TL;DR
This paper presents a "forensic hydrology" approach to reconstruct the catastrophic IANOS hurricane flood event in Karditsa, Greece, a data-scarce region. By integrating remote sensing data, limited ground observations, and coupled hydrological-hydrodynamic models (HEC-HMS and HEC-RAS), the study successfully simulated the flood extent and estimated its extreme return period, offering valuable insights for flood risk management.
What Problem Does This Paper Address?
Flooding is a major natural hazard, and its risk is increasing due to climate change and urbanization. Accurately reconstructing past flood events is crucial for effective flood risk management, but this is often hampered by a lack of reliable data, especially in data-scarce regions. Traditional flood studies rely heavily on gauged river flow and water level data, which are frequently unavailable or destroyed during extreme events. This paper tackles the challenge of reconstructing an extreme flood event in a data-scarce area by employing a "forensic hydrology" framework, which leverages a diverse range of data sources, including remote sensing and crowdsourced information, to compensate for the lack of traditional measurements.
Methods
The study employed a multi-faceted approach, combining statistical analysis, hydrological modeling, and hydrodynamic modeling, all within a forensic hydrology framework.
- Statistical Rainfall Analysis: Initial analysis focused on the spatial and temporal variability of rainfall, using both ground-based records and various remote sensing products.
- Hydrological Modeling (HEC-HMS): A rainfall-runoff model (SCS-CN unit hydrograph method) was used to simulate the hydrological response of the Kaletzis catchment. This model was coupled with a hydrodynamic model.
- Hydrodynamic Modeling (HEC-RAS): A 2D shallow water equations (2D-SWE) model was implemented to simulate flood propagation through the urban and peri-urban areas of Karditsa. This model was chosen for its ability to handle complex terrain and flow dynamics.
- Data Integration and Validation: The core innovation lies in the integration of diverse datasets for both model input and validation. Remote sensing data (NASA GPM-IMREG, SENTINEL) provided rainfall estimates and flood extent information, while crowdsourced data (social media videos and photos) offered crucial insights into flood evolution and water depths.
Datasets
The study utilized a comprehensive set of datasets:
- Rainfall Data:
- Remote Sensing: NASA GPM-IMREG (Early, Late, Final products), MSWEP, CMORPH, PERSIANN-CCS.
- Reanalysis: ERA5 land (ECMWF), MERRA land (NASA).
- Gauged: Plastiras dam (15-min records), Karditsa station (hourly records).
- Flood Extent and Depth Data:
- Remote Sensing: Copernicus EMS-Mapping products (SENTINEL-1 and SENTINEL-2 images from 20 and 24 September 2020).
- Crowdsourced: Georeferenced photos and videos posted on social media (Facebook) from around 12 hours after the peak flow.
- Topographic Data: A 5x5m resolution Digital Elevation Model (DEM) generated from aerial images.
- Land Cover Data: CORINE land cover data.
Key Insights and Ideas
- The Power of "Forensic Hydrology": This study demonstrates that even in data-scarce environments, a rigorous "forensic" approach, meticulously piecing together evidence from various sources, can yield reliable reconstructions of extreme flood events.
- Social Media as a Treasure Trove: Crowdsourced data from social media proved invaluable for validating flood extent and estimating water depths, especially when traditional monitoring systems were absent or compromised. This highlights the growing importance of social media in disaster response and research.
- Remote Sensing Limitations and Strengths: While remote sensing rainfall products generally underestimated the magnitude of the IANOS event compared to ground gauges, they provided crucial spatial and temporal information about the phenomenon's evolution, making them a valuable asset for understanding event dynamics.
- Coupled Modeling is Key: The successful integration of hydrological (HEC-HMS) and hydrodynamic (HEC-RAS) models was essential for capturing the complex flood processes, from rainfall-runoff generation to urban inundation.
- Extreme Event Probabilities: The study provides compelling estimates for the return period of the IANOS flood, suggesting it was a 1:400 to 1:1000 year event, underscoring its catastrophic nature.
Key Results
- Flood Extent Validation: The simulated flood extent showed satisfactory agreement with observed flood extents from SENTINEL satellite images, capturing the main flooding pathways.
- Water Depth Estimation: Modeled water depths generally fell within the range of estimated observed depths derived from crowdsourced data, with some instances of slight overestimation by the model.
- Return Period Estimates:
- Flows were estimated to be 1:400 years for low-lying catchments and 1:1000 years for upstream mountainous catchments.
- Flood volumes were estimated to be 1:1000 years for low-lying areas and up to 1:10,000 years for upstream catchments, highlighting the extreme nature of the event.
- Maximum Water Depths: Simulated maximum water depths reached approximately 5 meters in the main river channel and about 1 meter in the floodplain.
Limitations and Future Work
- Remote Sensing Rainfall Underestimation: The study noted that remote sensing rainfall products generally underestimated the event's magnitude compared to ground-based measurements. This suggests a need for further research into improving the accuracy of these products, especially for extreme events.
- DEM Resolution: While a 5x5m DEM was used, the authors acknowledge that DEM resolution can impact hydrodynamic modeling accuracy, particularly in areas with complex topography or hydraulic structures.
- Uncertainty in Crowdsourced Data: The depth estimations from photos and videos inherently carry uncertainty due to expert judgment and the difficulty in pinpointing exact locations.
- Future Work: The authors emphasize the need for continued expansion of ground meteorological networks and suggest that probabilistic flood mapping approaches and hybrid-stochastic modeling frameworks should be explored for future flood risk assessments, especially in complex scenarios. They also highlight the potential of integrating more advanced crowdsourced data analysis techniques.
Summary updated