Research datas and medhods on Habitat Quality and Biodiversity Conservation Functions of Land Use Transformation in the Yellow River Basin, China

This paper explores the interactions between land use, habitat quality and biodiversity via spatiotemporal habitat quality monitoring in the Yellow River Basin (2008–2023), and proposes targeted recommendations. We optimized the InVEST model with new parameters for the biodiversity maintenance capacity index was calculated using the NPP quantitative assessment method combined with the technical guidelines for delineating ecological protection redlines, pioneered standard deviation ellipses for trend and driver analysis, and applied SHAP-integrated machine learning for multiscale coupling analyses, thereby establishing a novel analytical framework for relevant research and monitoring.
To demonstrate the connection between habitat quality, ecosystem services, and biodiversity protection within the YSB, this study employs the basin as its research unit. It integrates natural and socioeconomic demographic data, such as elevation, net primary production (NPP), and land use. Landsat TM remote sensing imagery from 2008 to 2023 is used to evaluate land use data; additional data sources are provided in Table 1 below. ArcGIS software was used to resample and project all the data, resulting in a single spatial resolution of 1 km × 1 km in the WGS_1984_Albers coordinate system.
1.Habitat Quality Assessment—InVEST Model
Based on land-use type conversion, the InVEST model objectively captures habitat dynamics driven by such conversion and responds to variations in habitat quality and biodiversity abundance. Integrating regional land use patterns, the relative impact of biodiversity threats, the sensitivity of different habitat types to threats, and the distance between habitat grids and threat sources, the model quantifies landscape suitability, threat sensitivity, and threat source distance and weight. It then calculates the study area’s habitat quality index and generates a habitat quality map. In essence, habitat quality is a dimensionless function of habitat suitability and degradation degree across regional land use types, with the latter representing habitat degradation caused by the following threat factors:

The model's default parameter is, andis the semi-saturation parameter, which is equivalent to half the maximum degradation level.represents the habitat quality of grid cellwithin land use type,indicates the habitat suitability of land use type, andindicates the habitat degradation level of grid cellwithin land use type.

In this equation,stands for the habitat's threat source,for the grid cell inside threat source,for theth threat source's weight,for theth threat source's intensity inside grid cell,for theth threat source's influence attenuation function on grid cellwithin grid cell, andfor the accessibility of grid cell.indicates how sensitive land use j is to threat source,is the separation between grid cell(habitat) and grid cell(threat source), andis the threat source's maximum influence range.
The study region identified agriculture, urban land, rural residential land, industrial/mining and transportation land, and unutilized land as threat factors based on pertinent research and the Yellow River's actual situations. They were identified by their spatial decay types, weights, and maximum influence distances. Parameters including habitat suitability and susceptibility to threat variables were then determined. The habitat quality of the YSB was assessed by extracting threat factor layers.
2.Biodiversity Conservation Function Index
The biodiversity maintenance capacity index was calculated using the NPP quantitative assessment method combined with the technical guidelines for delineating ecological protection redlines. This index uses the following methods to describe the strength of biodiversity maintenance functions:

In the equation,denotes the biodiversity conservation service capacity index,represents net primary productivity of vegetation,indicates annual precipitation,signifies annual mean temperature, andreflects the altitude factor.
3.Standard Deviation Elliptical
The element's principal spatial distribution range is represented by the SDE's distribution range, and its mean center functions as the element's spatial distribution center of gravity. Important traits including the aggregation center, evolutionary dynamics, and spatial orientation are correlated with the ellipse's center position, area size, and rotation angle, respectively. To investigate the geographical evolution patterns of biodiversity conservation functions in the YSB, the following methodology uses SDE:

The centroid coordinates are represented byw andw in Equations (5) to (8); the weight is represented by; the elliptical azimuth is represented by; the coordinate deviations from the mean center for each study object are indicated byi andi respectively; the standard deviations along the x-axis and y-axis are indicated byand, respectively; and**is the elliptical area.
4.XGBoost Machine Learning
Based on the gradient boosting architecture, XGBoost is a potent machine learning algorithm created especially to improve prediction accuracy . It excels at handling large datasets, shows exceptional capabilities in regulating model complexity and reducing overfitting, and gradually builds a series of weak learners to improve model performance. This study uses the XGBoost model to accurately identify the intricate driving processes controlling the biodiversity maintenance function in the YSB. In particular:

The expected value of biodiversity conservation functionality in the YSB is represented by the equation's. The feature vector of sample, represented by, includes various factors that affect the functionality of biodiversity protection. The predicted value of theth base learner for sampleis. The set of all potential base learners is represented by, while the total number of base learners is represented by.
The following objective function, which focuses on reducing model error while managing complexity, is optimized to accomplish model training:

The error function in the equation,**, quantifies the difference between the expected valueand the actual coupling coordination degree. The regularization termis intended to limit the base learner's complexity to avoid overfitting and improve the model's performance on unknown inputs.
5.Shapley additive explanation
The XGBoost model cannot intuitively explain the precise direction and magnitude of each driver’s impact on biodiversity conservation functions. To address this black-box limitation, the SHAP approach is introduced, which draws on the Shapley value from game theory to assess the relative importance of individual features in machine learning models. The final effect value is derived by weighted averaging the marginal contribution of each feature across different combinations. As a highly interpretable tool applicable to both local and global explanations, SHAP’s calculation formula is presented below:

The contribution of driving factorto the final prediction is represented by the SHAP value,, in the formula.is the total set of all features,is a subset ofthat does not include feature i, andis the model prediction result for the combination of features in subset.
Conclusion
This study employs novel parameters, models, driving factors and machine learning methods to explore the spatiotemporal evolution of land use, habitat quality and biodiversity conservation in the Yellow River Basin. It enhances readers’ understanding of the basin’s ecological and biodiversity dynamics, prompts reflections on land use and ecological conservation practices, advances multi-scale coupling research in this field, addresses key issues of uncalibrated assessment models and discontinuous biodiversity datasets, and offers targeted solutions to pressing challenges.