Research on the risk of drought disasters in the middle reaches of the Yellow River coupled with ecosystem entity

Currently, drought risk assessments primarily focus on utilizing socio-economic indicators such as population, economy, and infrastructure as vulnerable entities, while neglecting the ecosystem (Ault 2020; Wu et al. 2020; Zhao et al. 2020; Hoque et al. 2021; Franke 2022). Ecosystems provide essential material foundations for human survival and development, and the health of ecosystems is closely related to human well-being. Frequent and severe droughts can negatively impact ecosystem health and services (such as providing nutritious food and clean water sources, species interactions, supporting crop pollination and soil formation, and creating recreational, cultural, and spiritual values) by inhibiting vegetation physiological processes (Weiskopf et al. 2020; Li et al. 2023; Yang et al. 2023). Functionally degraded ecosystems under drought stress can also trigger a series of land–atmosphere responses, including land degradation (Eckert et al. 2015), soil erosion (Lieskovský & Kenderessy 2014), increased atmospheric carbon dioxide concentration (Gatti et al. 2014), and vegetation damage (Gupta et al. 2020). Therefore, there is an urgent need to couple the ecosystem-human society vulnerable entities to comprehensively assess drought risk and conduct disaster prevention and mitigation work more targeted (At the interface between hydrology & ecology 2024).

As the breadth and depth of climate change impacts increase, ecosystem vulnerable entities are gradually being considered in the field of drought risk research, exploring and evaluating the exposure and vulnerability of ecosystems to climate change. This can lead to a deeper understanding of the mechanisms and evolution patterns of drought, thereby improving the accuracy of prediction and monitoring and enhancing drought resilience in agriculture, water resources management, and social planning (Haile et al. 2020; Werner et al. 2021; Zhao et al. 2021; Müller & Bahn 2022; Yin et al. 2023). The permeation and interaction between ecosystems and socio-economic elements form a complex structure known as the ecosystem-human society composite (Chang & Turner 2019), which represents a new approach for risk assessment based on vulnerable entities, focusing on comprehensive indicators of ecosystem-human society and emphasizing their overall integrity. However, there is still a lack of comprehensive assessment and multiple interpretations of the concept of ecosystem-human society vulnerable entities in academia (Dai et al. 2020).

Drought refers to a prolonged period of below-average precipitation and inadequate water supply, which has extensive and profound impacts on soil, water resources, and ecosystems. Many studies have been conducted on the evolution characteristics of drought indicators. Huang et al. (2015) introduced a non-parametric multivariate standardized drought index for studying the spatiotemporal characteristics of drought structures in the Yellow River Basin. Wang et al. (2019) explored the spatiotemporal evolution patterns of drought factors in the Yellow River Basin, while Zhu et al. (2018) used Copula functions to construct a joint distribution model of meteorological drought and hydrological drought to investigate the evolution patterns of drought events in the Yellow River Basin. The results of these studies indicate an increasing trend in seasonal drought severity in the Yellow River Basin.

The middle reaches of the Yellow River suffer from severe soil erosion, prompting the Chinese government to implement a series of ecological reconstruction projects to improve the local ecological environment. While achieving ecological benefits, these projects have significantly altered the underlying surface conditions, rendering the ecosystem more vulnerable. Against the backdrop of extensive ecological reconstruction in the middle reaches of the Yellow River, comprehensive assessment of drought risk by coupling human society and ecosystem vulnerable entities is of great practical significance. The current study focuses on (1) quantifying the exposure of ecosystem vulnerable entities, establishing a framework for calculating ecosystem vulnerability under drought stress, and coupling human society-ecosystem drought risk assessment vulnerable entities; (2) utilizing the standardized precipitation index (SPI) to characterize the probability of drought occurrence, identifying drought events using run theory, and constructing joint distributions of drought duration and intensity based on Copula functions, calculating the probability of drought events exceeding baseline drought events; and (3) conducting drought risk assessment from a multivariate perspective based on the Intergovernmental Panel on Climate Change's (IPCC's) definition of ‘risk = probability of occurrence × exposure × vulnerability’.

The hydrometeorological data for the middle reaches of the Yellow River from 1959 to 2019 were obtained from 47 meteorological observation stations. The data were sourced from the National Meteorological Information Center (http://data.cma.cn/), including daily precipitation, daily average temperature, daily maximum temperature, and daily minimum temperature datasets. Missing data were reasonably interpolated using hydrological analogy and linear interpolation methods.

GDEMV3 30-meter spatial resolution digital elevation data were sourced from the Geographic Spatial Data Cloud (https://www.gscloud.cn/). The data product is based on the Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) data and covers the entire land surface of the globe.

The NOAA Climate Data Record of normalized difference vegetation index (NDVI) contains global normalized vegetation index data spanning from 1981 to the present. The spatial coverage is global with a spatial resolution of 0.05°. The data were sourced from the National Centers for Environmental Information (https://www.ncdc.noaa.gov).

Land and water area datasets were sourced from the Socioeconomic Data and Applications Center (SEDAC) of NASA. The data are from 2010 with a spatial resolution of 30 arc-seconds.

Gridded global datasets for the gross domestic product (GDP) and human development index (HDI) contain data on the HDI and GDP. The spatial coverage is global with a spatial resolution of 1 km, spanning from 1990 to 2015. The data were sourced from Dryad (https://datadryad.org/stash/dataset/doi:10.5061/dryad.dk1j0).

The global human settlement layer (GHSL) contains data on population (POP), Settlement Model (SMOD), and built-up area (BUILT). The spatial coverage is global with a spatial resolution of 1 km, spanning from 1975 to 2030. The data were sourced from GHSL (https://ghsl.jrc.ec.europa.eu/download.php).

The global gridded monthly sectoral water use dataset contains monthly water consumption for various sectors. The spatial coverage is global with a spatial resolution of 0.5°, spanning from 1971 to 2010. The data were sourced from GHSL (https://ghsl.jrc.ec.europa.eu/download.php).

Considering that the exposure of the ecosystem can be characterized by the extent of vegetation coverage, among numerous vegetation indices, the NDVI is considered reliable for reflecting vegetation growth conditions and is suitable for assessing the adverse effects of climate extremes on ecosystem health. Therefore, this study utilizes NDVI to characterize the exposure of the ecosystem (Tucker et al. 2005; Pinzon & Tucker 2014).

When the exponential terms ⁠, ⁠, and vary within the range [1, +∞), the population component, GDP component, and development component of the exposure index will range from 0 to 1/3. Therefore, the resulting normalized exposure index will fall within the interval [0,1).

The mathematical expressions and explanations for each dimension are as follows:

Urban development dimension (dim(urben)): The DEGURBA is determined using the DEGURBA method developed by the European Space Agency (EuroSAT) (Schiavina et al. 2023). This method utilizes built-up area, volume, and population density to classify the degree of urbanization into eight categories for all grid units (Table 1).

The distribution of precipitation is not normally distributed but rather skewed. Therefore, the gamma distribution is used to describe the variation in precipitation when analyzing precipitation and predicting and assessing droughts. The SPI is employed to categorize the severity of droughts based on cumulative probability distributions of standardized precipitation (Asadi Zarch et al. 2015; Šebenik et al. 2017; Zhang et al. 2017; Table 2). This study evaluates the probability of drought occurrence based on a 3-month SPI index. The SPI index is a relatively simple calculation to perform, exhibits stability, and is flexible in its application. It offers advantages such as the ability to compare data from different time periods and locations, which makes it an appropriate tool for monitoring and assessing droughts at various spatial and temporal scales (Won et al. 2020; Liu et al. 2021; Laimighofer & Laaha 2022). The specific calculation method is as follows:

This study utilizes run theory to identify drought attributes (Endt et al. 1951; Moyé et al. 1988). Setting the SPI threshold to −0.5 and truncating the SPI sequence, we can identify drought events and extract their attributes. In a finite-length time series, a subset meeting the selection criterion (SPI ≤ − 0.5) forms a coherent short sequence, known as a ‘run’; the length of the run represents the duration D of the drought event. Within a run, the cumulative gap between the SPI and the threshold is the drought severity S.

In the event that the interval between two droughts is brief, we posit that the drought events are somewhat correlated rather than entirely independent, and thus merge these adjacent features. A series of interconnected drought events typically occurs within a longer drought period. When precipitation oscillates frequently, short-term precipitation exceeding the drought threshold results in the fragmentation of a complete drought process into non-truly independent drought events. This study employs the IC method to merge consecutively correlated drought events. If the interval period between adjacent drought events or the precipitation within the interval period is sufficiently short, meaning that the ratio of drought duration (or severity) to the average duration (or severity) of the drought event sequence is less than a threshold, then the drought event is considered a minor drought and is excluded from further analysis (Salvadori & De Michele 2015; Tu et al. 2016; Collet et al. 2018).

Based on Equation (17), the exceedance probability of the reference drought event is calculated at each station. When calculating the occurrence probability of the target drought event, typically four probabilities are considered under both univariate and bivariate perspectives:

• 1. Consider only the exceedance probability of a specific drought duration:

  

  (18)
• 2. Consider only the exceedance probability of a specific drought severity:

  

  (19)
• 3. Probability that at least one of the specific drought duration or severity is exceeded (OR relationship):

  

  (20)
• 4. Probability that both the specific drought duration and severity are exceeded (AND relationship):

  

  (21)

Vulnerability of the ecological system: The temporal distribution of the ecosystem vulnerability index of the middle reaches of the Yellow River exhibits a decreasing trend over time, while its spatial distribution exhibits a high value in the northern region and a low value in the southern region. The index is particularly high in the upper half of the mainstream of the middle reaches of the Yellow River, the upper part of the Wuding River, the upper part of the Fen River, and the upper part of the Beiluo River. In contrast, it is low in the lower half of the mainstream of the middle reaches of the Yellow River, the lower part of the Fen River, Jing River, Wei River, Yiluo River, and Qin River Basins. The vulnerability index of a small portion of the lower Wei River has exhibited a notable increase in recent years, becoming an area of high vulnerability (Figure 4(b1)–4(b3)).

Exposure of human society: The human social exposure index in the middle reaches of the Yellow River exhibits a yearly increase in temporal distribution, with correspondingly high exposure in the southeast and low exposure in the northwest and central parts of the country. The high-exposure areas are distributed in the lower half of the mainstream of the middle reaches of the Yellow River, Fen River, and Wei River basins. Conversely, the low-exposure areas are distributed in the upper half of the mainstream of the middle reaches of the Yellow River, Jing River, Beiluo River, and Wuding River basins. The higher the upstream location, the lower the exposure index. The increase in the exposure index in the middle reaches of the Yellow River is particularly pronounced, reaching a high level from 2006 to 2019 (Supplementary Appendix Figure S1(a1)–S1(a3)).

Vulnerability of human society: The human social vulnerability index in the middle reaches of the Yellow River has remained relatively stable over time, with a similar spatial distribution to that of human social exposure, which is generally high in the southeast and low in the northwest and central parts of the country. The high vulnerability area is distributed in the lower half of the mainstream of the middle reaches of the Yellow River, Fen River, and Wei River Basins. The higher the vulnerability index, the further downstream it is. The low-vulnerability area is distributed in the upper half of the mainstream of the middle reaches of the Yellow River, Jing River, Beiluo River, and Wuding River Basins. The higher the vulnerability index, the lower it is, and the closer it is to the upstream. The vulnerability index in the middle reaches of the Yellow River has exhibited a minimal change since 1959 (Supplementary Appendix Figure S1(b1)–S1(b3)).

In consideration of the intensity of the drought, under the 5a return period, the high-value areas in the study area are initially concentrated in the south and north, and subsequently in the west. From 1959 to 1990, the exceedance probability of the remaining areas was in the medium-high range, with the exception of the south of the middle reaches of the Yellow River, which exhibited a lower probability. From 1991 to 2005, the exceedance probability of the south and north was in the high range, while that of the upper reaches of the Jing River and the upper reaches of the Wei River was in the low range. From 2006 to 2019, the upper Beiluo River, the upper Jing River, and the upper Wei River exhibited higher probabilities, while all other areas exhibited lower probabilities (Figure 6(a1)–6(a3)). The high-value areas in the study area exhibited a similar trend under the 20-year recurrence period to that observed under the 5-year recurrence period. The high-value areas of the exceedance probability were sporadically distributed throughout the study area between 1959 and 1990 and between 1991 and 2005. With the exception of a few isolated areas in the eastern, southern, and northwestern regions, the exceedance probability was relatively low. From 2006 to 2019, the areas with high exceedance probability were intermittently distributed in the central and southern portions of the study area (Figure 6(b1)–6(b3)). The exceedance probability of each subarea under the 50-year and 100-year return periods was consistent with the 20-year return period (Figure 6(c1)–6(c3) and 6(d1)–6(d3)).

When considering only the drought calendar, it can be observed that under the 5-year return period, the drought risk in most areas of the study area shows an increasing trend. Among these areas, the most notable examples are the Wuding River, Beiluo River, and Jing River basins. In contrast, the Fen River basin exhibits a fluctuating pattern, with an overall increasing trend. The Fen River and the lower reaches of the Wei River exhibit the lowest drought risk, while the upper reaches of the Jing River exhibit the highest. Between 2006 and 2019, the entire region exhibited high drought risk, with the exception of the area near the mainstream of the Yellow River (Supplementary Appendix Figure S2(a1)–S2(a3)). In the 20-year recurrence period, the area of high drought risk shifts from the southeastern part of the region to the eastern and southern parts, and then to the central and southern parts. From 1959 to 1990, the drought risk was higher in the Wuding River, the Beiluo River, and the Beiluo River. The Wuding River, Beiluo River, and Jing River basins exhibited a lower risk, while other areas exhibited a higher risk. From 1991 to 2005, the high-risk areas were in the upper Fen River and the lower Wei River, while the rest of the areas exhibited a lower risk. From 2006 to 2019, the lower Wuding River and the lower Wei River exhibited a higher risk, while the rest of the areas exhibited a lower risk (Supplementary Appendix Figure S2(b1)–S2(b3)). The drought risk of the study area under the 50- and 100-year return periods was essentially equivalent to that under the 20-year return period, and it was not equivalent to that under the 20-year return period. The drought risk in the study area under 50- and 100-year return periods was essentially equivalent to that under a 20-year return period (Supplementary Appendix Figure S2(c1)–S2(c3) and S2(d1)–S2(d3)).

Considering only the drought intensity, under the 5a return period, the drought risk tends to migrate from the east to the west via the central part of the country, and after the migration, the risk is higher in the high-value areas and lower in the low-value areas. Downstream was the highest, and the Jing River upstream was the lowest; in 2006–2019, the areas with the highest risk values were in the upper reaches of the Jing River and the Beiluo River, and most of the other areas were lower (Supplementary Appendix Figure S3(a1)–S3(a3)). Under the 20a return period, the drought risk high-value areas were first distributed sporadically in the whole bureau, then concentrated in the northern and southern regions, and finally, clustered in a few areas in the central and southern parts of the country. Spatially, from 1959 to 1990, the risk was lower in the Wuding River, Beiluo River, and Jing River basins, and higher in the rest of the area; from 1991 to 2005, the high-risk areas were in the upstream of the Fen River and the downstream of the Wei River, and the rest of the area had lower risk; and from 2006 to 2019, the risk was higher in the downstream of the Wuding River and the downstream of the Wei River, and the rest of the area had lower risk (Supplementary Appendix Figure S3(b1)–S3(b3)). In the 50 and 100a return periods, the drought risk of the study area is basically the same as that of the 20a return period (Supplementary Appendix Figure S3(c1)–S3(c3) and S3(d1)–S3(d3)).

This study employed vegetation cover (NDVI) to quantify the ecosystem exposure. The observed trend of increasing exposure year by year is related to the policy of returning farmland to forests in the middle reaches of the Yellow River. The restoration of large areas of vegetation can increase the rate of soil and water conservation, alleviate soil and water erosion, and improve the ecological environment of the Yellow River. In response to the effects of climate change, vegetation restoration projects have been implemented in various regions, including the return of farmland to forests and grasslands, the establishment of the ‘Three-North’ protection forests and natural forest protection, and the enhancement of vegetation cover (Li et al. 2022; Mu et al. 2022; Shidong & Moucheng 2022). These initiatives have contributed to the reduction of ecosystem vulnerability in most areas, particularly in the southern region of the middle reaches of the Yellow River. The lower exposure in the northwestern part of the middle reaches of the Yellow River may be attributed to the fact that the annual precipitation and mean annual temperature decrease from southeast to northwest. This makes it challenging to meet the water and light conditions required for vegetation growth, which in turn inhibits vegetation growth. Moreover, the northwestern part of the Yellow River is situated within an area of complex land use, encompassing agricultural and animal husbandry activities. The predominant land use types in this region are cropland, grassland, and unutilized land, which exhibit lower NDVI values. Concurrently, the natural conditions in the northwest of the middle reaches of the Yellow River are suboptimal. Consequently, densely populated areas may overexploit land resources in order to satisfy their own development. This may include the overexploitation of oil and coal resources and the expansion of construction land, which destroys the connectivity of the vegetation, weakens the self-regulation mechanism of the ecosystem, and significantly increases the degree of vulnerability. Furthermore, the annual precipitation is low, which restricts vegetation growth. Additionally, the vegetation cover is low, and the ability to intercept post-rain runoff is weak, which is prone to cause soil erosion and other environmental issues.

This study proposes a probabilistic framework for calculating the vulnerability of ecosystems under drought stress. This framework calculates the probability of vegetation loss under the influence of drought, identifies the spatial extent of ecologically vulnerable areas, and visualizes and quantifies the vulnerability of ecosystems under drought stress. There are two significant advantages to this approach: ① The adverse impacts of drought on ecosystems can be assessed under any given drought scenario (often drought forecast information of interest) by deriving probabilistic expressions for the state of vegetation below specified quartiles (e.g., 30th, 20th, and 10th percentiles) of a long-term observation series. When the probability of loss is high, it can be inferred that the area in question has a high level of ecosystem vulnerability in the face of drought stress. Conversely, when the probability of loss is low, it can be concluded that the area in question has a low level of ecosystem vulnerability in the face of drought stress. ② If the natural geographic environment and socio-economic conditions of other regions are quite different from those of the middle reaches of the Yellow River, researchers can also replace the SPI and NDVI indicators in this paper with other indicators to explore the change of a certain indicator under that type of stress, e.g., replace the NDVI with an indicator related to erosion, and replace the SPI with an indicator related to runoff, to study the erosion of water and soil under the flooding stress.

From 1959 to 1990, the drought duration risk in the southwestern part of the study area (mainly in the Wei River basin) was higher than the drought intensity risk. This indicates that the region was vulnerable to droughts with long durations and low intensities during this time period. The distributions of drought duration and intensity were essentially identical in the period from 1991 to 2005, suggesting that droughts caused comparable damage and that droughts with long durations were associated with higher intensities. The droughts in the Wei River, Wuding River, Fen River, and Yiluo River basins in the period from 2006 to 2019 were higher than the drought intensity risk. In 2019, the drought calendar time risk was higher than the drought intensity risk in the Wei River, Wuding River, Fen River, and Yiluo River basins. This indicates that the situation of long-term, low-intensity drought has shifted to these areas.

In the case of a bivariate OR relationship, the entire study area is at high risk. This is due to the fact that the probability of exceeding the value of either attribute is easily exceeded during a drought event. Consequently, the probability of exceeding one of the two attributes is greatly increased. In the case of a bivariate AND relationship, the risk of the entire area is low from 1959 to 1990. However, the high drought risk area is concentrated in the southeast of the middle reaches of the Yellow River, including the lower reaches of the Jing River, the lower reaches of the Wei River, the Wuding River, the Fen River, and the Qin River from 1991 to 2005. However, the relative risk of these areas is low at the level of the 5-year return period, while the risk is relatively high at the level of the other return periods. The greater the return period, the more obvious the gap in drought risk becomes. The drought risk gap becomes more obvious, indicating that these areas are more prone to severe drought events. From 2006 to 2019, the risk shifted to the western and northern parts of the study area, mainly in the upper Wuding River, the upper Beiluo River, the upper Jing River, the upper Wei River, the upper Fen River, and the upper Yiluo River, which are far away from the mainstem of the Yellow River. The observed change in drought risk over time may be attributed to the fact that in the early days of China's development, the country's focus was on economic growth, with less attention paid to the protection of the ecological environment. The majority of regions with elevated levels of drought risk are characterized by high population densities and extensive economic activities. The implementation of the policy of returning farmland to forests has led to improvements in the ecological environment, with increased vegetation cover, enhanced water and fertilizer retention capacity of the soil, reduced soil erosion and erosion, and a reduction in the vulnerability of the ecosystem.

It is possible that different stakeholders may be concerned with different drought variables. In such a case, it is possible to provide personalized information to different stakeholders in order to support more effective decision-making. For instance, in the context of the agricultural sector, the duration of the drought may be of greater consequence, given its impact on crop growth cycles and irrigation demand. For water managers, drought intensity may be a more significant factor in decision-making because it directly reflects water scarcity. A multivariate approach to considering drought duration and intensity can provide a more comprehensive characterization of drought than a single variable. It can also provide a deeper understanding of the nature of drought and better reveal the complexity of drought events and associated ecosystem and human society impacts. Furthermore, it can provide a richer dataset for further research and management. Nevertheless, this study is not without limitations.

The vulnerability of disaster-bearing entities is a multifaceted concept. This study constructed a vulnerability index from the perspectives of human development, resource pressure, subsurface characteristics, and infrastructure conditions. However, it did not yet consider the significant influencing factors, such as governmental management (disaster mitigation planning and legislation on disaster risk management) and scientific and technological levels of disaster mitigation. In the subsequent study, it is intended to continue the collection of pertinent data and to seek more precise expressions of the exposure and vulnerability of disaster-bearing entities, driven by spatial big data and combined with physical models. This will be done with the aim of improving the reliability of risk assessment results. In the meantime, with regard to the data-driven calculation of ecosystem vulnerability and exposure, it is crucial to compare different sources of data for the same variable. For instance, vegetation status can be characterized by remote sensing data, such as NDVI, Leaf Area Index (LAI), and Gross Primary Productivity (GPP), respectively. This comparison may result in differences in assessment results. Some human responses to drought, such as the diversion of ecological flows for urban water supply, may exacerbate the risks to ecosystems. The difficulty of ecosystems in providing sufficient food, raw materials for production, and biomass energy to human societies as a result of water scarcity will also increase the level of risk to human societies under drought stress. This indicates that the two types of risks are interrelated and that there is a complex mutual feedback relationship. Consequently, future research will establish a risk transfer network and study the mutual feedback mechanism of ecosystem-human society risks, thereby providing a scientific basis for the integration of different disaster-bearing bodies and the development of integrated risk management techniques.

In this study, we employed NDVI to characterize the exposure of ecosystem vulnerability. We then established a probabilistic assessment framework for ecosystem vulnerability under drought stress, coupled ecosystems and human society vulnerability, and conducted a multivariate drought risk assessment of the middle reaches of the Yellow River. Finally, we developed a drought risk assessment methodology. In consideration of the interactions between ecosystems and human societies, multiple factors such as ecology, climate, and resources can be integrated, thereby enabling the comprehensive deployment of disaster prevention and mitigation strategies. A small portion of the middle reaches of the Yellow River (Wei River Basin and Fen River Basin) is vulnerable to drought events with long duration and low intensity, while the majority of the remaining areas exhibit a positive correlation between the risk triggered by duration and intensity. The overall drought risk in the study area tends to increase and gradually migrate from areas in proximity to the mainstream to those situated at a greater distance from the mainstream. The western and northern parts of the study area are more susceptible to drought events with long duration and high intensity. In light of the influence of global climate change, future research should employ a non-consistent drought probability calculation method to enhance the accuracy of the simulation. Additionally, the methodology can be extended to encompass other natural hazards, such as floods, in order to develop a comprehensive risk assessment model that encompasses multiple hazards.

This research was funded by the National Key Research Priorities Program of China (2023YFC320930304) and the Huang Committee Outstanding Young Talents Science and Technology Project (HQK-202305).

Data cannot be made publicly available; readers should contact the corresponding author for details.

The authors declare there is no conflict.