Spatiotemporal evolution of land use and habitat quality in the Three-River source region

Abstract

Investigating spatial-temporal variations in land use/cover change (LUCC) and habitat quality (HQ) is essential for evaluating ecosystem integrity and advancing sustainable development objectives. The Three-River Source Region (TRSR) has distinctive LUCC characteristics that reveal intricate relationships between land utilization patterns and environmental deterioration. Through an integrated PLUS-InVEST modeling framework, this research assesses historical LUCC and HQ transformations in the TRSR spanning three decades (1990–2020) while forecasting ecological patterns through mid-century projections (2050). Analysis indicates that grasslands dominate the TRSR landscape and constitute the core of its ecosystem. Over the past three decades, grassland, water body, and cropland areas increased by 14,593.43 km², 1,888.67 km², and 319.81 km², respectively, while the proportion of unused land declined from 22.24% to 17.79%. Projections forecast continued expansion of grassland and water body extents, accompanied by further reduction in unused land. Observed spatial shifts and reciprocal transformations among land-cover types underscore dynamic landscape processes. HQ exhibited an overall decline, with 0.67% of the region experiencing deterioration, which exceeded areas of improvement. In the absence of intervention, HQ degradation is projected to intensify, potentially doubling in spatial extent. Notably, HQ fluctuations occurred even in areas without LUCC, although more pronounced changes were detected where significant land-use conversions took place. These findings underscore the complex linkages between LUCC and HQ. Future strategies for the TRSR should prioritize limiting unused land expansion, conserving grasslands, and deepening understanding of regional ecological dynamics, thereby enabling more precise and targeted planning and management of unused land categories. Further elucidation of mechanisms relating land use to HQ will be critical for advancing research and guiding ecological restoration efforts.

Introduction

Person-made activities modify land use patterns through effects that considerably shape both ecosystem benefits and habitat integrity (HQ)¹. The societal evaluation and protection efforts rely on HQ as an ecosystem capability which maintains lasting habitats for viable species throughout various locations and time periods^2,3. HQ exists as an essential ecosystem characteristic which uses qualitative and quantitative evaluation techniques to show ecosystem status. Rare Ecosystem Habitat stands as a dominant assessment factor to evaluate biodiversity and ecosystem services patterns yet plays a decisive role in ecosystem sustainability and ecological integrity evaluations^4,5. Present day modifications in land use/cover function as main agents of Headquarter changes because they reflect human activities intensity⁶. Various improper ecological land management practices lead to significant LULC transformations which negatively impact HQ evolution and result in habitat disintegration and ecosystem degradation⁷. The necessary socio-ecological benefits of ecological habitats support human livelihood together with wildlife survival. Studies that focus on HQ evolution patterns along with investigation of habitat destruction patterns resulting from LULC shifts are essential for biodiversity conservation and sustainable land management in the region⁸. The research findings confirm the essential need to protect original habitats⁹.

The initial study of Habitat Quality used direct assessment approaches that measured biodiversity through methods like the Shannon-Wiener index as described by Kempton¹⁰ and Van¹¹. Ecological outcomes were measured by Russell through indicators such as species diversity as well as population density indices and reproductive success rates and survival metrics according to Zhang¹². Modern computational technologies combined with GIS and remote sensing tools have elevated composite index assessment methods to become prominent instruments for HQ evaluation^13,14. The research collaboration between Stanford University and The Nature Conservancy and WWF in 2013 built the InVEST model to conduct spatial ecosystem service quantification producing thematic distribution maps through geospatial analysis^15,16. Various ecosystem service modeling frameworks serve as essential assessment tools for headquarters evaluation studies¹⁷. Scientific research indicates that habitat quality suffers from urbanization patterns and social progress and environmental temperature changes mainly due to human-made constituents^5,18,19. Several scientific teams link land use changes with HQ parameters to model ecological regional patterns²⁰. The Sanjiang Plain experienced the largest conversion of 3,564.43 km² grassland from 1985 to 2017 according to Jin et al.‘s research that combined HQ indicators with land transition matrices at grid-level resolution. predominantly due to infrastructure expansion²¹. Nie²² conducted scenario-based analyses of habitat impacts from land development in the Lhasa River Basin. Land use patterns exhibit critical ecological significance, as evidenced by Kail²³ demonstrating robust associations between terrestrial utilization and freshwater species abundance in Natura 2000 protected zones. Given the substantial influence of land use alterations on HQ, researchers have investigated historical and projected LUCC patterns. Cellular Automata (CA) modeling has been widely implemented for land use simulation, as exemplified by Zhang²⁴ and Gomes²⁵. Gomes²⁶ applied CA modeling to assess Lithuanian HQ responses under multiple scenarios: status quo maintenance, urban growth, afforestation initiatives, and agricultural modernization. Integrating LUCC projections, Zhang²⁴ coupled CA-Markov modeling with InVEST to simulate habitat deterioration processes. In Jiangxi Province’s Nanchang City, Li²⁷ combined CA-Markov and Multi-Criteria Evaluation (MCE) models with InVEST for spatiotemporal habitat projections. The Patch-generating Land Use Simulation (PLUS) model addresses limitations in traditional CA frameworks regarding land conversion mechanisms and landscape evolution algorithms²⁸. Numerous studies have implemented PLUS-InVEST integrations for multi-scenario analyses²⁹. Gao³⁰ conducted land use evolution and ecological risk assessments for Nanjing using PLUS modeling, while Reheman³¹ employed InVEST-PLUS syntheses to simulate water yield variations in urban clusters. These methodological advancements complement existing HQ research across aquatic ecosystems³², watershed systems³³, protected areas³⁴, and rapidly urbanizing regions, enhancing regional HQ assessment frameworks³⁵. Land use changes alter material and energy flows between habitat patches, consequently affecting habitat quality and spatial patterns³⁶. Rapid economic development in China has intensified structural imbalances in land use. As crucial land reserves, unused lands play a vital ecological role in maintaining regional biodiversity and ecological equilibrium³⁷. Current studies predominantly attribute HQ decline to urban expansion, with limited examination of other land categories. This investigation specifically evaluates the ecological impacts of unused land expansion within the study area, identifying its spatial dynamics as a crucial determinant of HQ variation.

This investigation centers on the Three-River Source Region (TRSR), a protected area encompassing the largest natural reserve in China and globally significant high-altitude biodiversity. As the nation’s highest-elevation wetland ecosystem and the most ecologically vulnerable zone within the Three Rivers network, the TRSR demonstrates substantial sensitivity of HQ to land use alterations. The research aims to (1) quantify historical and projected land use change impacts on HQ and (2) establish spatiotemporal correlations between land use dynamics and HQ variations.(3) Reveal the correlation between land use and habitat quality, and elucidate the ecological significance of unused land in the Three-River Source Region.

Data and methods

Study area

The TRSR exists within southwestern Qinghai Province of northeastern Qinghai–Tibet Plateau China (Fig. 1) between 89° 45′ to 102° 23′ E and 31° 39′ to 36° 12′ N and contains the Yangtze river headwaters together with those of the Yellow and Lancang Rivers. The TRSR occupies approximately 382,000 km² space which contains glaciers alongside permafrost and high mountains and highland plains and hills that rise from 2,800 to 6,564 m. The freshwater source plays a vital role for China and all of Asia ensuring overall water security across the region³⁸. The combination of highland mountain climate and specific topography in this region has established an exclusive natural species collection which ranks among the world’s most biodiverse high altitude regions³⁹. Wetland environments control most of this area whose two principal plant groups consist of alpine steppe and alpine meadow vegetation. The Tibetan Rangeland Serma National Nature Reserve offers vital residency to seven nationally protected species of wildlife: snow leopard, Tibetan antelope, wild yak, Tibetan fox, Pallas’s cat, lynx and Tibetan wild ass as well as Tibetan gazelle.

Fig. 1

Location and topography of the Three-River source region. *This is an original map created using ArcGIS 10.7. https://www.esri.com/en-us/arcgis/products/arcgis-desktop/overview⁴⁰.

Data preparation

Historical land use changes were calculated using data on drivers of land-use and their changes. These drivers then informed the simulation of land use spatial distribution in 2050. HQ was estimated at five time points (1990, 2000, 2010, 2020, and 2050) using the appropriate models. Changes in HQ, its temporal trends, and spatial heterogeneity were subsequently analyzed to assess the relationship between landuse change and habitat quality. Data sources are summarized in Table 1. To satisfy the computational requirements of the PLUS and InVEST models, with reference to relevant literature^22,31, all spatial datasets, including base geographic layers, environmental variables, and socioeconomic factors, were projected to the Krasovsky_1940_Albers coordinate system and converted to 100 m×100 m raster format (.tif).

Table 1 Data information.

Methods

The methodological framework integrates four components (Fig. 2). First, remote sensing datasets (1990–2020) were analyzed through land use transfer matrices to identify landscape transitions. Second, the InVEST model calculated three-decade HQ variations. Third, the PLUS model projected land use patterns and HQ status for 2021–2050. Finally, synthesis of historical and predictive data revealed HQ response mechanisms to land use changes, informing targeted ecological management strategies. This systematic approach provides critical baseline data for regional conservation planning and ecosystem rehabilitation initiatives.

Fig. 2

Methodological flowchart.

PLUS model

Developed through collaborative efforts at the HPSCIL@CUG Laboratory of China University of Geosciences (Wuhan), the PLUS model overcomes critical constraints inherent in conventional CA frameworks regarding transformation rule optimization and landscape evolution simulation^28,41. This advancement has established the model as a preferred tool for predicting and simulating land use dynamics⁴². The system comprises two core components, the Land Expansion Analysis Strategy (LEAS) module and the Cellular Automata based on Multiple Random Seeds (CARS) framework, enabling comprehensive analysis of land use expansion. Operational procedures involve sequential phases: initial extraction of land expansion characteristics from multi-temporal datasets, followed by the LEAS module employing random forest algorithms to quantify correlations between land expansion patterns and environmental drivers, thereby generating categorical probability maps. The final phase integrates CARS mechanisms with stochastic seed allocation and adaptive threshold optimization to simulate prospective spatial configurations of land utilization within target regions.

LEAS

The LEAS reformulates the fundamental processes governing land utilization alterations as a dual-category identification task. Within the PLUS framework, exploration of relationships between spatial growth patterns in land allocation and multivariate causal elements is conducted via the Random Forest Classification (RFC) methodology. This machine learning approach capably manages datasets with numerous variables and intercorrelated predictors through aggregation of multiple decision trees cultivated from randomized subsamples of training data, ultimately generating development likelihoods for land use category k within spatial unit i. The mathematical representation can be expressed as:

$$\:\begin{array}{c}{P}_{i.k}^{d}\left(x\right)={\sum\:}_{n=1}^{M}I\left({h}_{n}\left(x\right)=d\right)/M \end{array}$$

(1)

where, x represents a multidimensional vector containing various driving factors; The binary variable d assumes values exclusively in {0,1}; When d equals 1, it signifies transformation from non-k land-use categories to type k, whereas 0 corresponds to transitions between other land-use classifications; The function h_n(x) characterizes the nth predictive category generated by decision trees processing vector x; The operator I(…) serves as a numerical indicator for decision node collections; The parameter M quantifies the aggregate count of constituent decision trees.

CARS

The CARS framework combines cellular automata-based stochastic seed creation and threshold relaxation approaches for predicting regional land utilization patterns over time. Functioning as a scenario-based analytical tool, this system addresses spatial-temporal variations in land allocation requirements at multiple scales. Through iterative computations, adjustment mechanisms govern resource competition among land categories to maintain consistency with projected requirements. The mathematical expression determining the composite likelihood for land category k development appears below:

$$\:\begin{array}{c}{OP}_{i,k}^{d=1,t}={P}_{i,k}^{d=1}\times\:{{\Omega\:}}_{i.k}^{t}\times\:{D}_{k}^{t} \end{array}$$

(2)

where, $\:{OP}_{i,k}^{d=1,t}$ signifies the expansion likelihood of spatial unit i transitioning to land-use category k. The neighborhood influence parameter $\:{{\Omega\:}}_{i.k}^{t}$ characterizes the spatial distribution ratio of category k within the specified neighborhood radius surrounding cell i. The adaptive driving coefficient $\:{D}_{k}^{t}$ quantifies regulatory impacts on projected land demand, being jointly determined by iteration count t and the gap between existing land allocation for category k and its predefined target. Probability distributions across all land categories are subsequently aggregated to establish a stochastic selection mechanism utilizing cumulative probability values, which governs spatial state transitions during iterative computations.

InVEST model

Employing the Habitat Quality module embedded in the InVEST modeling framework, the Habitat Quality Index (HQI) is computed to assess ecological habitat conditions across the research region, serving as an indicator of ecosystem functionality in delivering biodiversity conservation services¹⁵. This methodology establishes a correlation where superior habitat suitability corresponds with enhanced biological diversity preservation. The dimensionless HQI metric comprehensively evaluates regional habitat viability by integrating two critical dimensions: intrinsic environmental appropriateness and anthropogenic-induced deterioration levels across distinct land utilization categories. The computational procedure is mathematically expressed as:

$$\:\begin{array}{c}{Q}_{xj}={H}_{j}\left[1-\left({D}_{xj}^{z}/{D}_{xj}^{z}+{k}^{z}\right)\right] \end{array}$$

(3)

where, $\:{Q}_{xj}$ quantifies habitat quality metrics for raster cell x in landscape category j; $\:{H}_{j}$ specifies the habitat appropriateness rating (0–1 scale) associated with landscape classification j; Parameter z represents the dimensionless scaling factor conventionally assigned a value of 2.5; Variable k corresponds to the half-saturation coefficient, a user-defined parameter influenced by dataset granularity; $\:{D}_{xj}$ measures stress-induced ecological deterioration levels within the analyzed habitat unit. The mathematical relationship is expressed through the following equation:

$$\:\begin{array}{c}{D}_{xj}=\sum\:_{r=1}^{R}\sum\:_{y=1}^{{Y}_{r}}\left({\omega\:}_{r}/\sum\:_{r=1}^{R}{\omega\:}_{r}\right){r}_{y}{i}_{rxy}{\beta\:}_{x}{S}_{jr} \end{array}$$

(4)

where R represents the total count of threat factors. The variable $\:{Y}_{r}$ quantifies the aggregate grid cells occupied by stressors, while $\:{\omega\:}_{r}$ signifies their relative weighting coefficients. The term $\:{r}_{y}$ enumerates threat factors present within individual grid units. The accessibility parameter $\:{\beta\:}_{x}$ quantifies grid cell protection status through numerical scaling: maximum values (e.g., 1) denote strictly protected zones, minimum values (0) indicate exploitable areas, and intermediate values (0–1) represent varying protection intensities. Sensitivity coefficients $\:{S}_{jr}$ (0 ≤$\:{S}_{jr}$≤ 1) express landscape type j’s vulnerability to specific stressors. The stressor impact radius $\:{i}_{rxy}$ is determined through either linear or exponential decay formulations.

$$\:\begin{array}{c}{i}_{rxy}=1-{d}_{xy}/{d}_{rmax} \end{array}$$

(5)

$$\:\begin{array}{c}{i}_{rxy}=\text{exp}\left(-\left(2.99/{d}_{rmax}\right){d}_{xy}\right) \end{array}$$

(6)

where $\:{d}_{xy}$ is the linear distance between grid x and y; $\:{d}_{rmax}$ is the maximum effective distance of stressor r.

Essential data requirements for this module comprise: (1) contemporary land cover classification maps; (2) principal ecological stress factors; (3) weighting coefficients and spatial impact ranges of stress agents; (4) vulnerability parameters quantifying landscape responses to ecological pressures. Agricultural areas, urbanized zones, and undeveloped territories have been classified as ecological stress factors. Parameter configurations for the Habitat Quality module were developed based on established methodologies^6,8,15, with detailed specifications provided in Tables 2 and 3.

Table 2 Threat factors and their stress intensities.

Table 3 Sensitivity of land use type to habitat threat factors.

Results

Spatiotemporal analysis of LUCC characteristics in the TRSR

Spatiotemporal characteristics of LUCC

Following the terrestrial resource classification framework mandated by China’s statutory Land Administration Law, six primary land categories have been established within the research domain: agricultural fields, woodland areas, pasture zones, aquatic systems, built-up regions, and undeveloped territories. Through systematic application of ArcGIS 10.7 spatial analysis tools, core terrestrial resource datasets were obtained, enabling quantitative assessment of geospatial patterns characterizing land utilization across the TRSR throughout the 1990–2020 temporal spectrum.

Fig. 3

Spatial distribution of land use types in the TRSR from 1990 to 2020. *This is an original map created using ArcGIS 10.7. https://www.esri.com/en-us/arcgis/products/arcgis-desktop/overview⁴⁰.

Over the past 30 years, the areas of arable land, grassland, water bodies, and construction land in the TRSR increased by 319.81 km², 14,593.43 km², 1,888.67 km², and 234.32 km², respectively (Fig. 3). Grassland experienced the greatest expansion, consistent with the conclusions of³⁹ regarding the ecological fragility of the TRSR and reflecting the influence of the Three-River Source National Nature Reserve, established in 2003. Increases in arable land and water bodies are associated with the expansion of regional water resources, as arable land is water-dependent and typically located along riparian zones. In contrast, forest land and unused land decreased by 66.78 km² and 17,001.35 km², respectively, reflecting population growth, urbanization, economic development, deforestation, and the conversion of unused land to other uses. Spatially, the most substantial changes occurred in unused land, especially in the central counties of Qumarlai, Chengduo, and Maduo. Grassland expansion corresponded to areas of unused-land reduction, while construction-land expansion was concentrated in Gonghe County in the northeastern TRSR. Water bodies expanded mainly in Zhiduo, Golmud, Zaduo, and Maduo counties, whereas forest land increased in Gonghe, Nangqian, and Jiuzhi counties. Arable-land growth was most pronounced in Guinan, Tongde, Gonghe, and Xinghai counties.

Land use transition matrix analysis

Over the past 30 years, land-use transitions within the TRSR progressed from predominantly singular conversions to multiple shifts and, ultimately, diversified exchange patterns, although the extent of interconversion varied among categories, with distinct types serving as principal sources and sinks of change (Table 4). Between 1990 and 2000, transitions were largely unidirectional: grassland was chiefly converted to arable land, water bodies, and unused land, encompassing 120.09 km², 109.40 km², and 246.84 km², respectively. Water bodies, in turn, reverted primarily to grassland and unused land, covering 133.72 km² and 114.13 km². From 2000 to 2010, the diversity of transitions expanded markedly. Unused land exhibited the most substantial outflow, totaling 20,011.58 km², primarily into grassland. Grassland also experienced varied outflows into unused land, arable land, and water bodies. Notably, this period coincided with the establishment of national nature reserves and the rollout of ecological conservation and restoration initiatives. These initiatives included black soil beach management, desertification control, degraded grassland rehabilitation, wetland protection, and pest management, all contributing to the conversion of forest land and arable land into grassland. Between 2010 and 2020, land-use exchanges became even more dynamic: unused land remained the primary source of conversion, totaling 1,264.25 km², mainly into grassland and water bodies. Grassland continued to display the most diversified outflows, transitioning chiefly into forest land, water bodies, and unused land.

Table 4 Land use area change transfer matrix in the TRSR (km²).

Analysis of Spatiotemporal variation characteristics of habitat quality in the TRSR

The HQI assessment outcomes were analyzed through ArcGIS 10.7’s reclassification module, implementing a five-tier classification system via the natural breaks algorithm: minimal (0-0.2), reduced (0.2–0.5), moderate (0.5–0.6), elevated (0.6–0.8), and maximal (0.8-1.0) categories. Geospatial visualization outputs depicting habitat quality stratification were subsequently produced, accompanied by quantitative analysis of areal percentages and inter-category conversions across different quality levels, as documented in Table 5; Fig. 4.

Table 5 Proportion of habitat quality grades in the TRSR from 1990 to 2020.

As indicated in Table 5, between 1990 and 2022, minimal- and elevated-quality habitat zones exhibited area contraction, particularly pronounced in minimal-quality categories. By 2022, minimal-quality habitats attained their historical minimum coverage of 68,574.59 km², constituting 17.93% of total area, reflecting a 16,774.54 km² reduction and 4.39% proportional decrease over three decades. Elevated-quality habitats experienced a 17.33 km² diminishment. Conversely, maximal-, moderate-, and reduced-quality habitats demonstrated expansion across all categories. Of particular significance, moderate-quality habitats emerged as the predominant classification, both in absolute coverage (271,980.10 km² by 2020) and incremental growth, registering a 14,411.54 km² expansion relative to 1990 baseline measurements.

Habitat quality within the TRSR exhibits marked spatial heterogeneity. Moderate-quality areas are the most extensively distributed, whereas the Minimal‐quality zones coalesce into contiguous clusters. Elevated‐quality habitats occur in discrete patches dispersed throughout the region (Fig. 3). These elevated‐quality patches predominate in the south, east, and northwest, particularly around the Lancang River basin and adjacent montane landscapes. These areas are characterized by forest and aquatic land covers, which underpin elevated biodiversity and ecosystem integrity. Stringent policy frameworks and conservation measures have curtailed development, thereby minimizing anthropogenic disturbance. Surrounding the core elevated‐quality patches, zones of higher‐than‐average quality form protective buffers, chiefly in the southeastern and southern sectors. These buffers comprise shrubs, riparian beaches, and extensive grasslands with high vegetative cover. Intermediate‐quality habitats consist mainly of grassland communities with moderate to low canopy coverage. In contrast, minimal‐quality habitats, typically bordering built‐up areas, are dominated by croplands and exhibit greater prevalence in the eastern sector than in the west. The most degraded habitats concentrate in the northwest corner, eastern urban centers, and county seats, where vacant and constructed lands are prevalent. In the northwest, mountain ridges establish natural borders: south‐facing slopes support medium‐quality grasslands, whereas north‐facing slopes, characterized by exposed rock, scree, and bare ground, represent large expanses of lowest‐quality habitats, especially in Zhiduo County.

Overall, habitat quality across the TRSR exhibits a distinct east–west gradient, with higher values in eastern sectors, lower values in the west, and an overall predominance of intermediate-quality areas. Elevated‐quality habitats cluster tightly around Qinghai Lake and Zhaling Lake, forming discrete focal points within the lacustrine ecosystem network. This distribution pattern aligns with regional lake conservation priorities. As critical ecological resources, lakes are governed by environmental regulations that limit the conversion of adjacent territories to alternative land uses, thereby underscoring both the opportunities and constraints within the lake–land symbiosis ecological framework of the study area.

Fig. 4

Spatial distribution of HQ grades in the TRSR from 1990 to 2020. *This is an original map created using ArcGIS 10.7. https://www.esri.com/en-us/arcgis/products/arcgis-desktop/overview⁴⁰.

Figure 5 illustrates a net decline in habitat quality within the TRSR from 1990 to 2020. Degradation transitions predominantly occurred from higher to lower quality classes, whereas improvement transitions largely spanned from the lowest to intermediate classes. Specifically, during 1990–2000, degradation transitions affected 431.67 km² (0.11% of the study area), while improvements covered 687.56 km² (0.18%). Between 2000 and 2010, degradation transitions expanded to 22,443.8 km² (5.87%), and improvement transitions reached 6,017.34 km² (1.57%). From 2010 to 2020, degraded areas totalled 5,684.64 km² (1.49%), with improvements amounting to 4,882.70 km² (1.28%). Notably, the 2000–2010 decade experienced particularly pronounced degradation: although improvement transitions increased, their magnitude remained far lower than that of degradation, largely due to a significant reduction in grassland extent that precipitated sharp habitat-quality losses. Overall, habitat quality in the TRSR exhibited a clear downward trajectory over the 30-year period.

Fig. 5

Trend of Habitat Quality grade shift in the SRTR from 1990 to 2020.

Land use change simulation and forecasting

Implementation of the PLUS modeling framework forecasts TRSR’s 2050 land use configuration (Fig. 6), determining spatial distribution patterns and transition matrices across land categories. Projection results suggest the region will contain 2,356.85 km² cultivated areas, 16,772.88 km² forested zones, 280,908.50 km² grazing lands, 22,744.46 km² aquatic systems, 375.44 km² developed spaces, and 59,342.47 km² undeveloped territories by mid-century. Spanning 2020–2050, persistent reductions are projected for cultivation areas, forested regions, and undeveloped lands, with the latter demonstrating maximum contraction (-8,682.10 km²). Expansion patterns emerge in grazing territories (+ 7,671.81 km²) and aquatic systems (+ 1,381.34 km²), while the most substantial alterations occur in undeveloped lands, predominantly clustered within Geermu City, Zhiduo County, and Zaduo County, with minor concentrations in Guinan County’s central sector. Cumulatively, the 2020–2050 timeframe reveals declining trends in undeveloped and cultivated lands contrasting with aquatic and grassland increments, constituting predominant land transformation dynamics in TRSR, where ecological land restoration through undeveloped land conversion forms the principal axis of territorial change.

Fig. 6

Prediction of land use types in Three River 2050. *This is an original map created using ArcGIS 10.7. https://www.esri.com/en-us/arcgis/products/arcgis-desktop/overview⁴⁰.

Table 6 Land use transfer matrix in the TRSR from 2020 to 2050.

Between 2022 and 2030, Table 6 reveals that conversions to grassland within the TRSR originated predominantly from forest land (2,151.27 km²), water bodies (1,656.62 km²), and unused land (3,532.40 km²), totaling 7,593.72 km² of inflow. Grassland outflow was chiefly directed toward forest land (2,173.85 km²), water bodies (1,364.90 km²), and unused land (11,107.88 km²), amounting to 15,178.02 km². Concurrently, unused land received 12,706.60 km² of inflow from grassland and water bodies, while its outflow to those classes totaled 4,007.54 km². The most intense reciprocal conversion occurred between unused land and grassland, with the shift from unused land to grassland being most pronounced. These results suggest that stricter ecological protection measures will continue to drive the transformation of unused land into ecological land. Consequently, a sustained increase in ecological land is expected to remain a long-term trend in TRSR land-use change, marking a critical stage in the nationwide construction of the Three Rivers Source ecological barrier.

Habitat quality change simulation and forecasting

Employing projected 2050 landscape data for the TRSR, this study applied the InVEST modeling framework to simulate future ecological habitat conditions (Fig. 7) and quantify habitat quality tier transformations (Table 7). Analysis reveals substantial expansion in intermediate (7,539.63 km² increase) and elevated-quality (1,246.28 km² gain) habitat zones relative to 2020 baselines, constituting 73.08% and 4.88% of total regional coverage respectively. Moderately high and reduced-moderate habitat tiers demonstrated relative stability, exhibiting marginal variations of 116.81 km² and 92.88 km² over three decades, with higher tiers showing incremental gains and lower tiers slight reductions. Reduced-quality habitats contracted to 59,886.20 km², marking an 8,688.39 km² decrease, while intermediate habitats maintained dominant spatial proportions. Although significant habitat quality disparities persist across the TRSR, the fundamental spatial configuration remains largely consistent, with lacustrine systems persisting as elevated-quality habitat nuclei. The region maintains an east-west quality gradient (elevated eastern values, diminished western levels) with intermediate habitats predominating. Comparative analysis of 2020–2050 data confirms sustained spatial patterns, where major hydrological systems continue anchoring elevated-quality habitats within the characteristic “lake-grassland symbiosis” framework that defines regional ecological architecture.

Fig. 7

Spatial distribution of habitat quality grades in the TRSR in 2050. *This is an original map created using ArcGIS 10.7. https://www.esri.com/en-us/arcgis/products/arcgis-desktop/overview⁴⁰.

Table 7 Habitat quality grade transition matrix in the TRSR from 2020 to 2050 /km².

As detailed in Table 7, habitat quality degradation in the TRSR during 2020–2050 predominantly manifests as shifts from moderate-to-reduced (11,125.45 km²) and elevated-to-moderate (2,446.70 km²) quality tiers. Secondary degradation pathways include high-to-moderate, high-to-low, and moderate-to-reduced transitions, collectively accounting for 17,113.30 km² of degraded habitats, reflecting progressive expansion of reduced-quality zones. Quality improvements primarily involve reduced-to-moderate (3,602.85 km²) and moderate-to-maximal (2,377.01 km²) conversions, with minor contributions from reduced-to-elevated, moderate-to-elevated, and maximal-to-elevated transitions, totaling 7,945.8 km² of enhanced habitats. This pattern reveals that accelerated developmental processes and escalating environmental pressures will drive habitat degradation areas to significantly exceed improvement zones (17,113.30 km² vs. 7,945.8 km²) over the 50-year projection, establishing a persistent downward trajectory in regional habitat quality.

Response of habitat quality to land use change

Habitat contribution coefficients encompassing both positive and negative values, quantify the ecological benefits or detriments of land use modifications, where absolute magnitudes reveal intervention intensities. Limited-scale cross-category landscape transitions yield HCCs approaching neutral values (Fig. 8). Through integrated analysis of land use transition matrices and spatial patterns (Fig. 7), habitat quality dynamics were evaluated through enhancement and degradation lenses. Primary ecological enhancements stem from unused land transformations into grasslands (7,319.95 km²) and aquatic systems (610.65 km²) during 1990–2020, supplemented by constructive conversions of urban and agricultural lands to grasslands. Conversely, habitat degradation predominantly arises from high-to-low quality habitat transitions, including arboreal to grassland conversions, aquatic/grassland transitions to unused lands, and water-to-grassland modifications. The most extensive landscape modification involves grassland-to-unused land transitions (22,876.75 km²), representing the largest conversion category. Aquatic system conversions to grassland (1,978.68 km²) and unused land (1,981.60 km²) constitute additional significant transformations.

Fig. 8

Primary habitat quality contribution rate in the TRSR. 1. Cultivated land, 2. Forest land, 3. Grassland, 4. Waters, 5. Construction land, 6. Unutilized

During the 1990–2020 period, habitat quality contributions within the TRSR demonstrated a persistent downward trajectory. Projections for 2022–2050 indicate stabilization of cross-category landscape transformations across the region. Relative to the preceding three decades, conversions from unused land to aquatic systems show modest increments, generating limited positive impacts on habitat integrity. Comprehensive analysis reveals declining contribution coefficients across all land classifications. TRSR’s habitat quality oscillations exhibit strong correlations with landscape pattern alterations. Ecological preservation zones dominated by grassland and aquatic ecosystems serve as critical stabilizers, while unused territories inhibit ecological restoration potential. Both historical (1990–2020) and projected (2022–2050) data demonstrate that environmentally detrimental landscape modifications surpass beneficial conversions in spatial magnitude. This imbalance has driven continuous habitat degradation throughout the observation period, with models predicting sustained ecological decline through mid-century.

To systematically examine the influence of landscape modifications on habitat integrity fluctuations, this investigation employs quantitative variations in the HQI as analytical foundation. Land cover types demonstrating superior habitat integrity - including forested areas, grasslands, and aquatic systems - are aggregated into a unified classification designated “natural surfaces.” Unused territories exhibiting spatial transferability are maintained as independent classification units, whereas built-up areas displaying diminished habitat values are classified as “artificial surfaces.” This methodology produces a four-category land use classification map. Utilizing land cover imagery spanning the study period (1990–2020), spatial conversion patterns between land use types were extracted. Habitat quality changes were quantified through comparative analysis of HQI spatial distributions across the temporal sequence. Based on identified conversion relationships, statistical segmentation was performed to differentiate regions exhibiting habitat quality improvements, degradations, and stability, with corresponding areal extents and proportional distributions calculated. During the thirty-year study period, the transformation of unused territories into natural surfaces emerged as the predominant land conversion process, accounting for 6.53% of the total study area. Subsequent shifts were observed in natural surface conversions to unused land (2.09%) and arable land (0.15%), as detailed in Table 8.

Table 8 The relationship between habitat quality index change and land use conversion 1990–2020.

Land use patterns exhibit nonlinear associations with HQI variations. Even in regions maintaining stable HQIs, mutual transformations persist among cultivated land, natural surfaces, artificial surfaces, and unused land. The most substantial conversion involves unused land transitioning to natural surfaces (2,769.26 km²), followed by the reverse process where natural surfaces revert to unused land (1,742.16 km²). Notably, HQI fluctuations persist in non-conversion areas, with 1.09% of the study area demonstrating quality improvements and 0.67% showing declines, though both proportions remain substantially smaller than those observed in conversion-impacted zones. The most significant HQI enhancements originate from conversions of unused land and cultivated land to natural surfaces. Artificial surface-to-natural surface transitions and unused land-to-cultivated land conversions also demonstrate positive impacts. Conversely, the most severe HQI deterioration arises from natural surface conversions into unused land, cultivated land, or artificial surfaces, with secondary declines associated with cultivated land transitioning to artificial surfaces or unused land. Collectively, the dominant land use conversion pattern involves transitions from both unused and cultivated lands towards natural surfaces. While land use conversions significantly influence HQI variations, HQI alterations also manifest in non-conversion areas, underscoring the multifaceted interdependence between habitat quality and land use dynamics.

Discussion

Habitat quality forecasting based on the PLUS-InVEST coupled modeling approach

This investigation adopts an integrated PLUS-InVEST modeling framework to evaluate terrestrial utilization patterns and ecological habitat conditions within the TRSR. Developed as an enhanced iteration of the FLUS framework, the PLUS modeling architecture enables spatial projections of landscape utilization through 2050 by simulating potential landscape configurations derived from developmental probability matrices of terrestrial categories⁴³. Optimal model performance is achieved when baseline landscape patterns and spatial drivers exhibit strong congruence with empirical observations. For model calibration and verification, historical terrestrial data (2010–2020) coupled with associated spatial drivers were employed to predict 2020 landscape configurations, subsequently validated against empirical 2020 datasets to refine algorithmic parameters. Extending this calibrated framework using 2020 baseline data, we generated 2050 landscape projections. Subsequent application of the InVEST paradigm facilitated habitat condition simulations based on projected 2050 terrestrial patterns⁴². Spatiotemporal analyses of habitat integrity, as visualized through graphical and tabular outputs, confirm substantial consistency between model-derived habitat assessments and field observations. Given the ecological dominance of grassland ecosystems and underutilized territories in the SRSR, coupled with its hydrological significance as a critical conservation zone reinforced by China’s two-decade environmental protection initiatives, overall habitat integrity remains relatively stable. However, localized degradation trends and their spatial coalescence into larger degraded clusters demand focused attention. This research confirms the methodological reliability of the PLUS-InVEST framework for analyzing landscape dynamics and habitat condition trajectories.

Dynamic patterns and regional differences in habitat quality

In the study area, grassland ecosystems predominate significantly, accounting for nearly 70% of the landscape composition. Aquatic systems and forested areas represent other substantial natural cover types. These natural surface features demonstrate extensive spatial distribution across the territory, with their most concentrated contiguous distributions occurring in central and eastern sectors. Arboreal vegetation manifests dispersed patterns primarily in southeastern zones. Beyond substantial lacustrine formations, wetland ecosystems and fluvial networks permeate the entire research domain. The ecological configuration of this territory exhibits distinctive strengths and attributes, establishing it as a crucial environmental safeguard for China. Analysis of habitat quality contributions (Fig. 7) reveals that unused land expansion has partially displaced ecologically functional landscapes. Documented instances of anthropogenic activities and infrastructure development encroaching upon existing vegetated zones further confirm that both unused land proliferation and human interventions substantially influence ecological equilibrium⁴⁴. Concurrent research indicates that dependence on natural ecosystem regulatory capacities alone proves inadequate for ecological rehabilitation, necessitating active management strategies⁴⁵. Regarding TRSR’s ecological recovery and sustainability, scholarly proposals emphasize implementing vegetation structural rehabilitation through establishment of stratified plant communities mirroring natural assemblages, thereby ensuring environmental continuity³⁹. Future strategies for enhancing habitat quality in the Three-River Source Region should adopt a Nature-based Solutions (NbS) perspective, integrating socio-ecological systems and applying NbS theoretical frameworks to implement adaptive management. By combining natural restoration with active rehabilitation approaches, these efforts will strengthen the region’s resilience to climate change and human activities, thereby enhancing the capacity for coordinated and sustainable ecosystem development.

Regional habitat quality response to land use change

Characterized by biodiversity and intricate ecological networks, the investigated region maintains relatively intact ecosystem conditions. Nevertheless, ecosystem architecture and habitat conditions display substantial spatial variations stemming from inherent ecological intricacies and uneven regional socioeconomic development. Core conservation zones dominated by grassland vegetation demonstrate moderate-to-high habitat integrity levels, contrasting sharply with extensive low-quality habitats concentrated in northwestern mountainous territories where underutilized lands prevail. Similarly, diminished habitat values are observed in northeastern sectors dominated by agricultural and urbanized landscapes. A distinct correlation emerges between landscape utilization patterns and ecological habitat conditions: grassland-dominated areas correlate with superior habitat integrity, whereas urbanized and underutilized zones correspond to degraded conditions. Projections indicate future habitat deterioration will predominantly manifest in northwestern TRSR territories. Land transformation patterns reveal unused-to-water conversions as the dominant process by 2020, shifting toward unused-to-grassland transitions by 2050. Landscape modification emerges as the principal catalyst for habitat condition alterations, with modified areas experiencing substantial ecological shifts compared to stable regions showing minimal fluctuations. This pattern highlights the multifaceted interplay between habitat integrity and landscape transformations. As mathematically expressed in Eqs. (3)-(6), habitat integrity depends on three key parameters: ecological suitability coefficients, threat factor density/distance metrics, and habitat vulnerability indices⁴⁶. The precise mechanistic connections between landscape modifications and habitat integrity warrant deeper scientific exploration.

Conclusions

The TRSR and its network of national nature reserves play vital roles in ecosystem preservation. Leveraging three-decade land use/cover datasets combined with satellite observation technologies, this research employs the integrated PLUS-InVEST modeling framework to investigate historical and projected land use transformations and their ecological consequences. Key findings indicate: (1) From 1990 to 2020, grassland coverage, aquatic areas, and cultivated land increased by 14,593.43 km², 1,888.67 km², and 319.81 km² respectively, contrasting with shrinking unused territories. Continuation of current patterns predicts further reduction in unused lands alongside sustained expansion of grasslands and water resources over coming decades. (2) Ecosystem integrity has progressively weakened during the past 30-year period, showing net degradation exceeding recovery (0.67%). Unchecked development would accelerate habitat deterioration, potentially doubling affected zones within three decades. (3) While some locations demonstrate habitat quality variations without land use modifications, altered regions manifest more significant ecological shifts. Strategic conservation interventions must constrain unused land proliferation and address its cascading ecological impacts. Recent initiatives in Qinghai Province focus on restoring vulnerable ecosystems and priority conservation zones through comprehensive landscape management, encompassing terrestrial and aquatic systems while advancing studies on spatial ecological variations, driving factors, and underlying processes. Sustained implementation of protection and rehabilitation strategies remains crucial to counterbalance potential habitat quality declines from unused land expansion, harmonize biodiversity conservation objectives, and foster sustainable regional development. Near-term monitoring and fencing of critical unutilized land zones; Mid-term integration of Nature-based Solutions (NbS) into climate adaptation plans; Long-term establishment of multi-stakeholder platforms for participatory land-use planning—ensuring ecological, climatic, and community sustainability in the TRSR. Furthermore, this study incorporates the differential effects of land categories on habitat quality. Utilizing enhanced prediction parameters within the PLUS model, we simulated land use patterns in the Three-River Source Region for the next three decades. This methodology demonstrates superior performance compared to CA and Random Forest models, exhibiting reduced prediction errors, operational efficiency, and broad applicability across regional, national, and global-scale land type projections. However, constrained by conventional modeling frameworks and formulae, the research lacks validation through empirical measurements. Certain data limitations—whether due to accessibility constraints or inherent gaps—necessitated the use of estimation techniques, potentially introducing discrepancies between simulated outcomes and actual conditions. Future work should integrate field-measured data to refine model parameters.