Introduction

Cities, a human settlement of a notable size, reflect the great achievements of human civilization and are instrumental in shaping the future trajectory of global political and economic advancement1. The rapid pace of urbanization has catalyzed the evolution of urban economies and societies, drawing an increasing populace to urban regions. By 2030, it is projected that urban areas will accommodate 70% of the world’s population. Urban centers are anticipated to account for approximately 75% of global energy consumption and an equal share of carbon emissions2. The fast-paced urban growth has triggered extensive land cover change and urban sprawl, leading to a rise in impermeable surfaces, a decline in biodiversity, habitat fragmentation, and escalated environmental pollution3,4. Consequently, urban ecological security issues have surged, exerting significant strain on regional ecosystems and contributing to the increasing vulnerability of global and regional urban ecosystems5.

To address urban ecological security issues, scholars and policymakers have increasingly acknowledged the imperative of pursuing sustainable urban development. The core objectives of sustainable urban development are identified as security, comprehensiveness, harmony, and resilience6. Following Holling’s initial utilized the notion of “resilience” in the realm of ecology in 19737, its application has expanded significantly within urban studies, leading to the emergence of “urban resilience” as a distinctive concept8. Urban resilience is a critical component of green urban development and planning. It primarily concerns a city’s ability to mitigate disaster-related losses, withstand adversities, allocate resources efficiently, and swiftly recover from such events9. The concept of resilience has further evolved to include aspects of urban ecosystem studies, integrating economic, ecological, and sociological perspectives to promote sustainable urban growth10. Urban ecological resilience, as a subset of urban resilience, is recognized as a multidisciplinary concept that merges urban planning with ecological principles11, embodying social, environmental, and economic aspects of resilience. In this study, following the concept proposed by Shariff12, we define urban ecological resilience as the capacity of urban ecosystems to mitigate hazards, adjust to disturbances, and effectively recover from interruptions.

The resilience of urban ecosystems has gained significant attention in recent studies. Nathwani et al.13 contend that research on resilience of urban ecological should not only highlight the stable development of urban, natural, economic, social, and ecological systems but should also delve into the dynamic interactions between these subsystems. Currently, the quantitative evaluation of urban ecological resilience is considered a primary metric for assessing environmental harmony and the green development of a certain region14. Existing research on urban ecological resilience spans various domains, including urban construction and planning, emergency response to urban disasters, and the development of smart cities15,16. Methods employed for constructing evaluation indicators predominantly feature the Pressure-State-Response method17,18, the Driving-Force-State-Response model19, and the Driving-Force-Pressure-State-Impact-Response framework20. Associated analysis methods in this field commonly incorporate GIS, system simulation, and coupling coordination analysis21. Research topics are diverse, focusing on the mechanisms of resilience formation22, assessments of resilience23, policy strategies, and the impacts of urbanization, landforms, and resource allocation on urban ecological resilience13. The analysis and measurement of urban ecological resilience are pivotal for identifying pathways towards high-quality urban development. Insights from varied research contexts have contributed to the development of diverse methodologies and indicator frameworks. These studies collectively offer a robust theoretical and methodological foundation for further research in this area.

Urban ecological resilience has garnered significant research interest24, yet several research gaps persist in this field. Firstly, the complexity of urban ecological resilience introduces substantial uncertainties in its analysis and evaluation25. Existing evaluation systems frequently fail to capture the full essence of the urban ecological environment. To accurately reflect the unique characteristics of urban ecological resilience, a more targeted index system is necessary. Accordingly, current research often focuses on the ecological resilience of specific areas or individual cities, overlooking the spatiotemporal dynamics and variations across different regions26. This approach risks fostering a limited perception of urban ecological resilience, potentially hindering holistic and synergistic development efforts. Thirdly, as a statistical method to study regional spatial interaction and dependence degree, spatial autocorrelation analysis can explore the spatial layout of urban resilience state27. Investigations into urban ecological resilience often overlook spatial autocorrelation analysis to assess ecological resilience levels across regions. This omission neglects the spatial dependencies and correlations of resilience states. This oversight can lead to either an undue concentration or dispersion of urban ecological resources, thereby impacting sustainability. To bridge these research gaps, it is imperative to design a comprehensive evaluation framework that delve into the spatiotemporal differences and spatial correlation patterns of ecological resilience to ensure sustainable urban development.

The Yellow River Basin (YRB) is located in northern and central China, encompassing nine provinces, from east to west, including Shandong, Henan, Shanxi, Shaanxi, Inner Mongolia, Gansu, Sichuan, and Qinghai. Recognized as a critical agricultural hub and an ecologically vulnerable zone, the YRB faces severe ecological challenges such as river water shortages, land degradation, and broad ecosystem deterioration that threaten urban sustainability. This study investigates the spatiotemporal variations in urban ecological resilience within these nine provinces. Compared to urban centers, the diverse natural environments, geographical settings, and climatic conditions among the provinces significantly influence regional resilience. Therefore, taking the province as the unit to investigate the spatiotemporal disparities of ecological resilience can better consider the differences between regions and promote regional sustainable development.

The innovation of this research is manifest in several key aspects: First, it establishes a robust system of indicators for urban ecological resilience, incorporating the subsystems of resistance, recovery, and adaptability. This approach offers a more holistic representation of the dynamics within urban ecosystems, addressing the shortcomings of existing indicator systems. Second, it examines the spatial and temporal variations in urban ecological resilience on an expanded scale, a perspective crucial for leveraging ecological complementarity and fostering mutual benefits among adjacent regions. Third, it explores the spatial autocorrelation characteristics of regional resilience levels, identifying both spatial correlation and dependence relationships of resilience levels. These insights are crucial for strategic planning and decision-making in regional ecological management. The ultimate goal of this research is to enhance the stability of urban ecosystems, a fundamental prerequisite for sustainable development and achieving green growth across the YRB.

Methodology

Study area

The YRB, covering an area of 745,100 square kilometers, is the ecological and geographical region influenced by the Yellow River system. The Yellow River runs through nine provinces from its birthplace to its flow into the sea (Fig. 1). The YRB holds significant historical importance as one of the cradles of Chinese civilization, which is a significant region for the intersection of diverse cultures. Recent economic growth in the YRB has been remarkable, with the combined GDP of the nine provinces increasing from 15.582 trillion yuan in 2012 to 30.699 trillion yuan in 2022. During the same period, the total population of the YRB rose from 408.20 million to 420.62 million. The YRB is regarded as one of the crucial ecological functional regions in China, and the health state of its ecosystem has a profound influence on the ecological environment of the entire nation and even East Asia. However, due to climatic and geomorphic factors, the YRB is a high-risk region for floods and droughts, which pose a threat to the local ecosystem and human living environment.

Fig. 1
figure 1

Extent and location of YRB in China.

Full size image

Research framework

This study investigated the urban ecological resilience across the YRB during the years 2011–2020. The analysis process consists of the following steps (Fig. 2): (1) construct an indicator system using three subsystems: resistance, adaptability and recovery; (2) quantify the resilience level based on the improved Multi-Attribute Boundary Area Comparison (MABAC) model; (3) investigate the spatial distribution characteristics of ecosystem resilience state according to the spatial autocorrelation model; (4) explore the driving factors affecting the degree of ecosystem resilience applying the machine learning model; and (5) discuss the spatiotemporal differentiation and spatial distribution pattern of ecosystem resilience taking the YRB as a case study. The research framework is established based on Zhao et al.5.

Fig. 2
figure 2

Flow chart of this study.

Full size image

Indicator system and data

Indicator system establishment

Urban ecological resilience emphasizes the ability of ecological systems to withstand, adapt, and recover from uncertain risks before being reorganized into new structures, which is usually classified into three subsystems: resistance, adaptability, and recovery28. Among them, resistance emphasizes the level of ecosystem withstanding external forces, manifested as a response to waste and pollutant emissions in production, daily life, and manufacturing and farming production processes. Adaptability represents the capacity of ecosystems to handle air and water pollution caused by industrial construction and other activities in the process of urbanization. Recovery could improve urban ecosystem services and supply capacity by increasing urban green areas and water resource protection. The resilience theory framework based on resistance—adaptability—recovery capability helps to objectively and scientifically reveal the entire process of ecological risk response. Figure 3 illustrates the analytical framework for urban ecological resilience.

Fig. 3
figure 3

Analytical structure for urban ecological resilience.

Full size image

Given the aforementioned analysis, an indicator system was established from three subsystems: resistance, adaptability, and recovery. In these three subsystems, resistance is characterized by the vulnerability and sensitivity exhibited by the urban when subjected to strikes and challenges. Following previous findings29, the indicators of industrial waste gas emission, industrial smoke and dust emissions, and industrial wastewater discharged were selected to characterize the pollution status of urban industry; the total amount of domestic water consumption and industrial water consumption were used to characterize the consumption of urban water resources; and the indicators of urbanization rate and per capita GDP represented the degree of urban development. Urban adaptability is displayed as the level of self-adjustment of an urban in the face of interference. The proportion of days with good air quality and the urban air quality index were selected to reflect the urban adaptability. The sewage treatment rate and the total amount of sewage treated represented the absorption effect of urban water resources11. The coverage rate of green area and forest coverage rate represented the response effectiveness of the ecosystem. Resilience is reflected in the speed and degree of restoration of the city after the disturbance. The per capita domestic water consumption, average rainfall, total amount of water resources, and the length of water supply pipe were used to judge the state of urban water resources. The harmless treatment rate of consumption wastes, the amount of industrial smoke and dust removal, the area of the wetland, and the volume of gas supply were selected to measure the urban restoration effect. The index system is formed as depicted in Table 1.

Table 1 The assessment indicator system of urban ecological resilience.
Full size table

Data

In this study, data from different sources were utilized to diagnose the ecological resilience in nine regions of the YRB. The economic and social development data were derived from the China Statistical Yearbook and China Urban Statistical Yearbook. The resource and ecological environment data were collected from the China Eco-environment Statistical Yearbook and China Water Resources Bulletin. Land resource data were acquired by interpreting remote sensing information, and the remote sensing data used in this study was sourced from the website (https://www.globallandcover.com). What needs to be declared is that the value of a small amount of missing data was obtained through the function interpolation.

Model building

Standardization of the data

To remove the influences of range and feature disparities among the indexes, the standardized data model was used to preprocess the indicator data30:

$${\text{Positive indicators}}:\;\;y_{ij} = \left( {x_{ij} - \min \left\{ {x_{i} } \right\}} \right)/\left( {\max \left\{ {x_{i} } \right\} - \min \left\{ {x_{i} } \right\}} \right)$$
(1)
$${\text{Negative indicators}}:\;\;y_{ij} = \left( {{\text{max}}\left\{ {x_{i} } \right\} - x_{ij} } \right)/\left( {\max \left\{ {x_{i} } \right\} - \min \left\{ {x_{i} } \right\}} \right)$$
(2)

where xij is the original indicator value; and yij means the normalized value of xij, xij ∈ [0, 1].

Comprehensive weight method

The comprehensive weight of indicators was determined applying the criteria importance through the intercriteria correlation (CRITIC) approach and fuzzy cognitive map (FCM) technique. The CRITIC approach was utilized to identify the objective weight of the indicators. We selected the FCM technique to calculate the subjective weight. The CRITIC-FCM model was utilized to derive the comprehensive weight.

  1. 1.

    CRITIC approach.

  1. Step 1.

    Calculation of contrast intensity.

    $$\sigma_{i} { = }\sqrt {\frac{{\sum\nolimits_{j = 1}^{n} {(y_{ij} - y_{i} )} }}{n - 1}}$$
    (3)

    where \(\sigma_{i}\) is the contrast intensity.

  1. Step 2.

    Determination of conflict levels.

The conflict degree represents the level of correlation among different indicators. The greater the positive correlation between indicators, the smaller the conflict value.

$$f_{i} = \sum\limits_{j = 1}^{n} {(1 - r_{im} )}$$
(4)

where \(f_{i}\) is the level of contradiction between indicator i and other indicators, \(r_{im}\) represents the correlation degree linking indicator i with indicator m, where the Pearson correlation coefficient was adopted31.

  1. Step 3.

    Calculation of information bearing capacity.

    $$C_{i} = \sigma_{i} f_{i}$$
    (5)

    where \(C_{i}\) is the information bearing capacity of the indicator i.

  1. Step 4.

    Determination of objective weights.

    $$w_{i1} = \frac{{C_{i} }}{{\sum\limits_{i = 1}^{m} {C_{i} } }}$$
    (6)

    where \(C_{i}\) is the objective weight.

  1. 2.

    FCM model.

  1. Step 1:

    Determination of S0 and W0.

    Experts determine the starting value S0 and the starting causal weight W0 of the indicators based on their expertise and experience32.

  2. Step 2:

    Adjustment of the vertice status value.

    $$S_{i}^{t} = f\left( {S_{i}^{t - 1} + \sum\limits_{\begin{subarray}{l} j = 1 \\ \end{subarray} }^{N} {S_{j}^{t - 1} } w_{ji}^{t - 1} } \right)$$
    (7)

    where \(S_{j}^{{\text{t}}}\) means the state value of vertice \(t - 1\) at \(r_{i}^{t} = A_{i,real}^{t} - A_{i}^{t}\) time; \(S_{i,real}^{t}\) is the state value of vertice \(S_{i}^{t}\) at \(A_{i}^{t}\) time; \(A_{i,real}^{t}\) represents the impact level of vertice \(C_{j}\) at \(t - 1\) time.

  3. Step 3:

    Determination of systematic error.

    $$r_{i}^{t} = S_{i,real}^{t} - S_{i}^{t}$$
    (8)

    In expression, \(S_{i,real}^{t}\) is the actual system state value.\(r_{i}^{t}\) refers to the discrepancy between the model learning value \(S_{i}^{t}\) and the actual value of the system \(S_{i,real}^{t}\).

  4. Step 4:

    Judgment of the result.

    According to33, when met \(\left| {r_{i}^{t} } \right| < e_{1}\)(\(e_{1} = 0.001\)) and \(\left| {S_{i}^{t} - S_{i}^{t - 1} } \right| < e_{2}\)(\(e_{2}\) = 0.002), the iteration should be ceased, and the final causal association weight \(W^{t - 1}\) is output, Otherwise, continue to step 5.

  5. Step 5:

    Adjustment of the causal association weight.

    $$W_{ji}^{t} = (1 - \gamma )W_{ji}^{t - 1} + \left[ {\eta + \kappa r_{i}^{t} \left( {1 - r_{i}^{t} } \right)} \right]S_{j}^{t - 1} S_{i}^{t - 1} + \kappa r_{i}^{t} \left( {1 - r_{i}^{t} } \right)W_{ji}^{t - 1} S_{j}^{t - 1} - \eta W_{ji}^{t - 1} \left( {S_{j}^{t - 1} } \right)^{2}$$
    (9)

    Adjust the causal association weight value according to formula (5) and return to Step 2. The ultimate relational weight value among vertices is acquired by utilizing the aforementioned causal association weight learning algorithm.

  6. Step 6:

    Calculation of the subjective weight.

    $$Degreecentrality(C_{j} ) = In\deg ree(C_{j} ) + Out\deg ree(C_{j} )$$
    (10)
    $$w_{i2} = \frac{{Degreecentrality(C_{j} )}}{{\sum\limits_{j = 1}^{n} {Degreecentrality(C_{j} )} }}$$
    (11)

In the formula, \(In\deg ree(C_{j} )\) refers to vertice index value, which is calculated as the sum of the absolute values of the causal association weights that point to this vertice after each iteration. \(Out\deg ree(C_{j} )\) represents output value, obtained by summing the absolute values of the causal association weights pointing away from this vertice. \(w_{2i}\) denotes the subjective weight value.

  1. 3.

    Comprehensive weight model.

To take into account the impact of the objective station and the expertise of experts34, the combined method of the CRITIC model and FCM model is employed to compute the comprehensive weight of the evaluation indicator.

$$w_{i} = \frac{{\sqrt {w_{1i} w_{2i} } }}{{\sum\limits_{i = 1}^{m} {\sqrt {w_{1i} w_{2i} } } }}$$
(12)

where \(w_{i}\) is the comprehensive weight value.

Enhanced MABAC model

The MABAC model is a common management evaluation model, which seeks the optimal decision by judging the proximity of each alternative scheme to the approximate boundary region. However, the traditional MABAC technique sorts plans by a simple criterion distance summation35. When the criteria are close and distant, it is hard to make an exact evaluation of the scheme accurately. Thus, to address this issue, the enhanced MABAC technique regards the relative standard disparity as the evaluation criterion and determines the optimal decision by assessing the relative disparity between the plan and the negative and positive optimal solution36.

  1. Step 1.

    Calculation of the boundary approximation area (BAA).

    $$J_{ij} = w_{i} y_{ij}$$
    (13)
    $$B_{j} { = }\left( {\prod\limits_{i = 1}^{m} {w_{i} y_{ij} } } \right)^{1/m}$$
    (14)

    where \(B_{j}\) is the BAA.

  2. Step 2.

    Determination of the criterion gap between each plan and the BBA.

    $$d_{ij} { = }J_{j} - B_{j}$$
    (15)

In the formula, \(d_{ij}\) is the standard disparity.

Based on the value of \(d_{ij}\), we can make a judgment:

  1. 1.

    If \(d_{ij} < 0\), the plan is part of \(B^{ - }\)(LAA);

  2. 2.

    If \(d_{ij} = 0\), the plan is part of \(B\)(BAA);

  3. 3.

    If \(d_{ij}> 0\), the plan is part of \(B^{ + }\)(UAA).

Undoubtedly, if \(d_{ij}> 0\), the scheme is the optimal decision.

  1. Step 3.

    Determination of the urban ecological resilience index (UERI)

    $$\left\{ \begin{gathered} Q_{i}^{ + } = \sum\limits_{j = 1}^{n} {d_{ij} ,d_{ij}> 0} \hfill \\ Q_{i}^{ - } = - \sum\limits_{j = 1}^{n} {d_{ij} ,d_{ij} < 0} \hfill \\ \end{gathered} \right.$$
    (16)
    $$UERI = \frac{{Q_{i}^{ + } }}{{Q_{i}^{ + } + Q_{i}^{ - } }}$$
    (17)

    where \(Q_{i}^{ + }\) refers to the distance sum of positive criterion; \(Q_{i}^{ - }\) is the distance sum of negative criterion. The higher the UERI is, the nearer the plan is to the optimal solution, and the superior the scheme is.

UERI can be utilized to judge the extent of urban ecological resilience. According to the pertinent research findings37, the classification standards for UERI are displayed in Table 2.

Table 2 Classification criteria of UERI.
Full size table

Spatial autocorrelation analysis

The spatial autocorrelation analysis method is utilized to assess if geographic exhibit clustering in their distribution27, and it is a reliable approach for analyzing spatial patterns. Urban ecological resilience is linked to the spatial arrangement of natural geographic components, social and economic situations, and resource status. These factors are random and structural in space, and there is a certain spatial correlation. Moran’s I is commonly utilized to quantify the global indicator’s geographic autocorrelation state and the local indicator’s geographic autocorrelation state, which are both part of spatial pattern assessment. Here, global Moran’s I was employed to reveal the overall spatial average correlation degree. LISA, which adopts local indicators geographical autocorrelation, was utilized to show the clustering law of resilience level in local space.

Global indicators spatial autocorrelation is a description of the spatial characteristics of a variable value across an entire area. The global Moran’s I is utilized to indicate the general degree of correlation and variation of observations in regions adjacent or adjacent to space. The scope of Moran’s I value is from -1 to 1. When Moran’s I is positive, it represents that the spatial factors are positively correlated. When Moran’s I is negative, it illustrates that there is a negative correlation between spatial factors. When Moran’s I is 0, it explains that the spatial factors are not correlated. The global Moran’s I (Ia) is calculated as follows:

$$I_{a} = \frac{k}{{S_{o} }}*\frac{{\sum\limits_{a = 1}^{k} {\sum\limits_{b = 1}^{k} {W_{ab} \left( {o_{a} - \overline{o}} \right)} } }}{{\sum\limits_{i = 1}^{k} {\left( {o_{a} - \overline{o}} \right)^{2} } }}$$
(18)

In the formula, \(o_{a}\) is the observed value for the i-th region, \(W_{ab}\) expresses the neighboring relationships of spatial elements and functions as a weight matrix for binary adjacency space. When area a is contiguous to area b, \(W_{ab}\) = 1; When area a is not contiguous to area b, \(W_{ab}\) = 0. A positive \(I_{a}\) value means that spatial clusters of similar values (low or high) appear around the spatial units.

$$LocalMoran^{\prime}s\;I_{a} = \left( {\frac{{o_{a} - \overline{o}}}{p}} \right)\sum\limits_{b = 1}^{k} {W_{ab} } \left( {o_{a} - \overline{o}} \right)$$
(19)
$$p = \left( {\sum\limits_{b = 1,b \ne a}^{k} {o_{b}^{2} } } \right)/(k - 1) - \overline{o}^{2}$$
(20)

Random forest model

The random forest model (RFM) was utilized to explore the indicator system, and the important factors influencing the resilience of urban ecological networks were identified from 21 indicators. The Gini importance index was utilized to identify influencing factors38. Nembrini et al.39 have described the particular methods.

Results and analysis

Spatiotemporal dynamic evolution characteristics

The resilience levels of ecological systems from 2011 to 2020 across nine provinces within the YRB are depicted in Fig. 4. Analysis of Fig. 4 reveals that throughout the study period, except for Henan Province, which exhibited notable fluctuations, the remaining provinces demonstrated a consistent upward trajectory in resilience. Noteworthy is Gansu Province, which, despite displaying the most significant growth trend, remained at a resilience level lower than that of the other eight provinces. From a temporal viewpoint, the evolution of the UERI value progressed through three distinct phases: a period of rapid development from 2011 to 2014, followed by a slower growth phase from 2015 to 2017, and culminating in a phase of stable growth from 2018 to 2020. Spatially, the distribution of UERI values manifested a gradient, with higher degrees in the southeastern provinces and lower degrees in the northwestern provinces, suggesting a geographical disparity in ecology system resilience across the YRB.

Fig. 4
figure 4

The UREI in every province of the YRB.

Full size image

Time series characteristics

Over the decade from 2011 to 2020, the UERI value in the YRB consistently ranged between 0.43 and 0.87. The ecological resilience status of the nine provinces within the YRB is displayed in Fig. 5. Temporally, the evolution of ecological resilience in the YRB can be delineated into three phases: initially, an “ordinary resilience” phase characterized by rapid development during 2011–2014; subsequently, a “good resilience” phase marked by slower growth from 2015 to 2017; and finally, a continuation of the “good resilience” phase, albeit with stable development, from 2018 to 2020. This reflects a progression in the resilience level of the region. The progression of ecology system resilience in the YRB from 2011 to 2020 can be segmented into the following three phases:

  1. 1.

    2011 to 2014: Emerging Resilience. Initially, the ecological resilience within the YRB was categorized as being at a medium state. In 2011, except for Inner Mongolia and Gansu which were identified at low levels of resilience, the remaining seven provinces were assessed at an ordinary resilience level. The subsequent year saw the adoption of a pivotal strategy focusing on urban construction guided by ecological civilization principles, leading to strengthened ecological protection across the YRB. This period was marked by rapid development in the ecological system toughness of the nine provinces, culminating in all reaching at least the ordinary resilience level by 2014.

  2. 2.

    2015 to 2017: Consolidating Resilience. During this interval, the pace of growth in the UERI decelerated compared to the preceding period, characterized by gradual development in ecological resilience across the YRB. Initially, in 2015, only Shanxi, Qinghai, and Ningxia provinces were classified at a good resilience level. Nonetheless, the intensified efforts in environmental and water resources conservation catalyzed a continuous enhancement in resilience levels, with all provinces except Gansu elevating to a good resilience status by 2017.

  3. 3.

    2018 to 2020: Stabilizing Resilience. The UERI values of the nine provinces demonstrated a steady upward trend, reflecting a generally stable state of urban resilience along the YRB. During this period, except for Gansu was in an ordinary state, and other provinces always was in a good state. Notably, according to the classification criteria of UERI, the UERI value of Shandong was 0.87 by 2020, reaching an good resilience level.

Fig. 5
figure 5

The ecological resilience state in all provinces of the YRB.

Full size image

These phases highlight the dynamic evolution of ecological resilience across the YRB, showcasing the effects of targeted environmental policies and initiatives on enhancing the urban ecological fabric of the region.

Spatial evolution patterns

The spatial distribution of ecological resilience in the YRB exhibits significant disequilibrium, characterized by distinct zonal features over the decade of 2011–2020 (Fig. 6). Areas of low resilience, notably Gansu and Inner Mongolia, contrast with regions of ordinary resilience levels, which include Henan, Shanxi, Shaanxi, Qinghai, and Sichuan. Meanwhile, zones with good resilience are sporadically located in Shandong and Ningxia. This spatial distribution reveals a clear gradient of ecological resilience, with degrees diminishing from the southeast towards the northwest. Consequently, a discernible low-resilience belt has formed in the northwest, while a corridor of higher resilience values distinguishes the southeast. This examination highlights not only the regional disparities in resilience but also points to potential areas for targeted ecological improvement initiatives and policy interventions, aiming to balance the spatial inequities and enhance the overall ecological stability of the YRB.

Fig. 6
figure 6

Spatial distribution of UERI.

Full size image

The spatial distribution of ecological resilience within the YRB exhibits pronounced temporal dynamics: (1) In 2011, a geospatial pattern divergence in ecological resilience was evident, with the north of the YRB displaying lower resilience levels compared to the southern regions. This disparity was attributed to suboptimal urban planning and inadequate environmental protection measures, leading to the lowest UERI values in Gansu and Inner Mongolia, located in the basin’s north. (2) In 2014, the status of ecological resilience evolved to exhibit higher levels in the eastern and central areas, contrasting with lower levels in the western parts of the YRB. This period noted an overall improvement in the basin’s urban ecological resilience. Notably, Henan, Shanxi, and Ningxia experienced the most significant increases, achieving a good resilience level. (3) In 2017, the count of provinces recorded at good resilience levels saw a notable increase from three in 2014 to eight in 2017. This widespread enhancement in resilience was closely tied to the robust promotion of ecological civilization construction post-2015, markedly improving air, water, and soil quality across the region. (4) By 2020, the southeastern portion of the YRB observed substantial improvements in ecological resilience, with Shandong, situated in the eastern part of the YRB, achieving the highest UERI value and being classified at the level of excellent resilience.

These stages highlight the evolving nature of urban ecological resilience within the YRB, demonstrating significant progress in environmental management and urban planning efforts across the region, ultimately fostering enhanced resilience against ecological challenges.

Characteristics of the spatial correlation pattern

Global correlation feature

The changes in global Moran’s I value for the YRB from 2011 to 2020 are documented in Fig. 7. Figure 8 illustrates the results of the global autocorrelation distribution. In 2011, the Moran’s I index was observed to be positive at 0.405, which indicates a significant spatial correlation among provinces within the YRB in terms of ecosystem resilience. Specifically, it means that provinces with similar degrees of urban ecological resilience were more likely to be located near each other, thus evidencing a spatial agglomeration effect. This effect is further visualized in Fig. 8a, which maps the global autocorrelation distribution, underscoring the clustered nature of ecological resilience across the area. The level of ecological resilience had a spatial agglomeration effect.

Fig. 7
figure 7

The results of Global Moran’s I.

Full size image
Fig. 8
figure 8

Global autocorrelation distribution results.

Full size image

As shown in Fig. 7, the P value exceeded 0.01 for the timeframe from 2012 to 2020, which indicates that the Moran’s I value for urban ecological resilience within the YRB did not attain statistical significance during this period. This means that the level of ecological resilience was distributed randomly across the region during these years (Fig. 8), indicating an absence of a spatial agglomeration effect. Further analysis from 2016 to 2020 reveals an M-shaped pattern in the changes of the Moran’s I value. This pattern signifies that the spatial distribution of the ecosystem resilience state underwent repeated fluctuations, highlighting a clear spatial disequilibrium. Such oscillations point to significant variances in the distribution of ecological resilience across different areas across the YRB, with periods of increased similarity between neighboring locations being followed by periods where the resilience states were more disparate. In conclusion, between the years 2012–2020, the spatial distribution of ecosystem resilience in the nine provinces was discrete, and the degree of dispersion fluctuated repeatedly in the latter half of the period. This means that resilience levels were randomly distributed among provinces in the YRB.

Local correlation feature

The Geoda software was utilized to construct the Moran scatter map for ecological resilience within the YRB. In 2011, the analysis of local Moran’s I indices revealed that urban ecological resilience within the YRB exhibited a positive spatial correlation. This means that the development of ecological resilience across the studied area was strongly interrelated. Contrary to a random spatial distribution, there was an inherent correlation present, signifying that areas of similar levels of urban ecological resilience tended to cluster together spatially (Fig. 8a). However, the situation altered in subsequent years. Specifically, in 2014, 2017, and 2020, scatterplots derived from local Moran’s I analysis demonstrated that urban ecological resilience transitioned to a state of negative spatial correlation and marked spatial disequilibrium. This shift indicates that, unlike in 2011, areas with high levels of resilience were more likely to be adjacent to areas with lower resilience levels, leading to a more dispersed and uneven distribution of ecological resilience across the YRB.

The aggregation status of the nine provinces was categorized into four categories: “low–low” homogeneity category (L–L), “high–high” agglomeration category (H–H), “low–high” collapse category (L–H), and “high-low” polarization category (H–L)40. On the whole, the four categories of local autocorrelation clustering patterns in the study area had good spatial stability, but their quantitative structure changed greatly.

  1. 1.

    In the H–H quadrant, provinces falling into this quadrant display a minimal spatial difference in ecological resilience, characterized by both the urban area in question and its surrounding ones having higher levels of resilience and a significant positive correlation among them. Initially, in 2014, Shandong, Henan, and Sichuan were part of this quadrant, with Shanxi added by 2017, indicating a cluster of areas with relatively uniform and high ecological resilience.

  2. 2.

    The L–H quadrant is marked by significant spatial heterogeneity, with the ecological toughness of the region being lower than that of surrounding urban areas, thus presenting a negative relationship with these higher-value surroundings. By 2017, Sichuan and Gansu were identified within this quadrant, highlighting the pronounced discrete level of ecological resilience in these locations.

  3. 3.

    The L-L quadrant identifies areas where both its own urban area and its neighboring urban areas exhibit low levels of ecological resilience. The small spatial heterogeneity degree signifies minor variations within these areas and their proximity, leading to significant spatial negative correlations. Gansu and Inner Mongolia were classified under this category in 2014, with Qinghai joining them by 2020, suggesting a consistent pattern of low ecological resilience.

  4. 4.

    The H–L quadrant shows a large spatial difference in ecological resilience, where the urban area itself has a high level of resilience, but the neighboring areas exhibit lower resilience levels, indicating a negative correlation with these low-value areas. Provinces, such as Shanxi, Ningxia, and Shaanxi, exhibiting a “point-like scattered, cluster-like agglomeration” distribution, have been allocated to this category. The number of provinces within this quadrant varied, with more cities identified in 2014 and 2017, and fewer in 2011 and 2020.

Influence factors analysis

In the RFM method, the Gini importance index was employed to explore the significant indicators that impact the ecological toughness degree of both the entire YRB and the nine provinces. The influence level of 21 factors on ecological resilience was calculated by using the machine learning method. Table 3 reveals notable variations in the factors influencing urban ecological resilience across the nine provinces studied. Despite these differences, the indicators for domestic water consumption (R2) and the total amount of sewage treated (D3) stood out, being identified five times each as the most influential across the provinces. This recurrence underscores their significance, positioning these two indicators as paramount in affecting the ecological resilience levels within the provinces analyzed.

Table 3 Calculation results of impact factors.
Full size table

The results of the impacting factors in the whole YRB show that the indicators of the amount of wastewater discharged (R4), sewage treatment rate (D2), and harmless treatment rate of consumption wastes (A6) were the key indicators impacting the overall ecosystem resilience in the YRB. The amount of wastewater discharged was a main factor hindering the strengthening of ecosystem resilience. Urban population growth and industrial advancement have led to an increase in the total amount of wastewater, exerting significant stress on the water ecosystem. At the same time, sewage treatment capacity and technology are important links to improve water quality and hold a key position in improving ecological resilience. Finally, waste harmless treatment can effectively reduce or eliminate waste pollution, which is crucial for boosting the level of ecosystem resilience in the YRB.

Discussion

Through the dynamic analysis of the ecological resilience degree in the YRB, we found that the resilience state of 9 provinces in the YRB displayed an improving trend. The research findings of this paper are in line with the existing research outcomes (e.g., Zhao et al.5). This is primarily due to the ecological and environmental protection policies introduced by the national government after 2010, especially the major strategy of “ecological civilization city construction” proposed in 2012. During this period, the relevant government successively issued a series of ecosystem conservation and environmental restoration regulations and strengthened the environment management and supervision of the cities in the YRB. For example, a red line system for ecological protection has been established, which limits the scale and intensity of construction in important ecological functional areas. However, we also found that UERI in Gansu rose the slowest throughout the study period and remained at an ordinary resilience level in 2020. Limited by geographical conditions and ecological environment, Gansu, which is located in the northwestern border region of China, is relatively short of water resources and has poor natural environment conditions. These conditions make it difficult to increase its ecological resilience in Gansu. Even, Gansu’s economic development mainly relies on traditional resource-based industries, such as coal, oil, natural gas, and mineral deposits. The excessive dependence on resource-based industries has resulted in heightened strain on the environment’s health and limited progress in the advancement of ecological resilience in Gansu.

In the YRB, the degree of ecological resilience exhibited a low level in the northwest region and a high level in the southeast region. The outcome of this study contradicts the findings reported by Zhang et al.22. This trend may be attributed to the climatic and geographical conditions as well as technical and financial support in the southeast region. Firstly, compared with Gansu and Ningxia in the northwest, Henan, Shandong, and Sichuan in the southeast of the YRB have a wetter climate, more abundant water resources, and better land quality. These favorable climatic and geographical conditions provide a better ecological foundation for cities in the southeast region. Secondly, the urban economy of the southeast region is relatively developed. The economic structure is more diversified and modern, and the development of ecological industries is more mature. In contrast, the northwest region mainly relies on the traditional resource-based industry, and the economic structure is relatively unitary. The economic growth of the northwest region exerts great stress on the ecological environment.

The spatial feature of the ecosystem resilience state in the YRB was highly unbalanced and negatively correlated. The four types of local autocorrelation cluster patterns all displayed the characteristics of relatively good spatial stability but large changes in quantity structure. The spatial dispersion of ecosystem resilience in the YRB can be due to the following causes: (1) The YRB spans many provinces and has diverse geographical conditions, including different areas such as mountains, plains, and deserts. Different areas have their own specific natural environment and resource endowments; thus, the degree of ecological resilience between regions would vary. (2) The difference in natural disaster type among cities in the YRB is also the primary factor contributing to the uneven spatial distribution. The northwest area of the YRB is insufficient in water resources, and drought disasters and sandstorms often occur. The southeast region is located in the downstream area of the Yellow River and often suffers from floods. (3) The administrative management level and environmental governance capacity among cities in the YRB differ greatly11, which affects the synchronized development of ecological resilience among regions to a certain degree, resulting in the discrete spatial pattern of resilience development levels among cities.

Previous research has identified economic growth, technological advancement, regional GDP, and land use as main factors hindering sustainable urban development11,41,42. However, our findings reveal significant impacts of wastewater discharge volumes and sewage treatment rates on urban ecological resilience. Specifically, increased wastewater discharge negatively affects urban water environments by introducing more pollutants, thereby destabilizing aquatic ecosystems. Conversely, improvements in sewage treatment rates, which result in more effective handling of sewage, reduce the pollutant load directly discharged into water bodies. This enhancement in treatment facilitates the optimization of water quality and maintain the stability of aquatic ecosystems. In addition, in order to bolster urban ecological security, optimizing the harmless treatment of consumption waste and pattern of urban ecological land use are also key measures toprotect water quality and the ecological environment in the YRB42.

Conclusion

Research conclusion

In this study, we used the CRITIC-FCM-MABAC model and machine learning method to explore the ecological resilience degree in the YRB. The core research findings are presented as follows:

  1. 1.

    Spatio-temporal patterns: Temporally, the ecological resilience of the YRB underwent distinct phases. From 2011 to 2014, the region experienced rapid development from a medium resilience level, followed by a slower phase of improvement at a good resilience level between 2015 and 2017, and finally, a period of stable development at a good resilience level from 2018 to 2020. Spatially, a clear geographical pattern, with resilience levels being lower in the northwestern part of the YRB and higher in the southeastern part, was observed. This spatial disequilibrium and zonal characteristic suggest variable ecological resilience across the YRB, reflecting differences in environmental, socio-economic, and governmental policy effects.

  2. 2.

    Spatial correlation: In 2011, spatial autocorrelation analysis revealed a positive correlation among the provinces within the YRB in terms of ecological resilience, suggesting a spatial agglomeration effect. This implies that regions with high levels of resilience were clustered together, as were regions with lower levels. Contrarily, from 2012 to 2020, the spatial distribution entered a state of randomness with no apparent spatial agglomeration effect. This underscores a fragmented and uneven development in ecological resilience across the YRB.

  3. 3.

    Influencing factors: The factors of domestic water consumption and the total amount of sewage treated emerged as the most frequent critical indicators, appearing five times. Additionally, the amount of wastewater discharged, sewage treatment rate and the harmless treatment rate of consumption waste were identified as key factors affecting the overall ecosystem toughness level in the YRB. These findings emphasize the significant role of water resource management, waste treatment efficacy, and sustainable consumption practices in enhancing or undermining ecological resilience.

Overall, these insights into the ecosystem resilience of the YRB through both temporal and spatial lenses, along with the identified influencing factors, provide a in-depth understanding of the region’s ecological resilience state within the YRB.

Recommendation

In light of the survey results of this research, the following policies and suggestions are given:

  1. 1.

    The government should to formulate targeted environmental protection institutions according to the environmental characteristics and resource endowments of the provinces in the YRB. For tourism-oriented cities (Shaanxi and Shanxi), the administration should encourage the advancement of green buildings and eco-tourism. Water resource preservation and efficient utilization should be given priority in Gansu and Henan (water-shortage provinces).

  2. 2.

    The higher government should further promote the cooperation and exchanges between the southeast and northwest regions, to promote the complementary and sharing of resource advantages and promote the synergetic development of the area. For example, the planning department can formulate relevant incentive measures to attract investment and talent to the northwest region.

  3. 3.

    It is suggested to adopt the regional management model for urban ecological environment governance. For the central and northwest areas with low resilience degrees, future urban planning should focus on the development ahd protection of ecological zone to strengthen the resilience of urban ecosystems. The southeastern region with developed technology and economy should pay attention to industrial upgrading and transformation to decrease the affect of high-emission enterprises on the environment.

  4. 4.

    The provincial government should further promote and introduce advanced sewage treatment technologies and processes, such as biological treatment technology, membrane technology, and ultraviolet purification, to enhance sewage treatment efficiency and water quality compliance rate. In addition, investment in harmless treatment facilities for consumption wastes should be increased, including incineration facilities, disinfection equipment, and landfills.

Limitations and future work

This study evaluates urban ecosystem resilience using a framework centered on resistance, adaptability, and recovery. However, due to its complex nature, the current study cannot include every single aspect that impacts on urban ecosystem resilience. Specifically, the incomplete integration of socio-economic factors, which are crucial for understanding the dynamics of urban ecosystems, might influence the results obtained. Urban areas, as centers of human activity, are profoundly influenced by human behaviors. The interaction between humans and ecological systems thus becomes a critical component of urban ecological resilience. Future research focusing on the interplay between socio-economic factors and urban ecological systems will not only provide a more comprehensive understanding of urban resilience but also offer valuable insights into devising more effective urban management strategies. By recognizing the significance of human interventions, future studies can contribute to developing more resilient, sustainable, and livable urban environments. In addition, a potential direction for future research would be examining the trends in urban ecological resilience over extended time periods.