Flood Extent Analysis Using SAR and Google Earth Engine

Case Study: Joso City Flooding Event, September 2015

Authors
Affiliations

Waruth POJSILAPACHAI

English Engineering Education Program (e3), Hokkaido University

Prof. Tomohito YAMADA

Hokkaido University

Published

November 7, 2025

Research progress.

English Engineering Education Program (e3)

Hokkaido University

Introduction

Background: The 2015 Joso City Flooding Event

In September 2015, Joso City, located in Ibaraki Prefecture, Japan, experienced catastrophic flooding caused by Typhoon Etau. The torrential rainfall caused the Kinugawa River to overflow its banks, inundating large portions of the city and resulting in:

Widespread damage to homes, buildings, and infrastructure Severe agricultural losses, particularly to rice fields Mass evacuations and emergency rescue operations International attention to disaster response strategies This analysis employs Synthetic Aperture Radar (SAR) imagery and Google Earth Engine to map and quantify the flood extent, demonstrating modern geospatial techniques for disaster assessment and response.

Objectives

Map the spatial extent of flooding using Sentinel-1 SAR data Implement semi-automated flood detection through change detection analysis Quantify affected areas and assess damage to various land cover types Demonstrate the integration of multiple geospatial datasets for comprehensive hazard mapping

Methodology

Hazard Mapping Framework Hazard mapping involves identifying, assessing, and analyzing various hazards to understand their potential impact on a particular area. Key components include:

Data Collection: Utilizing satellite imagery, elevation models, and historical records

Risk Assessment: Evaluating vulnerability of populations and infrastructure Spatial Analysis: Applying GIS techniques for pattern recognition and modeling Emergency Planning: Supporting mitigation strategies and response protocols This study employs data from multiple authoritative sources including the Geospatial Information Authority of Japan (GSI) and global satellite archives.

Synthetic Aperture Radar (SAR) for Flood Mapping

SAR technology offers unique advantages for flood detection:

Why SAR for Flood Mapping?

All-weather capability: Operates day or night, penetrates cloud cover High spatial resolution: 10m resolution enables detailed mapping Vegetation penetration: Detects water beneath canopy cover Temporal consistency: Insensitive to solar illumination angles

Polarization Selection

This analysis uses VV polarization (vertical transmit, vertical receive) because:

Water surfaces produce strong specular reflection in VV Effective discrimination between flooded and non-flooded areas Optimal for flat water detection in agricultural landscapes Standard availability in Sentinel-1 archive Data and Study Area Global Floodplain Dataset The Global Floodplain dataset provides baseline context for identifying historically flood-prone areas. This 250m resolution dataset integrates:

River discharge models Digital elevation data Historical flood records Land use patterns

Semi-Automated Thresholding of Derived Flood Extent Using GEE-SAR

During the flooding event in September 2015 in Joso City, Japan, mapping and assessment of the flooded areas were likely conducted using aerial imagery, satellite data, or drones. These sources provided high-resolution images of the affected areas, which were then analyzed to delineate the extent of flooding. GIS techniques, such as image classification and change detection, were likely employed to process the imagery and generate flood extent maps. These maps would have provided valuable information for disaster response and management efforts, aiding in the identification of affected areas and the allocation of resources for relief and recovery.

Explore online databases and repositories that specialize in geospatial data and disaster management. Websites like the Japan Platform for Disaster Reduction (JPDR) or the Geospatial Information Authority of Japan (GSI) may provide access to relevant datasets and maps.

Hazard mapping involves identifying, assessing, and analyzing various hazards to understand their potential impact on a particular area. This process includes identifying natural and human-made hazards, assessing the vulnerability of the area and its inhabitants, conducting risk analysis to evaluate potential damages, and collecting spatial data using GIS technology. Modeling and simulation help predict hazard behavior, while emergency preparedness and response planning ensure effective mitigation measures are in place. Community engagement, continuous monitoring, and policy integration are essential for successful hazard mapping, along with education and awareness programs to build resilience within communities.

Global Floodplain Dataset

The Global Floodplain dataset is a comprehensive geospatial resource that maps flood-prone areas worldwide using high-resolution hydrological and topographical data. Developed to support flood risk assessment, environmental planning, and climate resilience strategies, it integrates river discharge models, elevation data, and historical flood records to delineate floodplain extents across diverse terrains. This dataset enables researchers, policymakers, and disaster response teams to identify vulnerable regions, prioritize mitigation efforts, and improve land-use planning. Its global coverage and standardized format make it especially valuable for comparative studies and cross-border water management initiatives.

Figure 1: Global Floodplain and Elevation Analysis of Joso City Study Area

Regional Hazard Mapping in case of Flooding and Inundation

Regional hazard mapping for flooding and inundation plays a crucial role in disaster risk reduction and management. By identifying flood-prone areas and assessing their vulnerability, authorities can develop effective mitigation strategies and emergency response plans. Hazard maps provide valuable information for land-use planning, infrastructure development, and policy-making to minimize the impact of flooding on communities, infrastructure, and the environment. Additionally, regional hazard mapping helps raise awareness among residents about flood risks, promotes community preparedness and resilience, and facilitates early warning systems to mitigate the loss of life and property during flood events. Overall, regional hazard mapping is essential for enhancing the resilience of communities and reducing the socio-economic impact of flooding and inundation.

Synthetic Aperture Radar (SAR)

Synthetic Aperture Radar (SAR) images are advantageous for flood mapping due to their all-weather capability, high spatial resolution, and ability to penetrate vegetation cover. SAR sensors can operate day or night and penetrate cloud cover, enabling continuous monitoring even in adverse weather conditions. With their spatial resolution, SAR images provide detailed information on flood extent and dynamics, allowing for accurate delineation of flooded areas. Moreover, SAR’s ability to penetrate vegetation allows for the detection of submerged areas beneath dense canopy cover. Multi-temporal analysis of SAR data enables the tracking of flood progression over time, while the insensitivity to illumination angle ensures consistent results regardless of solar geometry. These characteristics, combined with operational flexibility and broad coverage, make SAR imagery a valuable tool for regional hazard mapping and flood management.

Polarization

Polarization in Synthetic Aperture Radar (SAR) plays a crucial role in detecting floodwater by influencing how radar waves interact with different surfaces. SAR systems typically transmit and receive radar signals in two orthogonal polarizations: horizontal (H) and vertical (V). When SAR signals encounter water, they experience significant changes in polarization due to the smooth and specular nature of water surfaces. The backscattering properties of water vary depending on the polarization of the incident radar waves and the incidence angle. For instance, in HH polarization, where both the transmitted and received signals are horizontally polarized, the radar waves are more likely to penetrate the water surface and interact with underlying terrain or objects. In contrast, in VV polarization, where both signals are vertically polarized, the radar waves are more likely to be reflected off the water surface. By analyzing the differences in backscatter between different polarization channels, SAR data can effectively discriminate between flooded and non-flooded areas, enabling accurate flood mapping and monitoring.

The choice of polarization in SAR for enabling accurate flood mapping and monitoring depends on various factors, including the characteristics of the floodwater and the surrounding terrain. However, in many cases, the cross-polarization, specifically the combination of H transmit polarization with V receive polarization (referred to as HV polarization), is often preferred for flood mapping. HV polarization offers a good balance between penetration capability and sensitivity to surface roughness, making it suitable for distinguishing flooded areas from other land cover types. This polarization configuration can effectively detect the presence of water by exploiting the unique back-scattering characteristics of water surfaces, which exhibit strong reflection in VV polarization and reduced reflection in HV polarization due to the specular nature of water. Additionally, HV polarization can provide valuable information about flood extent and dynamics while minimizing interference from other surface features. Ultimately, the choice of polarization should be tailored to the specific requirements and characteristics of the flood mapping application.

Providing R connect to GEE

Google Earth Engine (GEE) is a cloud-based platform developed by Google that offers a comprehensive suite of tools and datasets for analyzing large-scale geospatial data. Leveraging Google’s cloud infrastructure, GEE enables users to access and process a vast array of satellite imagery, climate data, and terrain data for various applications, including environmental monitoring, land cover mapping, and agriculture. With its extensive collection of analysis algorithms and machine learning integration, GEE provides users with the capability to perform advanced geospatial analysis tasks, such as image classification, change detection, and time series analysis, at scale. The platform also offers interactive visualization tools for exploring and sharing geospatial insights, making it a valuable resource for researchers, scientists, educators, and developers in understanding and addressing complex Earth-related challenges.

Using rgee package is an R interface for interacting with Google Earth Engine (GEE) from within the R environment. Google Earth Engine is a cloud-based platform for planetary-scale environmental data analysis, providing access to a vast amount of remote sensing and geospatial data and tools for analysis. The rgee package allows R users to access and utilize the capabilities of GEE directly within R, enabling them to perform geospatial analysis, apply remote sensing algorithms, and visualize results using familiar R syntax and functions.

Loading a shapefile from GEE assets. There is a shapefile of Joso City in Japan.

Joso City is located in Ibaraki Prefecture, Japan. It is situated in the northern part of the Kanto region on the eastern coast of Honshu, the largest island of Japan. Joso City is known for its agricultural industry, particularly rice cultivation, and it is home to various historical and cultural landmarks. However, it gained international attention in September 2015 when it was severely affected by flooding caused by heavy rainfall brought about by Typhoon Etau. The flooding resulted in significant damage to infrastructure, homes, and agricultural fields in the area.

The torrential rain caused the Kinugawa River to overflow its banks, inundating large areas of the city with floodwaters in 2015. The flooding resulted in widespread damage to homes, buildings, roads, and agricultural fields. Many residents were forced to evacuate, and there were reports of stranded individuals requiring rescue by emergency services. The disaster prompted a large-scale response effort, including search and rescue operations, as well as relief and recovery initiatives to assist affected residents and rebuild the affected areas.

Calculated Threshold: 2.696306 
Polling for task <id: 3EKYZTGI556QLZVG7H4GZEMR, time: 0s>.
Polling for task <id: 3EKYZTGI556QLZVG7H4GZEMR, time: 5s>.
Polling for task <id: 3EKYZTGI556QLZVG7H4GZEMR, time: 10s>.
Polling for task <id: 3EKYZTGI556QLZVG7H4GZEMR, time: 15s>.
Polling for task <id: 3EKYZTGI556QLZVG7H4GZEMR, time: 20s>.
State: RUNNING

Fetching histogram data for before flooding image...

Structure of histogram data:
[1] "VV"

Plotting histogram for BEFORE flooding...
Number of bins: 30 
Backscatter range: -28.03821 to 28.49688 dB


Fetching histogram data for after flooding image...

Plotting histogram for AFTER flooding...
Number of bins: 30 
Backscatter range: -26.19172 to 30.9297 dB


Fetching histogram data for difference image...

Plotting histogram for DIFFERENCE image...
Number of bins: 12 
Difference range: -12.26014 to 8.974965 dB


Creating combined comparison plot...


Histogram analysis complete!

FLOOD DETECTION AND AREA CALCULATION


=== Applying threshold to detect flooded areas ===
=== Masking permanent water bodies ===

#Discussion

Key Findings

  1. Automated Detection: The statistical threshold approach successfully identified flood extent without manual digitization
  2. Spatial Patterns: Flooding concentrated in low-elevation agricultural areas adjacent to the Kinugawa River
  3. Data Integration: Combining SAR, elevation, and historical water data provides robust flood mapping

Methodological Advantages

  1. Reproducibility: Fully automated workflow applicable to other regions
  2. Timeliness: Near-real-time processing capability for emergency response
  3. Multi-temporal: Change detection approach minimizes classification errors
  4. Cloud-independent: SAR penetrates cloud cover during storm events

Limitations and Considerations

  1. Speckle noise: Requires filtering that may reduce spatial detail
  2. Urban areas: Complex backscatter from buildings may affect detection
  3. Vegetation: Dense canopy can partially obscure water signals
  4. Temporal resolution: Sentinel-1 revisit time may miss rapid flood dynamics

Conclusions

This study demonstrates an effective workflow for rapid flood extent mapping using Sentinel-1 SAR data and Google Earth Engine. The semi-automated approach successfully detected r format(round(floodAreaHa, 2), big.mark=“,”) hectares of flooding in Joso City during the September 2015 event.

Applications:

  1. Emergency response and resource allocation
  2. Damage assessment for insurance and recovery planning
  3. Validation of hydraulic models
  4. Long-term hazard mapping and land use planning

Future Work

  1. Integration with land cover data for damage quantification
  2. Population exposure analysis
  3. Comparison with official flood surveys
  4. Adaptation for real-time flood monitoring systems

Data Availability

##Satellite Data:

  1. Sentinel-1 SAR: Copernicus Programme (free and open access)
  2. JRC Global Surface Water: European Commission Joint Research Centre
  3. ALOS DEM: Japan Aerospace Exploration Agency

Code Repository: xxxx

References

  1. Pekel, J.F., et al. (2016). High-resolution mapping of global surface water and its long-term changes. Nature, 540, 418-422.
  2. ESA Sentinel-1 Mission: https://sentinel.esa.int/web/sentinel/missions/sentinel-1
  3. Google Earth Engine: https://earthengine.google.com/
  4. Geospatial Information Authority of Japan: https://www.gsi.go.jp/ENGLISH/