Flood inundation amplified by large-scale ground subsidence funnel under the ongoing global climate change

Abstract

Global climate change is leading to more severe rainfalls, greatly exacerbating the inundation risk of large-scale subsidence funnels worldwide. Here, we present a case of flood disaster in a large ground subsidence funnel at the Liaohe plain of China in 2022, triggered by continuous heavy rainfall under the ongoing global climate change. We conducted a retrospective investigation and analysis of the disaster event and its impacts through the comprehensive utilization of satellite remote sensing, numerical simulation, and risk assessment techniques. We observed the scale of the subsidence funnel for nearly 50 years, quantitatively analyzed the impact of subsidence on the flood disaster, and assessed the flood disaster risk. The results indicated that the ground subsidence has partially compromised the flood protection capacity of rivers, elevating flood risk in this region by 28.24%. Additionally, the subsidence has also damaged the flood control dam, leading to a breach and a secondary flooding disaster. Our work demonstrates that it’s possible to characterize the inundation risk of large-scale subsidence funnels by spaceborne remote sensing in advance. This approach could enhance the capacity to cope with extreme weather events in similar ground subsidence regions worldwide.

Introduction

Due to global climate change, extreme weather is becoming more frequent, and an increasing number of extreme hydrological events are rapidly becoming a major threat to the development of contemporary human society^1,2,3,4,5. Floods, as one of the most common and destructive natural disasters^6,7,8, have drawn significant attention due to their devastating impacts^9,10. In this century, climate change will have a severe impact on the world’s water resources through the rise in sea levels, melting of glaciers, more frequent and severe floods, droughts, and snow disasters. Predictions suggest that, by 2035, the frequency of extreme floods in Europe, with a recurrence interval exceeding 100 years, will double on average, with further increases projected by the end of the century^8,11,12. Additionally, population-based assessments of flood-related losses indicate that, by 2050, the global risk of river and coastal flooding will have doubled compared to 2010¹³. Floods account for over $40 billion in losses worldwide each year, affecting over 250 million people globally^6,14. Among them, China is one of the countries frequently affected by major flood disasters^15,16,17,18. The floods that occurred in 1991, 1994, 1996, 1998, 2016, and 2020 represented losses accounting for 3% to 4% of China’s Gross Domestic Product (GDP)¹⁹. In 2024, heavy rainfall occurred in many regions of southern China. For instance, the flood disaster in Pingyuan County, Guangdong Province, claimed 38 deaths and 2 people missing. In Huarong County, Hunan Province, a major breach occurred at the Tuanzhou dike of Dongting Lake, leading to the evacuation of more than 5,000 people, posing a serious threat to life safety and socio-economic development.

Beyond climate change, existing studies have demonstrated that large-scale land surface degradation caused by both anthropogenic and natural factors—such as ground subsidence—can amplify the impact of extreme hydrological events^{20,21,22,23,24}. Although subsidence is a gradual process, it can permanently diminish the water storage capacity of aquifer systems and elevate flooding risks in alluvial basins or coastal plains. Even minor subsidence, on the order of a few centimeters, can significantly increase the likelihood of flooding in these regions²⁵. In China, by 2021, ground subsidence had affected over 8% of the country’s land area, approximately 790,000 km^2 26. A recent study using satellite-based InSAR examined ground subsidence (2015–2022) in 82 major Chinese cities, including those with over 2 million people, all provincial capitals, and key industrial cities. It found that 45% of the land is subsiding faster than 3 mm/year, with 16% exceeding 10 mm/year. This affects 29% and 7% of the urban population, respectively²⁷. In Iran, satellite data from 2014 to 2020 indicate that approximately 40,000 km² of land is subsiding at rates exceeding 1 cm/year, with more than 25,000 km² subsiding faster than 2 cm/year, and nearly 3000 km² sinking at rates surpassing 10 cm/year²⁸. Similar patterns are evident in India’s Delhi region, where between 2014 and 2019, continuous deformation affected around 11 km², with the Kapashera area experiencing peak subsidence rates of over 17 cm/year during 2018–2019²⁹. In Afghanistan, an assessment from 2015 to 2022 identified multiple regions with severe subsidence, exhibiting annual rates greater than 5 cm/year³⁰. In Mexico City, by 2021, 9% of urbanized areas—roughly 70.7 km²—were exposed to high or very high land subsidence risk, impacting approximately 12.6% of the city’s population³¹. Beyond these regions, subsidence is also prevalent in coastal regions globally, including Hawaii³² and the Mekong Delta³³, contributing to accelerated relative sea level rise (SLR) and its associated impacts^34,35. In O’ahu’s South Shore, localized subsidence is occurring at a much faster rate than global sea level rise. By 2050, flood exposure in the Mapunapuna industrial district may increase by 53%, with subsidence compressing flood preparedness timelines by up to 50 years³². In Northern Jakarta, land subsidence is projected to account for 88% of the 110.5 km² increase in coastal flood extent by 2050, far outweighing the contribution of global SLR^36,37.

This widespread and accelerating phenomenon of land subsidence significantly alters local topography and hydrology, fundamentally reshaping flood hazards in vulnerable regions. The combined pressures of global climate change and this ongoing subsidence dramatically increase the exposure of populations and infrastructure to flooding. Consequently, accurate and reliable flood simulation in subsidence-prone areas has become increasingly critical for effective risk assessment and management. However, previous studies have primarily relied on models to simulate the relationship between flood inundation and ground subsidence, often without validating these models against real-world events. This discrepancy between model requirements and available data is a key reason why flood simulation models are not yet widely adopted by decision-makers in flood management³⁸. To address this issue, improving of modeling algorithm and validation using representative cases can significantly enhance the reliability of flood simulation tools³⁹. In practice, evaluating simulation results against observed runoff data and actual flood events provides a robust means of assessing model accuracy, offering key insights for advancing flood-related research and practical applications.

This study reports on a flood disaster in a large ground subsidence funnel, triggered by continuous heavy rainfall amid ongoing global climate change. From July to August 2022, the Liaohe Plain in Liaoning Province, China, experienced intense rainfall, leading to substantial water accumulation in the subsidence funnels. This triggered flooding that damaged oil extraction infrastructure and precipitated widespread flood disasters in the area. The situation worsened when a flood control dam in the subsidence zone breached, inundating large swathes of farmland and leading to significant economic losses. This event, a typical example of extreme rainfall in the context of global climate change, was further exacerbated by large-scale ground subsidence, which serves as an important warning for other similar regions worldwide.

Study region

The Liaohe Plain in China is rich in oil and gas resources and hosts the country’s fourth-largest oilfield⁴⁰. Within this oilfield, the Shuguang Oil Production Plant is the largest oil extraction factory, serving as a key center for regional energy production. Due to the direct impacts of oil extraction, severe ground subsidence has occurred in this region⁴¹. Moreover, the study area is situated within the Liaohe River Delta, where the water system is complex, including the Liaohe, Raoyang, Daling, and Xiaoling River systems. The main river channels extend over a total length of 326 km, draining an area of approximately 10,400 km². The Raoyang River which is a tributary of the Liaohe River flows through the study area, and there are more than 2800 oil wells densely distributed on both banks of the Raoyang River, as shown in Fig. 1. This region has been continuously conducted underground oil and gas resource extraction since 1975. As a result, large areas of underground oil and gas reservoirs have been emptied, and the effective stress in the rock stratum produces compression and deformation and thus manifests itself as ground subsidence. After nearly 50 years of continuous subsidence, a large subsidence funnel has been formed, covering an area of about 44 km². This subsidence area has become an increasingly severe geological disaster region within the Liaohe Plain⁴¹.

Fig. 1: Location map of the study region.

The red dots represent oil wells, the blue pentagon indicates the position of dam breach, blue lines represent rivers, red lines represent roads, the yellow line indicates the dam, the area enclosed by white lines is the subsidence region, and the light blue shaded area represents the inundation region of the secondary flood disaster triggered by the dam breach.

Since the rainy season began in July 2022, Liaoning Province has experienced extremely heavy rainfall. The large-scale ground subsidence funnel at the Shuguang Oil Production Plant in Panjin City has caused the local area to become low-lying. In addition, the Raoyang River meanders through this region. These factors together have contributed to one of the most severe flood disasters in Liaoning Province. The subsidence region has converged a large amount of floodwater due to the extremely heavy rainfall, coupled with the destruction of the flood dam due to ground subsidence, leading to the breach of the dam and the occurrence of a secondary disaster. More than 8100 people were evacuated due to the dam breach, devastating important infrastructures such as oil wells of the Liaohe Oilfield, villages, and surrounding farmland. Figure 1 shows the areal extent of the subsidence funnel, the distribution of oil wells, and the inundation region of the secondary flood disaster triggered by the dam breach.

Ground subsidence monitoring and analysis

Long-term ground subsidence monitoring using multi-source data

In order to reconstruct long-term surface subsidence in the study region, we utilized multi-source data to acquire the deformation results from 1977 to 2022 since the exploitation of oil and gas resources, as shown in Fig. 2. We used time-series InSAR technique to obtain subsidence rates from 2007 to 2011 and from 2015 to 2022, but lacked satellite data from 2011 to 2015. In addition, the leveling measurements, GNSS CORS (continuously operating reference station) monitoring, and digital elevation model data were also utilized to obtain the ground subsidence results in the study area. Existing monitoring results indicate that the deformation trend in the study region exhibits a roughly linear trend (Fig. S1). Therefore, an average subsidence rate was calculated for each grid cell, and linear interpolation was employed to estimate deformation during years without direct observations⁴². By integrating these interpolated data with multi-source obervation datasets, we reconstructed a long-term surface subsidence over the past 50 years from 1977 to 2022, as depicted in Fig. 3. Based on this fusion processing, the final 3D topography of this region was generated, as shown in Fig. 4. The large-scale subsidence funnel in the study area is situated within the complex water systems of the Raoyang River, where the river channel is both curved and characterized by a gentle gradient. During heavy rainfall, the river water level rises rapidly, significantly increasing flood control pressures and heightening the risk of flood inundation⁴³.

Fig. 2: Long-term ground subsidence monitoring using multi-source data.

The leveling measurement data were obtained from 1977 to 1985, 1982 to 1994, 2013 to 2014, and 2020 to 2022. The terrain change detection data were obtained from 1977 to 2007. The GNSS CORS monitoring data were obtained from 2019 to 2022. The ALOS data were obtained from 2007 to 2011, processed using PS-InSAR, and the Sentinel-1 data were obtained from 2015 to 2022, processed using SBAS-InSAR. The “Ascending” of orange represents data from the ascending orbit of the SAR satellite. The “Descending” of green represents data from the descending orbit of the SAR satellite. The “All” of blue indicates SAR data from both ascending and descending orbits.

Fig. 3: Topography of the study region after the addition of the subsidence funnel.

All other features remain consistent with those presented in Fig. 1.

Fig. 4: The 3D topography of the study region with superimposed subsidence funnel.

The thick blue line indicates the Liao River, while the thin blue line denotes the Raoyang River.

Subsidence damage to dam

The perennial underground extraction of oil and gas resources in the study area has significantly altered the original local topography and undermined the flood control system within this region. Under the combined influence of extreme weather and ground subsidence, on August 1, 2022, a serious leakage occurred in a section of the dam on the left bank of the Raoyang River within the subsidence area. As the leakage site developed into the breach, the width of the breach expanded from several meters to about fifty meters, as shown in Fig. 5c. The floodwater continuously gushed out from the left bank of the dam, flooding villages, farmland, and oilfields, leading to a severe secondary flood disaster.

Fig. 5: The location of selected points on the dam for analysis of dam subsidence, time series subsidence of selected points on the dam, and photograph of the dam breach.

a the location of selected points on the dam for time series InSAR analysis with the red triangles from top to bottom labeled from A to O, and the blue pentagon indicates the breach location; b time series subsidence for all the selected points on the dam from June 6, 2020, to June 20, 2022, with the black pentagon (★) representing the breach location; c photograph of the dam breach.

There are several key factors probably contributing to the dam breach event, including 1) severe ground subsidence, which compromises the safety and stability of the dam, heightening the risk of structural failure; 2) the region’s low-lying terrain and restricted downstream outflow, which elevate floodwater levels and increase inundation depths; and 3) the presence of numerous pipelines passing through the dam without anti-seepage treatment. Prolonged flooding significantly raises the likelihood of seepage damage at these pipeline locations⁴³. In order to analyze the causes of dam leakage, we select 16 points (points A ~ O and the “Breach” point labeled with a blue pentagon is the location where the breach occurred) on the dam and analyzed their time series InSAR results from June 2020 to June 2022, as shown in Fig. 5a. The time series InSAR results shown in Fig. 5b indicate that these 16 points have experienced different degrees of subsidence from June 2020 to June 2022. The subsidence magnitude that appeared at the breach point is −0.309 m, and the subsidence rate is −154.5 mm/yr. The breach point is located where the dam has the maximum subsidence and the fastest subsidence rate. The different subsidence rates have created varying subsidence gradients along the entire dam. Long-term uneven subsidence has led to a reduction in the dam’s height, weakening its protective capacity. At the same time, severe uneven deformation can also trigger cracks in the dam, increasing the risk of dam breach⁴⁴. In addition, since July 2022, the dam has been under the influence of low-lying topography of the subsidence funnel, inducing a long-term state of soaking by rainwater. Soaking will dissolve and weaken the chemical composition of the soil, leading to the loss of soil suction, thickening of the water film, and will lead to deterioration of soil mechanical properties, thereby weakening the stability of the dam⁴⁵.

Flood inundation simulation and analysis

In order to dynamically reconstruct the process of the flood inundation, we collected multi-source satellite remote sensing images from July to August 2022, including Gaofen-3 satellite (GF-3) with 10 m resolution, and Gaofen-1 satellite (GF-1) with 16 m resolution. By using the Dual-Polarized Water Index (SDWI) and Normalized Difference Water Index (NDWI) to extract the flood inundation area, we obtained the dynamic changes of the flood in the Liaohe Plain from July 1 to August 28, 2022, as shown in Fig. 6.

Fig. 6

The dynamics process of the flood in the study area obtained from multi-source satellite remote sensing images from July 1, 2022, to August 28, 2022, where the red border represents the boundary of the subsidence funnel.

As shown in Fig. 6, the first stage of the flood disaster in July was mainly caused by extremely heavy rainfall events in the Raoyang River Basin⁴³. This rainfall led to a significant increase in river flow. As a result, multiple monitoring stations recorded water levels exceeding the warning threshold for an extended period. These conditions ultimately gave rise to a flood inundation disaster. The second stage began on August 1, with the breach of the flood control dam, triggering the floodwaters to inundate a large area of surrounding farmland. From the satellite images on August 2 and 4, it can be seen that the relative height around the study area formed a natural dam. The inundation area generated by the entire flood disaster began to gradually decrease from August 6, as the breach was successfully closed by the government. However, the inundation area in the subsidence funnel area has not seen a significant reduction.

To evaluate the role of ground subsidence in amplifying flood impacts, high-resolution satellite imagery was used to extract the inundation boundary, serving as the reference for hydrodynamic modeling. The HEC-RAS model was applied to simulate this flood event under four different scenarios, considering that the simulation results were strongly influenced by topography, subsidence funnel, and dam breach. These scenarios use the high-precision DEM data from 2007 or the 2007 DEM data overlaid with the subsidence funnel (that is the subsidence state of this study region in July 2022) in the region. The four scenarios are 1) the inundation area based on the 2007 DEM data, assuming no subsequent ground subsidence in the area and no dam breach occurs; 2) the inundation area based on the 2007 DEM data, assuming no subsequent ground subsidence in the area, but the dam breach occurs, leading to the secondary flood disaster; 3) the inundation area based on the 2007 DEM data adding the subsidence funnel, but no dam breach occurred; and 4) the inundation area based on the 2007 DEM data adding the subsidence funnel, and the dam breach occurs, leading to the secondary flood disaster, respectively. By comparing the inundation areal extents across all four scenarios, the influence of ground subsidence on flood behavior was quantitatively assessed.

As illustrated in Fig. 7, simulations across four scenarios demonstrate the significant amplifying effect of subsidence funnels on flood disasters. Table S1 summarizes the flood inundation area for each scenario. In the absence of substantial subsidence, a dam breach increases the inundation area by 21.94%. Incorporating the 2022 subsidence funnel, the inundation area expands by 9.79% due to terrain changes alone, and by 28.24% when combined with dam breach. Without pronounced subsidence, floodwaters would gradually recede, substantially mitigating disaster impacts. However, the presence of a large subsidence funnel has fundamentally altered flood dynamics, transforming the region into a low-lying basin. Furthermore, the river’s deep, narrow, and highly sinuous channel restricts water outflow, promoting stagnation in adjacent low-lying areas, thereby exacerbating flood risk⁴³. These findings confirm that large-scale ground subsidence significantly intensifies flood hazards, a concern that is increasingly critical under ongoing global climate change.

Fig. 7: The simulation results of flood inundation under four different scenarios, along with a comparative analysis of the inundation areas.

Left: Simulation results of flood inundation for four different scenarios: a represents the inundation area based on the 2007 DEM data, assuming no subsequent ground subsidence in the area and no dam breach occurs; b represents the inundation area based on the 2007 DEM data, assuming no subsequent ground subsidence in the area, but the dam breach occurs, leading to the secondary flood disaster; c represents the inundation area based on the 2007 DEM data adding the subsidence funnel, that is the subsidence state of this study region in July 2022, but no dam breach occurred; d represents the inundation area based on the 2007 DEM data adding the subsidence funnel, that is the subsidence state of this study region in July 2022, and the dam breach occurs, leading to the secondary flood disaster. Right: Comparison of the flood inundation areas and corresponding growth rates under the four scenarios of the left Figure. We compare the inundation areas in five different pair cases, i.e. (a vs b), (a vs c), (a vs d), (b vs d), and (c vs d).

Risk assessment

Accurately assessing the risk of flood disasters is essential for disaster prevention, as it is crucial for enhancing the capacity to withstand such disasters and mitigate the associated losses. We integrate the components of hazard, exposure, and vulnerability according to their interrelationships, and provide a comprehensive evaluation of regional flood disaster risk from the perspective of the interaction of multiple factors^46,47. Firstly, based on past experiences, the occurrence of flood disasters usually did not have clear boundaries. Therefore, nine indicators including rainfall, terrain wetness index, river network density, soil erodibility factor (K), population density, GDP, DEM, slope, and NDVI were selected to establish Model-1(Fig. S2). During this flood disaster, the surrounding inundation areas were constrained by continuous relative height, such as roads and flood dams, which not only impeded the spread of the flood but also provided protection to adjacent regions. To better align the risk assessment with actual conditions, we incorporated the relative height indicator into Model-2, yielding enhanced performance compared to Model-1.

Among the three indicators of the hazard of the disaster-inducing factors—rainfall, river network density, and terrain wetness index—rainfall data plays a dominant role. In the Raoyang River basin, a tributary of the Liaohe River, three major rainstorm events occurred between June and July. The total rainfall during this period averaged 541.0 mm, which is significantly higher than the annual average of 234.1 mm and represents the highest recorded amount for this timeframe. Notably, from July 28 to 30, the average rainfall across the area was 112.0 mm. These high-intensity rainfalls occurred rapidly, significantly impacting the rivers surrounding the study area. Coupled with the higher river network density in the region, the rivers converge more quickly, leading to a sharper process of flood formation. However, the relative height surrounding prevents rapid drainage of the basin, trapping a large amount of floodwater being unable to drain out of the study region. At the same time, the higher terrain wetness index indicates that this study area has a larger catchment capability and gentler slopes, increasing the likelihood of flood disaster occurrence.

Among the five indicators of the exposure of the disaster-inducing environment (NDVI, soil erodibility factor (K), DEM, slope, and relative height), the NDVI in the study area is relatively low. This indicates low vegetation coverage and a lack of surface protection from vegetation, which results in weak soil erosion resistance. The lower soil erodibility factor (K) also suggests that the soil in this region is easily eroded. These two indicators indicate that the study region has poor water retention capacity, making it more likely to form surface runoff and increase the risk of flood disasters. Moreover, from the DEM and slope data of this study region, it can be seen that there is a large-scale subsidence funnel with significantly lower elevation than the surrounding regions. The substantial ground subsidence leads to significant elevation loss, which alters the gradient, affecting both the direction and velocity of the river’s flow. Additionally, the narrow, winding channel of the Raoyang River slows flood discharge and promotes water retention within the subsidence funnel area. This elevation loss also diminishes the original flood control capacity designed for the river’s dam. The distribution of relative height around the study region hinders the spread of floodwaters, exacerbating the impact of flood disasters. In terms of the two indicators of the vulnerability of disaster-bearing body (GDP and population density), the study region, with a large number of oil wells, has a higher GDP, but the population density is relatively low. Therefore, the economic losses attributable to flood disasters are significant, but there are no casualties.

The flood disaster risk index for the study region can be calculated by inputting the nine indicators represented by flood disaster hazard, exposure, and vulnerability, along with their calculated weights, into the risk assessment Model-1. As shown in the left figure of Fig. 8, the results indicate that if there is no relative height surrounding the study region, the high risk regions are mainly distributed in three regions, labeled as a, b, and c in Fig. 8. Region a is characterized by subsidence funnels; Region b is located in the north upstream of the Raoyang River; Region c is in the northeastern upstream of the Raoyang River. Based on the performance of the nine indicators within the study region, it is evident that the region a faces a high flood risk. This risk is mainly attributed to several factors: high rainfall, a dense river network, and the low-lying topography of the subsidence funnel. In addition, the presence of gentle slopes with converging tributaries, land use dominated by cultivated land, low vegetation coverage, and weak soil erosion resistance further exacerbate the flood risk. Regions b and c in the northern upstream and northeastern upstream of the Raoyang River are prone to flood disasters due to the presence of the mainstream, with the river cross-section areas showing low and flat terrain, gentle slopes, and a lack of vegetation protection.

Fig. 8: Risk assessment results of Model-1 and Model-2.

Left: risk evaluation map generated by the risk evaluation Model-1, with a, b, and c indicating high-risk regions; Right: risk evaluation map generated by the risk evaluation Model-2 with the addition of the indicator of relative height compared to the risk evaluation Model-1.

Due to the relative height within the study region significantly affecting the spread of floods, therefore the indicator of relative height was added to the risk assessment Model-1 to update the risk assessment Model-2. As can be seen from the right figure of Fig. 8, the high risk regions remain in the subsidence funnel area and the upstream of the two tributaries. The relative height impedes the spread of floodwaters, thereby increasing the flood risk in its own region. Apart from the urban area on the right, which is assessed as generally high risk due to its higher vulnerability, both the high risk and generally high risk regions are within the range of relative height. In addition, the risk is classified as high, generally high, and medium based on other indicators within the range of relative height intersected by rivers. Moreover, there are regions of relative height without rivers running through but are close to high risk regions, thus exhibiting medium risk and generally high risk, which corresponds to the secondary flood disaster region in this study.

Discussion

After an in-depth exploration of the various factors affecting the risk of flood disasters, we next focus on the amplifying effect of ground subsidence on flood disasters. This study region, as the largest oil extraction plant in the entire Liaohe Plain, cannot ignore the impact of ground subsidence stemming from long-term oil extraction activities on the flood disaster. The distribution of oil wells in the study region is dense, and the continuous extraction has been ongoing with annual crude oil production at about 10 million tons. The extraction areal extent is gradually expanding, and the area and amount of ground subsidence also gradually increase over time. Meanwhile, due to the deep burial of oil and gas reservoirs in the study region and the vast area of mined-out regions, the region primarily experiences slow and progressive subsidence, characterized by large-scale, wide areal extent, and significant hazards. Additionally, a large area within the subsidence region is planted with crops such as rice and corn, which require a massive amount of water during their growth cycles. Long term groundwater extraction has led to a continuous decline in the underground confined water level in the subsidence area, thereby exacerbating ground subsidence⁴⁸.

The results of this study indicate that ground subsidence poses a significant threat to riverine areas, particularly in regions with long-duration and large-scale subsidence. It reduces the effective level of protection provided by the dam, increases maintenance costs, and amplifies the consequences of flood control infrastructure failure. For instance, since 1959, subsidence in Shanghai has required the flood control walls along the Huangpu River to be raised four times, with the total height of the flood control walls in the Bund area increased by 3.4 meters, incurring substantial costs and increasing risks. The Tokyo Bay area experienced substantial ground subsidence (exceeding 4.5 meters) due to excessive groundwater extraction. In response, the Tokyo government began working on the Gentle-Slope Levee Project in 1980 and the Super Levee Improvement Project in 1985. As a result of such endeavours, by the end of 2008, 14.4 km of river dam had already been protected using super-levees⁴⁶. The flood disaster that occurred in the large subsidence funnel triggered by this extreme weather serves as a warning case for other similar regions around the world. This study quantitatively analyzed the impact of ground subsidence on flood disasters based on a real case that occurred under the ongoing global climate change. On one hand, we comprehensively simulated the amplifying effect of ground subsidence on the flood disaster. On the other hand, we deeply analyzed the secondary flood disasters induced by the fissures or cracks in the flood control dam due to significant subsidence. With the continuation of activities such as oil and groundwater extraction, if effective measures are not taken, these issues will become even more severe in the future.

Statistical studies show that encouraging trends in vulnerability to floods and droughts have been observed as a result of effective risk management⁴⁹. Globally, from 1980–1989 to 2007–2016, human and economic vulnerability decreased by about 6.5 times and 5 times, respectively, highlighting the importance of risk management⁵⁰. In this study, the Analytic Hierarchy Process (AHP) was applied to develop two risk assessment models. The analysis began with an evaluation of three key factors: hazard, exposure, and vulnerability. Subsidence-induced topography was assigned the highest weight, while the following indicators were individually weighted: rainfall, river density, terrain wetness index, slope, NDVI, soil erodibility factor, relative height, GDP, and population density. A consistency check was performed to verify the reliability of the weighting process. Ultimately, established two risk assessment models to describe the spatial distribution of flood disaster risk as accurately as possible. The models both consider anthropogenic environmental changes, such as ground subsidence or the construction (or destruction) of dams. Comparison with actual situation has proven this to be a feasible method for flood risk assessment, particularly for basin-scale flood risk assessment in the context of subsidence.

In this study, we performed long-term surface deformation monitoring using multi-source data. Due to the 3-year time gap within the 50-year multi-source observation datasets, we assume a linear subsidence rate between 2011 and 2015, based on the linear deformation pattern observed in the Sentinel-1 data. This assumption is supported by historical subsidence trends in the study area. Although some errors are present, they remain relatively small compared to the total cumulative deformation observed over the whole time period. Future study should focus on large-scale, long-term terrain predictions to more accurately capture subsidence trends. In the flooding mapping, due to the limited number of available satellite images, we used both optical and C-band SAR images for flood mapping. This sometimes resulted in misclassification of dry areas near dams, requiring manual correction based on field investigation²⁸. In flood disaster risk assessment, no single evaluation method fully captures actual conditions⁵¹. This limitation also applies to the approach proposed in this study. However, we highlight the strong consistency between the results from Model-2 and actual scenarios, which can offer valuable insights for assessing flood risks in subsidence-prone areas. Future work will explore the incorporation of regional adaptation indicators to address the unique challenges of different geographical regions.

Conclusion

Given the current situation of global climate change, we analyze the quantitative relationship between large-scale ground subsidence driven by human activities and the risk of flood disasters. We take the severe flood disaster that occurred in Liaoning Province of China in 2022 as an example. By integrating leveling measurement techniques, terrain change detection, GNSS CORS monitoring, SBAS-InSAR, and PS-InSAR technologies, we obtained nearly 50 years of ground subsidence from 1977 to 2022 in the study area, which was used to generate the current topography. The results indicate that the terrain in the study region has been significantly altered due to subsidence, with the lowest elevation reaching −2.2 m, presenting the entire study region as a large subsidence funnel. However, the subsidence not only expanded the areal extent and reduced the elevation of the study region but also compromised the flood dam due to long-term uneven subsidence, thereby increasing the risk of flood disasters.

Simultaneously, the multi-source remote sensing technique was used to obtain the dynamic process of this flood disaster, serving as the basis for the flood disaster simulation. The results demonstrate that when ground subsidence progressed without causing dam breach, the inundation area under 2022 terrain conditions increased by 9.79% compared to 2007. However, when ongoing subsidence led to dam breach, the inundation area expanded significantly, with an increase of 28.24%. In order to mitigate the effects of such flood disasters, we further combine remote sensing technology with risk assessment to evaluate the risk of the study region under two scenarios: one with the presence of relative height and one without surroundings. The results confirm that in nearly flat alluvial basins or coastal plains, relative height (dams, railway, and road) plays a key role in determining the areal extent of flooding. Therefore, in risk assessment Model-2, we introduce the indicator of relative height around the study area with reference to the remote sensing image data at the time of flooding. The resulting risk evaluation aligns closely with the actual flood conditions observed. Therefore, our study is of great significance for flood prevention and control in the Liaohe Plain region, where significant subsidence occurred along the river, and can provide policy recommendations for the government based on the results of risk evaluation, especially in the context of extreme weather.

In summary, our study demonstrates that long-term ground subsidence, by enlarging the subsidence area and compromising the stability of flood control dam, not only elevates flood risk but also intensifies flood impacts. This effect is further amplified by the increasing frequency and severity of extreme weather events driven by ongoing global climate change. The occurrence of the flood disaster in Liaohe Plain further underscores the necessity to globally prioritize the attention on subsidence funnels and conduct timely risk assessments, which could provide a scientific reference for the prevention and mitigation of flood disasters under extreme weather conditions.

Methods

Method 1: monitoring of the subsidence funnel

In order to obtain the long-term time series monitoring results of ground subsidence in the study area, we have integrated terrain change detection technique, leveling measurement technique, and time-series InSAR technique to acquire the deformation results from 1977 to 2022 since the exploitation of oil and gas resources (Table S2). The ground subsidence from 1977 to 2007 was obtained by the terrain change detection technique using two large-scale topographic maps produced in 1977 and 2007, respectively. The ground subsidence from 1977 to 1985, 1982 to 1994, 2013 to 2014, and 2020 to 2022 was obtained by the leveling measurement technique. The ground subsidence from 2007 to 2011 and from 2015 to 2022 was obtained by PS-InSAR and SBAS-InSAR techniques, respectively. At the same time, the ground subsidence from 2019 to 2022 was obtained by GNSS CORS monitoring, and these GNSS monitoring data were also used to validate the accuracy of the InSAR data. The existing monitoring data indicate that the deformation pattern in the study region is roughly linear, which is consistent with previous studies conducted in the same region⁴¹. Although linear interpolation was used to fill the 2011–2015 data gap based on adjacent period trends, this method introduces certain uncertainty since it assumes a constant deformation rate without considering possible short-term variability. The possible sources of error include changes in oil and gas production activities and groundwater extraction. Nevertheless, previous studies and the observed linear trends before and after this period suggest that the associated uncertainty is limited. Therefore, linear interpolation of the average subsidence rate of each grid cell is used to calculate the deformation of vacant years (2011–2015), and combined with multi-source data to reconstruct the long-term (1977–2022) surface subsidence in the study area.

The elevation accuracy of the topographic maps, which are used to detect the change of terrain, is about 0.1 m. The accuracy of the leveling measurement is very high, reaching 0.5 mm. To verify the reliability of the InSAR monitoring results, we compared the quality of the time series InSAR deformation monitoring results with GNSS CORS monitoring data from the same period and found a good consistency between the two methods, with a mean error of 7.9 mm.

Method 2: flood mapping using multi-source remote sensing data

This study integrates the complementary advantages of optical and Synthetic Aperture Radar (SAR) imagery to delineate the spatiotemporal dynamics of surface water bodies using multi-source remote sensing data (Table S2). For optical imagery, the Normalized Difference Water Index (NDWI) is employed to extract water areal extents. For SAR data, the Dual-Polarized Water Index (SDWI) is utilized to identify water areal extents.

• 1) The Normalized Difference Water Index

  The Normalized Difference Water Index (NDWI) was proposed by McFeeters in 1996 and can be used to extract water bodies based on optical images. Water exhibits strong reflectance in the green band and minimal reflectance in the near-infrared band. Calculating the ratio of these two bands effectively enhances the contrast between water and non-water features. This process suppresses background objects and accurately extracts the areal extent of water bodies. The NDWI is calculated as follows:

  $${NDWI}=\frac{{\rho }_{G{{\rm{reen}}}}-{\rho }_{{NIR}}}{{\rho }_{G{{\rm{reen}}}}+{\rho }_{{NIR}}}$$

  (1)

  where ${\rho }_{G{{\rm{reen}}}}$ and$\,{\rho }_{{NIR}}$ represent the radiance values of the green band and the near-infrared band, respectively.

• 2) The Dual-Polarized Water Index

Based on the microwave reflectivity and scattering mechanism of radar imagery, water bodies exhibit low backscattering coefficients in SAR images. The Dual-Polarized Water Index (SDWI) enhances water features and suppresses soil and vegetation features by multiplying dual-polarized data. The study area in this paper is a plain region without mountain shadow effects, where the SDWI performs well for water body extraction. The SDWI is calculated as follows:

$${SDWI}={ln}\left(10* {VV}* {VH}\right)$$

(2)

where ${VV}$ denotes vertical transmission and vertical reception, while ${VH}$ denotes vertical transmission and horizontal reception.

Method 3: simulation of the process of flood inundation

We utilize the HEC-RAS hydrodynamic model to simulate the process of flood inundation. The HEC-RAS model is a hydrodynamic computation program developed by the Hydrological Engineering Center of the United States Army Corps of Engineers. It is suitable for one-dimensional or two-dimensional unsteady flow hydraulic calculations, and can numerically simulate the two-dimensional progression of floods^52,53. The HEC-RAS hydrodynamic model includes the continuity equation and the momentum equation for two-dimensional simulations. The equations are as follow:

The continuity equation:

$$\frac{\partial \rho }{\partial t}+\frac{\partial \left(\rho {\mu }_{i}\right)}{\partial {x}_{i}}=0$$

(3)

The momentum equation:

$$\frac{\partial {\mu }_{i}}{\partial t}+{\mu }_{j}\frac{\partial {\mu }_{i}}{\partial {x}_{i}}={f}_{i}-\frac{1}{P}\frac{\partial P}{\partial {x}_{i}}+\lambda \frac{{\partial }^{2}{\mu }_{i}}{\partial {x}_{j}\partial {x}_{i}}$$

(4)

where $\rho$ is the fluid density in kg/m³; $t$ is the time in s; ${\mu }_{i}$ and$\,{\mu }_{j}$ are the flow velocities at cross-sections ${i}$ and $j$, respectively, in m/s; ${f}_{i}$ is the mass force in m/s²; $P$ is the pressure in N/m²; $\lambda$ is the fluid’s dynamic viscosity coefficient in (N·s)/ m²; ${x}_{i}$ and ${x}_{j}$ are the distances between cross-sections $i$ and $j$ and the next cross-section in m.

The main processes of flood inundation simulation using the HEC-RAS hydrodynamic model are as follow:

• (1) Input the topographic data: the HEC-RAS hydrological model depends on high precision digital topographic data. In this simulation, the topographic data of the study area with added subsidence funnel was obtained based on the Method 1.

• (2) Construct the simulated region: we use the maximum inundation area observed on August 4, 2022, as the inundation boundary for model simulation. In order to improve simulation accuracy, a grid resolution of 30 m × 30 m was used near the river channel, while a grid resolution of 100 m × 100 m was used in other regions. This model includes 59196 calculation units. The inundation region was classified into two zones based on the presence or absence of dam breach. Zone 1, representing the area unaffected by dam breach, comprises 53,218 units. Zone 2 corresponds to the area impacted by the secondary flooding disaster, consisting of 5978 units. To simulate the dam breach scenario, a 52-meter-wide dam breach was introduced in the simulation.

• (3) Setup the boundary conditions: the boundary conditions refer to the physical states or parameters at the interface between a flow system or model and its surroundings, which directly affect flow characteristics and simulation accuracy. The model uses the flood inflow processes from the Dujia water station in Raoyang River and the Panshan water station in Liao River as inputs. The model’s output is the water level process measured at Zhangmingjia station downstream of the confluence of Liao River and Raoyang River⁴³, as shown in Fig. S3.

• (4) Set the roughness parameters: the roughness is a key indicator that describes the resistance experienced by a fluid (such as water flow) when flowing on a solid surface. It reflects the flow obstruction caused by the unevenness or roughness of the solid surface to the fluid. In hydrological and hydraulic modelling, roughness significantly influences flow behavior and distribution. When constructing models, the determination of roughness requires comprehensive consideration of actual land cover types, as shown in Table S3.

• (5) Validate the model accuracy: the highest measured water level at the Yuhegai station in the Raoyang River was 4.93 m, with a discrepancy of just 0.01 m from the model-calculated water level of 4.92 m. Simultaneously, we use CSI (critical success index) to quantitatively analyze the simulation results, and the equation is as follow:

$${CSI}=\frac{{F}_{m}\cap {F}_{o}}{{F}_{m}\cup {F}_{o}}$$

(5)

where ${F}_{m}\cap {F}_{o}$ is the intersection of the modeled and observed flood extent, or number of correct forecasts, and ${F}_{m}\cup {F}_{o}$ is the union of modeled and observed extent. The CSI ranges from 1 (best) to 0 (worst). Due to the occurrence of the dam breach on August 1, 2022, and was successfully closed on August 6, 2022, this study combines remote sensing images from August 2 and 4 to depict the maximum inundation area. The resulting inundation area was then compared with the simulated flood extent for the same period, as shown in Fig. 9. The calculated Critical Success Index (CSI) is 75.54%, suggesting that the simulation results are accurate and reliable⁵⁴. These error values are within the required accuracy standards, indicating that the model parameters are well-configured and appropriate for the simulation tasks in this study.

Fig. 9: The intersect and union areas between simulation result and real scenario.

a The intersect area between the simulation result and the real scenario, b is the union area between the simulation result and the real scenario, and c is the overlay display of two images.

Method 4: risk assessment of regional flood disaster

We employ the Analytic Hierarchy Process (AHP) to determine the weights of indicators for regional flood disaster risk assessment. The AHP is a multi-criteria evaluation method that assigns weights to criteria based on expert judgment. It is widely applied in disaster risk assessment to systematically quantify the relative importance of influencing factors. The calculation steps are as follows^55,56.

• (1) To derive the relative weights of these factors, a pairwise comparison matrix $A=[{a}_{{ij}}]$ is constructed using the Saaty 1–9 scale method (Table S4). Each element ${a}_{{ij}}$ represents the relative importance of factor $i$ over factor $j$, satisfying ${a}_{{ij}}$ = 1/${a}_{{ji}}.$

• (2) A consistency check is performed to ensure the logical coherence of the pairwise comparisons. The consistency index (CI) is computed as:

  $${CI}=\frac{1}{{n}\! - \!{1}}({\lambda }_{\max }-n)$$

  (6)

  where $n$ is the order of the matrix, ${\lambda }_{\max }$ is the maximum value of the eigenvector matrix. The consistency ratio (CR) is then calculated as:

  $${CR}=\frac{{CI}}{{RI}}$$

  (7)

  where ${CR}$ is the random consistency ratio of the comparison matrix; ${RI}$ is the average stochastic consistency index of the comparison matrix, and the value is shown in Table S5. When ${CR}$ is less than 0.1, the consistency is considered acceptable. If the consistency requirement is not met, the comparison matrix must be revised and adjusted until satisfactory consistency is achieved.

• (3) The unit eigenvector $W$ corresponding to the maximum eigenvalue of the comparison matrix $A$ was calculated, where each component of $W$ represents the importance order of each evaluation factor, the sum of components of $W$ equal to 1.

We conducted flood disaster risk assessment from three different levels: goal, criteria, and alternatives. The overall risk of flood disasters was set as the goal level. Considering the importance of hazard, exposure, and vulnerability in achieving the overall goal, they were set as the criteria level, and a judgment matrix was constructed to calculate the first-level weights. Indicators such as rainfall were set as the alternatives level, being significant influencing factors, and the second-level weights were calculated. Finally, the consistency checks were performed on both the criteria and alternatives level judgment matrices to ensure the rationality of the model. Detailed information can be found in Table 1. The calculation formulas are as follow:

Table 1 Weight of risk indicators

The hazard index:

$$H\left(x\right)={RF}\times {H}_{{rf}}\left(x\right)+T{\times H}_{t}\left(x\right)+{RN}\times {H}_{{rn}}\left(x\right)$$

(8)

The exposure index:

Model−1:

$$S\left(x\right)=N\times {S}_{n}\left(x\right)+K\times {S}_{k}\left(x\right)+D\times {S}_{d}\left(x\right)+{SP}\times {S}_{{sp}}\left(x\right)$$

(9)

Model-2:

$$S\left(x\right)=N\times {S}_{n}\left(x\right)+K\times {S}_{k}\left(x\right)+D\times {S}_{d}\left(x\right)+{SP}\times {S}_{{sp}}\left(x\right)+{RH}\times {S}_{{rh}}\left(x\right)$$

(10)

The vulnerability index:

$$E\left(x\right)=P\times {E}_{p}\left(x\right)+G\times {E}_{g}\left(x\right)$$

(11)

The risk index:

$${FDRI}=H\left(x\right)\times {w}_{H}+S\left(x\right)\times {w}_{S}+E\left(x\right)\times {w}_{E}$$

(12)

where ${FDRI}$ refers to the comprehensive flood disaster risk index. $H\left(x\right)$, $S\left(x\right)$, $E(x)$ denote the hazard index related to disaster-inducing factors, the exposure index related to the disaster-inducing environment, and the vulnerability index related to disaster bearing body, respectively. ${w}_{H},\,{w}_{S},\,{w}_{V}$ represent weights for the evaluation factors. According to the theory of natural disaster risk, three main aspects are considered: the hazard of disaster-inducing factors, the exposure of disaster-inducing environment, and the vulnerability of disaster-bearing body. A total of ten indicators are selected, including rainfall (RF), terrain wetness index (T), river network density (RN), NDVI (N), soil erodibility factor (K), DEM (D), slope (S), relative height (RH), as well as socio-economic factors including population density (P) and GDP (G) (Table S2). The corresponding weights for these indicator factors are represented by ${H}_{{rf}}\left(x\right),{H}_{t}\left(x\right),{H}_{{rn}}\left(x\right),{S}_{n}\left(x\right)$, ${S}_{k}\left(x\right),{S}_{d}\left(x\right),{S}_{{sp}}\left(x\right),{S}_{{rh}}(x)$, ${V}_{p}\left(x\right),{V}_{g}\left(x\right),$ respectively. Considering the varying roles of various influencing factors in flood disaster events, they are classified into hazard, exposure, and vulnerability, and in conjunction with the actual conditions of the study area, assessing the importance of indicators at each level. All indicator data are normalized before computation. A multi-level judgment matrix is constructed, and the weights at two levels are calculated separately. These weights are then processed with a tiered weighting approach, followed by a consistency check.

• 1) The hazard of disaster-inducing factors

  In this paper, the hazard of disaster-inducing factors includes rainfall, river network density, and terrain wetness index. Rainfall, particularly its high-intensity and persistent nature, serves as the primary meteorological factor for the occurrence of flood disasters⁵⁷. From July to August 2022, the study region experienced continuous heavy rainfall. The rainfall data used in this study is derived from field rainfall observation stations. The river network density reflects the sparsity of a region’s river network and is one of the indicators of runoff development in a river basin. A higher river network density indicates a greater degree to which the basin has been dissected by floodwaters, with runoff converging more quickly⁵⁸. Conversely, a lower river network density means runoff converges slowly, leading to poor drainage in the basin. The TWI (terrain wetness index) acts as a physical gauge that signifies the impact of regional topography on the flow direction and accumulation of runoff. This index facilitates the identification of rainfall runoff patterns, regions likely to experience an increase in soil moisture content, and areas susceptible to water pooling.

• 2) The exposure of disaster-inducing environment

  The disaster-inducing environment is predominantly influenced by the characteristics of the underlying surface within a watershed. We select five indicators, namely slope, DEM, NDVI, soil erodibility factor (K), and relative height, to evaluate the exposure of the disaster-inducing environment in the watershed. Notably, the indicator of relative height is only incorporated into the risk assessment Model-2. The formation and spatial distribution of flood disasters are intrinsically linked to topography features. Regions with lower elevations and gentler slopes exhibit greater capacities for runoff generation and collection, thereby conferring a higher risk of flood occurrence⁵⁹. The ground subsidence in the study area has driven significant elevation loss, transforming the region into a large-scale subsidence funnel. This has diminished the river’s flood discharge capacity, making it more susceptible to overflow during high-intensity rainfall. As a result, there is a heightened risk of flooding, with water more likely to accumulate within the subsidence funnel, increasing the potential for flood disasters. The soil erodibility factor (K) is an indicator of soil resistance to erosion under the influence of climatic conditions and human activities⁶⁰. A higher K value signifies greater soil vulnerability to erosion and easily resulting in an elevated risk of flood disasters. Surface vegetation possesses significant water retention capabilities, and its coverage can mitigate the impact force of raindrops, improve soil structure, and effectively reduce flood risk. Higher vegetation coverage is associated with lower surface flood disaster exposure. The study area is encircled by continuous relative height with rivers meandering through. In this region, extreme weather events are prone to triggering flood disasters. Simultaneously, the presence of relative height impedes the dispersal and recession of floodwaters, intensifying the effects of flood disasters. Once the flood breaks through the surrounding relative height, the secondary disaster is likely to occur in the adjacent regions.

• 3) The vulnerability of disaster-bearing body

Flood disaster is a significant threat to people’s lives and property of inhabitants within the impacted regions. In this study, two indicators, that is, the population density⁶¹ and GDP, are utilized to evaluate the vulnerability of the region⁶². It is assumed that regions characterized by higher population densities and greater GDP are likely to bring more substantial losses in the event of a flood disaster.