIS Server enterprise.

Figure 3 Figure 1 Figure 4
Figure 3 ArcGIS model for filtering (Step 1) and geocoding (Step 2) of a single

Figure 3
Figure 3 ArcGIS model for filtering (Step 1) and geocoding (Step 2) of a single COSMO-SkyMed image.

Figure 5
Figure 5 Filtering (Step 1) with the ENVI SARscape Gamma Distribution-Entropy Maximum A Posteriori filter.The filter is called via an IDL batch routine (below).This routine is linked to a customized ArcGIS Toolbox via a Python script (above).