Damage detection using SAR coherence statistical analysis, application to Beirut, Lebanon
Introduction
On August 4th, 2020 at 15:08 UTC, the city of Beirut, capital of Lebanon, was shocked by a massive explosion at the city’s port. A large amount of ammonium nitrate which was stored in warehouse 12 exploded causing more than 6500 human casualties, nearly 15 billion US dollars of property damage, and almost 300,000 people homeless. It was estimated that the cause of the explosion was the improper storage of nearly 2750 tons of ammonium nitrate which caused an explosion equivalent to around 1155 tons of TNT (Devi, 2020).
These unexpected catastrophes raise the demand for rapid damage identification and assessment for better disbursing of the limited available resources to contain the effects of these events. One of the most used techniques for prompt damage detection and hazard assessment is synthetic aperture radar (SAR). This technique has the advantage of all-weather day and night observational capabilities over optical remote sensing. SAR-based damage (or change) detection technique can be categorized into an incoherent and coherent application (Preiss et al., 2006).
Incoherent change detection compares the backscattering amplitude of SAR images acquired before and after the event to identify target changes based on the observed amplitude differences (Dekker, 1998, Rignot and Zyl, 1993). To improve the accuracy of incoherent change detection, amplitude normalized difference, and log-ratio techniques were introduced (Bovolo and Bruzzone, 2005, Takeuchi et al., 2000). Also, an automated damage detection method based on amplitude correlation discriminant analysis was proposed (Matsuoka and Yamazaki, 2004), and to identify buildings with lower damage level Matsuoka and Yamazaki, (2006) proposed using ALOS/PALSAR time series amplitude images to calculate correlation difference. Guida et al. (2010) identified destroyed buildings from L’Aquila earthquake using double bounce changes from COSMO-SkyMed images.
On the other hand, coherent change detection techniques (CCD) use the interferometric correlation between SAR images to identify target changes based on the phase decorrelation rate of radar signals observed before and after the event (Yonezawa and Takeuchi, 2001). Several researchers have applied CCD techniques to detect damages related to natural disasters such as earthquakes (Arciniegas et al., 2007, Takeuchi et al., 2000), floods (Li et al., 2019), soil liquefaction (ElGharbawi and Tamura, 2015), and volcanos (Jung et al., 2016). Yonezawa and Takeuchi (2001) presented the relation between phase decorrelation and severity of damage using pre- and post-event coherence differences. Hoffmann (2007) used a coherence change index to quantitatively interpret urban damage induced by the Bam earthquake. Arciniegas et al. (2007) used coherence and amplitude correlation to detect earthquake damages and found that earthquake damage causes both increases and decreases in amplitude while coherence decreases with increasing damage severity (Dong and Shan, 2013).
The application of CCD extended beyond the study of two coherent SAR pairs to coherence time series analysis which utilizes the estimated phase correlation between several SAR images acquired along a certain time frame to identify the temporal behavior of the imaged targets based on the stability of their returned radar signals. Gamba et al. (2007) proposed using multitemporal SAR with Markov random field classifier and ancillary data to identify the damages resulted from the 2003 earthquake in Bam, Iran. Jung et al. (2016) formulated temporal decorrelation model to identify areas with decorrelation level higher than the expected amount due to naturally changing elements.
Our main target in this paper is presenting a simple yet robust change detection technique that allows prompt identification of the damages resulted by the industrial explosion in Beirut city, Lebanon. Our approach utilizes single master SAR stack with only one image acquired after the event to estimate coherent and incoherent correlation maps. The pre-event correlation maps are utilized in a temporal analysis to identify the natural statistical behavior of each pixel, then a co-event correlation map is utilized in a null hypothesis test based on student distribution to identify the damaged regions with 99%, 97.5%, and 95% confidence levels.
We also propose using spatial phase filter applied in the frequency domain to increase the accuracy of the estimated coherence maps by removing the dominant spatial phase component without affecting the nuisance components of the phase. This can be achieved by utilizing only the high coherent (stable) pixels in the proposed filter. We applied our technique to the city of Beirut using Sentinel 1 data and the results are presented and compared against damages map issued by ARIA and against optical satellite images.
This paper is organized as follows: Section 2 describes the study area, used data, and the main challenges in this analysis. Section 3 presents the proposed analysis approach. Section 4 discusses the obtained damage map of Beirut city, Lebanon and finally, Section 5 is dedicated to conclusions.
Section snippets
Study area
In this paper, we focus our analysis on the city of Beirut, Lebanon (Fig. 1), which was unfortunate to endure the devastating industrial explosion in its main port on August 4th, 2020 at 15:08 UTC. The port of Beirut serves as the main maritime entry point to Lebanon and a vital piece of infrastructure for importation. The port includes four basins, sixteen quays, twelve warehouses, a large container terminal, and a large grain silo (Fig. 2a) (www.portdebeyrouth.com).
One of the port’s
Damage detection methodology
This section illustrates the details of our analysis to identify the damaged regions in Beirut city. We start by presenting the basic idea and rationale of the analysis. Then, we illustrate the background and calculations of coherent and incoherent correlation maps. Finally, we present the statistical inference used in hypothesis testing to identify damaged regions.
Detected damages of Beirut City
For classifying the damage severity of the study region, we applied three significant level thresholds to the estimated combined P-value map (Fig. 8c) . The used significant level represents pixels that have a probability of to be presenting correlation value smaller than the pre-event estimated correlation, i.e. probability to be damaged.
We present our suggested three levels of damage severity in Fig. 9a overlaid on the post-event optical satellite
Conclusions
We present our analysis to detect damaged regions of Beirut city, Lebanon, as a result of the industrial explosion that occurred on August 4th, 2020. The proposed approach utilizes a SAR stack with one post-event image and nine pre-event images to identify the natural pre-event correlation behavior of each pixel and statistically classify the damaged pixels based on probability analysis.
We propose using a spatial phase filter applied in the frequency domain to remove the spatial decorrelation
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgement
We are grateful to the European Space Agency (ESA) for providing Copernicus Sentinel 1 data (2020), and to the Planet Team (2020) for providing all the optical satellite images.
References (30)
Lebanon faces humanitarian emergency after blast
The Lancet
(2020)- et al.
A comprehensive review of earthquake-induced building damage detection with remote sensing techniques
ISPRS J. Photogramm. Remote Sens.
(2013) - et al.
Estimating deformation due to soil liquefaction in Urayasu city, Japan using permanent scatterers
ISPRS J. Photogramm. Remote Sens.
(2015) - et al.
Urban flood mapping with an active self-learning convolutional neural network based on TerraSAR-X intensity and interferometric coherence
ISPRS J. Photogramm. Remote Sens.
(2019) - et al.
Lessons to be learned from an analysis of ammonium nitrate disasters in the last 100 years
J. Hazard. Mater.
(2014) - [dataset] Advanced Rapid Imaging and Analysis (ARIA), 2020....
- et al.
Coherence- and amplitude-based analysis of seismogenic damage in Bam, Iran, using ENVISAT ASAR data
IEEE Trans. Geosci. Remote Sens.
(2007) - et al.
A detail-preserving scale-driven approach to change detection in multitemporal SAR images
IEEE Trans. Geosci. Remote Sens.
(2005) - [dataset] Copernicus Sentinel data, p. b....
Speckle filtering in satellite SAR change detection imagery
Int. J. Remote Sens.
(1998)
Rapid damage detection in the Bam area using multitemporal SAR and exploiting ancillary data
IEEE Trans. Geosci. Remote Sens.
SAR interferometry: a “Quick and dirty” coherence estimator for data browsing
IEEE Trans. Geosci. Remote Sens.
Monitoring of collapsed built-up areas with high resolution SAR images
Mapping damage during the Bam (Iran) earthquake using interferometric coherence
J. Remote Sens.
Cited by (13)
Burnt-Net: Wildfire burned area mapping with single post-fire Sentinel-2 data and deep learning morphological neural network
2022, Ecological IndicatorsCitation Excerpt :With the coming new series of satellite imagery (i.e. Landsat-9, Sentinel-2) in addition to improving data quality, the temporal resolution of the dataset has been significantly improved (McCorkel et al., 2018). Accordingly, the use of remote sensing has been considered in many applications such as classification (Ahmad et al., 2021; Sellami and Tabbone, 2022), active fire monitoring (Wang et al., 2022; Xu et al., 2021b), flood mapping(Luppichini et al., 2022; Muñoz et al., 2021), and damage mapping (ElGharbawi and Zarzoura, 2021; Pilger et al., 2020). Due to the increase in the availability of remote sensing datasets, wildfires have received significant attention in past years (Morresi et al., 2022; Smith-Ramírez et al., 2022).
Spatial Correlation-Constrained Low-Rank Modeling for SAR Image Change Detection
2024, IEEE Transactions on Geoscience and Remote SensingA Review of Blast Loading in the Urban Environment
2023, Applied Sciences (Switzerland)Leveraging involution and convolution in an explainable building damage detection framework
2023, European Journal of Remote Sensing