Conference Agenda

Multi-agent Awareness No-retracing Integrated Q-Learning (or MANIQ) for Anomaly Detection utilises multiple agents to select from possible legal actions using a boolean screening process. Increased awareness is supplied by a kernel and an awareness field with padding = 1. The kernel allows for 8 agents in 2 dimensions; each of which are assigned an instance with their designation; path; state; awareness field and activity status (TRUE or FALSE). While agents remain active in the field, they execute either explorative or exploitative policies on the environment. The status of an agent is determined by the dot product of the goal and the number of agents still active in the field divided by the total number of instances. Once an agent reaches its goal, it is summarily deactivated. Once all agents are deactivated, a new round begins. At each round, the start point of the agent, the field and legal actions and screenings are reset. A significant barrier in all earlier versions of Q-Learning anomaly detection was accounting for multiple rewards (of the same value) on the same Q-Table. MANIQ AD solves this by segmenting each datapoint into its own environment. The start point for each agent is the data point under consideration. The agent works backwards towards the zero-line and backpropagates the reward value, which is equal to the square of the normed and rounded value of the data point. If the integral of the Q-Learning Surface is greater than a predefined value, the data point is classified as an anomaly; otherwise normal. The MANIQ algorithm is a significant improvement in terms of both time complexity and accuracy when compared against earlier versions of the algorithm, including; AKAD, MRQL, MRQLV2, FUQL and IQAD, and its performance is comparable to state-of-the-art algorithms such as Matrix Profiles and HBOS. The algorithm has been tested on VFM SWARM data. Its goal is to detect anomalies in the Earth’s geomagnetic field in the months and years preceding several seismic events including the Aegean Sea (2020), Croatia (2020) and Mexico (2017) earthquakes.

The collapse of buildings is the main cause for casualties after earthquakes. Real-time and accurate positioning of the building areas are crucial to make an effective implementation of emergency rescue after an earthquake. Synthetic Aperture Radar (SAR) possesses advantages such as all-weather and all-day capabilities, as well as resilience to lighting and weather conditions. Therefore, the use of SAR imagery has garnered significant attention in various fields, including post-earthquake rescue, damage estimation, and urbanization studies, and so on. The single-polarimetric spaceborne SAR data are very difficult to decipher, because their information is simple and abstract, and the spatial resolution of radar satellite data is limited. In this case, the accuracy of recognizing building earthquake damage information using only one post-earthquake single-phase SAR data is low. In order to ensure the accuracy of building damage information extraction as much as possible, the project utilizes deep learning network to fuse multiple feature parameters to identify building damage from post-earthquake SAR images. By comparing the classification accuracies of deep learning networks with different data and feature parameters, we found that the deep learning method combining spatial domain and frequency domain texture features can more accurately recognize collapsed buildings and non-collapsed buildings. The method demonstrates a robust capability to identify collapsed and intact buildings. Taking the area of Kahramanmaras, Turkey hit by the 6 February, 2023 Turkey-Syria earthquake as the case study, the region severely affected by the earthquake, this project incorporates both spatial and frequency domain features into the deep learning network for classification. Experimental results show that the proposed method achieves a classification accuracy of 80.98%, significantly surpassing the classification accuracy of 47.84% for the original SAR image. Moreover, the accuracy of 80.98% is higher than using only spatial domain features (73.30%) or only frequency domain features (73.42%). The proposed method in this study can provide fundamental support for post-earthquake disaster assessment and situational awareness.

The Tianshan orogenic belt (TSOB) represents one of the most dynamically active regions in Eurasia. The distal effects stemming from the late Cenozoic collision between the Indian and Eurasian tectonic plates have triggered renewed tectonic activity within the TSOB, fostering intracontinental orogenesis. Concurrently, the TSOB has extended into the foreland basins flanking its peripheries, giving rise to multiple series of fold belts associated with décollements and faults at the basin-mountain interface. Global Positioning System (GPS) data indicates a gradual reduction in the north-south shortening rate across the TSOB, diminishing from approximately 20 mm/yr in the western segment to around 8 mm/yr in the eastern part. Nevertheless, the internal distribution of deformation within the TSOB remains a subject of debate. In this study, we ascertain the contemporary kinematics of the principal structural zones by leveraging Interferometric Synthetic Aperture Radar (InSAR) data acquired from the Sentinel-1 satellite constellation.

Our analysis incorporates Synthetic Aperture Radar (SAR) data gathered from 7 ascending tracks (T27;T129;T56;T158;T85;T12;T114) and 6 descending tracks (T107;T34;T136;T63;T165;T92) of the Sentinel-1A/1B satellites spanning the period between November 2014 and December 2020. Through the utilization of Gamma software, we generated a total of 1509 single-reference interferometric pairs, covering a geographical extent of 1290 km in length and 490 km in width within the TSOB. Subsequently, the InSAR time series data underwent processing using the StaMPS software package. Long-wavelength atmospheric errors and elevation-dependent variations were mitigated through the application of the TRAIN package in conjunction with the ECWMF ERA5 models.

By integrating InSAR observations with GPS measurements, our study demonstrates a heterogeneous distribution of tectonic deformation across the TSOB. The convergence rate along the Tianshan ranges between 15–24 mm/yr, with the most pronounced deformation gradient localized at the junction between South Tianshan and Pamir, accounting for approximately 68% of the overall convergence deformation. South Tianshan exhibits relative stability with limited deformation gradients, while residual deformation is discernible within the intermontane faults and basin systems north of South Tianshan. Notably, the Kashi fold-thrust belt emerges as a highly active unit, with deformation predominantly concentrated in the Mushi, Kashi, and Atushi folds, as well as the interfold faults like Kashi, Atushi, and Toth Goubaz faults. The Maidan fault, serving as the boundary fault separating South Tianshan and the Tarim basin, exhibits a discernible deformation gradient. Within the Keping nappe, deformation primarily manifests around the Keping hill and Kepingtag fault at the nappe's forefront. The Kuche foreland hosts several prominent deformation zones, with deformation patterns in the northern sector of South Tianshan dispersed across a series of active intermontane structures and depression basins, contrasting the southern region where deformation primarily concentrates on thrust folds.

Strain rate analyses within the Tianshan Zone highlight prominent strain accumulation zones at the pre-subduction front of the Pamir, followed by the southern margin of Issyk-Kul lake and the southern periphery of the eastern segment of South Tianshan. The Mw7.1 earthquake on January 23, 2024, in Wushi, Xinjiang, coincided with areas of heightened surface contraction strain and shear strain. Using InSAR data, we derived and inverted the fault slip distribution associated with the coseismic deformation field of this event, revealing it to be a thrust and strike-slip earthquake in alignment with the regional strain context.

The goal of this work is to examine the applicability of Generative Adversarial Networks (GANs) Neural Networks (NNs) in creating synthetic geomagnetic residual spatial temporal datasets, comparable to real data, as acquired by ESA’s SWARM A and C satellites. The long-term aim is to utilize the synthetic data for detecting potential geomagnetic pre-earthquake anomalies and investigate to which extent this generation can help in addressing the problem of sparsely collected real data points. Geomagnetic data collected in a period of one year before and during a major M6.0 earthquake episode in Arzak, China, on 19^th January 2020, was used for this study. Geomagnetic datasets collected by SWARM A and C have been merged for this period, and a 5-day window has been found to provide the optimal spatial coverage. A Feedforward Discriminator and Generator and Long Short Term (LSTM) GAN NN architectures have been used for the purpose of generating synthetic 5-day data blocks. Initial results appear promising, and a comparison study is currently underway.

The Li-Kan Road landslide (LKRL)，located on the right bank of the Lijia Gorge Reservoir (LJGR) in northwestern China, is a typical active slope induced by a road engineering project. The long-term creep of the landslide has resulted in frequent repairs to the infrastructure. More significantly, it poses a severe threat to the safety of the nearby hydropower station located approximately one kilometer away. In this study, Sentinel-1 datasets are utilized to analyze the spatiotemporal evolution characteristics of surface deformation in the LKRL since 2014. Additionally, geometric parameters of the landslide are inverted to provide an estimation of its thickness and volume. Firstly, line-of-sight (LOS) surface displacement velocities are derived from both ascending and descending orbit images in the LJGR area using time series InSAR technology. This approach reveals the presence of multiple active slopes in the region. Subsequently, by integrating dual-orbit observations with the aspect-parallel flow model, three-dimensional (3D) displacement time series of the LKRL are retrieved. The results indicate that the maximum cumulative horizontal and vertical displacements of LKRL during the period from 2014 to 2023 exceed 3.2 m and 1.0 m, respectively. Additionally, the landslide exhibits spatial heterogeneity in its deformation pattern. Following this, using the 3D displacement rate and the mass conservation equation, the sliding geometry of the landslide is inverted, revealing a sliding surface with an average depth of 10.6 m and a volume of approximately 1.45×107 m3. The landslide is characterized by an uneven thickness distribution. Finally, the deformation mechanism of the LKRL is discussed and it is concluded that the landslide is currently in a phase of constant motion without an imminent risk of catastrophic failure. This study represents the first systematic mapping and analysis of LKRL movement using satellite imaging geodesy methods. The findings contribute to the mechanistic interpretation and risk management of the LKRL.

The pixel offset tracking method has been widely applied to monitor the horizontal displacement of landslides through orthophotos. However, it predominantly relies on cross-correlation matching and sliding windows for offset estimation, which may introduce bias or boundary effects due to lower cross-correlation. In this paper, we utilized a two-stage method that combines cross-correlation matching with the pyramid Lucas-Kanade optical flow method to retrieve the horizontal displacement from optical images. The former ensures the initial global alignment of images, while the latter calculates the local horizontal displacement. In this work, we chose Upper Tena Valley in Spain as the study area because many landslides were developed in this place, and open-access orthophotos covering the region were also collected. First, the horizontal surface displacements of landslides in the area were recovered through the two-stage method. Subsequently, the three-dimensional displacement was retrieved based on a prior model assumption. Our results demonstrate the effectiveness of this method in mitigating the boundary effect associated with traditional cross-correlation offset estimation methods, providing robust technical support for landslide deformation monitoring.

In large hydropower projects, first impoundment leads to substantial increases in reservoir water levels, significantly transforming the geoenvironment and activating associated geohazards. The Baihetan Hydropower Station, ranked as the world's second-largest hydropower project, beginning its first impoundment on April 6, 2021. Unlike other large hydropower projects, the first impoundment of the Baihetan Hydropower Project has altered the geoenvironment more drastically, experiencing a rise in reservoir water levels of approximately ~120 m in just seven months. To reveal the impact of extensive and first impoundment on the dynamics of reservoir slopes in Baihetan, we employed an integrated spaceborne time series interferometric SAR (InSAR), optical remote sensing, Airborne UAV Photogrammetry, and field investigation. This approach led to the detection of 24 unstable landslides. We also introduced an innovative method for identifying displacement patterns of detected slopes, including acceleration and deceleration patterns. Our findings indicate that subsequent to the first impoundment, numerous reservoir slopes experienced acceleration, primarily due to the buoyancy effect of the reservoir water on their resisting sections. Additionally, the observed transition from acceleration to deceleration in the displacement of some reservoir slopes can be primarily linked to the buoyancy effect within the driving sections of these slopes. Our research has shed light on the profound impact exerted by both extensive and first impoundment on the dynamic behavior of reservoir slopes. Moreover, our study highlights the remarkable capability of the time series InSAR technique in accurately identifying and characterizing the displacement patterns of reservoir slopes. This knowledge is crucial for geohazard prevention and geoenviroment protection, especially in the context of constructing large hydropower projects.

Floods is a natural risk that threat with significant impacts in flood-prone regions, particularly those with complex geomorphological and hydrological characteristics. The present study investigates the flood dynamics in the Alto Guadalentín valley, an orogenic tectonic depression experiencing flash floods and land subsidence due to groundwater withdrawal. Employing 2D flood event models, two scenarios for 1992 and 2016 were simulated and compared, using the HEC-RAS 2D hydraulic model. Differential SAR Interferometry (DInSAR) techniques allowed quantifying ground deformations for the period 1992-2016; such data were combined with the Digital Elevation Model (year 2009) in order to obtain the topography for the two scenarios. The study demonstrates that in regions with simultaneous land subsidence and flooding, flood risks can be significantly altered. The hydraulic modelling results indicate a considerable rise in flood depth and in the extent of inundation in 2016 when compared to 1992. There were notable alterations in the water surface elevation throughout the 14-year span, resulting in a 2.04 km² expansion of areas experiencing water depths surpassing 0.7 m and indicating a critical interaction between subsidence and flooding. Additionally, changes in flood propagation patterns, supported by velocity maps, indicate a migration of inundation areas, influencing flood risk distribution. The study also discusses the economic implications of subsidence-induced flood risk changes, highlighting the importance of considering subsidence in flood risk evaluations. The impact on drainage capacity and economic costs emphasises the need for comprehensive flood risk management strategies. Flood hazard maps, integrating subsidence data, provide essential information for local authorities to identify high-risk areas, formulate effective management strategies, and evaluate hazard and economic costs.

Keywords: floodplain dynamics, hydrogeological modelling, InSAR, land subsidence, risk evaluation.

1. Introduction

InSAR technology currently stands as a prominent field within remote sensing monitoring techniques, finding widespread application in diverse areas such as ground subsidence monitoring and geological hazard investigation, among others [1-2]. The processing workflow of InSAR data involves a blend of knowledge across multiple disciplines, including computer science, geodesy, signal processing, and remote sensing. The theoretical models underlying InSAR are intricate, and the steps involved in processing are numerous. Furthermore, the sheer volume of data necessitates significant computing power, storage resources, and other hardware devices. Harnessing the vast SAR remote sensing data and cloud computing resources available on the "AI Earth" Cloud platform, we have introduced a cloud-based InSAR comprehensive computation system. This system effectively dismantles barriers and complexities associated with InSAR technology, empowering users to directly apply InSAR results to address diverse industry applications without becoming overly focused on the intricacies of the technology itself. The algorithms employed by the system encompass a range of application scenarios, satisfying the requirements for comprehensive surveys and the emergency monitoring of natural disasters.

2. GPU-assisted InSAR processing

This paper aims to employ Graphic Processing Unit (GPU) techniques to accelerate the deformation extraction process from big Interferometric Synthetic Aperture Radar (InSAR) datasets, using Geospatial Cyber-Infrastructure (GCI) form AI Earth Geoscience Cloud Platform to bridge GPU with InSAR domain knowledge. We choose GPU technology because this technique has rapid development, and it offers local parallel data processing with low costs and latency. Our GCI integrates GPU and parallel deformation extraction algorithms to achieve near real-time extraction of deformation areas from big Sentinel-1 InSAR images, which also provides a high-performance computing platform for other InSAR data analytics in future studies. Using GPU technology and high-performance computing cluster to realize massive InSAR data calculation issues in a wide area is a critical technical problem, including efficient registration and high-precision ESD estimation methods [3].

(1) Rapid registration of SAR images. This study will use GPU technology to improve the efficiency of SAR image registration based on the Cross-Correlated method in traditional SRA image registration.

(2) ESD (Enhanced Spectral Diversity) estimation. This study combines GPU high-performance computing based on existing ESD theory to improve the efficiency of ESD estimation.

(3) We use traditional methods to realize the InSAR time series analysis process. This GCI will accelerate the deformation

extraction and could be applied to address significant data challenges for other InSAR studies.

3. Easy-to-use service cloud-based computing system

For different types of disasters and application researchers, the customizable service is designed to remove the application obstacles the user may encounter, including the constraints problems of full process data processing. The service applied a GPU rapid processing technology to provide users with quasi-real deformation analysis results in time with limited parameter input. The customized service system is built with adaptive parameters to reduce the difficulty for primary users of InSAR and considers different application scenarios to meet various monitoring requirements of natural hazards. The system provides wide-area crustal deformation monitoring results and fine deformation monitoring in critical areas [4].

The InSAR comprehensive computation system, built on cloud services, surmounts traditional limitations such as data downloading, storage, parameter configuration, and computing resources. It simplifies the input of parameters for various types of surface deformation and applications, providing users with rapid deformation analysis outcomes . At present, the InSAR processing environment has been successfully launched on the AI Earth cloud platform. Users can promptly submit tasks and leverage its capabilities via the platform (https://engine-aiearth.aliyun.com/).

4. Conclusion

This paper constructs Easy-to-use and fast-processing algorithm for InSAR data based on GPU technology and a high-performance computing cluster in AI Earth Geoscience Cloud Platform. The GPU acceleration optimization is carried out for the relatively time-consuming steps in processing InSAR data to achieve the fast calculation of InSAR deformation of massive Sentinel-1 SAR data. Based on the massive computing resources of the AI Earth Geoscience Cloud Platform, a customizable InSAR service is built for different application demands. Based on this system, it can realize targeted early identification of geological hazards, hidden dangers, dangerous situation, and disaster identification and provide efficient technical means for rapid post-disaster emergency response.

References

[1] Dai K, Li Z, Xu Q, Bürgmann R, Milledge DG, Tomas R, Fan X, Zhao C, Liu X, Peng J, Zhang Q (2020) Entering the era of Earth-Observation based landslide warning system. IEEE Geosci Remote Sens Magaz 8(1):136–153

[2] Cigna F, Tapete D (2021) Sentinel-1 big data processing with P-SBAS InSAR in the geohazards exploitation platform: an experiment on coastal land subsidence and landslides in Italy. Remote Sens 13(5):885

[3] Yu Y, Balz T, Luo H, Liao M, Zhang L (2019) GPU accelerated interferometric SAR processing for Sentinel-1 TOPS data. Comput Geosci 129:12–25. https://doi.org/10.1016/j.cageo.2019.04.010

[4] Li, Y., Jiang, W. & Zhang, J. A time series processing chain for geological disasters based on a GPU-assisted sentinel-1 InSAR processor. Nat Hazards 111, 803–815 (2022). https://doi.org/10.1007/s11069-021-05079-9

Sentinel-1 satellite provides free access to dual-polarization (VV and VH) images. The integration of information from both VV and VH channels in polarimetric persistent scatterer interferometry (PolPSI) techniques is expected to enhance the accuracy of ground deformation monitoring as compared to conventional PSI techniques, which utilize only the VV channel for Sentinel-1.

Persistent scatterer (PS) and distributed scatterer (DS) points play a crucial role in the PSI techniques. PSs with high phase qualities are commonly found in urban areas. As a complementary for PSs, DS points whose phase is affected by noise are commonly present in rural areas.

In this study, the identification and selection of PS and DS is based on an optimal channel created by combining the two polarimetric channels. PS candidates are selected through the amplitude dispersion (DA) criterion. To jointly utilize both PS and DS points, an adaptive speckle filtering based on the selection of homogeneous pixels (HP) was applied to the coherency matrix. Then, DS candidates were identified by using the average coherence criterion. Finally, using both PS and DS points, the Coherent Pixels Technique (CPT) was employed as the Persistent Scatterer Interferometry (PSI) processing method.

To analyze how the introduction of the VH channel helps improve the deformation measurement results, an experiment over Barcelona in Spain was carried out. The dataset consists of 189 dual-polarization SAR images acquired between December 2016 and January 2021. A wide variety of scenarios are present in this region, i.e., airport, harbor, and urban areas which exhibit diverse orientations of streets and buildings with respect to the acquisition geometry. Additionally, ground deformation is expected over some areas due to settlement of recent constructions and in the harbor.

Regarding PS, there are two cases in which the VH data contribute to improve the PS density. The first corresponds to scatterers that are oriented with respect to the incidence plane. The VH amplitude value of those scatterers are higher than VV channel. The second case appears more frequently than the first case and corresponds to pixels in which the VH amplitude is low but stable. Through the application of PolPSI technique, the VH channel can contribute to the selection of high-quality pixels by reducing the presence of peaks and fluctuations present in the VV channel, thus enabling the selection of pixels with good quality which would not have been identified if only VV data were processed (Luo, et al., 2022).

Instead of increasing the density, the contribution of VH channel for the identification of DS points is associated with a more accurate selection of HP (Luo, et al., 2023). The polarimetric information enables the differentiation of pixels that belong to different targets but have similar amplitude values in the VV channel. This results in a more reliable deformation measurement, as the HP group becomes more accurate.

A comparison with experimental data and all cases (single- and dual-pol) serves to illustrate and evaluate the performance of PolPSI in this domain.

Reference:

Luo, J., Lopez-Sanchez, J. M., De Zan, F., Mallorqui, J. J., & Tomás, R. (2022). Assessment of the Contribution of Polarimetric Persistent Scatterer Interferometry on Sentinel-1 Data. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 15, 7997-8009.

Luo, J., Lopez-Sanchez, J. M., De Zan, F. (2023). Analysis of the performance of polarimetric PSI over distributed scatterers with Sentinel-1 data. International Journal of Applied Earth Observation and Geoinformation, 125, 103581.

The heavy industrial district in the Shenyang municipality in Northeast China plays a relevant role in the economic and social development. The hard mining activities have a strong impact on local environment due to continuous ground excavations related to coal and iron extraction. Therefore, Shenyang is subject to a multi-hazard exposure including subsidence, landslides, ground fissure and building inclination. In particular, starting from the ESA DRAGON-4 project we begun to study the Shenyang city and the Fushun open pit mine by means of multi-source remote sensed data. One of the most important adopted methodology consisted on the use of the Advanced InSAR (A-InSAR) technique able to provide ground velocity and displacement time series with millimetric accuracy per year. Then, in the framework of the Dragon-5 project, we went on to monitor such areas, and to achieve this goal, a new COSMO-SkyMed (CSK) images dataset, operated by the Italian Space Agency (ASI), was required to extend the investigated period using the Persistent Scatterers Interferometry (PSI) technique. In fact, results from the previous Dragon-4 project indicated landslides around open-pit mines, building instability and structural damages. Furthermore, the tunnel construction of underground lines in Shenyang has caused surface fissuring, subsidence and sinkholes.

The city of Shenyang is covered by two distinct descending CSK frames along the descending orbit. One frame covers the western part, whereas the other covers the eastern part. The request of CSK images was carried out by submitting a project card to the Italian Space Agency (ASI). The project card ID 896 – DRAGON-5 was submitted via the ASI portal (https://portal.cosmo-skymed.it/CDMFE/home#).

The selected CSK images have been acquired along the descending orbit in the STR_HIMAGE mode. A total of 71 images were captured for the western part of Shenyang city, spanning from April 7, 2019 to November 11, 2023, while 94 images were acquired for the eastern part, interesting the Fushun open pit mine area, and covering the period from October 13, 2018 to December 30, 2023.

In this work, we show the updated results for the Shenyang and Fushun areas retrieved through the Enhanced PS technique, especially focusing on the pit mine site and the urban infrastructures (i.e. bridges, underground lines, embankments, etc.).

Acknowledgments

The COSMO-SkyMed data are provided by ASI through the project card ID 896.

Satellite Interferometric Synthetic Aperture Radar (InSAR) is widely used for topographic, geological and natural resource investigations. However, most of the existing InSAR studies of ground deformation are based on relatively short periods and single sensors. This paper introduces a new multi-sensor InSAR time series data fusion method for time-overlapping and time-interval datasets, to address cases when partial overlaps and/or temporal gaps exist. A new Power Exponential Knothe Model (PEKM) fits and fuses overlaps in the deformation curves, while a Long Short-Term Memory (LSTM) neural network predicts and fuses any temporal gaps in the series. Taking the city of Wuhan (China) as experiment area, COSMO-SkyMed (2011-2015), TerraSAR-X (2015-2019) and Sentinel-1 (2019-2021) SAR datasets were fused to map long-term surface deformation over the last decade. An independent 2013-2019 InSAR time series analysis based on TerraSAR-X scenes was also used as reference for comparison. The correlation coefficient between the results of the fusion algorithm and the reference data is 0.87 in the time overlapping region and 0.97 in the time-interval dataset. The correlation coefficient of the overall results is 0.78, which fully demonstrates that the algorithm proposed achieves a similar trend as the reference deformation curve. Based on the long time series settlement results obtained by fusion, we analyze the causes of settlement in detail for several subsidence zones. The subsidence in Houhu is caused by soft soil consolidation and compression. Soil mechanics are therefore used to estimate when the subsidence is expected to finish and to calculate the degree of consolidation for each year. The InSAR results indicate that the area has entered the late stage of consolidation and compression and is gradually stabilizing. The subsidence curve found for the area around Xinrong shows that the construction of an underground tract of subway Line 21 caused large-scale settlement in this area. The temporal granularity of the InSAR time series also allows precise detection of a rebound phase following a major flooding event in 2016. The experimental results demonstrate the accuracy of the proposed new fusion method to provide robust time series for the analysis of long-term land subsidence mechanisms and unveil previously unknown characters of land subsidence in Wuhan, thus clarifying the relationship with the urban causative factors.

In the framework of the research line that the Dragon 5 project n. 58113 “SARchaeology: Exploiting Satellite SAR for Archaeological Prospection and Heritage Site Protection” has dedicated to the improvement of satellite techniques for monitoring and conservation of archaeological sites, the present study has aimed to trial a multi-temporal approach based on optical multispectral imagery as a complement of Synthetic Aperture Radar (SAR) based investigations. The scope was to assess the impacts due to climate change-induced transformations of water bodies and marshlands in semi-arid environments that may threaten the conservation of archaeological sites.
The demonstration site is located in central-southern Iraq, encompassing a number of water bodies (lakes, swaps, streams and springs) that have been largely affected by changes over the last decade, including significant influences due to climate change.
Recent studies converge on warning about climate change impacts on Iraq’s natural resources, including a water shortage crisis. With surface waters projected to dry up within the next 20 years, environmental consequences are also likely to affect local cultural landscapes and heritage. Indeed, Iraqi archaeological sites and historic settlements, canals and palaeo-landscapes, are often located in proximity to water bodies and marshlands, thus shrink-swell cycles due to drought and flooding caused by extreme events, and consequent surface water run-off and accumulation, may accelerate cultural heritage deterioration, up to severe damage and disappearance.
On the other side, there is a growing literature and several national and international efforts are being conducted to document the rich cultural heritage of Iraq and improve the digital recording and databases of sites. Therefore, an analysis of the impacts due to climate change on Iraqi archaeological sites supported by Earth Observation data is more than timely and can rely on abundant authoritative geospatial datasets.
In order to assess the scale of recent impacts, a multi-temporal back-analysis was performed across a 8.000 km2 area, spanning from south of Baghdad to Basra. Multispectral Sentinel-2 images were processed to estimate annual changes in water level and marshland surface extent of both permanent and ephemeral lakes (Hammar Lake, Najaf Sea, Hor Al-Shuwaija), artificial reservoirs (Razzaza, El Delmej) and the Ahwar of Southern Iraq UNESCO World Heritage Site. Normalized Difference Vegetation (NDVI), Normalized Difference Water (NDWI) and Moisture Indexes enabled automatic per-pixel image classification, followed by thresholding and, when needed, manual refinement, to assess surface extent changes and spatio-temporal trends of water bodies and marshlands in 2015-2023. The observed divergent behaviour between the analysed bodies highlights a diverse range of situations, and thus different risk levels for heritage assets conservation. The integration with environmental and contextual data, and information from UNESCO reports, confirms the current challenges in preserving historical marshlands and the relationships with anthropogenic activities. The present paper therefore provides a spatio-temporal account of the evolving situation across a wide cultural landscape, and attempts first considerations about future projections should the same water-cycle dynamics continue as per the Sentinel-2 based observations.