Long-term mapping and landscape dynamics of major riverine wetlands in the Yellow River Basin from 1986 to 2022

Abstract

Wetlands in the Yellow River Basin face the composite pressures of climate change and human disturbances. However, analyzes of the major riverine wetlands remain far less than those of the headwater and estuarine regions in the Yellow River Basin. Based on the Google Earth Engine platform, we combined a sample migration strategy and Random Forest classification to produce an annual 30 m riverine wetland dataset (1986–2022) for the Yellow River Basin (excluding the headwater and estuarine regions). We then analyzed the spatiotemporal change and driving factors of water bodies, mudflats, and marshes. Results show that long-term wetland maps exhibit high reliability, with overall accuracy above 90%, and an average Kappa coefficient exceeding 0.88. The expansion of water bodies and mudflats failed to offset marsh degradation, leading to a net riverine wetland loss of 1326.30 km², with the total wetland area in 2022 decreasing to 95% of its 1986 level. Increasing temperature and runoff at the Lanzhou station are associated with marsh core loss and fragmentation, accompanied by partial conversion to water bodies and mudflats, thereby threatening wetland biodiversity and carbon sequestration capacity. Water body expansion and increased connectivity are primarily associated with runoff at the Lanzhou station, average temperature, and precipitation. Landscape change of mudflats in the midstream and downstream is associated with average temperature, sediment discharge at the Huayuankou station, runoff at the Toudaoguai station, and regional human activities. This study provides critical data support and spatiotemporal evidence for riverine wetland conservation and restoration in the Yellow River Basin.

Keywords

riverine wetland mapping marsh landscape pattern driving factors sample migration

Introduction

Wetlands are among the most critical ecosystems on Earth. Despite occupying only 2.6% of the terrestrial surface, wetlands account for more than 20% of global organic carbon storage and provide indispensable ecological functions, including flood regulation, water purification, and biodiversity conservation (Bao et al., 2023; Bar-On et al., 2018; Osland et al., 2022). However, wetlands are highly sensitive and vulnerable to environmental changes. Under escalating climate change and intensifying human disturbance, wetlands face growing risks of degradation and loss driven by hydrological extremes, sediment accumulation, and agricultural and pastoral expansion (Sandi et al., 2020; Schuerch et al., 2018), which directly or indirectly constrain progress toward 75 Sustainable Development Goal targets (Wetlands, 2021). Although a series of conservation and restoration efforts have been undertaken, including the Everglades restoration, mangrove rehabilitation in Indonesia, and large-scale wetland restoration programs in China (Liu et al., 2024; Schneider et al., 2024; Suhardi et al., 2024), global wetlands have continued to decrease since 2000, particularly in China, Europe, and the United States (Ai et al., 2023; Fluet-Chouinard et al., 2023). Considering the large geographical differences among the world’s big rivers, there is a clear practical requirement to map the spatiotemporal changes of wetland area and landscape patterns, and to identify their regional dominant driving factors based on the specific geographic condition of the target river.

Accurately identifying the wetland types and spatial locations is fundamental to dynamic wetland monitoring. However, spectral similarity between wetland and terrestrial vegetation, as well as among wetland subtypes, substantially complicates accurate classification (Adam et al., 2010; Taddeo et al., 2019). Meanwhile, regional heterogeneity driven by topography and hydrology constrains the spatial transferability of sample points and classification models (Wijesingha et al., 2024). Although various optimization strategies have been proposed, including multi-temporal classification models incorporating vegetation phenology (Sun et al., 2023), zonal mapping approaches (Mao et al., 2020), and automated sample extraction based on multi-source data (Wang et al., 2023), long-term annual wetland mapping remains constrained by the limited availability of sufficient and reliable historical training samples. Recent advances have used the Google Earth Engine (GEE) to automate training sample migration across time, under the assumption that spectrally similar pixels represent unchanged land cover (Fekri et al., 2021; Huang et al., 2020). However, thresholds derived solely for optimal classification accuracy do not guarantee temporal consistency of land-cover types, leading to decreased robustness when applied over extended time periods. Therefore, developing sample migration strategies tailored to the unique ecological characteristics and spatiotemporal heterogeneity of different wetland types remains particularly important for long-term mapping.

The Yellow River Basin encompasses 3.91 million hectares of wetlands, which play a vital role in maintaining the regional water supply, supporting human livelihoods, and conserving rare and endangered species (Yun et al., 2022). Nevertheless, the Yellow River Basin is located in a high human pressure area, where wetland degradation is widespread, and exhibits pronounced spatial heterogeneity (Wang et al., 2025). In the headwater region, the Ruoergai wetlands experienced severe degradation, with 38.9% of the wetland area degraded by the late twentieth century (Gao et al., 2016), primarily driven by grazing activities, warm-season temperature, and annual precipitation variability (Mao et al., 2022). In the estuarine region, wetland dynamics are primarily driven by human activities, including reclamation projects, agricultural expansion, and urbanization, which intensify landscape fragmentation and disrupt hydrological connectivity (Zhang et al., 2015). However, compared with the extensive attention given to the headwater and estuarine regions (Bai et al., 2013; Xiao et al., 2010; Zhang et al., 2021), the major riverine wetlands of the Yellow River basin (excluding the headwater and estuarine regions) remain insufficiently studied. In addition, the lack of long-term and consistent wetland datasets hinders a comprehensive understanding of the relative contributions of hydraulic engineering, climate variability, and human activities to the major riverine wetland landscape change in the Yellow River Basin.

To reveal the spatiotemporal change and dominant drivers of riverine wetland of the Yellow River Basin, we conducted the following analyses: (1) Constructed a 30 m resolution riverine wetland classification dataset for the Yellow River Basin from 1986 to 2022; (2) Analyzed the spatiotemporal change in the areas and landscape patterns for three riverine wetland types, including water bodies, mudflats, and marshes; (3) Quantitatively assessed the relative contributions of natural and anthropogenic drivers to changes in wetland area and landscape patterns. Collectively, these analyses provide spatiotemporal evidence to support riverine wetland landscape conservation and adaptive management strategies in the Yellow River Basin.

Materials and methods

Study area

We focused on the riverine wetlands of the Yellow River Basin, excluding alpine peatlands in the headwater region and delta wetlands in the estuarine region, since the wetland types in the headwater and estuarine regions differ and have already been mapped. The study area spans from 102°42′ to 119°09′ E and from 31°20′ to 40°48′ N (Figure 1(a)), covering elevations from −8 to 3805 m, with an average annual temperature of 9°C and annual precipitation ranging from 300 to 600 mm. Based on the locations of the Toudaoguai and Huayuankou hydrological stations, the study area was subdivided into the upper, middle, and lower reaches of the Yellow River Basin (Figure 2).

Figure 1.

Situation of the study area. (a) Location and elevation of study area; (b) Number of remote sensing image pixels and images distribution in the Yellow River Basin (excluding the headwater and estuarine regions).

Figure 2.

Maximum boundaries for riverine wetland extraction in the Yellow River Basin (excluding the headwater and estuarine regions). The boundaries were used only to separate riverine wetland sections associated with different hydrological stations and do not represent formal hydrological or administrative boundaries of the basin.

Riverine wetlands in the Yellow River Basin (excluding the headwater and estuarine regions) are primarily composed of water bodies, mudflats, and marshes (Cherry, 2011). Water bodies are defined as natural or artificial areas characterized by permanent or seasonal inundation. Mudflats are flat or gently inclined surfaces formed by the deposition of fine-grained sediments under low-energy hydrological conditions and are alternately exposed and submerged, contributing to flow regulation, wave attenuation, bank stabilization, and the filtration of suspended sediments. Marshes are typical low-elevation wetlands occurring along riverbanks and provide essential ecosystem services, such as habitat provision, water regulation, carbon sequestration, and soil conservation. These three wetland types were identified as the primary targets for riverine wetland mapping in the Yellow River Basin (excluding the headwater and estuarine regions).

Data source and preprocess

Remote sensing images for the Yellow River Basin were obtained from the Landsat-5 TM, Landsat-7 ETM+, and Landsat-8 OLI sensors at a spatial resolution of 30 m and are freely available through the GEE platform. We used the Level-2 Collection 2 Tier 1 products, which have undergone radiometric and geometric corrections by the United States Geological Survey (USGS) and can be directly used for supervised wetland classification following cloud removal and spatial clipping. In total, 16,762 remote sensing images acquired during the growing season (May–October) from 1986 to 2022 were selected for the study area. Specifically, 10,620 images were obtained from Landsat-5 TM (1986–2010), 1219 from Landsat-7 ETM+ (2011–2013), and 4923 from Landsat-8 OLI (2014–2022) (Figure 1(b)). Given the relatively low cloud cover and high image quality, we performed median compositing using images from adjacent years during the growing season. Because the Landsat image acquired in May 2011 exhibited pronounced striping, only median composites generated from June to October were used for that year.

Additional statistical datasets were used to analyze the drivers of riverine wetland area and landscape change. Climatic variables include the monthly average temperature and precipitation datasets at 1 km resolution from the National Earth System Science Data Center (geodata.cn), which were clipped to the study area for the period 1986–2022 using Python 3.10. Hydrological variables include the runoff and sediment discharge data from the Lanzhou, Toudaoguai, and Huayuankou gauging stations (1986–2022), which were obtained from the Yellow River Conservancy Commission.

Extraction of the maximum riverine wetland boundary

We selected the maximum riverine wetlands boundary from 1986 to 2022 in the Yellow River Basin (excluding the headwater and estuarine regions) as the spatial extent for wetland classification. We first extracted the annual water boundary ( $C_{w a t e r}$ ) based on the following algorithm:

C_{w a t e r} = (m N D W I > E V I o r N D W I > N D V I) a n d (E V I < 0.1)

(1)

N D V I = \frac{ρ_{n i r} ‐ ρ_{r e d}}{ρ_{n i r} + ρ_{r e d}}

(2)

EVI = 2.5 \times \frac{ρ_{n i r} ‐ ρ_{r e d}}{ρ_{n i r} + 6 \times ρ_{r e d} ‐ 7.5 \times ρ_{b l u e} + 1}

(3)

N D W I = \frac{ρ_{g reen} ‐ ρ_{n ir}}{ρ_{g reen} + ρ_{n ir}}

(4)

mNDWI = \frac{ρ_{g reen} ‐ ρ_{s w i r 1}}{ρ_{g reen} + ρ_{s w i r 1}}

(5)

where NDVI is the Normalized Difference Vegetation Index (Huang et al., 2021), EVI is the Enhanced Vegetation Index (Gurung et al., 2009), NDWI is the Normalized Difference Water Index (Gao, 1996), and mNDWI is the modified Normalized Differential Water Index (Xu, 2006). ρ_blue, ρ_green, ρ_red, ρ_nir, and ρ_swir1 represent the blue, green, red, near-infrared, and shortwave infrared 1 bands of Landsat 5/7/8 sensors, respectively. Spatial overlays of annual water body extents were performed to derive the maximum water boundary for the study area from 1986 to 2022. We removed water patches smaller than 1 km², then created a 500 m inward buffer to eliminate internal void patches, followed by a 500 m outward buffer. The final result represents the maximum extraction boundary of riverine wetlands in the study area (Figure 2). Wetland samples of water bodies, marshes, and mudflats were collected within this boundary for subsequent classification.

Riverine wetland mapping of the Yellow River Basin

Reference-year wetland mapping provides the baseline for long-term mapping and critically influences the dataset accuracy. First, Landsat images were preprocessed in GEE through cloud masking removal, temporal median compositing, and spatial clipping. Visual interpretation was conducted using false-color composite images and selected riverine wetland sample points. A Random Forest classification model was then trained using spectral bands, spectral indices, texture features, elevation, and phenological metrics derived from these samples. We used 70% of the sample points to train the classification model, with the remaining 30% used for accuracy assessment. Through visual interpretation, we constructed a reference-year sample database for water bodies, mudflats, and marshes. The sampling process is shown in Supplemental Figure 1. In false-color composite images, marshes appear dark red, mudflats yellow-brown, and water bodies blue. In total, we selected 300 water body samples, 100 mudflat samples, and 130 marsh samples. Samples and model parameters were then iteratively refined using Google Maps and Sentinel-2 images at 10 m resolution to generate the optimal reference-year riverine wetland map.

We employed an approach combining training sample migration with Random Forest classification for long-term riverine wetland mapping. The sample migration strategy is one of the most effective approaches for long-term remote sensing mapping (Huang et al., 2020), based on evaluating spectral similarity between samples in the target year and the reference year to assess whether land-cover change has occurred. If no significant change is detected, samples can be directly migrated to the target year. Euclidean distance (ED) is widely used as an effective measure of spectral similarity. It represents the overall difference between spectral vectors, with an ED value of 0 indicating identical spectra (Fekri et al., 2021). The ED is calculated as follows:

E D = \sqrt{{\sum_{i = 1}^{N} (X_{i (t 1)} ‐ Y_{i (t 2)})}^{2}}

(6)

Where ED is the Euclidean Distance, which ranges between 0 and 1; X_i is the reference spectra at time t₁; Y_i is the target spectra at time t₂; i is the spectral band; and N is the total number of bands. Therefore, we assessed the suitability of migrating samples from the reference year to the target year by calculating spectral ED.

To ensure classification accuracy, we first supplemented samples for target years through visual interpretation and established a reference sample database at 10-year intervals in areas exhibiting pronounced spectral variability. Following the optimal ED threshold determination strategy proposed in our previous work (Qiu et al., 2024a), we set the sample migration threshold for water bodies and mudflats as ED < 0.02. As marsh samples were not rational in the given threshold interval, the Land Surface Water Index (LSWI), which reflected the moisture content of vegetation soil, was selected as the screening condition. When LSWI < 0, it was considered that the land cover of the marsh sample had changed. The LSWI is calculated as follows:

L S W I = \frac{ρ_{n i r} ‐ ρ_{s w i r 1}}{ρ_{n i r} + ρ_{s w i r 1}}

(7)

However, mudflats exhibited pronounced interannual variability, and sample migration caused substantial sample loss. Therefore, additional sample points must be supplemented to meet classification demands. We added 332 sample points during 1989, 1995, 2006–2008, and 2011–2015, thereby establishing a complete sample database for riverine wetland classification in the Yellow River Basin (excluding the headwater and estuarine regions) from 1986 to 2022. Finally, we input the migrated samples into a Random Forest classification model to generate annual riverine wetland maps from 1986 to 2022 and evaluated classification performance using overall accuracy, the Kappa coefficient, user’s accuracy, and producer’s accuracy.

Changes of riverine wetland landscape patterns

To characterize long-term changes in riverine wetland spatial patterns, we quantified landscape pattern indices at both the landscape and class levels using FRAGSTATS. The combined use of these two categories of indices allows us to capture not only changes in total wetland extent, but also shifts in fragmentation, aggregation, connectivity, and shape complexity of wetland patches (Riitters et al., 1995). Because multiple landscape pattern indices may reflect similar aspects of spatial configuration, we selected a set of complementary indices rather than relying on a large number of highly correlated indices.

At the landscape level, we employed patch density (PD), contagion index (CONTAG), and splitting index (SPLIT) to characterize riverine wetland landscape patterns. PD measures the number of patches per unit area and reflects overall landscape fragmentation. CONTAG measures the degree of spatial aggregation and contagion among riverine wetland patches, with higher values indicating stronger landscape dominance and continuity. SPLIT measures landscape fragmentation by integrating patch area and number.

At the class level, we quantified riverine wetland landscape patterns using class percentage of landscape (CPLAND), largest patch index (LPI), connectivity index (CONNECT), aggregation index (AI), patch density (PD), and landscape shape index (LSI), covering dimensions of area characteristics, spatial structure, and landscape shape. LPI and CPLAND characterize the quantitative features of riverine wetland areas and indicate the stability of core patches. CONNECT, AI, and PD assess functional connectivity, spatial clustering, and patch density of riverine wetland patches from a spatial structure perspective. LSI captures landscape shape complexity and reflects the extent to which riverine wetland edges are affected by natural or human disturbances. The calculation methods for the above landscape pattern indices are presented in Supplemental Table 1.

We employed single linear regression to analyze temporal trends in riverine wetland area. Given the complex change of riverine wetland landscape patterns, a single linear model fails to accurately describe their temporal trends. Accordingly, we employed multivariate nonlinear regression to capture nonlinear dynamics in riverine wetland landscape change.

The driving factor of riverine wetland landscape change

Previous research indicates that the spatiotemporal distribution of precipitation has a strong influence on the ecosystem dynamics of the Yellow River Basin (Zhang et al., 2024; Zhou et al., 2025). Therefore, we focused on analyzing the relationships between riverine wetland area and hydrological variables, including precipitation and runoff, while integrating temperature and other climatic variables to explain the change in spatiotemporal landscape patterns. We initially employed Pearson correlation coefficients to quantify the relationships between changes in the riverine wetland area and potential driving factors, and evaluated statistical significance using T-tests (Carreiras et al., 2006; Legendre et al., 2002).

To further elucidate the effects of various drivers on riverine wetland change, we developed stepwise multiple linear regression models to quantify their relative contributions to changes in riverine wetland area and landscape patterns. First, we identified drivers significantly associated with riverine wetland changes based on correlation analysis results. Second, we applied stepwise regression to iteratively include or exclude variables, thereby optimizing model structure and minimizing multicollinearity. We constructed the regression model using the ordinary least squares method, expressed as

Y = β_{0} + β_{1} X_{1} + β_{2} X_{2} + \dots + β_{p} X_{p} + ϵ

(8)

where Y is the dependent variable, including riverine wetland area and landscape pattern indexes;

X_{1}, X_{2}, \dots, X_{p}

are the selected driving factors,

β_{0}

is the intercept,

β_{1}, β_{2}, \dots, β_{p}

are the regression coefficients, and

ϵ

is the random error term. We assessed multicollinearity using the Variance Inflation Factor (VIF) and excluded variables with VIF > 10. We further assessed model performance using the adjusted R² and tested overall and individual variable significance using F-tests and T-tests, respectively.

Results

The accuracy of riverine wetland mapping

Riverine wetland mapping for the Yellow River Basin (excluding the headwater and estuarine regions) in 2020 achieved an overall accuracy of 98.73% and a Kappa coefficient of 0.98, indicating high reliability of the classification result (Supplemental Table 2). Meanwhile, user’s and producer’s accuracies for water bodies, mudflats, and marshes all exceeded 90%. For long-term riverine wetland mapping, overall accuracy ranged from 90.00% (1992) to 99.28% (2021) (Figure 3(a)), with an average of 92.98%, and the Kappa coefficients ranged from 0.83 (2012) to 0.99 (2021), with an average of 0.88, indicating that the sample migration strategy and classification model consistently demonstrate strong performance across the time series.

Figure 3.

Riverine wetland mapping in the Yellow River Basin (excluding the headwater and estuarine regions). (a) Number of training samples and classification accuracy for riverine wetlands from 1986 to 2022; (b) Spatial distribution of riverine wetlands in the Yellow River Basin in 2020 (reference year); (c). Distribution of wetland area across the upstream, midstream, and downstream of the Yellow River in 2020.

Results showed that riverine wetlands within the Yellow River Basin (excluding the headwater and estuarine regions) covered an area of 6778.61 km² in 2020 (Figure 3(b)). Marshes constituted the dominant wetland type, accounting for nearly half of the total wetland area (3090.84 km², 45.25%). Water bodies constituted the second largest wetland type, covering nearly one-third of the total wetland area (2224.00 km², 32.88%), whereas mudflats accounted for the smallest share (1463.78 km², 21.87%). Spatially, water bodies and mudflats were mainly concentrated in the upstream, exhibiting a gradual decline in area from the upstream to the downstream. Water bodies covered an area of 924.59 km² in the upstream. The upstream, midstream, and downstream accounted for 41.57%, 34.71%, and 23.72% of the total water body area, respectively. In contrast, marshes were mainly concentrated in the middle reaches, covering 1591.66 km² and accounting for 51.50% of the total marsh area.

The spatiotemporal distribution of riverine wetlands

From 1986 to 2022, the area of water bodies within riverine wetlands exhibited a significant increasing trend (R² = 0.47). Relative to 1986, the water body area increased by 950.35 km², of which the midstream accounted for 52.85% (Figure 4(b)). In contrast, changes in mudflat area were relatively modest (R² = 0.0037), with a net increase of 379.33 km², primarily driven by an increase in the upstream, which accounted for 86.08% of this increase (Figure 4(c)). Despite these increases, the total riverine wetland area in the Yellow River Basin (excluding the headwater and estuarine regions) decreased by 1284.98 km² over the past 36 years (Figure 4(a)), primarily driven by a significant decrease in marsh coverage (R² = 0.74). By 2022, the marsh area had decreased by nearly 30.76% compared with 1986, with an average annual decrease of 36.94 km² (Figure 4(d)), and the upstream and midstream accounted for 82.15% of this decrease. Expansion of water bodies in the midstream and mudflats in the upstream and downstream replaced former marsh wetlands (Supplemental Figure 2).

Figure 4.

Changes in riverine wetland area in the Yellow River Basin (excluding the headwater and estuarine regions) from 1986 to 2022. (a) Riverine wetland area changes in the upstream, midstream and downstream; (b) Riverine wetland area changes in the midstream; (c) Riverine wetland area changes in the upstream; (d) Riverine wetland area changes in the downstream.

Riverine wetland landscape patterns exhibited an increase in fragmentation between 1986 and 2022 (Figure 5). Marshes exhibited substantial landscape degradation, as evidenced by decreases in LPI, CPLAND, CONNECT, and AI (Figure 5). These trends indicate decreased connectivity and aggregation among marsh patches. In contrast, LPI, CPLAND, CONNECT, and AI for water bodies exhibited a fluctuating increase trend (Figure 5), indicating increased dominance of core water body patches and enhanced connectivity and aggregation. Concurrent decreases in PD and LSI suggest increasing spatial stability and concentration of water body landscapes. Although the landscape pattern indices of mudflats fluctuated considerably, LPI, CPLAND, CONNECT, and AI increased after 2019 or 2021 (Figure 5). Moreover, PD and LSI generally showed increasing trends but declined sharply in 2022.

Figure 5.

Changes in landscape indices of riverine wetlands in the Yellow River Basin (excluding the headwater and estuarine regions) from 1986 to 2022.

Driving factors for riverine wetland area and landscape pattern change

The primary drivers of water body area change in riverine wetlands were within May–October runoff at the Lanzhou station, average temperature, total precipitation, and sediment discharge at the Lanzhou station (Table 1). Stepwise linear regression analysis showed that these factors explain 69.2% of the changes in water body area (

R_{a d j}^{2} = 0.692

), with no evidence of multicollinearity (VIF < 10) or strong autocorrelation (DW = 1.311). Runoff at the Lanzhou station emerged as the dominant driver (β = 0.538, p < 0.001), showing a significant positive effect. The marsh area was mainly associated with average temperature and May–October runoff at the Lanzhou, Toudaoguai, and Huayuankou stations (Supplemental Table 3). Among these factors, average temperature (β = −0.451) showed a significant negative effect on marsh area and explained 42.6% of its change (Table 1). Notably, stepwise regression and correlation analyses revealed no significant associations between mudflat area and runoff, sediment discharge, temperature, or precipitation at the basin scale (Supplemental Table 3). In contrast, the mudflat area in the midstream exhibited a significant positive correlation with sediment discharge at the Huayuankou station (Supplemental Table 3, Pearson’s r = 0.33). However, this relationship remains relatively weak, explaining only 8.2% of mudflat area change (

R_{a d j}^{2}

= 0.082).

Table 1.

Driving factors of area changes in water bodies, marshes, and mudflats in the Yellow River Basin (excluding the headwater and estuarine regions).

Dependent variable: Water body area	$R_{a d j}^{2} = 0.692$	DW = 1.311	Significance P	VIF
Independent variable	Standardized coefficient $β$	Pearson’s r	Significance P	VIF
Runoff at Lanzhou (May–October)	0.538***	0.647**	0.000	1.198
Sediment discharge at Lanzhou	−0.250*	−0.347*	0.020	1.228
Average temperature (May–October)	0.326**	0.528**	0.003	1.195
Total precipitation	0.253*	0.473**	0.016	1.157

Dependent variable: marsh area	$R_{a d j}^{2} = 0.426$	DW = 1.43	Significance P	VIF
Independent variable	Standardized coefficient $β$	Pearson’s r	Significance P	VIF
Average temperature (May–October)	−0.451**	−0.505**	0.001	1.014

Dependent variable: Mudflat in the middle reach	$R_{a d j}^{2} = 0.082$	DW = 0.67
Independent variable	Standardized coefficient $β$	Pearson’s r	Significance P	VIF
Sediment discharge at Huayuankou	0.204*	0.33*	0.017	1.442

Note. *p < 0.05, **p < 0.01, ***p < 0.001. The reported significance values (P) refer to the standardized coefficients in the stepwise linear regression models.

Precipitation and sediment delivery at the Huayuankou station were the primary drivers of changes in CONNECT for water bodies (Figure 6). Precipitation showed a significant positive effect on CONNECT. In contrast, sediment discharge showed a negative effect on CONNECT and positive effects on PD and LSI, indicating that sediment accumulation causes increased water body fragmentation and increased patch shape complexity, especially in the downstream. Mudflat landscape patterns were mainly associated with average temperature, precipitation, runoff, and sediment discharge (Figure 6). Average temperature showed a negative effect on CONNECT and a positive effect on PD and LSI. In addition, sediment discharge at the Huayuankou station positively affected AI, whereas precipitation and runoff at the Toudaoguai station showed a negative effect. Marsh connectivity and patch shape complexity are mainly affected by the average temperature and runoff at the Lanzhou station (Figure 6), indicating that increasing temperature and runoff decreased marsh connectivity and were accompanied by lower shape complexity.

Figure 6.

Driving factors of changes in landscape indices of water bodies, marshes, and mudflats in the Yellow River Basin (excluding the headwater and estuarine regions).

Discussion

Imbalance of riverine wetland protection and restoration

In this study, we employed a previously improved sample migration strategy to map long-term riverine wetland in the Yellow River Basin (excluding the headwater and estuarine regions). The resulting wetland maps exhibited high classification accuracy and consistency (Kappa_min = 0.83, Kappa_average = 0.88). These results indicate that the improved sample migration strategy performs robustly across highly heterogeneous regions, characterized by complex climatic gradients, variable human disturbance, and diverse ecological functions. The method therefore provides a reliable framework for fine-resolution, long-term wetland mapping in large river basins.

The mapping results reveal a complex trend of mixed blessings in riverine wetland area changes across the Yellow River Basin (excluding the headwater and estuarine regions). Since the early 2000s, wetland conservation and restoration programs, including cropland-to-wetland conversion, ecological water replenishment, wetland reserve establishment, and riverine restoration projects (Fu et al., 2025; Mao et al., 2018; Wang et al., 2021), have promoted the expansion of water bodies and the recovery of mudflats, thereby enhancing hydrological connectivity in parts of the basin. However, a concerning pattern emerges for marshes, which represent a core component of riverine wetlands in the Yellow River Basin, as their area is experiencing a significant decrease, particularly pronounced over the past decade (1314.02 km²). Unlike water bodies that primarily enhance connectivity or mudflats that support wave attenuation and food production, marshes constitute the dominant carbon pool of riverine wetlands (Tang et al., 2024; Zhang et al., 2025). Previous research demonstrates that marshes store more carbon than water bodies or mudflats, especially in temperate arid and semi-arid regions, where average biomass density reaches 300.5 ± 73.2 g C m⁻² (Shen et al., 2021). Therefore, the extensive degradation of marshes and their conversion to water bodies and mudflats (1092.35 km²) in the midstream and upstream indicate a rapid loss of carbon stocks, thereby intensifying the regional carbon imbalance.

Landscape pattern analysis further reveals the changing characteristics among the three riverine wetland types. Results show that the landscape connectivity of water bodies has increased significantly. Although mudflats experienced complex fluctuations, their landscape patterns have recently exhibited increased aggregation and reduced fragmentation. In contrast, marshes, which support basin biodiversity, exhibit increasing patch density and fragmentation, accompanied by decreasing core areas. Marshes in the upstream and downstream provide critical habitats for endangered birds, including the black stork, white-tailed eagle, and great bustard, supporting migration, breeding, and foraging, and sustaining intact regional food-web networks (Li et al., 2025). Increasing marsh fragmentation may result in habitat loss and disrupt genetic exchange at the basin scale (Liu et al., 2025a; Sounapoglou et al., 2025). Collectively, these results highlight the lack of targeted wetland management across the entire basin, where current measures remain insufficient to curb the marsh degradation.

The different driving factors for changes in three riverine wetland types

Wetland area dynamics and landscape pattern change across the Yellow River Basin are strongly regulated by hydrological and climatic drivers. Runoff at the Lanzhou station acts as a critical hydrological gate, regulating water body area and landscape patterns by directly controlling water availability and providing fundamental habitat conditions for multiple riverine wetland types. This regulation effect is particularly evident in the arid and semi-arid upstream and midstream of the Yellow River Basin, where limited water resources make runoff a primary constraint on the water body expansion (Gao et al., 2025; Wang et al., 2025; Yu et al., 2021). Temperature and precipitation further enhance water body connectivity and aggregation by regulating basin-scale hydrological cycles and seasonal water redistribution. In contrast, increased sediment discharge at the Huayuankou station increases the spatial complexity of water bodies through sediment accumulation while undermining hydrological continuity (Zhang et al., 2023).

Unlike water bodies, marsh degradation and fragmentation are significantly associated with rising temperature and increasing runoff at the Lanzhou station, indicating a distinct response mechanism to hydroclimatic forcing. One possible explanation is that marshes are highly dependent on soil moisture, and climate warming exacerbates soil water stress by increasing evapotranspiration rates (Hu et al., 2025; Li et al., 2024; Liu et al., 2025b). These processes may cause the loss of edge habitats and shift amphibious and benthic organisms toward patch cores, leading to reduced connectivity and simplified patch shapes. On the other hand, increased runoff at Lanzhou station may cause intensified reservoir releases driven by concentrated precipitation events in the upstream. These changes disrupt the seasonal shallow inundation regimes to which marshes are adapted, altering inundation patterns and constraining marsh habitat availability (Fluet-Chouinard et al., 2023; Mei et al., 2023). This response further reflects the high sensitivity of marshes to climate warming.

Mudflat area shows a significant positive association with sediment discharge at Huayuankou station, indicating that sediment accumulation is the primary driver of mudflat formation. In terms of landscape patterns, sediment discharge causes aggregation of scattered mudflat patches, whereas warming and changes in runoff inhibit mudflat aggregation and increase the shape complexity. Importantly, water-sediment processes explain only 8.2% of the changes in mudflat landscape patterns. Therefore, mudflat landscape change might be primarily influenced by human disturbance. River regulation projects, mudflat reclamation, and other human activities may regulate mudflat area and landscape patterns through sediment redistribution (Dou et al., 2023; Qiu et al., 2024b; Wei et al., 2021), helping to explain the strong interannual fluctuations observed over the past 36 years.

Limitations and prospects

Due to the lack of field-based observations, the 30 m resolution wetland maps derived from limited training samples could not be independently validated. In addition, cloud contamination and sensor heterogeneity in Landsat time-series data may introduce uncertainty, leading to fluctuations in classification accuracy and potential error propagation. Although this study examined the effects of climate change and hydrological processes on wetland area and landscape pattern change, several important driving factors remain unaccounted for. In particular, discrete flood events and short-term high-magnitude hydrological pulses were not explicitly incorporated in this study. While interannual hydrological variability is partially represented through precipitation- and runoff-related indicators, such proxies may not fully capture the rapid geomorphological reshaping effects induced by extreme flooding events. Therefore, future studies should incorporate key natural and anthropogenic drivers, including event-scale flood records, groundwater dynamics, ecological restoration initiatives, hydraulic engineering, and agricultural irrigation. Furthermore, the use of annual-scale data limits the ability to resolve seasonal wetland dynamics. Multi-scale temporal analyses should be employed in future work to resolve seasonal dynamics and regional lag effects, thereby providing scientific support for targeted wetland conservation, ecological restoration, and coordinated management in the Yellow River Basin.

Conclusions

Wetlands play a critical role in maintaining hydrological balance, biodiversity, and carbon cycling in the Yellow River Basin, which is threatened by climate change and human disturbances. However, the long-term change patterns and driving factors of riverine wetland in the Yellow River Basin (excluding the headwater and estuarine regions) remain understudied. By combining a sample migration strategy and Random Forest classification on the GEE platform, we produced the first annual 30 m riverine wetland dataset (1986–2022) for the Yellow River Basin (excluding the headwater and estuarine regions), achieving an average Kappa coefficient of 0.88 and overall accuracy above 90%. We found that riverine wetlands in the Yellow River Basin (excluding the headwater and estuarine regions) have undergone pronounced spatial reorganization under combined climate and anthropogenic pressures, showing an overall pattern of degradation accompanied by localized expansion. Over the past 36 years, water bodies and mudflats increased, whereas marshes, the dominant carbon pool and an important habitat for endangered species, declined markedly and became more fragmented. These contrasting trends were driven by distinct factors: runoff at the Lanzhou station dominates water body expansion, climate warming and increased runoff promote marsh degradation. Mudflats in the midstream are mainly affected by sediment discharge at Huayuankou station and human activities. Management should therefore shift from basin-wide uniform measures toward differentiated, reach-specific strategies, with priority given to marsh conservation and restoration through protection of core patches and maintenance of hydrological connectivity. This study advances long-term riverine wetland landscape research, provides methodological support for long-term and high-resolution mapping, and highlights the necessity of differentiated management strategies across wetland types.

Supplemental material

Supplemental material - Long-term mapping and landscape dynamics of major riverine wetlands in the Yellow River Basin from 1986 to 2022

Supplemental material for Long-term mapping and landscape dynamics of major riverine wetlands in the Yellow River Basin from 1986 to 2022 by Chenxu Wang, Mengqi Qiu, and Yanxu Liu, Suai Wang in Progress in Physical Geography: Earth and Environment

Footnotes

ORCID iDs

Chenxu Wang

Yanxu Liu

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was supported by the National Natural Science Foundation of China (No. 42522105), Ordos Key Science and Technology Program Project (No. YF20250267), the Fundamental Research Funds for the Central Universities of China, and the Ordos Desertification Control and Sustainable Development Academician Expert Workstation.

Declaration of conflicting interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Supplemental material

Supplemental material for this article is available online.

References

Adam

Mutanga

Rugege

(2010) Multispectral and hyperspectral remote sensing for identification and mapping of wetland vegetation: a review. Wetlands Ecology and Management 18: 281–296. https://doi.org/10.1007/s11273-009-9169-z

, et al. (2023) Tracking economic-driven coastal wetland change along the East China Sea. Applied Geography 156: 102995. https://doi.org/10.1016/j.apgeog.2023.102995

Bai

Wang

, et al. (2013) Landscape pattern evolution processes of alpine wetlands and their driving factors in the Zoige Plateau of China. Journal of Mountain Science 10: 54–67. https://doi.org/10.1007/s11629-013-2572-1

Bao

Jia

(2023) Weakening greenhouse gas sink of pristine wetlands under warming. Nature Climate Change 13: 462–469. https://doi.org/10.1038/s41558-023-01637-0

Bar-On

Phillips

Milo

(2018) The biomass distribution on earth. Proceedings of the National Academy of Sciences 115: 6506–6511. https://doi.org/10.1073/pnas.1711842115

Carreiras

JMB

Pereira

JMC

Pereira

(2006) Estimation of tree canopy cover in evergreen oak woodlands using remote sensing. Forest Ecology and Management 223: 45–53. https://doi.org/10.1016/j.foreco.2005.10.056

Cherry

(2011) Ecology of wetland ecosystems: water, substrate, and life. Nature Education Knowledge 10: 16.

Dou

Guo

Zhang

, et al. (2023) Dynamic landscapes and the influence of human activities in the Yellow River Delta wetland region. Science of the Total Environment 899: 166239. https://doi.org/10.1016/j.scitotenv.2023.166239

Fekri

Latifi

Amani

, et al. (2021) A training sample migration method for wetland mapping and monitoring using sentinel data in Google Earth Engine. Remote Sensing 13: 4169. https://doi.org/10.3390/rs13204169

10.

Fluet-Chouinard

Stocker

Zhang

, et al. (2023) Extensive global wetland loss over the past three centuries. Nature 614: 281–286. https://doi.org/10.1038/s41586-022-05572-6

11.

Wang

, et al. (2025) Integrating hydrological impacts for cost-effective dryland ecological restoration. Communications Earth & Environment 6: 667. https://doi.org/10.1038/s43247-025-02649-8

12.

Gao

(1996) NDWI—A normalized difference water index for remote sensing of vegetation liquid water from space. Remote Sensing of Environment 58: 257–266. https://doi.org/10.1016/s0034-4257(96)00067-3

13.

Gao

(2016) Wetland and its degradation in the. In: Brierley

Cullum

, et al. (eds) Yellow River Source Zone. Landscape and Ecosystem Diversity, Dynamics and Management in the Yellow River Source Zone. Springer International Publishing, pp. 209–232.

14.

Gao

Zhou

, et al. (2025) Climate change and human activities amplify runoff variability risks in lower reaches of large rivers. Communications Earth & Environment 6: 794. https://doi.org/10.1038/s43247-025-02759-3

15.

Gurung

Breidt

Dutin

, et al. (2009) Predicting Enhanced Vegetation Index (EVI) curves for ecosystem modeling applications. Remote Sensing of Environment 113: 2186–2193. https://doi.org/10.1016/j.rse.2009.05.015

16.

Xie

Liu

, et al. (2025) Mapping the primary factors driving spatiotemporal variations of surface soil moisture in the Yellow River Basin of China. Scientific Reports 15: 12359. https://doi.org/10.1038/s41598-025-96129-w

17.

Huang

Wang

Liu

, et al. (2020) The migration of training samples towards dynamic global land cover mapping. ISPRS Journal of Photogrammetry and Remote Sensing 161: 27–36. https://doi.org/10.1016/j.isprsjprs.2020.01.010

18.

Huang

Tang

Hupy

, et al. (2021) A commentary review on the use of normalized difference vegetation index (NDVI) in the era of popular remote sensing. Journal of Forestry Research 32: 1–6. https://doi.org/10.1007/s11676-020-01155-1

19.

Legendre

Dale

MRT

Fortin

, et al. (2002) The consequences of spatial structure for the design and analysis of ecological field surveys. Ecography 25: 601–615. https://doi.org/10.1034/j.1600-0587.2002.250508.x

20.

Zhang

Bevacqua

, et al. (2024) Future increase in compound soil drought-heat extremes exacerbated by vegetation greening. Nature Communications 15: 10875. https://doi.org/10.1038/s41467-024-55175-0

21.

Wang

Yan

, et al. (2025) Avian species and functional diversity in the Yellow River Wetland Nature Reserve of Henan Province. Global Ecology and Conservation 57: e03352. https://doi.org/10.1016/j.gecco.2024.e03352

22.

Liu

Dong

Jiang

, et al. (2024) Agricultural flood resistance enhanced after returning farmlands to lakes. Proceedings of the National Academy of Sciences 121: e2410967121. https://doi.org/10.1073/pnas.2410967121

23.

Liu

Chen

Wang

, et al. (2025a) Habitat fragmentation differentially affects invasive and native plant diversity in a human-dominated wetland island system. Plant Diversity 47: 824–832. https://doi.org/10.1016/j.pld.2024.09.003

24.

Liu

Wang

Zhao

, et al. (2025b) Evapotranspiration stress intensifies with enhanced sensitivity to soil moisture deficits in a rapidly greening China. Hydrology and Earth System Sciences 29: 3379–3404. https://doi.org/10.5194/hess-29-3379-2025

25.

Mao

Luo

Wang

, et al. (2018) Conversions between natural wetlands and farmland in China: a multiscale geospatial analysis. Science of the Total Environment 634: 550–560. https://doi.org/10.1016/j.scitotenv.2018.04.009

26.

Mao

Wang

, et al. (2020) National wetland mapping in China: a new product resulting from object-based and hierarchical classification of Landsat 8 OLI images. ISPRS Journal of Photogrammetry and Remote Sensing 164: 11–25. https://doi.org/10.1016/j.isprsjprs.2020.03.020

27.

Mao

Yang

Wang

, et al. (2022) Reverse the hidden loss of China’s wetlands. Science 376: 1061. https://doi.org/10.1126/science.adc8833

28.

Mei

Wang

, et al. (2023) Restore the Yangtze River’s floodplains. Science 382: 1369–1370. https://doi.org/10.1126/science.adn0769

29.

Osland

Chivoiu

Enwright

, et al. (2022) Migration and transformation of coastal wetlands in response to rising seas. Science Advances 8: eabo5174. https://doi.org/10.1126/sciadv.abo5174

30.

Qiu

Liu

Chen

, et al. (2024a) Spatio-temporal changes and hydrological forces of wetland landscape pattern in the Yellow River Delta during 1986–2022. Landscape Ecology 39: 51. https://doi.org/10.1007/s10980-024-01850-y

31.

Qiu

Liu

Tian

, et al. (2024b) Marsh decrease was much faster than the water increase among the Yellow River Source wetlands during 1986–2022. Science of the Total Environment 947: 174377. https://doi.org/10.1016/j.scitotenv.2024.174377

32.

Riitters

O'Neill

Hunsaker

, et al. (1995) A factor analysis of landscape pattern and structure metrics. Landscape Ecology 10: 23–39. https://doi.org/10.1007/bf00158551

33.

Sandi

Rodriguez

Saintilan

, et al. (2020) Resilience to drought of dryland wetlands threatened by climate change. Scientific Reports 10: 13232. https://doi.org/10.1038/s41598-020-70087-x

34.

Schneider

Crandall

Baiser

(2024) The large-scale restoration of fire and water regimes in Everglades National Park reveal little change in plant diversity along an elevational gradient. Restoration Ecology 32: e14138. https://doi.org/10.1111/rec.14138

35.

Schuerch

Spencer

Temmerman

, et al. (2018) Future response of global coastal wetlands to sea-level rise. Nature 561: 231–234. https://doi.org/10.1038/s41586-018-0476-5

36.

Shen

Jiang

, et al. (2021) Aboveground biomass and its spatial distribution pattern of herbaceous marsh vegetation in China. Science China Earth Sciences 64: 1115–1125. https://doi.org/10.1007/s11430-020-9778-7

37.

Sounapoglou

Janeček

Bartoš

, et al. (2025) The significance of reproduction-related traits for wet meadow species survival in a fragmented landscape. Flora 330: 152771. https://doi.org/10.1016/j.flora.2025.152771

38.

Suhardi

Rahardi

Shih

, et al. (2024) A review of the mangrove ecosystem in Indonesia: biodiversity, conservation, and challenges in sustainable management. Ecological Genetics and Genomics 32: 100282. https://doi.org/10.1016/j.egg.2024.100282

39.

Sun

Liu

, et al. (2023) Tracking annual changes in the distribution and composition of saltmarsh vegetation on the Jiangsu coast of China using Landsat time series–based phenological parameters. Remote Sensing of Environment 284: 113370. https://doi.org/10.1016/j.rse.2022.113370

40.

Taddeo

Dronova

Depsky

(2019) Spectral vegetation indices of wetland greenness: responses to vegetation structure, composition, and spatial distribution. Remote Sensing of Environment 234: 111467. https://doi.org/10.1016/j.rse.2019.111467

41.

Tang

Wang

Tian

, et al. (2024) Exploring the factors influencing the carbon sink function of coastal wetlands in the Yellow River Delta. Scientific Reports 14: 28938. https://doi.org/10.1038/s41598-024-80186-8

42.

Wang

Xiao

, et al. (2021) Rebound in China’s coastal wetlands following conservation and restoration. Nature Sustainability 4: 1076–1083. https://doi.org/10.1038/s41893-021-00793-5

43.

Wang

Mao

Wang

, et al. (2023) Wetland mapping in East Asia by two-stage object-based random forest and hierarchical decision tree algorithms on Sentinel-1/2 images. Remote Sensing of Environment 297: 113793. https://doi.org/10.1016/j.rse.2023.113793

44.

Wang

Song

Zhang

, et al. (2025) Anthropogenic impacts on the Yellow River Basin. Nature Reviews Earth & Environment 6: 656–671. https://doi.org/10.1038/s43017-025-00718-2

45.

Wei

Cai

Zhan

(2021) Impact of anthropogenic activities on morphological and deposition flux changes in the Pearl River Estuary, China. Scientific Reports 11: 16643. https://doi.org/10.1038/s41598-021-96183-0

46.

Wetlands

RCO

(2021) The global wetland outlook: special edition 2021. 13.

47.

Wijesingha

Dzene

Wachendorf

(2024) Evaluating the spatial–temporal transferability of models for agricultural land cover mapping using Landsat archive. ISPRS Journal of Photogrammetry and Remote Sensing 213: 72–86. https://doi.org/10.1016/j.isprsjprs.2024.05.020

48.

Xiao

Tian

, et al. (2010) Landscape patterns and their changes in Sichuan Ruoergai Wetland National Nature Reserve. Acta Ecologica Sinica 30: 27–32. https://doi.org/10.1016/j.chnaes.2009.12.005

49.

(2006) Modification of normalised difference water index (NDWI) to enhance open water features in remotely sensed imagery. International Journal of Remote Sensing 27: 3025–3033. https://doi.org/10.1080/01431160600589179

50.

Han

Wang

, et al. (2021) The impact of runoff flux and reclamation on the spatiotemporal evolution of the Yellow River estuarine wetlands. Ocean & Coastal Management 212: 105804. https://doi.org/10.1016/j.ocecoaman.2021.105804

51.

Yun

Liu

, et al. (2022) Assessing changes in the landscape pattern of wetlands and its impact on the value of wetland ecosystem services in the Yellow River Basin, Inner Mongolia. Sustainability 14: 6328. https://doi.org/10.3390/su14106328

52.

Zhang

Yin

, et al. (2015) Impacts of human activities on the evolution of estuarine wetland in the Yangtze Delta from 2000 to 2010. Environmental Earth Sciences 73: 435–447. https://doi.org/10.1007/s12665-014-3565-2

53.

Zhang

Wang

Xue

, et al. (2021) Dynamic landscapes and the driving forces in the Yellow River Delta wetland region in the past four decades. Science of the Total Environment 787: 147644. https://doi.org/10.1016/j.scitotenv.2021.147644

54.

Zhang

Zhao

(2023) Impact of changes in river network structure on hydrological connectivity of watersheds. Ecological Indicators 146: 109848.

55.

Zhang

Zhao

Zhu

, et al. (2024) The dynamic patterns of critical ecological areas in the Yellow River Basin are driven primarily by climate factors but threatened by human activities. Journal of Environmental Management 371: 123282. https://doi.org/10.1016/j.jenvman.2024.123282

56.

Zhang

Fang

, et al. (2025) Carbon sequestration potential of wetlands and regulating strategies response to climate change. Environmental Research 269: 120890. https://doi.org/10.1016/j.envres.2025.120890

57.

Zhou

Wang

Lyu

, et al. (2025) Response of vegetation to extreme climate in the Yellow River Basin: Spatiotemporal patterns, lag effects, and scenario differences. Remote Sensing 17: 3967. https://doi.org/10.3390/rs17243967

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.94 MB