InVEST Model

The habitat quality module of the InVEST model assumes that areas with higher habitat quality can support greater species richness, and habitat quality changes will lead to biodiversity changes [39]. The calculation of habitat quality requires four main components: the relative impact of each threat factor, the sensitivity of each land use type to each threat factor, the distance between land use and the source of the threat, and the degree of legal protection. The InVEST model equation is as follows [40]:

(linear decay)

(exponential decay)

where D_xj, R, W_r, Y_r, and r_y represent the habitat degradation index, the number of threat factors, the weight of threat factor r, the number of threat factor grids, and the value of threat factors on the grid, respectively. i_rxy represents the distance between habitat and threat sources and the spatial impact of threats. β_x is a factor that mitigates the effects of threats on habitats through various conservation policies. S_jr represents the sensitivity of habitat type j to threat factor r. d_xy is the straight-line distance between grids x and y, and d_rmax is the maximum influence range of threat factor r. The habitat quality can be estimated as follows:

where Q_xj represents the habitat quality index of grid x for land use type j; H_j represents the habitat suitability of habitat type j, with values ranging from 0 to 1; k is the half-saturation constant, typically set to half of the maximum habitat degradation value; and z is a normalization constant, usually set to 2.5 [40].

The inputs for this module primarily include land use data, major threat factors, the weights and impact distances of threat factors, and the sensitivity of each land use type to each threat source. Based on the InVEST model user guide and previous research [41- 44], we specifically considered the distribution of nature reserves, the implementation of reforestation and grassland restoration policies, and the processes of industrialization and urbanization in the study area to ensure the model's applicability and accuracy. The selected threat factors include paddy fields, dryland, urban land, rural residential areas, other construction land, and unused land. The habitat suitability and threat sensitivity for each land use type are listed in Table 2, along with the relative weight of each threat factor, the maximum impact distance, and the type of spatial decay (Table 3).

FLUS Model

The FLUS model was employed to simulate the spatial distribution of land use. This model improves the traditional Markov chain model and incorporates suitability probability calculations, neighborhood factor calculations, adaptive inertia coefficient calculations, transition cost settings, and comprehensive probability calculations [30]. The suitability probability calculation uses a BP-ANN algorithm to fit the baseline land use types with various spatial driving factors, obtaining the suitability probability of different land use types [29, 30]. The expression for this is:

where sp(p, k) is the suitability probability, W_{j, k} is the weight between the hidden layer and the output layer, N_i(p, q) represents the signal received by neuron j in the hidden layer from the input layer, x_i(p, q) is the input value of cell p for neuron iii at iteration q, and W_{i, j} is the weight between the input layer and the hidden layer. The neighborhood factor represents the interaction between land use types and land use cells within the neighborhood range. Its expression is:

Fig. 2.

Land use structure change in the YRB.

Fig. 3.

Land use transfer of the YRB from 2000 to 2020.

Fig. 4.

Land use distribution map from 2000 to 2020.

where is the neighborhood influence factor of cell p at time t; represents the total number of cells of land use type k within the Moore neighborhood of N × N in the last iteration at time t - 1. W_k represents the neighborhood factor parameter for each land use type, with values ranging from 0 to 1, and the value is positively correlated with land use expansion capability.

Geodetector

The Geodetector is a new statistical method used to detect spatial heterogeneity and uncover the driving factors behind it. The single-factor and interaction detection methods can measure spatial heterogeneity, explore explanatory factors, and analyze the interaction between variables [45, 46]. Its expression is:

where q represents the explanatory power of a factor on the spatiotemporal variation of habitat quality; h is the number of classifications or partitions of the factor; L refers to the stratification of the independent or dependent variable; N_h and N represent the number of units in layer h and the total number of units in the study area, respectively; and σ_h² and σ² are the variances of the independent variable factor and the dependent variable habitat quality spatiotemporal variation, respectively. The range of q values is [0, 1], with larger q values indicating stronger explanatory power of the independent variable factor for the spatiotemporal variation of habitat quality and smaller values indicating weaker explanatory power.

Considering the natural geographical characteristics and the intensity of human activities in the YRB, and based on relevant literature and expert opinions, we selected 14 driving factors: altitude (X1), slope (X2), aspect (X3), terrain niche index (X4), RDLS (X5), soil type (X6), vegetation coverage (X7), annual average temperature (X8), annual precipitation (X9), land use (X10), population density (X11), GDP (X12), impervious surfaces (X13), and road distance (X14). A systematic sampling of the YRB was conducted with a grid size of 500 m, resulting in a total of 3,255,852 sample points.

Habitat Quality Analysis

The equal interval method was used to divide the scores into 5 levels of habitat quality status, including lower (0 ~ 0.2), low (0.2 ~ 0.4), middle (0.4 ~ 0.6), high (0.6 ~ 0.8), and higher (0.8 ~ 1.0). The average habitat quality values in 2000, 2010, and 2020 were 0.5052, 0.5069, and 0.5068, respectively, all within the middle range and showing an overall improving trend. Habitat quality of the YRB was predominantly at a middle level, followed by low and high levels. From a temporal perspective, the areas of high and higher habitats continuously increased, while those of low habitat quality decreased steadily. The area of middle habitat quality first decreased and then increased, while the area of lower habitat quality initially increased and then decreased. From a spatial distribution perspective, the overall habitat quality of the study area exhibited a pattern of high values in the western and southeastern regions, middle values in the central region, and low values in the northern region, which largely aligned with the distribution of land use (Fig. 5). The areas of higher value were primarily located in the southeastern part of the YRB, dominated by forestland. The area of high value was mainly distributed in the southwestern part of the basin, characterized by forestland, grassland, and water bodies. In contrast, lower and low habitat quality areas were concentrated in the eastern and northern parts of the basin, where urban and rural living land, industrial and mining land, and unused land were prevalent. The middle habitat quality area was mainly concentrated in the central part of the basin, primarily in grassland regions.

Fig. 5.

Habitat quality classification map of the YRB from 2000 to 2020.

Fig. 6.

Habitat quality degradation level map of the YRB from 2000 to 2020.

Fig. 7.

Interactive detection values of habitat quality factors in the YRB.

From 2000 to 2010, areas with declining habitat quality were mainly concentrated around the Weihe River Basin, the Yellow River Delta, and parts of Shaanxi, Sichuan, Qinghai, and Shanxi. Regions where habitat quality improved were scattered. From 2010 to 2020, areas with declining habitat quality were distributed more extensively but showed improvement compared to the 2000-2010 period. In particular, habitat quality in the Yellow River Delta showed improvement. Over the entire period from 2000 to 2020, the habitat quality index of the YRB initially increased and then stabilized, with some areas maintaining constant habitat quality. Regions experiencing severe degradation were mainly located in the Qilian Mountains, parts of the plains in the lower reaches of the Yellow River, the Guanzhong Basin, and the Taihang Mountains. Areas where habitat quality improved were distributed across the Loess Plateau, the Lanzhou New Area, the Ningxia Plain, and the transitional area of the Mu Us Desert.

Habitat Degradation Analysis

The habitat degradation index was also assessed, with degradation categorized into 5 levels: lower (0), low (0 ~ 0.03), middle (0.03 ~ 0.06), high (0.06 ~ 0.09), and higher (> 0.09). The average degradation indices for 2000, 2010, and 2020 were 0.01653, 0.01650, and 0.01644, respectively, indicating a gradual decrease. This suggests that habitat quality has improved slightly over the study period.

From a spatial distribution perspective, the overall spatial pattern of habitat degradation across the three periods remained relatively stable (Fig. 6). Habitat degradation showed a spatial distribution characterized by "high in the center, low at the periphery, high in the east, and low in the west". Areas with lower habitat degradation were primarily forestland and unused land. Low degradation areas were mostly grassland and water bodies, with a more continuous spatial distribution. Medium degradation areas were mainly cultivated land, while areas with high and higher levels of habitat degradation were construction land and rural residential areas, showing a scattered distribution, primarily located in the lower reaches of the Yellow River. From a temporal perspective, between 2000 and 2020, the areas with lower and middle habitat degradation levels first increased and then stabilized, increasing by 0.56% and 0.61%, respectively. The areas with high and low degradation decreased by 0.17% and 1.07%, respectively. The proportion of areas with high degradation levels was very small, and the changes in these areas were also minimal.

Accuracy Validation

Based on land use data from 2000 and 2010, the CA-Markov model within the FLUS framework was used to calculate the number of land use grid cells for 2020. These results were compared with the actual land use in 2020. Across 4 scenarios, the average prediction accuracy of the Markov chain prediction was 0.9192, indicating that the model could accurately predict land use quantities. Furthermore, the FLUS model's simulation of the production-living-ecological space coordination priority scenario for 2020 (Fig. 8) was compared with the actual land use data for 2020. The Kappa coefficient was 0.8927, and the overall accuracy was 0.9253. This demonstrates that the FLUS model has a high degree of accuracy and can effectively reflect future land use conditions in the YRB.

Fig. 8.

Land use status and simulation results in the YRB in 2020

Fig. 9.

Simulated prediction of land use in the YRB by 2030 under various sce-narios.

Future Land Use Scenario Analysis

Based on the actual conditions of the YRB, simulations were conducted for the years 2030 (Fig. 9) and 2050 (Fig. 10) under 4 scenarios: production space priority, living space priority, ecological space priority, and production-living-ecological space coordination. The results show significant differences in land use changes and habitat quality between the scenarios. In the production space priority scenario, cultivated land and industrial and mining construction land increase, leading to decreased ecological space. Urban and rural living land expands rapidly in the living space priority scenario, leading to significant habitat degradation. In contrast, the ecological space priority scenario shows an increase in ecological space, particularly in forestland and grassland areas, with habitat quality improving significantly. The production-living-ecological space coordination scenario balances the demands of production, living, and ecological spaces, resulting in moderate improvements in habitat quality.

(1) In the production space priority scenario, forestland, grassland, and unused land are the primary types of land being converted. The main goals of this scenario are to protect cultivated land and ensure sufficient industrial and mining construction land. The model achieves this by increasing the probability of other land types converting to cultivated land and reducing the probability of cultivated land converting to other land types. Additionally, the demand for production space increases by the target years, raising the cost of converting cultivated land to other types. Conversely, the cost of converting other land types into cultivated land, except for urban and rural living land and industrial and mining construction land, is reduced to prioritize production space. From 2020 to 2030, the production space within the "Three-Life Space" of the YRB is projected to increase by 7.02×10³ km², with cultivated land increasing by 6.97×10³ km² and industrial and mining construction land increasing by 50 km². The living space will also expand by 1.05×10² km², while the ecological space is expected to decrease by 7.13×10³ km². As production and living spaces continue to grow, ecological space is projected to decrease further by 1.57×10² km² between 2030 and 2050;

Fig. 10.

Simulated prediction of land use in the YRB by 2050 under various scenarios.

Fig. 11.

Habitat quality simulation results in the YRB by 2030 under various scenarios.

Fig. 12.

Habitat quality simulation results in the YRB by 2050 under various scenarios.

Fig. 13.

Simulation results of habitat quality degradation in the YRB by 2030 under various scenarios.

(2) In the living space priority scenario, cultivated land, grassland, and unused land are the primary land types being converted. The primary land type for living space is urban and rural living land. In the model, this scenario is realized by increasing the probability of other spaces converting to living space while restricting the conversion of living space land types to other categories, achieving the goal of prioritizing living space development. Under this scenario, ecological space continuously decreases. From 2020 to 2030, ecological space is expected to decrease by 2.89×10³ km²; from 2030 to 2050, it will decrease by 1.42×10² km². Regarding land expansion capability, industrial and mining construction and urban and rural living land have strong expansion potential with similar land use characteristics. Cultivated land, grassland, and forestland are the main land types being converted. As urban areas expand, on the one hand, the continuous expansion of living space is observed, with an increase of 1.97×10³ km² in living space from 2020 to 2030 and 3.37×10³ km² from 2030 to 2050. On the other hand, transportation infrastructure and industrial and mining construction land should also increase and improve to facilitate more convenient living conditions. From 2020 to 2030, industrial and mining construction land is projected to increase by 1.33×10³ km², greater than the decrease in cultivated land by 4.11×10² km², resulting in an overall increase in production space. However, from 2030 to 2050, industrial and mining construction land is expected to continue increasing by 1.24×10³ km², while cultivated land will decrease by 4.47×10³ km², leading to a slight overall decline in production space;

Fig. 14.

Simulation results of habitat quality degradation in the YRB by 2050 under various scenarios.

(3) In the ecological space priority scenario, cultivated and unused land are the primary types being converted. The main goal is to protect ecological space from encroachment while encouraging the conversion of other land types into ecological space to restore ecosystems. The model achieves this by strictly controlling the conversion of ecological space-related land types into living and production space while moderately increasing the conversion of production and living spaces into ecological space. From 2020 to 2030, production space is expected to decrease by 4.3×10³ km², living space will increase by 1.97×10³ km², and ecological space will increase by 2.33×10³ km². From 2030 to 2050, production space is projected to further decrease by 1.75×10³ km², living space will increase by 3.37×10³ km², and ecological space will decrease by 1.62×10³ km². This indicates that production space will become the primary land type being converted under this scenario. Between 2020 and 2030, production space shifted to ecological and living spaces. However, from 2030 to 2050, a large amount of unused land will be converted into living space, leading to an increase in living space and a reduction in ecological space;

(4) The production-living-ecological space coordination scenario shares similar land use conversion patterns with the living space priority scenario, though the changes are somewhat more moderate. This scenario aims to balance the multiple demands of optimizing production space, prioritizing living space, and prioritizing ecological space. The model achieves this by following the principle of conversion costs, increasing the conversion cost for industrial and mining construction land, and controlling the conversion of ecological space land types. The results indicate that the land use area proportions in the production-living-ecological space coordination scenario are very similar to those in the ecological space priority scenario. In terms of area changes, from 2020 to 2030, production space is expected to decrease by 4.27×10³ km², ecological space will increase by 2.30×10³ km², and living space will increase by 1.97×10³ km². From 2030 to 2050, production space is projected to decrease by 1.78×10³ km², ecological space will decrease by 1.59×10³ km², and living space will increase by 3.37×10³ km². The pattern of change is similar to that in the ecological space priority scenario.

Future Habitat Quality Analysis

The habitat quality of the YRB in 2030 (Fig. 11) and 2050 (Fig. 12) was simulated and evaluated under 4 different scenarios. The results show that ecological space exhibits a growth trend in both the ecological space priority scenario and the production-living-ecological space coordination scenario, while it decreases in the other 2 scenarios. Production space shows a growth trend only in the production space priority scenario and a decreasing trend in the other 3 scenarios, with the decrease being particularly significant. Living space shows a growth trend in all scenarios, but the growth is smaller in the production space priority scenario.

In 2030, the average habitat quality values for the YRB under the 4 scenarios (production space priority, living space priority, ecological space priority, and production-living-ecological space coordination) were 0.52612, 0.52687, 0.52882, and 0.52880, respectively. By 2050, the values were 0.53006, 0.52586, 0.53110, and 0.53111, respectively. Compared to 2020, the average habitat quality in 2030 and 2050 shows variation across different scenarios. In the living space priority scenario, the average habitat quality of the YRB is projected to decline further in the future, while in the production space priority, ecological space priority, and Three-Life Space coordination scenarios, the average habitat quality is expected to improve.

(1) In the production space priority scenario, compared to 2020, the habitat quality index of the YRB improved in both 2030 and 2050. The area of lower habitat quality decreases by 0.3% and 0.95%, respectively, while the area of higher habitat quality increases by 3.31% and 3.32%. As a result, there is a slight improvement in the habitat quality index. However, expanding industrial and mining construction land leads to further degradation of the local ecological environment, causing the habitat quality in 2030 under this scenario to be lower than that of the other scenarios;

(2) In the living space priority scenario, the accelerated urban expansion and rapid socioeconomic development result in a significant increase in urban and rural living land, while other land types experience substantial reductions. This leads to a further increase in lower habitat quality areas, particularly concentrated in the northern part of the YRB and the middle and lower reaches of the river. Compared to 2020, the lower habitat quality area will increase by 0.43% by 2050. Additionally, between 2030 and 2050, habitat quality will decline further, with a decrease of 1.1%;

(3) In the ecological space priority scenario, there is a significant improvement in habitat quality across the region. By 2030 and 2050, the area of higher habitat quality in the YRB will increase by 3.8% and 4.06%, respectively, compared to previous years. The area of lower and low habitat quality decreases. The regions experiencing the greatest increase in higher habitat quality are mainly located in areas where forestland has expanded. Forestland, as a key habitat type, receives better protection under this ecological priority scenario;

(4) In the production-living-ecological space coordination scenario, habitat quality is very similar to that in the ecological space priority scenario. The areas of lower and low habitat quality decrease, while those of higher habitat quality increase.

Overall, the spatial distribution of lower habitat quality areas across the 4 scenarios aligns closely with the distribution of urban and rural living land, industrial and mining construction land, and unused land. This indicates that expanding urban and rural living land, industrial and mining construction land, and unused land directly affect the distribution of lower-value habitat areas within the basin. The distribution of low habitat-quality areas corresponds to the distribution of cultivated land. Additionally, water bodies consistently fall within higher or high habitat quality areas across all scenarios. Although the quantity and spatial distribution of forest land and grassland vary in the simulations under different scenarios, they consistently remain in the higher habitat quality areas.

Future Habitat Degradation Analysis

In 2030, the average habitat degradation index under the 4 scenarios (production space priority, living space priority, ecological space priority, and production-living-ecological space coordination) was 0.02462, 0.02448, 0.02421, and 0.02422, respectively. By 2050, the values were 0.02469, 0.02464, 0.02438, and 0.02439, respectively. Overall, the ranking of habitat degradation is as follows: production space priority > living space priority > production-living-ecological space coordination > ecological space priority. Compared to 2020, the average habitat degradation index in 2030 and 2050 shows variations across the different scenarios.

According to the evaluation results (Fig. 13, Fig. 14), during the periods of 2020-2030 and 2030-2050, the overall changes in habitat degradation were relatively small, but the habitat degradation index showed a gradual increase over time. Although the focus of each simulation differs among the 4 scenarios, the higher habitat degradation areas are consistently concentrated in the eastern part of the basin. These regions have many cities and experienced rapid economic development, leading to significant human disturbance and lower habitat quality. This pattern closely aligns with the spatial distribution of urban and rural living land, indicating that expanding urban-rural residential areas is a key factor driving the increase in lower habitat quality areas. Overall, from 2020 to 2050, the spatial distribution of habitat degradation continues to exhibit the pattern of "high in the center, low at the periphery, high in the east, and low in the west".