Introduction

Ecosystem assets, defined as continuous spatial areas characterized by distinct biotic and abiotic components and their interactions1, are foundational to the provision of ecosystem services2,3, directly affecting the benefits derived by humans4,5. In recent decades, the Yellow River Basin, significantly impacted by human activities6,7, has undergone substantial ecological transformations, primarily due to increasing anthropogenic pressures8. Urban expansion and agricultural development have prompted shifts in ecosystem types and reductions in ecological land9,10,11, affecting both the extent and quality of ecosystem assets.

The growing intensity and complexity of human activities globally present diverse impacts on ecosystems12,13,14. In response to escalating environmental threats in the Yellow River Basin15, the Chinese government has initiated extensive ecological conservation projects to mitigate environmental degradation and promote ecosystem restoration16,17. these efforts have resulted in significant improvements in vegetation coverage, ecological quality, water yield, and habitat quality across the basin18,19,20,21, suggesting a positive trajectory for the condition of ecosystem assets22,23,24. However, a thorough evaluation of these assets, considering both their spatial extent and quality, requires the formulation of an integrated index25,26.

Previous research highlight that while ecological conservation efforts can enhance ecosystem assets, human development frequently leads to their degradation. Therefore, human activities exert a dual impact on ecosystem assets within the Yellow River Basin. To understand these impacts, it’s crucial to employ indicators that reflect the aggregate effects of various human pressures. The human footprint metric, which quantifies the combined impact of urbanization, agriculture, and transportation networks27,28, provides a comprehensive measure of human influence beyond individual land use activities29,30. Despite its potential, the human footprint has rarely been utilized to fully assess the anthropogenic effects on ecosystem assets.

While previous research has highlighted the dual impact of human activities on ecosystem assets in the Yellow River Basin, a comprehensive understanding of their long-term spatiotemporal dynamics remains limited. Most studies have focused on individual aspects of ecosystem changes or human influences, without fully integrating these factors to assess their complex interplay over an extended period. Furthermore, the use of human footprint data to quantify aggregate anthropogenic effects on ecosystem assets has been underutilized in this region. This study aims to address these gaps by:

  1. 1.

    Constructing a comprehensive ecosystem asset index for the Yellow River Basin, spanning from 2001 to 2020, to capture both spatial extent and quality of ecosystem assets.

  2. 2.

    Integrating this index with human footprint maps to analyze the spatiotemporal dynamics of ecosystem assets and human activities within the basin.

  3. 3.

    Employing advanced landscape pattern metrics to quantify changes in landscape patterns of ecosystem assets under varying levels of human modification.

  4. 4.

    Investigating the complex relationship between human activities and ecosystem assets through detailed global and local correlation analyses.

By addressing these objectives, this study provides a nuanced assessment of ecosystem assets dynamics in the Yellow River Basin over two decades. Our findings offer valuable insights for quality development in the region and may provide a scientific basis for watershed ecological protection and sustainable management in areas heavily influenced by human activity but on a path to recovery. This comprehensive approach contributes to a deeper understanding of socio-ecological systems in human-modified landscapes, with potential implications for similar regions globally.

Materials and methods

Study area overview

The Yellow River Basin of China, extending across longitudes 95°52’ to 119°05’E and latitudes 32°10’ to 41°50’N (Fig. 1), with a total basin area of 795,000 km2, stands as the country’s second-largest river system and ranks as the fifth largest globally. As an expansive basin, it forms a natural bridge between the Qinghai-Tibet Plateau, the Loess Plateau, and the North China Plain. Its terrain gradually descends from west to east, marked by a dramatic altitudinal gradient that encompasses three climatic zones, eight climate regions, and 15 sub-climate regions, fostering a diverse array of ecosystem types. The basin spans a geographically vast area in eastern and western China. The upper reaches are water conservation areas, which belong to national key ecological protection zones with sparse human population. While the middle and lower reaches cover three large-scale urban clusters: Guanzhong, Zhongyuan, and Shandong Peninsula, where there is a huge difference in human activities. Grasslands dominate the ecosystem’s land cover within the basin, while forests are predominantly located in the southern areas, and majority of population resides in the central and lower regions.

Fig. 1
figure 1

Study area (showing the major ecosystem types in China’s Yellow River Basin in 2020). The map was created by the authors using ArcGIS Pro 3.1 (https://www.esri.com/en-us/arcgis/products/arcgis-pro/).

Full size image

Data collection and sources

The China land cover data (CLCD) for the years 2001 to 2020 utilized in our study developed by Wuhan University, based on Landsat images from Google Earth Engine31. Overall accuracy of CLCD reached 79.31% based on 5463 visually interpreted samples. The dataset covering the distribution of vegetation biomass in China for the same period was sourced from the Pixel Information Expert Engine (https://engine.piesat.cn/), with dataset retrieval id being “EMDO/CHINA_AGB”. Its overall prediction accuracy is R2 = 0.72, RMSE = 33.90 Mg/ha. Furthermore, data on vegetation coverage for the years 2001 to 2020 was sourced from the National Tibetan Plateau Data Center (https://cstr.cn/18406.11.Terre.tpdc.300330)32. Additionally, the human footprint data, which provides a global annual overview of terrestrial human impact for these years, was obtained from an extensive dataset that tracks environmental impacts caused by humans33. Validation against visually interpreted samples yielded an improved correlation, with an R2 of 0.62.

Analytical framework

This study utilized land use, vegetation biomass, and vegetation cover datasets to analyze ecosystem dynamics in the Yellow River Basin from 2001 to 2020 (refer to Fig. 2). We constructed an ecosystem asset index and incorporated the human footprint as a proxy for human activities. Employing trend analysis, landscape pattern analysis, and comprehensive correlation analyses, we assessed the spatiotemporal dynamics of ecosystem assets in the Yellow River Basin under the influence of human activities during this two-decade period.

Fig. 2
figure 2

Analytical framework.

Full size image

Quantifying ecosystem asset variations

Ecosystem assets can be measured using both physical and monetary units34, and the choice depends on the research objective. Our study focuses on the physical dimensions of forest, shrub, and grassland ecosystems, specifically their area and quality. Initially, the land use raster dataset is segmented into seven principal land use types, which serve as proxies for ecosystem categories, as depicted in Fig. 1. The extent of these land use types from 2001 to 2020 is meticulously calculated to represent the spatial distribution and dynamics of the corresponding ecosystems. Subsequent analysis of ecosystem type transitions delineated areas of ecological degradation and recovery (Table 1)1,35. To assess ecosystem quality, pixel-based relative biomass density was employed as a primary indicator for forests and shrublands, while vegetation coverage served as the metric for grasslands 1. Ecosystem assets were then categorized into five quality levels: excellent, good, medium, poor, and bad, based on specific evaluation criteria and grading standards (Table 2). All data processing and visualization were conducted using the Google Earth Engine (GEE) Python API, ensuring a streamlined and effective analysis.

Table 1 Classification of ecological degradation and restoration types.
Full size table
Table 2 Evaluation indicators, methods, and grading standards for ecosystem asset quality.
Full size table

Construction of the ecosystem asset index

Ecosystem asset index can comprehensively evaluate the condition of forest, shrubland, and grassland ecosystem assets26, offering a more detailed perspective on the regional ecosystem assets than merely examining their area or quality. In our study, based on approximately 800,000 1 km grid cells, the ecosystem asset index for the Yellow River Basin covering the period from 2001 to 2020 was computed using the Google Earth Engine (GEE) Python API. This comprehensive ecosystem asset index is used as proxy for ecosystem assets and classified into five categories: excellent, good, medium, poor, and bad (Table 3). It facilitates a thorough evaluation of ecosystem assets within our study area by considering both the spatial distribution and quality of these essential ecological resources.

Table 3 Calculation methods and grading standards for ecosystem asset index.
Full size table

Human activity proxy via human footprint

Human footprint is used as proxy for human activities, quantifying the overall impact of direct human influences on the environment. This metric incorporates data on built environments, population density, nighttime lights, agriculture and pasture lands, and transportation infrastructure (roads, railways, waterways), with assigned scores reflecting their relative impacts28. The human footprint can be utilized to gain a deeper understanding of the impacts human activities have on the planet36, helping better actions for biodiversity conservation37,38, landscape planning39,40, and resources recycling41. Human footprint data spans an extensive temporal range and maintains a high degree of temporal consistency, facilitating analysis of global scale changes33. However, it is essential to acknowledge that the human footprint does not encompass all human pressures and that its measurement of human impact on nature may not directly correlate with ecosystem or biodiversity responses, which can vary spatially. To explore dynamics of human activity, different types areas are classified based on human footprint. These classifications included intact areas (human footprint < 4), low modified areas(4 ≤ human footprint < 10), moderate modified areas(10 ≤ human footprint < 20), high modified areas(20 ≤ human footprint < 30) and extreme modified areas(30 ≤ human footprint ).

Trend analysis in human activities and ecosystem assets

To accurately assess the trends of human activities and ecosystem assets throughout the study period, a rigorous statistical methodology was implemented. The Sen’s Slope Estimator, a non-parametric method, was utilized for determining the trends in human activities and ecosystem assets42. It facilitates the calculation of temporal trends without assuming a particular distribution for the data. Furthermore, the Mann-Kendall test43, a widely-used non-parametric test, was employed to determine statistical significance of observed trends. By coupling Sen’s slope estimator with Mann-Kendall test by the Google Earth Engine (GEE) Python API, a robust and statistically sound assessment of change trends and their significance levels was achieved. A significant change was indicated by a trend with |Z| < 1.96 at the 95% confidence level. Sen’s slope estimation calculation formula is:

$$\:S=median\left(\frac{\left({X}_{j}-{X}_{i}\right)}{j-i}\right),\left(2001\le\:i<j\le\:2020\right)$$
(1)

where S is utilized to represent changing trend, Xi and Xj represent values in years and .

Mann–Kendall nonparametric statistical test described as follows:

$$\:Z=\:\left\{\begin{array}{c}\frac{S-1}{\sqrt{Var\left(S\right)}},\:S>0\\\:0,S=0\\\:\frac{S+1}{\sqrt{Var\left(S\right)}},S<0\end{array}\right.,$$
(2)
$$\:S=\sum\:_{i=1}^{n-1}\:\sum\:_{j=i+1}^{n}sign\left({X}_{j}-{X}_{i}\right),$$
(3)
$$\:Var\left(S\right)=\frac{n(n-1)(2n+5)}{18},$$
(4)
$$\:sign({X}_{j}-{X}_{i})=\left\{\begin{array}{c}1,{X}_{j}-{X}_{i}>0\\\:0,{X}_{j}-{X}_{i}=0\\\:-1,{X}_{j}-{X}_{i}<0\end{array},\right.$$
(5)

where statistic Z can evaluate significance.

Landscape pattern assessment

To illustrate changes in landscape patterns tied to different classifications of ecosystem assets and human activities, and to evaluate the dynamics of anthropogenic impacts on ecosystem assets, five key indices were chosen from landscape pattern metrics: Area (AREA), Number of Patches (NP), Area Means (AREA-MN), Largest Patch Index (LPI), and Landscape Shape Index (LSI)44,45,46. These indices were computed to reveal the landscape pattern dynamics linked to different grades (types) of ecosystem assets and human activities, facilitating a detailed examination of the changing anthropogenic effects and general trends of ecosystem assets in our study area. The Google Earth Engine (GEE) Python API, along with pylandstats-2.4.2, was utilized for the computation of these five landscape pattern indices (Table 4).

Table 4 Landscape pattern indices and descriptions.
Full size table

Correlation analysis between human activities and ecosystem assets

To investigate the relationship between human activities and ecosystem assets, the Pearson correlation analysis method was employed to quantify overall correlation between these two crucial variables across study area. Additionally, to understand the spatial variation in the relationship more clearly, cross-correlation analysis was also conducted. It facilitated a pixel-by-pixel comparison of similarity between two raster datasets, revealing localized dynamics of correlation47,48. The spatial correlation between human activities and ecosystem assets was mapped with precision at a 1 km grid scale, offering insights into the intricate dynamics influencing their interaction.

Results

Spatial changes in ecosystem composition (2001–2020)

Between 2001 and 2020, we mapped the distribution of various ecosystem types across the Yellow River Basin (refer to Fig. 3). During these two decades, grasslands, farmlands, and forests emerged as the predominant ecosystems in terms of basin. Both forested and urban areas exhibited a discernible expansion, with the former increasing by 12,973.87 km2 and the latter by 10,401.07 km2. On the contrary, farmland and other land categories experienced a significant contraction, with reductions of 19,342.02 km2 and 8,857.66 km2, respectively.

Grasslands consistently accounted for over half of the basin’s total area (56.70–58.20%), thereby representing the most prevalent ecosystem. The combined coverage of the three largest ecosystems—grasslands, farmlands (23.19–25.87%), and forests (9.78–11.38%)—consistently made up more than 90% of the area, though a slight downward trend was observed. Meanwhile, the proportion of wetlands experienced a small increase (+ 0.22%), whereas shrublands saw a minor decline (− 0.19). Despite the ongoing urbanization, the areas covered by forests and grasslands remained relatively stable. However, the reduction in farmland might highlight potential concerns for food security in the region.

Fig. 3
figure 3

Area and proportion of ecosystems in the Yellow River Basin from 2001 to 2020. The line graph shows change in area of different ecosystems, while bar graph shows change in proportion of different ecosystem areas.

Full size image

Ecosystem transformation and restoration dynamics

Ecosystem transformations within the Yellow River Basin from 2001 to 2020 identified that the exchange between grasslands and farmlands represented the most significant portion of change, comprising 52.4% of the total ecosystem alterations (Fig. 4a). Specifically, the conversion from farmlands to grasslands surpassed the opposite direction by 8,791.6 km2. Additionally, farmlands were the main contributors to the expansion of urban areas, with 8,541.89 km2 of farmlands being repurposed for urban development. Overall, returning farmland emerged as primary driver of basin-wide ecosystem changes, accounting for 41.22% of total change area. Urban expansion and grassland restoration processes were predominantly driven by returning farmland.

Evaluation of ecosystem degradation and restoration dynamics indicated that area of ecosystem degradation was 48,723.27 km2, whereas area of ecosystem restoration was 69,299.15 km2, with the latter substantially exceeding the former degradation extent (Fig. 4b,c). Farmland cultivation was identified as the main cause of ecosystem degradation, accounting for 58.44% of total degraded area, predominantly in the middle and lower basin areas. Following this, urban expansion and vegetation deterioration accounted for 39.2% of the total degradation, with urban growth focused in valley regions and vegetation decline evident across the Loess Plateau (Fig. 4b). In contrast, returning farmland facilitated 57.89% of total ecosystem restoration, mainly around urban outskirts in the middle and lower basin areas. Afforestation and vegetation restoration followed, contributing to 38.39% of total restored area, mainly taking place on the Loess Plateau. The dynamics of ecosystem degradation and recovery were concentrated in the basin’s middle and lower reaches, with farmland changes being primary driving factor (Fig. 4c).

Fig. 4
figure 4

Transition and spatial distribution of ecosystem types in the Yellow River Basin from 2001 to 2020. (a) shows changes in ecosystem types; (b) and (c) show spatial distribution and proportion of ecosystem degradation and restoration, respectively. The map was created by the authors using ArcGIS Pro 3.1 (https://www.esri.com/en-us/arcgis/products/arcgis-pro/).

Full size image

Improvements in ecosystem asset quality

To understand the variations in the quality of ecosystem assets across forests, shrublands, and grasslands within the Yellow River Basin between 2001 and 2020, a detailed spatial analysis was conducted. Results showed that, from 2001 to 2020, the quality of forest ecosystem assets increased annually, with 69.5% of basin experiencing improvement concentrated primarily in the southern middle reaches (Fig. 5a). The area classified as excellent grade expanded by 22,968.17 km2, representing 95.6% of the change in forest areas and reflecting an overall positive condition of forest assets (Fig. 5d).

Shrubland ecosystems similarly experienced marked quality improvements, with 57.9% of their area showing positive changes. Improvement and degradation areas were mixed within the middle reaches (Fig. 5b). Despite a decrease in total area of shrubland ecosystem assets by 1245.52 km2, proportion of excellent grade escalated to 98.7% (Fig. 5e). It suggests that although the quantity of shrubland assets has decreased, their quality has significantly improved.

Grassland ecosystems across the basin also exhibited significant quality improvements, with 80.5% of the basin experiencing enhancement widely distributed across middle and upper reaches (Fig. 5c). The extent of areas rated as excellent grade increased by 101,321.1 km2, raising from 49.6 to 67.3% (Fig. 5f). Meanwhile, good and medium quality categories represented 16.3 and 10.9% of the grassland areas, respectively. Overall, a large-scale improvement in the quality of grassland ecosystem assets was indicated, demonstrating a positive trend in the condition of grassland ecosystems across the Yellow River Basin.

Fig. 5
figure 5

Changes in ecosystem asset quality and grades in the Yellow River Basin from 2001 to 2020. (a)–(c) represent spatial distribution of changes in quality of forest, shrubland, and grassland assets, respectively. (d)–(f) show area changes of different quality grades for forest, shrubland, and grassland assets, respectively. The map was created by the authors using ArcGIS Pro 3.1 (https://www.esri.com/en-us/arcgis/products/arcgis-pro/).

Full size image

Human activity trends and their ecological impact

From 2001 to 2020, human activity trends and their impacts on the ecosystem assets of the Yellow River Basin were quantified. A rising human footprint (HFP) across basin during study period was revealed (Fig. 6a,d). Mean HFP increased from 12.58 to 14.75, marking a 17.25% uptick, with 66.3% of the basin (21.1% at 95% confidence level) showing intensified anthropogenic activity, especially in the middle and lower reaches. Only 5.1% remained unchanged, while 28.5% (1.1% at 95% confidence) experienced weakening impacts, primarily across the Loess Plateau.

Meanwhile, the ecosystem asset quality within the basin, as depicted by the comprehensive ecosystem asset index (EQ), showed significant improvements (Fig. 6b,e). The mean EQ value rose from 62.24 to 76.31, marking a growth of 22.61%. This enhancement was observed across 68.9% of the basin (46.9% at 95% confidence), especially in the middle to upper reaches. Meanwhile, 27.2% of the region, mainly located in the southeast and southwest sectors, remained stable, and a negligible 3% (1.1% at 95% confidence level) underwent degradation, indicating an overall enhancement of ecosystem conditions across the basin.

Spatial distribution of concordance between human activity and ecosystem trends demonstrated that approximately 45.88% of the basin experienced a simultaneous intensification of both human activities and ecosystem asset quality (Fig. 6c,f). A contrasting pattern emerged in the Loess Plateau, where a decrease in human activities coincided with an enhancement in ecosystem assets, affecting 20.48% of the area. Furthermore, 17.61% of the basin, primarily in the middle and lower reaches, experienced heightened human activities alongside stable ecosystem conditions. Overall, the trends in human activities and ecosystem assets exhibited relatively strong alignment, suggesting coordinated dynamics between these factors.

Fig. 6
figure 6

Trends of human activities and ecosystem assets in the Yellow River Basin from 2001 to 2020, where “significant” trends indicate a confidence level of 0.05. (a) and (b) show spatial distribution of trends in human activities and ecosystem assets, respectively; (d) and (e) show annual changes in human activities and ecosystem assets, respectively; (c) and (f) show spatial distribution and proportion of the consistency between human activity and ecosystem asset trends. The map was created by the authors using ArcGIS Pro 3.1 (https://www.esri.com/en-us/arcgis/products/arcgis-pro/).

Full size image

Landscape patterns influenced by human activities and conservation efforts

From 2001 to 2020, we assessed the influence of human activities and conservation efforts on the landscape patterns within the Yellow River Basin, utilizing a suite of indices including area (AREA), number of patches (NP), area means (AREA-MN), the largest patch index (LPI), and the landscape shape index (LSI). Results revealed that dynamic shifts across various land modifications due to human actions (refer to Fig. 7a).

Areas subject to extreme modification increased significantly (+ 42,148.53 km2, 105% growth), while high modified areas extents increased slightly (+ 19,893.81 km2, 18.57%). In contrast, regions with moderate disturbance initially experienced a reduction (2001–2013) but demonstrated growth from 2013 to 2020. Areas with minimal or intact landscapes experienced reductions in size (−50,826.78 km2, −15.60% and − 11,323.60 km2, −15.13%, respectively), reflecting the extensive reach of human influence. The trajectories for NP and LSI across different intensities of modification (Fig. 7c,f) generally aligned, with extreme and high modification areas experiencing slight increases, moderate and low intensity areas showing a decline post-2018, and intact areas remaining constant. The AREA-MN (Fig. 7d) for intact areas varied, while areas of moderate and low intensity saw expansion post-2018. LPI (Fig. 7e) for moderate modified areas fluctuated significantly, low modified areas displayed a downward trend, and extreme modified areas saw an uptick in 2018. Overall, human activities showed an intensifying trend, especially in areas undergoing the most severe modifications.

Regarding ecosystem assets (as depicted in Fig. 7b), excellent grade expanded markedly (+ 217,332.36 km2, 74.54% growth) while good (−61,912.18 km2, −39.32%), medium (−63,021.89 km2, −42.06%), poor (−72,612.27 km2, −66.52%), and bad (−19,786.02 km2, −19.62%) quality grades experienced contractions. NP showed varying declines across the excellent, medium, and poor grades (Fig. 7g). Variability in the AREA-MN and LPI for excellent quality areas increased (Fig. 7h,i), suggesting a broadening of high-quality ecological spaces. The LSI for grades other than good moderately decreased (Fig. 7j), indicating an enhancement in ecological integrity. The considerable growth and spatial prevalence of excellent quality ecosystem assets, presenting a broad trend of ecological improvement across the basin’s terrestrial ecosystems over the last two decades.

Fig. 7
figure 7

Changes in landscape patterns of different grades of human activities and ecosystem assets in the Yellow River Basin from 2001 to 2020. (a) and (b) show changes in area (AREA) of different grades of human activities and ecosystem assets, respectively; (c−f) show changes in NP, AREA-MN, LPI, and LSI for different grades of human activities, respectively; (g–j) show changes in NP, AREA-MN, LPI, and LSI for different grades of ecosystem assets, respectively.

Full size image

Interactions between human activities and ecosystem assets

The Pearson correlation analysis of the relationship between human activities and ecosystem assets in the Yellow River Basin from 2001 to 2020 (Fig. 8a) revealed a negative correlation, which displayed an increasing but fluctuating trend. The correlation coefficient shifted from − 0.19 to −0.25 over the period. Results from the cross-correlation test indicated that the frequency of negative correlations also followed an upward trend (Fig. 8b), moving from 47.03% in 2001 to 53.27% in 2020. Nevertheless, the average value of the cross-correlation coefficients exhibited a declining trend, moving from − 0.27 to -0.23. In terms of the spatial distribution of these cross-correlation coefficients (Fig. 8c,d), negative correlations were predominantly found in the middle and lower reaches of the basin, with the area showing coefficients lower than − 0.5 diminishing over time. When analyzing the data density and distribution trends of the Human Footprint (HFP) and Ecosystem Quality (EQ) indices (Fig. 8e,f), it was observed that low HFP typically corresponded to high EQ. However, the high-density distribution area of both variables shifted rightward, indicating an improvement in ecological quality in areas with moderate human activity. This shift suggests that even in regions with general human presence, ecosystem assets have increased, reflecting an enhanced coordination between human activities and ecosystem quality.

Fig. 8
figure 8

Correlation between human activities and ecosystem assets in the Yellow River Basin. (a) shows Pearson correlation coefficients and p-values from 2001 to 2020; (b) shows distribution and proportion of cross-correlation coefficients; (c) and (d) show spatial distribution of cross-correlation coefficients in 2001 and 2020, respectively; (e) and (f) show data density and distribution trends of two variables in 2001 and 2020, respectively. The map was created by the authors using ArcGIS Pro 3.1 (https://www.esri.com/en-us/arcgis/products/arcgis-pro/).

Full size image

Discussion

Gains and losses of ecosystem assets

Investigating gains and losses of ecosystem assets under anthropogenic pressures is crucial for informing ecological conservation and promoting sustainable socioeconomic development. Our study examined spatiotemporal changes in extent and quality of basin ecosystems from 2001 to 2020 (Figs. 3, 4). Chinese government has dedicated efforts to restoring degraded lands, notably through the Grain for Green program since 1999, facilitating large-scale conversion of farmlands to forests, shrublands, and grasslands49,50. Throughout study period, grasslands persisted as the predominant ecosystem (Fig. 3), while farmlands experienced the most pronounced changes (Fig. 4a). Farmlands reversion to ecological land uses (forests, shrublands, and grasslands) increased ecosystem extent by 15,994 km2 (3% growth). Simultaneously, farmland cultivation also emerged as the principal driver of ecosystem degradation across the basin (Fig. 4b), potentially reflecting Chinese government’s " balance of cultivated land occupied and complimented” policy aimed at cultivated land conservation51,52. However, we found that urban expansion occupied 8541.89 km2 of farmlands, while farmlands occupied a total of 27,891.79 km2 of ecological land uses, with compensation share of farmlands accounting for only 30.63%, indicating policy factors alone cannot fully explain these dynamics.

The evolving landscape of farmland is critical for regional development and exhibits a general decreasing trend (Fig. 3). While this reduction may raise concerns about food security, it is important to consider the complex factors driving these changes. Some farmlands may have been abandoned due to unsuitable soil conditions, and advancements in agricultural technology over the past two decades have mitigated the impact on food production53. For instance, precision agriculture and improved crop varieties have significantly boosted productivity, potentially offsetting the loss of farmland54. Additionally, more than 8791.6 km² of farmland was converted to grassland compared to the reverse (Fig. 4), primarily due to policies promoting ecological restoration and natural succession of abandoned farmlands. Farmland also played a significant role in urban expansion, with 8541.89 km² repurposed for urban development (Fig. 4). This substantial shift from old farmlands to urban areas underscores the dynamic interplay between agricultural land use and urbanization. Urban expansion often leads to the loss of high-quality agricultural land; however, the negative impacts of such losses can be mitigated through farmland reclamation in other regions and improvements in agricultural production conditions55. Moreover, this trend reflects the socio-economic development and urbanization patterns driven by population growth and economic activities in the region56.

The observed trends emphasize the need for integrated land management strategies that balance ecological conservation, urban development, and agricultural productivity demands. Effective policies should prioritize optimizing land use to ensure the protection of ecosystem assets, enhance agricultural production through sustainable practices, and accommodate necessary urban development while minimizing negative impacts.

Impacts of human activities on ecosystem assets

Our findings indicate a growing human footprint across a significant portion (66.34%) of the Yellow River Basin during 2001–2020 (Fig. 6a,d), exerting escalating pressures on ecological integrity57. The expansion and fragmentation of highly modified areas (Fig. 7a,c) may severely disrupt ecosystem connectivity, affecting biodiversity conservation and degrading ecosystem services58,59. Paradoxically, despite rising anthropogenic stress, only 3% of basin ecosystems exhibited declining asset quality. More intriguingly, 45.88% of the region experienced simultaneous increases in both human activities and ecosystem assets (Fig. 6f). This unexpected trend highlights the effectiveness of local conservation efforts and underscores the complex, non-linear relationships between human activities and ecosystem health.

Spatial distribution of ecosystem changes reveals distinct patterns that reflect the basin’s socio-economic and geographic diversity. Urban expansion and vegetation deterioration, accounting for 39.2% of total degradation, were concentrated in valley regions and across the Loess Plateau, respectively (Fig. 4b). In contrast, returning farmland facilitated 57.89% of total ecosystem restoration, primarily around urban outskirts in the middle and lower basin areas. This spatial difference likely reflects varying levels of economic development and policy implementation across the basin.

Valley regions, characterized by favorable topography and infrastructure, attract urban development and industrial activities, leading to ecosystem degradation60. The Loess Plateau, despite being the focus of extensive ecological restoration projects, still faces challenges in maintaining vegetation cover, possibly due to its fragile ecosystem and historical land use practices61,62. The concentration of farmland restoration around urban outskirts suggests a “suburbanization” of ecological restoration efforts, potentially driven by land use policies, economic incentives for peri-urban green spaces, and the need to balance urban expansion with ecological preservation63,64,65.

Dynamics of ecosystem degradation and recovery were predominantly concentrated in the basin’s middle and lower reaches, with farmland changes being the primary driving factor (Fig. 4c). This regional contrast highlights the complex interplay between socio-economic development and ecological conservation. The middle and lower reaches, historically more developed and densely populated, experience more intense land use transitions, facing stronger pressures for urban expansion and agricultural intensification while simultaneously being the focus of targeted restoration efforts66. In contrast, the upper reaches, characterized by more pristine ecosystems and lower population densities, may experience less dramatic changes in land use and ecosystem assets67.

Over two decades, improvement in ecosystem asset quality by 22.61%, compared to a 3% increase in total area, emphasizes the value of boosting ecological integrity over merely expanding areas. This is also well illustrated by interesting trends in scrub ecosystems, which have improved in quality despite a decrease in their total area. Success of targeted conservation efforts focusing on enhancing the ecological integrity of existing shrublands rather than expanding their area. Natural succession processes where less robust shrublands are replaced by other ecosystem types (e.g., forests), while the remaining shrublands become more established and ecologically valuable68. This trend in shrubland dynamics exemplifies the importance of considering both quantity and quality in ecosystem asset assessments.

The temporal dynamics of landscape patterns reveal intriguing shifts, particularly around 2018, that warrant further examination. These shifts around 2018 likely reflect the cumulative effects of several key policy initiatives implemented in the preceding years. The continuous promotion of policies such as returning farmland to forest and grassland, ditch control, and the integrated management of mountains, rivers, forests, fields, lakes, and grasslands, along with the launch of rural revitalization projects after 2017, have played a crucial role in optimizing the landscape pattern of the Yellow River Basin7,69,70. These observed changes highlight the importance of policy continuity in driving regional ecological restoration. The lag between policy implementation and observable landscape changes suggests that sustained, long-term commitment to environmental policies is crucial for achieving significant improvements in ecosystem assets.

Correlation analysis revealed an inverse relationship between human activities and ecosystem assets, which has become more pronounced over time (Fig. 8a). Despite a slight improvement in the average cross-correlation coefficient, the proportion of the area with significant negative correlations grew, reflecting a better alignment between socioeconomic progress and ecological measures. However, the persistent negative correlation highlights the adverse effects of increasing human activities, such as urban expansion and intensive farming, which encroach upon and deteriorate natural habitats71,72. Moreover, under a changing climate, high-intensity disturbances could synergistically amplify vegetation stress73,74,75. Consequently, even though there is still increase in degradation, the quality of the ecosystems is improving.

Limitations and future research directions

While we provide a comprehensive spatiotemporal analysis of ecosystem asset dynamics in the Yellow River Basin, several limitations merit consideration for future work. The sole reliance on the human footprint metric to quantify human activities may oversimplify diverse pressures caused by human influence. Incorporating additional indicators could provide more comprehensive insights. Additionally, biomass and vegetation coverage alone are simplified assessments of ecosystem asset quality, and subsequent studies should use more comprehensive factors. Wetland ecosystems play an important role in the watershed and should be taken into consideration.

Moreover, ongoing monitoring and predictive analysis of ecosystem asset trends, considering factors such as climate change and demographic shifts, will be essential for proactive environmental management. To further illuminate the intricate dynamics between human activities and ecosystem health, the integration of advanced analytical methodologies, such as machine learning, is imperative. These tools offer the potential to uncover hidden patterns, predict future trends, and optimize management strategies. By harnessing the power of machine learning, researchers can delve deeper into the complexities of ecosystem asset change, identify early warning signals of degradation, and explore novel approaches to restoration.

Ultimately, a multidisciplinary approach that combines ecological, social, and economic perspectives will be essential to unlocking the full potential of these advanced techniques for achieving sustainable ecosystem management and human well-being.

Conclusions

This study conducted a comprehensive analysis of ecosystem asset dynamics in the Yellow River Basin from 2001 to 2020, elucidating the complex interplay between human activities and ecological integrity. By constructing an ecosystem asset index and integrating it with human footprint data, we revealed significant spatiotemporal changes in the region’s ecological landscape.

Our findings demonstrate that despite intensifying human activities across 66.34% of the basin, ecosystem asset quality improved by 22.61%, with a 3% increase in total ecological area. This paradoxical trend highlights the effectiveness of regional conservation policies while also underscoring the persistent challenges in balancing human development with ecosystem preservation. The conversion of farmlands to ecological land uses, resulting in a 15,994 km² increase, emerged as a key driver of positive change, reflecting the impact of initiatives like the Grain for Green program. Importantly, our research revealed significant spatial heterogeneity in ecosystem changes, with degradation and restoration processes concentrated primarily in the middle and lower reaches of the basin. This spatial variation emphasizes the need for targeted, location-specific conservation strategies that account for local socio-economic and geographic factors. The complex, often non-linear relationships observed between human activities and ecosystem assets challenge simplistic narratives of human-nature interactions. Our finding that 45.88% of the region experienced simultaneous increases in both human activities and ecosystem quality suggests potential for sustainable co-existence, given appropriate management practices.

In conclusion, this study provides crucial insights for promoting high-quality development in the Yellow River Basin and similar regions globally. By demonstrating the importance of integrating spatial and quality considerations in ecosystem assessment and management, our research contributes to a more nuanced understanding of socio-ecological systems in human-modified landscapes. These findings can inform policy-making and land management strategies aimed at achieving a balance between ecological conservation and sustainable socio-economic development.