A Multi-Scenario Analysis of Urban Vitality Driven by Socio-Ecological Land Functions in Luohe, China

Abstract

Urban Vitality (UV) is a critical indicator for measuring sustainable urban development and quality. It reflects the dynamic interactions and supply–demand coordination within urban systems, especially concerning the human–land relationship. This study aims to quantify the UV of Luohe City, China, for the year 2023, analyze its spatial characteristics, and investigate the driving patterns of socio-ecological land functions on UV intensity and heterogeneity under different scenarios. Utilizing multi-source data, including human mobility data from Baidu Location-Based Services (LBSs), Landsat-9, MODIS, and diverse geo-information datasets, we conducted factor screening and comprehensive assessments. Firstly, Self-Organizing Maps (SOMs) were employed to identify typical activity patterns, and the Urban Vitality Index (UVI) was calculated based on Human Mobility Intensity (HMI) data. Subsequently, a framework for quantity–quality–structure assessments weighted and aggregated sub-indicators to evaluate the Land Social Function (LSF) and Land Ecological Function (LEF). Following the screening process, a Multi-scale Geographically Weighted Regression (MGWR) was applied to analyze the scale and driving relationships between UVI and the land assessment sub-indicators. The results were as follows: (1) The UV distribution in Luohe City was highly uneven, with high vitality areas concentrated within the built-up regions. (2) UV showed significant correlations with both LSF and LEF. The influence of LSF on UV was stronger than that of LEF, with the effectiveness of LEF relying on the well-established provisioning of LSF. (3) Artificial Surface Ratio (ASR) and Corrected Night Lights (LERNCI) were identified as key drivers of UV across multiple scenarios. Under the weekend scenario, the Green Space Ratio (GSR) and the Vegetation Quality (VQ) notably enhanced the attractiveness of human activities. (4) The impacts of drivers varied at the urban, township, and street scales. The analysis focuses on factors with significant bandwidth changes across multiple scenarios: VQ, Remote-Sensing-based Ecological Index (RSEI), GSR, ASR, and ALSI. This study underscores the importance of socio-ecological land functions in enhancing urban vitality, offering valuable insights and data support for urban planning.

1. Introduction

Vitality is a critical driver of sustainable urban development [1]. It serves as an essential indicator for evaluating the quality of urban development [2,3]. Vibrant cities exhibit continuous and efficient development [4]. Conversely, a lack of vitality can lead to urban decay or sprawl, as exemplified by China’s “ghost cities” and Japan’s “edge villages”. In the context of the United Nations’ 2030 Agenda and Sustainable Development Goals (SDGs) [5], China’s urbanization rate is projected to reach 73% by 2030, necessitating sustained urban land development [6]. To optimize human settlements and construct “production-living-ecological” spaces, indicators as effective management tools are crucial. Quantifying Urban Vitality (UV) and revealing its responses to various urban subsystems, particularly the land system, is of great significance for improving human–land relationships.

Jane Jacobs first introduced “vitality” in urban planning, linking it to street activity and diversity [7]. Later, Alexander (1965) suggested that natural cities’ inherent qualities shape vitality [8], while Kevin Lynch (1984) expanded the idea, stating that vitality indicates how well urban forms support essential functions, ecological needs, and human capabilities [9]. Maas (1984) described UV as a spatial quality arising from varied commercial opportunities and a diverse, dense pedestrian population [10]. Montgomery (1998) saw vitality as a hallmark of successful urban areas, manifested in the high levels of human and street activity [11,12]. Chinese scholars, such as Jiang (2007) [13], constructed a framework based on social, economic, and cultural dimensions, positing that UV is the lifeblood of the city, reflecting its ability to provide a humanized living environment for residents. While academia lacks a precise definition of UV, it generally includes two main aspects: the intensity and richness of human activities. Scholars have identified typical indicators like human “activity”, “aggregation”, and “mobility” to operationalize UV [4,14,15]. Secondly, the concept highlights the city’s role in sustaining human life, emphasizing the interactions between people and urban spaces. Drawing on the theory of complex systems, this broad concept of “vitality” reveals the coupled characteristics of cities as dynamic, organic entities composed of multiple systems [16]. Long and Zhou [17] employed big data to construct a framework analyzing factors influencing vitality at the urban street level and discussed the influence of these factors on different types of functional streets through regression analysis. Additionally, scholars have analyzed the mechanisms influencing regional vitality using multi-source data and methods such as correlation and machine learning. These analyses have explained the impact of various forms and elements and proposed corresponding optimization strategies [18,19].

Early studies on UV primarily focused on constructing evaluation indicators, constrained by the limitations of data acquisition. Methods such as field surveys and questionnaires [20,21] could not provide multidimensional data with sufficient temporal and spatial precision. With the advent of the information and digital age, new urban images have emerged through the use of OpenStreetMap (OSM) GPS data, social media check-in data, and smart card data [22,23,24]. However, these data often necessitate corrections using additional data, such as road accessibility [25], due to spatial sampling biases. Mobile movement data and human mobility data based on Location-Based Services (LBS) have gained significant attention due to their balanced spatial coverage and fine-grained temporal resolution [19,26]. Overall, UV research has transitioned from qualitative theoretical explorations to quantitative-method-based measurements. The continuous refinement of data sampling’s spatiotemporal dimensions drives a shift towards more diversified and human-centered research perspectives.

Despite these advancements, many studies still handle raw data in a relatively coarse manner. A common practice is to directly utilize the human mobility quantity or intensity of sampling areas as a singular representation of UV. However, this approach fails to fully investigate the temporal characteristics of the data [26,27]. Additionally, research on the driving mechanisms behind UV tends to focus more on the driving coefficients of various factors or the spatial heterogeneity of vitality, with less emphasis on human–land interactions, scale dependence, and scenario comparisons [19,28]. From an urbanization perspective, urban land development and the aggregation of production and living elements inevitably lead to corresponding changes in human activities. If urban expansion and land development intensity are not balanced with human activity patterns, urbanization may face significant sustainability challenges [29,30].

The paper is divided into three sections: First, it introduces a new UV assessment method using Baidu human mobility data. Unlike traditional methods that directly depend on activity intensity, this approach analyzes the data’s characteristics and uses time-based self-calibration for more accurate UV measurement. Secondly, we assess land functions using diverse data sources. By applying the socio-ecological subsystem framework, we gain a comprehensive understanding of the spatial variability in land sub-functions across quantity, quality, and structure, enabling a detailed and holistic view of land functions. Finally, we examine the human–land relationship. Spatial correlation analysis shows UV’s dependency on various land functions. Using Multi-scale Geographically Weighted Regression (MGWR), we study the driving relationships and bandwidth changes in urban land functions and their sub-indicators on UV across multiple scenarios. Unlike OLS and GWR models, MGWR accounts for geographic characteristics and specific impact scales. The analysis identifies two categories of driving factors: Scale-Invariant and Scale-Variable. Notably, Scale-Variable factors emerge as more noteworthy regulators of the human–land relationship.

2. Research Materials

2.1. Study Area

Luohe City (113°27′–114°16′ E, 33°24′–33°59′ N) is located in the south-central part of Henan Province in central China. The terrain is predominantly flat, bisected by the Sha Li River. This division creates three distinct zones within the urban area: the old town in the south, the new town in the north, and the peninsula area in the west (Figure 1). Luohe has a medium level of economic development (ranking 15th out of 18 cities in Henan Province’s GDP in 2022) but boasts a rapid growth rate (ranking 3rd out of 18 cities in GDP growth rate in 2022).

The economic and urbanization levels of Luohe City are representative of many typical yet often overlooked cities in China. Therefore, the study of Luohe can provide a prototypical case for addressing issues related to land development and urban vitality in these cities. Additionally, Luohe’s relatively flat terrain, lack of prominent natural landscapes and historical sites, and limited peri-urban green spaces present a unique research opportunity. This specific context allows for a more focused analysis of the relationship between urban land development and UV based on human activity.

2.2. Data Acquisition and Preprocessing

2.2.1. Data Sources and Collection

Human mobility data were sourced from Baidu Maps. Although this dataset does not directly reflect the actual number of human activities, it has been proven to be a reliable indicator of the real-time intensity distribution of human activities [26]. The total data collection period spanned four weeks (encompassing four activity cycles), distributed across four seasons: 11–16 April 2023; 11–16 July 2023; 11–16 October 2023; 15–21 January 2024. During the sampling period, there were no extreme weather events. The data boast a high temporal precision (hourly) and spatial resolution (200 m × 200 m).

Remote sensing images were acquired from the atmospherically corrected Landsat-9 surface reflectance collection (L9 SR, Dataset: LANDSAT/LC09/C02/T1_L2, Resolution: 30 m). Land Surface Temperature (LST) was derived from the L9 SR Thermal Infrared (TIR) bands using the single-channel empirical algorithm [31]. Gross Primary Productivity (GPP) data were obtained from preprocessed MODIS satellite datasets (GPP Dataset: MODIS/061/MOD17A2H, Resolution: 500 m). Nighttime Light (NTL) data were extracted from the monthly composite data of the NPP-VIIRS satellite (Dataset: VIIRS Nighttime Day/Night Band Composites Version 1, Resolution: 460 m). Population data (POP) were sourced from the WorldPop dataset (Dataset: WorldPop/GP/100m/pop, Resolution: 100 m). The Digital Elevation Model (DEM) data were sourced from the Copernicus DEM dataset (Dataset: COPERNICUS/DEM/GLO30, Resolution: 30 m).

The above remote sensing datasets were processed using the Google Earth Engine platform (GEE, https://code.earthengine.google.com/, accessed on 4 April 2024). Road network data were obtained from OSM (https://www.openstreetmap.org/, accessed on 4 April 2024) and subsequently manually corrected using Google Maps. Text and yearbook data were sourced from the Henan Provincial Bureau of Statistics (https://tjj.henan.gov.cn/tjfw/tjsj/, accessed on 4 April 2024).

2.2.2. Data Processing

Cloud Removal and Composite of Remote Sensing Data: The 2023 L9 SR dataset was utilized, and pixels corresponding to Dilated Cloud (Bit 1), Cloud (Bit 3), and Cloud Shadow (Bit 4) were excluded based on the QA_PIXEL band. To address null values and ensure data stability, the annual median was calculated for each pixel. This process combined multiple scene images into a single annual composite dataset.

Population Data Adjustment and Correction: As the WorldPop population dataset extends only to 2020, an adjustment approach was employed to account for the time gap. Considering the relative stability of population growth and data consistency, we constructed a spatiotemporal cube using historical data from 2000 to 2020. Subsequently, a linear fitting model was built per pixel to predict the 2023 population spatial distribution. To linearly correct the raster data, the district-level permanent population data for 2023, obtained from urban yearbooks, were used to compute a scaling factor.

GPP/NTL Composite and Correction: The raw NTL data were provided as monthly composites, and GPP data were provided as eight-day composites. Both datasets were processed into annual data by calculating the mean value for each pixel. Given the common “blooming” effect in NTL/GPP data, a mask was generated based on the 30 m land classification data (described in detail later) to exclude NTL/GPP values over the water surface.

Identification of Built-up Area Boundaries (199.09 km²): We use three datasets, including NTL, which shows the level of local urban development; Normalized Difference Building Index (NDBI), which captures radial mutations in urban building density; LST, which helps to identify industrial and warehousing zones on the periphery of the city. The built-up zone boundaries were determined based on the Expectation Maximization (EM) clustering method [32].

3. Methods

3.1. Urban Vitality Assessment

3.1.1. Human Mobility Intensity and Self-Organizing Map

Based on traditional human mobility indicators, this paper introduces the application of the Self-Organizing Map (SOM) model. It is an unsupervised neural network technique for visualizing and interpreting large, high-dimensional datasets [33]. In this study, the SOM model was used to segment the 24 h day into distinct periods, revealing the city’s typical behavioral pattern characterized by a 16 h active period and an 8 h rest period. This method is combined with the K-means algorithm for a two-stage clustering process. In the first stage, SOM acts as an initial clustering method, determining a reasonable range for cluster number N and assigning initial cluster centers. Subsequently, based on sample similarity, K-means is employed to optimize the clustering achieved by SOM [34]. The clustering effectiveness is then evaluated using the Davies–Bouldin Index (DBI).

The analysis was performed using the R-kohonen package. The specific process was as follows: (1) Hourly Human Mobility data for 24 h across four weeks (a total of 28 days) were input into the SOM model. To achieve optimal network performance, the number and combination of neurons were selected based on the lowest quantization error (QE) and topographic error (TE). (2) The weights obtained from the SOM clustering results were used as the initial cluster centers. The optimal number of clusters for K-means was determined using the DBI. Finally, Human Mobility Intensity (HMI) for both workdays (Monday to Friday) and weekends (Saturday and Sunday) was calculated as the hourly mean of human mobility during active periods:

$$\mathrm{H}\mathrm{u}\mathrm{m}\mathrm{a}\mathrm{n} \mathrm{M}\mathrm{o}\mathrm{b}\mathrm{i}\mathrm{l}\mathrm{i}\mathrm{t}\mathrm{y} \mathrm{I}\mathrm{n}\mathrm{t}\mathrm{e}\mathrm{n}\mathrm{s}\mathrm{i}\mathrm{t}\mathrm{y}:\mathrm{H}\mathrm{M}\mathrm{I}={\displaystyle \frac{\sum _{\mathrm{i}=1}^{\mathrm{n}}\hspace{0.17em}{\mathrm{v}}_{\mathrm{i}}}{\mathrm{n}}}$$

(1)

where v_i is the human mobility data at the ith hour of the active period, and n is the total number of hours within the active period (16 h in this study). The HMI value reflects the average level of human mobility during active periods, calculated separately for both workdays and weekends, with higher values indicating higher population density and activity intensity within the area.

3.1.2. Atypical Activity Adjustment Coefficient

Using HMI as the base value for UV assessment, it is important to note that human activities in certain urban areas may not align with the typical behavioral patterns identified by SOM. This discrepancy is particularly common in night-shift workplaces, such as some industrial zones or nighttime recreational venues like bars and night markets:

$$\mathrm{A}\mathrm{c}\mathrm{t}\mathrm{i}\mathrm{v}\mathrm{i}\mathrm{t}\mathrm{y} \mathrm{R}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{o}:\mathrm{A}\mathrm{R}={\displaystyle \frac{\frac{\sum _{\mathrm{t}=1}^{\mathrm{m}}\hspace{0.17em}{\mathrm{v}}_{\mathrm{t}}}{\mathrm{m}}}{\mathrm{H}\mathrm{M}\mathrm{I}}}$$

(2)

where v_t is the HMI at the t_th hour of the rest period, and m is the total number of hours in the rest period (8 h in this study). The term ($\sum _{\mathrm{t}=1}^{\mathrm{m}}\hspace{0.17em}{\mathrm{v}}_{\mathrm{t}}$)/m reflects the average human mobility during the rest period on workdays and weekends. An AR value greater than 1 indicates that the area does not follow the typical activity pattern. The larger the AR value, the more the area tends toward activity during the rest period. The Atypical Activity Adjustment Coefficient (AAAC) was then calculated to adjust the HMI accordingly. The AAAC for the active period on workdays and weekends was computed as follows:

$$\mathrm{A}\mathrm{t}\mathrm{y}\mathrm{p}\mathrm{i}\mathrm{c}\mathrm{a}\mathrm{l} \mathrm{A}\mathrm{c}\mathrm{t}\mathrm{i}\mathrm{v}\mathrm{i}\mathrm{t}\mathrm{y} \mathrm{A}\mathrm{d}\mathrm{j}\mathrm{u}\mathrm{s}\mathrm{t}\mathrm{m}\mathrm{e}\mathrm{n}\mathrm{t} \mathrm{C}\mathrm{o}\mathrm{e}\mathrm{f}\mathrm{f}\mathrm{i}\mathrm{c}\mathrm{i}\mathrm{e}\mathrm{n}\mathrm{t}:\mathrm{A}\mathrm{A}\mathrm{A}\mathrm{C}=\left\{\begin{array}{c}if\mathrm{AR}\le 1,1\\ if\mathrm{AR}>1,\mathrm{AR}\end{array}\right.$$

(3)

Therefore, for areas with an AR less than or equal to 1, the AAAC is set to 1, indicating no adjustment to the HMI. For areas with an AR greater than 1, the HMI is adjusted proportionally to the AR value.

3.1.3. Temporal Characteristics Weighting of HMI

In addition to HMI, three indicators were calculated to comprehensively measure the spatiotemporal patterns of UV: Leisure Activity Ratio (LAR), Human Mobility Variability (HMV), and Human Mobility Consistency (HMC). Tang and Ta [3] used the night rate to indicate the tendency for leisure activities within an area. A higher night rate signifies more human activities during the night, reflecting the extent of nighttime economic and social activities. However, in this study, the intensity and proportion of nighttime activities are better represented by the AAAC. The night rate indicator mainly shows the tendency for leisure activities during post-work or evening periods on rest days. Therefore, the indicator was renamed LAR. LAR for workdays and weekends was calculated as follows:

$$\mathrm{Leisure}\mathrm{Activity}\mathrm{Ratio}:\mathrm{L}\mathrm{A}\mathrm{R}={\displaystyle \frac{\sum _{21}^{23}\hspace{0.17em}\mathrm{v}}{\sum _{21}^{23}\hspace{0.17em}{\mathrm{v}}_{\mathrm{i}}+\sum _9^{11}\hspace{0.17em}{\mathrm{v}}_{\mathrm{i}}}}$$

(4)

Activity variability is defined as the difference in human mobility numbers at different times, represented by the Standard Deviation (STD) of hourly human flow data throughout a 24 h period [14]. This is considered a negative indicator, as higher STD values indicate greater fluctuations in activity intensity across the day. To account for significant differences in average human mobility numbers across different areas, the Coefficient of Variation (CV) was used to enhance data comparability. HMV for active periods on workdays and weekends was calculated as follows:

$$\mathrm{Human}\mathrm{Mobility}\mathrm{Variability}:\mathrm{H}\mathrm{M}\mathrm{V}=\mathrm{C}\mathrm{V}={\displaystyle \frac{\mathrm{S}\mathrm{T}\mathrm{D}}{\mathrm{M}\mathrm{e}\mathrm{a}\mathrm{n}}}$$

(5)

where Mean represents the average human activity over 24 h.

Lastly, the measurement of activity consistency utilized 2.5 times the Median Absolute Deviation (MAD) to calculate the number of hours with outlier activity intensity at each sampling point daily. This value is then normalized by the total number of global outliers for that specific day [25]. HMC for active periods on workdays and weekends was calculated as follows:

$$\mathrm{Human}\mathrm{Mobility}\mathrm{Consistency}:\mathrm{H}\mathrm{M}\mathrm{C}={\displaystyle \frac{\mathrm{R}\mathrm{e}\mathrm{g}\mathrm{i}\mathrm{o}\mathrm{n}\_\mathrm{o}\mathrm{u}\mathrm{t}\mathrm{l}\mathrm{i}\mathrm{e}\mathrm{r}\mathrm{s}}{\sum \mathrm{D}\mathrm{a}\mathrm{i}\mathrm{l}\mathrm{y}\_\mathrm{o}\mathrm{u}\mathrm{t}\mathrm{l}\mathrm{i}\mathrm{e}\mathrm{r}\mathrm{s}}}$$

(6)

3.1.4. CRITIC Model and Multi-Scenario Urban Vitality Intensity

The issue of multicollinearity among indicators needs careful consideration. Therefore, the Criteria Importance Through the Intercriteria Correlation (CRITIC) model is used to weight the aforementioned indicators [35]. By weighting and combining the three indicators, we obtain the final weighted coefficient for the temporal characteristics of HMI. This method imposes a penalty on highly correlated data. The specific formula is as follows:

$${\mathrm{W}}_{\mathrm{j}}={\displaystyle \frac{{\mathrm{C}}_{\mathrm{j}}}{\sum _{\mathrm{j}=1}^{\mathrm{p}}\hspace{0.17em}{\mathrm{C}}_{\mathrm{j}}}}$$

(7)

The CRITIC method, based on Multi-Criteria Decision-Making (MCDM), determines the weights by combining indicator variability S_j (i.e., the degree of dispersion of this indicator), indicator conflict R_j (i.e., the irrelevance between different indicators), and the information-carrying capacity C_j (obtained by multiplying S_j and R_j). These indicator features are calculated within one data group (p group) of the j indicator. The method is highly advanced in addressing covariance issues among different urban study indicators.

After weighting, the HMI is corrected to derive the Urban Vitality Index (UVI):

$$\mathrm{U}\mathrm{V}\mathrm{I}=\mathrm{H}\mathrm{M}\mathrm{I}\times \mathrm{A}\mathrm{A}\mathrm{A}\mathrm{C}\times ({\mathrm{W}}_1\times {\mathrm{n}\mathrm{o}\mathrm{r}}^{+}\left(\mathrm{L}\mathrm{A}\mathrm{R}\right)+{\mathrm{W}}_2\times {\mathrm{n}\mathrm{o}\mathrm{r}}^{-}\left(\mathrm{H}\mathrm{M}\mathrm{V}\right)+{\mathrm{W}}_3\times {\mathrm{n}\mathrm{o}\mathrm{r}}^{+}\left(\mathrm{H}\mathrm{M}\mathrm{C}\right))$$

(8)

where W₁, W₂, and W₃ are the weights corresponding to each coefficient, nor⁺ represents the normalization of positive indicators, and nor⁻ represents the normalization of negative indicators:

$${\mathrm{n}\mathrm{o}\mathrm{r}}^{+}={\displaystyle \frac{\mathrm{X}-{\mathrm{X}}_{\mathrm{m}\mathrm{i}\mathrm{n}}}{{\mathrm{X}}_{\mathrm{m}\mathrm{a}\mathrm{x}}-{\mathrm{X}}_{\mathrm{m}\mathrm{i}\mathrm{n}}}}$$

(9)

$${\mathrm{n}\mathrm{o}\mathrm{r}}^{-}={\displaystyle \frac{{\mathrm{X}}_{\mathrm{m}\mathrm{a}\mathrm{x}}-\mathrm{X}}{{\mathrm{X}}_{\mathrm{m}\mathrm{a}\mathrm{x}}-{\mathrm{X}}_{\mathrm{m}\mathrm{i}\mathrm{n}}}}$$

(10)

X is the original data point, X_min is the minimum value of that dataset, and X_max is the maximum value of that dataset.

Compared to a single dependent variable, the UVI in this study is distinguished by its occurrence time and is categorized into UVI under non-volitional human activity scenarios (i.e., workday UVI) and UVI under volitional human activity scenarios (i.e., weekend UVI).

3.2. Land Function Assessment

Land use often manifests in both explicit and implicit forms [36,37]. During periods of low human activity, area serves as the primary driving factor and represents the explicit outcome of human activities. However, as human activity increases, the forms and functions of land use, the social/ecological benefits generated per unit of land, and the patch morphology, distribution, and connectivity at the class scale undergo significant changes, which are implicit expressions of human activity outcomes. Based on this understanding, we constructed a comprehensive framework to evaluate the spatial heterogeneity of land functions, considering both the social and ecological functions of land. This framework integrates quantity, quality, and structure (

Table 1. Land Function Assessment Framework and Indicators.

Goal Layer	Framework	Criteria Layer	Indicators
Land Social Function (LSF)	Quantity	Spatial Service	Artificial Surface Rate (ASR)
	Quality	Urbanization Level	LST and EVI regulated NTL city index (LERNCI)
		Transportation Accessibility	Weighted Integration (WI)
		Population Density	WorldPop Population Density (POP)
	Structure	Artificial Surface Patch Morphology	Artificial Surface—Landscape Shape Index (ALSI)
		Artificial Surface Patch Aggregation	Artificial Surface—Patch Cohesion Index (ACO)
Land Ecological Function (LEF)	Quantity	Basic Green Space Service	Green Space Rate (GSR)
	Quality	Vegetation Cover	Vegetation Quality (VQ)
		Ecological Benefits	Remote Sensing Ecological Index (RSEI)
	Structure	Green Space Patch Morphology	Green Space—Landscape Shape Index (GLSI)
		Green Space Patch Aggregation	Green Space—Patch Cohesion Index (GCO)

).

3.2.1. Land Classification/Ratio and Landscape Pattern Index

Random Forest has been widely applied in land classification studies across global cities [38,39]. Using the annual composite data from L9 SR, we first compressed the B2–B7 spectral bands into three principal component bands via Principal Component Analysis (PCA) as spectral features. Subsequently, we calculated the Normalized Difference Vegetation Index (NDVI), NDBI, Modified Normalized Difference Water Index (MNDWI), and Soil Index (SI) as index features. Texture features were derived from the higher spatial resolution panchromatic band (B8, resolution: 15 m) using the Gray Level Co-occurrence Matrix (GLCM), from which we extracted the GLCM mean and GLCM variance. These three feature sets were input into the Random Forest classifier, resulting in the Land Use and Land Cover (LULC) for 2023, categorized as (1) Artificial Surface, (2) Green Space, (3) Water, and (4) Farmland (overall accuracy = 89.95%, kappa accuracy = 86.43%). Based on HMI data, a 200 m-scale fishnet grid was created, converting the classified LULC data into continuous land use rates as quantity indicators for land function assessment.

Based on previous studies, the scale, shape, quantity, type, and spatial configuration of landscapes significantly impact ecological security [40] and correlate with human activities [27,41]. Therefore, we initially screened indices such as the Aggregation Index (AI), Patch Density (PD), Largest Patch Index (LPI), Landscape Shape Index (LSI), Edge Density (ED), Effective Mesh Size (MESH), and Patch Cohesion Index (COHESION) to reflect the primary characteristics of landscape patterns [42]. Using FRAGSTATS 4.2 [43], we calculated these indices with a moving window method and selected those with contributions greater than 0.90 to the first principal component (PC1) through PCA (

Table 2. Factor Loading Factors.

	PC1	ED	LPI	LSI	MESH	PD	AI	COHESION
Factor Loadings	Artificial Surface	0.852	0.731	0.902	0.584	0.666	0.915	0.964
	Green Space	0.916	0.719	0.912	0.52	0.723	0.861	0.949

, Bartlett’s test of sphericity passed). Ultimately, the LSI was chosen to describe patch morphology and COHESION to describe patch aggregation.

3.2.2. Vegetation Quality

The assessment of Vegetation Quality (VQ) in this study follows the National Ecological Environment Standard (HJ 1172-2021, China) [44]. It integrates three indicators: Fractional Vegetation Cover (FVC), Leaf Area Index (LAI), and Gross Primary Productivity (GPP), each given equal weight:

$$\mathrm{V}\mathrm{Q}={\displaystyle \frac{{\mathrm{n}\mathrm{o}\mathrm{r}}^{+}\left(\mathrm{F}\mathrm{V}\mathrm{C}\right)+{\mathrm{n}\mathrm{o}\mathrm{r}}^{+}\left(\mathrm{L}\mathrm{A}\mathrm{I}\right)+{\mathrm{n}\mathrm{o}\mathrm{r}}^{+}\left(\mathrm{G}\mathrm{P}\mathrm{P}\right)}{3}}$$

(11)

This study employs the pixel binary model to calculate the FVC, which expresses remote sensing information of a pixel as a combination of vegetation and bare soil information [45] derived from reprocessed NDVI data. LAI is calculated using the empirical formula developed by Boegh et al. [46] through linear computation of the Enhanced Vegetation Index (EVI). The GPP data were directly obtained from well-established MODIS datasets. Higher VQ values represent better green volume, healthier vegetation, and more efficient ecosystems.

3.2.3. Remote-Sensing-Based Ecological Index

The Remote-Sensing-based Ecological Index (RSEI) is valued for its visibility, objectivity, and comparability [47], making it one of the widely used ecological assessment methods. RSEI combines greenness (NDVI), wetness (WET), dryness (NDBSI), and heat (LST) indicators from May to October 2023. After processing these indicators through PCA, the first principal component (PC1) is extracted and normalized to obtain the objective ecological index:

RSEI = 1 − nor⁺ (PC1 (NDVI, WET, NDBSI, LST))

(12)

3.2.4. LST and EVI Regulated NTL City Index

NTL images have long been used to monitor urban areas and their expansion characteristics [48]. However, due to pixel saturation and low spatial resolution, it is often necessary to apply cross-satellite gain and correction to extract more spatial information. This study employs the LST and EVI Regulated NTL City Index (LERNCI) as a weighting method [49]. Compared to other common indices, LERNCI introduces non-vegetation information, specifically thermal bands, which better correlate with economic and human activities:

$$\mathrm{L}\mathrm{E}\mathrm{R}\mathrm{N}\mathrm{C}\mathrm{I}={\mathrm{n}\mathrm{o}\mathrm{r}}^{+}\left(\right({\displaystyle \frac{\mathrm{L}\mathrm{S}\mathrm{T}}{\mathrm{L}\mathrm{S}\mathrm{T}\mathrm{A}\mathrm{V}\mathrm{G}\mathrm{s}}}+{\displaystyle \frac{\mathrm{E}\mathrm{V}\mathrm{I}\mathrm{A}\mathrm{V}\mathrm{G}\mathrm{s}}{\mathrm{E}\mathrm{V}\mathrm{I}}})\times \mathrm{N}\mathrm{T}\mathrm{L})$$

(13)

LSTAVGs and EVIAVGs represent the average LST and EVI values of all pixels in the urban core area (regions where NTL exceeds the 95th percentile).

3.2.5. Weighted Road Density

Road networks are the lifeblood of cities, often positively correlating with the quality of commercial, medical, cultural, and educational services. However, road grades and locations are not entirely homogeneous. Therefore, we introduce the road Integration Index (I_i) based on the space syntax–segment model, weighted by road length, to obtain Weighted Integration (WI). This process is conducted using Depthmap10.14.00b software, and vector road network files are converted to raster using the line density tool in ArcGIS Pro 2.8. Integration, a specific concept in space syntax, consists of the number of path networks (N) and the topological depth (TD) of the path [25,50]:

$$\mathrm{W}\mathrm{I}={\displaystyle \frac{\mathrm{N}\left[{\mathrm{l}\mathrm{o}\mathrm{g}}_2\left(\frac{\mathrm{N}+2}{3}-1\right)+1\right]}{(\mathrm{N}-1)|\mathrm{T}\mathrm{D}-1|}}\times \mathrm{W}\mathrm{e}\mathrm{i}\mathrm{g}\mathrm{h}\mathrm{t}\mathrm{e}\mathrm{d}$$

(14)

Areas with high WI typically correspond to major urban roads with high traffic volume. The weighted model enhances the calculation of road “corner” costs, better aligning with the continuous movement mechanism of both pedestrian and vehicular traffic.

3.2.6. Comprehensive Indicators and Weight Calculation

In the comprehensive assessment of land functions, it is also necessary to address the multicollinearity. Additionally, considering that residents are the ultimate service recipients of urban land development, subjective weights are incorporated. Subjective weights are calculated using the Analytic Hierarchy Process (AHP), a systematic decision analysis tool that employs minimal mathematical information in decision processes to provide a simple method for multi-objective, multi-criteria, or unstructured, complex decision problems [51]. By using SPSS Pro 1.1.27 for pairwise comparison judgment matrices, the weights of each indicator are derived. The consistency of the AHP model is evaluated using the Consistency Ratio (CR), represented by the λ indicator. The λ indicator measures the rationality of the AHP model, where a value less than 0.1 indicates an acceptable level of consistency and, thus, the model’s usability. The formula for calculating the final valuation result is as follows:

$$\mathrm{L}\mathrm{S}\mathrm{F}={\mathrm{W}}_1\times \mathrm{A}\mathrm{S}\mathrm{R}+{\mathrm{W}}_2\times \mathrm{P}\mathrm{O}\mathrm{P}+{\mathrm{W}}_3\times \mathrm{L}\mathrm{E}\mathrm{R}\mathrm{N}\mathrm{C}\mathrm{I}+{\mathrm{W}}_4\times \mathrm{W}\mathrm{I}+{\mathrm{W}}_5\times \mathrm{A}\mathrm{L}\mathrm{S}\mathrm{I}+{\mathrm{W}}_6\times \mathrm{A}\mathrm{C}\mathrm{O}$$

(15)

$$\mathrm{L}\mathrm{E}\mathrm{F}={\mathrm{W}}_1\times \mathrm{G}\mathrm{S}\mathrm{R}+{\mathrm{W}}_2\times \mathrm{V}\mathrm{Q}+{\mathrm{W}}_3\times \mathrm{R}\mathrm{S}\mathrm{E}\mathrm{I}+{\mathrm{W}}_4\times \mathrm{G}\mathrm{L}\mathrm{S}\mathrm{I}+{\mathrm{W}}_5\times \mathrm{G}\mathrm{C}\mathrm{O}$$

(16)

3.3. Factor Selection and MGWR Model

The UVI is used as the dependent variable Y. The independent variables X are based on various sub-indicators from the land socio-ecological function assessment. Unlike the Ordinary Least Squares (OLS) model, which constructs a global regression relationship, the advantage of Geographically Weighted Regression (GWR) is its ability to account for spatial variability in regression modeling. However, if the phenomenon involves multiple spatial processes with varying spatial scales, a fixed spatial scale becomes ineffective [52]. The MGWR model extends GWR by reconstructing it as a generalized additive model to obtain the standard errors of local parameter estimates. This allows the relationship between the response and explanatory variables to vary spatially and across different scales, addressing overfitting under uniform bandwidth and resolving certain nonlinear relationships [53]:

$${\mathrm{y}}_{\mathrm{i}}={\sum }_{\mathrm{j}=1}^{\mathrm{k}}\hspace{0.17em}{\mathsf{\beta }}_{\mathrm{b}\mathrm{w}\mathrm{j}}\left({\mathrm{u}}_{\mathrm{i}},{\mathrm{v}}_{\mathrm{i}}\right){\mathrm{x}}_{\mathrm{i}\mathrm{j}}+{\mathsf{ϵ}}_{\mathrm{i}}$$

(17)

For each sampling point i, MGWR constructs the coordinates (u_i, v_i). y_i is the value of the dependent variable at point i; bwj represents the bandwidth used for the regression coefficient of variable j. β is the regression coefficient, and ϵ_i is the random disturbance term. This study employs the quadratic kernel function and the AIC criterion. Preliminary indicators from the land assessment are screened using Spearman correlation, OLS, and the Variance Inflation Factor (VIF) test, and the qualifying factors are then input into the MGWR model to calculate driving coefficients and spatial bandwidth.

3.4. Spatial Autocorrelation Analysis

Moran’s I is a commonly used method to determine spatial autocorrelation, including both global and local Moran’s I values. For datasets with spatial attributes, the global Moran’s I value measures the distribution characteristics and degree of spatial autocorrelation, ranging from −1 to 1. Positive values indicate spatial clustering, negative values indicate spatial dispersion, and the absolute value shows the extent of the corresponding trend. A value of 0 indicates random distribution. The significance of the calculation is represented by the Z-value: when the absolute value of the Z-value exceeds 2.58, it corresponds to a 99% confidence level; when it exceeds 3.29, it corresponds to a 99.9% confidence level.

Local Moran’s I is used to test the spatial autocorrelation of local data [54], thereby generating statistically significant clusters of high–high (HH), low–low (LL), high–low (HL), and low–high (LH) spatial outliers.

4. Results and Analysis

4.1. From HMI to UVI and Their Distributions

4.1.1. Identification of Typical Urban Activity Patterns

The distribution of HMI across different seasons showed similar hourly patterns. In Luohe City, activity fully resumed around 07:00, reaching a stable state before gradually declining after 22:00, with the lowest point occurring around 04:00. Notably, there were no significant differences in activity patterns between workdays and weekends. However, as depicted in Figure 2, the highest HMI was observed in spring, with the intensity gradually decreasing in summer, autumn, and winter.

Identifying Classical Human Activity Patterns in Cities by SOM. Through iterative testing, it was revealed that the SOM achieved optimal training performance with 24 neurons (4 × 6) (Figure 3c). This configuration resulted in a QE of 0.230 and a TE of 0.133. As shown in Figure 3b, when the number of clusters was set to five, the DBI reached its first low point. Further reclassification based on HMI identified two main clusters. Cluster 1 covered the active period from 08:00 to 23:00, with an average activity level of 2.650. Cluster 2 represented the rest period from 00:00 to 07:00, with an average activity level of 1.696 (Figure 3a).

4.1.2. Results of Multi-Scenario UVI Calculations

The AAAC indicates that most regions maintained the classic activity pattern. Compared to workdays, weekend periods exhibited a slight increase in areas with AAAC exceeding one. The LAR highlighted areas with a tendency for recreational activities, while HMV identified areas with high fluctuations. Both LAR and HMV served to mitigate the “road prosperity” effect caused by traffic flow on national and provincial highways. The HMC metric emphasized regions with sustained high vitality (Figure 4a1,2). The Moran’s I (0.905/0.897) suggested significant spatial clustering of the UVI. Based on the profile analysis, major human activities were concentrated within the built-up area. Comparing workdays and weekends, although the mean and STD showed minimal changes (Mean = 1.396/1.395; STD = 1.087/1.089), there were notable increases in local vitality distribution, such as in the peninsula region (Figure 4b1,2).

4.2. Land Assessment and Spatial Characteristics

4.2.1. Factors and Land Assessment Results

Based on the land function assessment framework, various factors influencing the land function are illustrated in Figure 5. Additionally, comprehensive weights and spatial heterogeneity for each factor were calculated (

Table 3. Eleven Factors’ Weights and Corresponding Moran’s I.

Goal Layer	Framework	Name	Objective Weight	Subjective Weight	Comprehensive Weight	Moran’s I	Z
LSF	Quantity	ASR	0.258	0.150	0.204	0.767	133.838
	Quality	LERNCI	0.037	0.407	0.222	0.843	144.786
		POP	0.098	0.130	0.114	0.812	136.931
		WI	0.109	0.176	0.142	0.981	167.147
	Structure	ACO	0.308	0.055	0.181	0.736	131.183
		ALSI	0.191	0.082	0.137	0.700	129.338
LEF	Quantity	GSR	0.133	0.444	0.288	0.641	114.039
	Quality	VQ	0.253	0.132	0.192	0.461	77.415
		RSEI	0.130	0.191	0.160	0.415	67.175
	Structure	GCO	0.282	0.131	0.206	0.711	125.488
		GLSI	0.203	0.102	0.153	0.731	135.439

). In the LEF/LSF system, the AHP evaluation consistency was λ = 0.068 ≤ 0.1, indicating that the subjective weight assessment had passed the consistency test and the weights were valid for further analysis.

4.2.2. LSF and LEF Assessment Results

The LSF exhibited a strong aggregation trend (Moran’s I = 0.869). However, compared to UVI, the spatial distribution of LSF was more balanced within the built-up area (Figure 6a). The LEF showed a weaker aggregation trend (Moran’s I = 0.689). Notably, within the built-up area, core urban areas exhibited lower ecological supply, with regions of higher ecological value forming a ring around the urban core (Figure 6b).

The Local Indicators of Spatial Association (LISA) map (Figure 7) showed that HH and LL clusters were predominant, indicating a spatial polarization phenomenon. The spillover effect of land socio-ecological functions was evident, with high/low-function areas significantly influencing land development in surrounding regions. The radar chart further highlighted spatial distribution patterns. LSF showed high functionality in the S/SE/E directions, indicating a vertical distribution trend of “southeast high, northwest low.” In contrast, LEF showed a nearly balanced distribution of ecological supply across all directions, indicating a more horizontal distribution.

4.2.3. Spatial Correlation Analysis

Both LSF/LEF and workdays/weekends UVI showed a significant correlation, with LSF demonstrating a stronger correlation with UVI (Figure 8a). Based on the IQR test (threshold = 1.5), UVI was classified into six levels, from H1 (extremely low) to H6 (extremely high). The changes in UVI levels corresponded to changes in land ecological/social functions (Figure 8b1,2). UVI’s dependence on LSF was evident, with H2 to H5 levels of UVI highly concentrated in areas with high LSF (≥0.50) and moderate to high LEF (≥0.30). The H6 level UVI responded only to high LSF.

The distribution trends of UVI on workdays and weekends were generally consistent. Using weekend UVI as an example, we examined its spatial response to land functions (Figure 8c). LSF and LEF were divided into low, medium, and high categories based on percentiles. This combination created nine service scenarios, such as high LSF and high LEF (HH), high LSF and medium LEF (HM), and so on. Spatially, extremely high UVI (H6) mainly appeared in the HH scenario (H6 Number = 661), followed by HM (H6 Number = 621), and lastly, HL (H6 Number = 499). This indicated that UV relied on comprehensive LSF, with a further demand for LEF. This trend was more evident when LSF was at a median level. UVI values in the H5 to H4 range primarily occurred in MH scenarios, followed by MM and ML scenarios. When LSF supply was insufficient, LEF also lost attractiveness, with UVI values in the H3 to H2 range mainly appearing in LH, LM, and LL scenarios.

4.3. Weighted Regression Results

4.3.1. MGWR Calculation Results and Bandwidth

The correlation analysis between various factors and UVI is summarized in

Table 4. Spearman Correlation Analysis.

	ASR	GSR	WI	LERNCI	POP	RSEI	VQ	GCO	ACO	GLSI	ALSI
Workdays	0.481	0.616	0.631	0.781	0.486	−0.380	−0.445	0.512	0.610	0.295	0.433
Weekends	0.466	0.604	0.621	0.777	0.477	−0.376	−0.428	0.498	0.599	0.281	0.424

, with all results statistically significant at the 99.9% level.

Following data cleaning procedures, factors exhibiting weak driving relationships (p ≥ 0.05) and high collinearity (VIF ≥ 10) were excluded using OLS/VIF. Notably, this procedure resulted in the exclusion of the cohesion variable (encompassing both GCO and ACO). The dataset was then sampled using a 600 m scale fishnet, resulting in a total of 1710 sampling points. A comparison of goodness-of-fit statistics across MGWR, Spatial Lag Model (SLM), Spatial Error Model (SEM), and OLS showed MGWR as the model with the best fit and the lowest AIC value (

Table 5. Comparison of Model Fit Accuracy.

Accuracy Indicator	MGWR		SLM		SEM		OLS
	R2	AIC	R2	AIC	R2	AIC	R2	AIC
Workdays	0.844	1980.841	0.833	1996.862	0.840	1989.556	0.678	2928.674
Weekends	0.842	2011.303	0.829	2044.219	0.835	2034.661	0.668	2977.909

).

In the MGWR model, bandwidth indicates the spatial range within which a factor exerts a uniform influence, directly representing the scale of its influence. A larger bandwidth means that the factor’s influence remains relatively consistent over a larger area. This study incorporated land function variables as explanatory factors, with the intercept term capturing the influence of other uncontrolled variables such as location, resident culture, and city history. Based on the Jenks natural breaks method for bandwidth classification, we divided these locational factors into three distinct categories: (a) Global scale (bandwidth = [959, 1710]), covering 56.082–100% of the study area; (b) Semi-global factors (bandwidth = [290, 479]), covering 16.959–28.012%; (c) Local factors (bandwidth = 57), covering 3.333%.

From workdays to weekends, four factors maintained consistent bandwidths: LERNCI, POP, WI, and GLSI. These persistent global factors served as constant drivers of UVI, while persistent local factors were significant drivers of UVI heterogeneity and remained unaffected by various scenarios. The bandwidth of the remaining five factors significantly changed depending on human activity scenarios (

Table 6. Bandwidth Changes in Driving Factors on Workdays/Weekends.

NO.	Name	Workdays	Weekends	Change
X1	ASR	479	57	Semi-global → Local
X2	LERNCI	57	57	Local → Local
X3	POP	1680	959	Global → Global
X4	WI	1710	1710	Global → Global
X5	ALSI	1710	57	Global → Local
X6	GSR	57	1710	Local → Global
X7	VQ	330	1710	Semi-global → Global
X8	RSEI	290	1710	Semi-global → Global
X9	GLSI	1710	1710	Global → Global
/	Intercept	57	1710	Local → Global

).

ASR and LERNCI were the main driving forces, with LSF also playing a significant role (LSF Mean(abs) = 0.148 > LEF Mean(abs) = 0.045). When disregarding factors with a change rate lower than 0.005, it was observed that in autonomous activity scenarios (Weekends), factors comprising LEF (GSR, VQ, RSEI) and factors comprising LSF (ASR) exhibited an increase in attractiveness, whereas the attractiveness of LERNCI and POP within LSF diminished (

Table 7. Driving Forces Mean Changes and Directions of Driving Forces on Workdays/Weekends.

System	Variable	Workdays	Weekends	Change	Direction
LSF	ASR	0.207	0.217	0.010	+
	LERNCI	0.424	0.356	−0.068	−
	POP	−0.071	−0.080	−0.009	−
	WI	−0.034	−0.032	0.002	/
	ALSI	0.004	0.072	0.068	+
Mean (abs)		0.148	0.151	0.003	+
LEF	GSR	0.002	0.024	0.022	+
	VQ	−0.078	−0.024	0.054	+
	RSEI	0.091	0.090	−0.001	/
	GLSI	−0.054	−0.057	−0.003	/
Mean (abs)		0.045	0.039	−0.059	−
Intercept		−0.136	−0.195	−0.059	−

).

4.3.2. Coefficient Analysis: Scale-Invariant Factors

(1) Persistent Global Factors

From workdays to weekends, all three persistent global factors exhibited negative driving forces (

Table 8. Changes in Driving Forces of Persistent Global Factors on Workdays/Weekends.

	Workdays					Weekends
Variable	Mean	STD	Min	Median	Max	Mean	STD	Min	Median	Max
POP	−0.071	0.002	−0.074	−0.071	−0.068	−0.08	0.017	−0.099	−0.086	−0.032
IW	−0.034	0.001	−0.037	−0.034	−0.031	−0.032	0.001	−0.035	−0.032	−0.03
GLSI	−0.054	0.003	−0.058	−0.056	−0.048	−0.057	0.004	−0.061	−0.058	−0.05

). UVI decreased with increases in population, accessibility, and the complexity of green space patch morphology. The influence of WI was mild, while POP exerted the most significant impact.

(2) Persistent Local Factors

The only persistent local factor was LERNCI, acting as a positive driver (

Table 9. Changes in Driving Forces of Persistent Local Factors on Workdays/Weekends.

	Workdays					Weekends
Variable	Mean	STD	Min	Median	Max	Mean	STD	Min	Median	Max
LERNCI	0.424	0.214	−0.735	0.405	1.187	0.356	0.246	−1.861	0.379	1.073

). This indicates that UVI generally increased alongside the degree of urbanization. Notably, the driving force associated with LERNCI exhibited the highest mean value among all factors. This finding underscores the attraction of population aggregation and activities facilitated by well-developed infrastructure.

Figure 9 illustrates the spatial distribution of the LERNCI coefficient. The coefficient of LERNCI showed both positive and negative driving effects. Within the urban core (Area 3), increasing LERNCI values were associated with rising UVI. In the peninsula area (Area 2), LERNCI showed minimal impact on UVI. However, on the urban fringe, especially in logistics and warehousing areas (Area 4) and northern rural clusters (Area 1), increased light appeared to be linked with decreased human activity.

4.3.3. Coefficient Analysis: Scale-Variable Factors

(1) Factors with Increased Bandwidth

An increased bandwidth implies an expanded spatial range of homogeneous influence. Notably, VQ and RSEI, previously categorized as semi-global factors of LEF, transitioned to global factors. Meanwhile, the GSR and the intercept exhibited a transition from local to global influence (

Table 10. Changes in Driving Forces of Increased Bandwidth Factors from Workdays to Weekends.

	Workdays					Weekends
Variable	Mean	STD	Min	Median	Max	Mean	STD	Min	Median	Max
VQ	−0.078	0.063	−0.25	−0.056	−0.011	−0.024	0.001	−0.026	−0.025	−0.023
RSEI	0.091	0.04	−0.012	0.093	0.165	0.09	0.002	0.086	0.09	0.094
GSR	0.002	0.109	−0.352	−0.009	0.79	0.024	0.001	0.023	0.024	0.027
Intercept	−0.136	0.185	−0.624	−0.16	0.712	−0.195	0.002	−0.199	−0.194	−0.192

).

Overall, the driving force of RSEI remained stable across both workday and weekend scenarios. VQ’s negative impact on UVI decreased significantly during weekends. The green space quantity factor, GSR, consistently acted as a positive driver of UVI, with a significant increase in its influence during weekends.

The most notable change observed in the coefficients of RSEI, VQ, and GSR was the disappearance of cold spots within the built-up area during weekends (Figure 10). The localized negative driving effects of RSEI and GSR observed during workdays disappeared on weekends, resulting in an overall positive influence on UVI. Although VQ remained a negative driver, its impact weakened. This indicates a shift in green space function from landscape to recreational use. Notably, the increased demand for green spaces during weekends resulted in higher UVI in areas characterized by high LEF.

During workdays, the intercept served as a local factor, representing locational attributes such as the constraints imposed by commercial or work areas on human activity and UV. During weekends, the absence of these constraints under volitional human activity scenarios led to significant bandwidth changes in the intercept.

(2) Factors with Decreased Bandwidth

Several factors exhibited a decrease in bandwidth. ASR transitioned from a semi-global to a local factor, while ALSI shifted from global to local influence. ASR exhibited a stronger driving force compared to ALSI, but ALSI demonstrated greater variations in influence across different scenarios (

Table 11. Changes in Driving Forces of Decreased Bandwidth Factors from Workdays to Weekends.

	Workdays					Weekends
Variable	Mean	STD	Min	Median	Max	Mean	STD	Min	Median	Max
ASR	0.207	0.042	0.155	0.192	0.287	0.217	0.192	−0.083	0.154	0.892
ALSI	0.004	0.001	0.001	0.004	0.007	0.072	0.221	−0.616	0.024	1.462

).

Both on workdays and weekends, ASR showed a strong positive driving force in the core areas, with increased heterogeneity and a significant increase in the maximum driving force (from 0.287 to 0.892), highlighting a preference for urban built-up areas. During weekends, UVI was more attracted to the ASR of the built-up area, which suppressed surrounding townships. This also explained the change in ALSI coefficients, transitioning from global positive driving to significant local positive driving (maximum from 0.007 to 1.462), with the emergence of local negative driving effects (Figure 11).

5. Discussion

5.1. UV Growth Pole and Regulation

The spatial distribution of UVI was highly uneven, with high values concentrated in built-up areas closely related to LSF. Combined with Moran’s I, high UV regions have a significant positive effect on their neighbors. The HMV indicator shows that Luohe lacks a significant agglomeration of consistently high human activity within the built-up area, i.e., it lacks a core pole of vitality growth. The HMC indicator indicates that while roads and parks attract temporary traffic, they struggle to sustain long-term vitality.

To tackle this issue, it is suggested that growth poles be created within built-up areas, such as commercial complexes and multi-purpose recreational facilities, especially in regions with high LSF potential. According to Jacobs, a good mix of functions in urban areas ensures the flow and density of people and activities [7]. This localized construction can prioritize the local drivers (e.g., ASR and LERNCI) in the MGWR for site selection. Building growth poles in regions where they are positively driven (e.g., the northeastern areas along the river, especially Area 3 shown in Figure 9) will help stabilize and cluster high human activity and promote regional development [3].

In the Old Town area, high UV is aggregated but remains in the “HL” (high LSF-low LEF) category due to insufficient LEF. There is a growing trend among scholars and urban managers to promote compact city development and urban densification/reuse strategies, aiming to guide vitality growth through construction [25,55]. However, the compact city paradox, where high-density areas struggle with inadequate social and ecological services per capita, needs to be addressed [56,57]. Therefore, it is advised that vegetation quality be improved and green spaces should be created at the street scale to enhance the living environment. The negative spatial response of UV to LEF subfactors (GLSI/VQ) should be used to regulate UV and balance human–land functions [58].

5.2. Dependency on Land Functions under Different Scenarios

MGWR analysis highlighted LSF’s significant impact on UV, with ASR and LERNCI as key drivers, indicating LEF’s secondary role [59]. Human activity showed some dependence on the presence of green spaces and vegetation, which can enhance UV, but this enhancement must be based on a robust foundation of social functions [25].

In Luohe City, LSF is mainly concentrated in the urban core area, and the high-value LEF area forms a ring around the built-up area, reflecting the characteristics of “high edge, low center” distribution [53]. The different clustering characteristics of LSF and LEF indicate that the spatial provision of the two is not synchronized. Therefore, from a land system hierarchy perspective, non-built-up areas (LL, LM, and LH Areas in Figure 8) with low LSF should prioritize developing social functions like road networks, community service facilities, and so on. Establishing secondary growth poles will enhance these areas’ UV and promote balanced urban–rural development [60]. For built-up areas (HL Area in Figure 8) that already have a complete LSF, improving ecological functions by densifying vegetation, planting trees, and reorganizing green spaces is a long-term goal.

5.3. Driving Forces and Bandwith

This study prioritized the analysis of UVI under autonomous human activity conditions. Human activities on weekends are more responsive to changes in land function than on weekdays, which are constrained by workplace demands, suggesting that more attention should be paid to optimizing land functions in planning. Global factors should inform policy as macro-parameters, while semi-global and local factors, showing spatial differences, should guide specific construction needs [32].

On weekends, the hierarchy of global factors and their driving forces was as follows: RSEI (+) > POP (−) > GLSI (−) > WI (−) > GSR (+) = VQ (−). The positive influence of RSEI on UV was notably significant at a global level. Increasing green space areas and constructing micro community parks with low vegetation density are effective strategies [26]. Although GLSI and VQ were negative driving factors, it is crucial to consider the dependence of LEF on LSF for effectiveness. Further analysis is required to distinguish whether green spaces with a high LSI lack social function support or are located on the urban fringe with inadequate supporting facilities, thus failing to effectively enhance overall UV [61]. The complexity and size of green patches attract people, but low actual usage frequency leads to their negative driving effects, which are issues that urban managers need to address [25,50].

The average bandwidth of semi-global factors (366) corresponded to an area of 42.651 km². As the most basic administrative unit in China, the average area of the current townships in the study area was 31.740 km², which was very close to the scale of semi-global factors. Notably, semi-global scale factors only appeared in workday scenarios. In contrast, the bandwidth of local factors (57) corresponded to an area of about 6.636 km², which was closer to the street scale. Among local factors, driving forces were ranked as follows: LERNCI (+) > ASR (+) > ALSI (+). This demonstrates that the quality of construction holds greater influence than quantity, followed by patch structure complexity. It also illustrates how the location of the urban core affects UVI, with land use, function, and morphology within the built-up area playing a significant role in this process.

The overall planning recommendation is as follows: Strategic Growth Pole Development: Develop mixed-use commercial centers and sites within built-up areas to stimulate neighborhood development by leveraging areas with high UV. Coordination of LSF/LEF: Prioritize social function development in non-built-up areas with insufficient LSF (LL/LM/LH areas), focusing on building social infrastructure like roads, community services, and night markets to boost UV. In built-up areas with strong LEF but poor ecological functions (HL areas), focus on enhancing the ecological functions to balance human activity and improve residential quality. Improving the Planning Quality: Focus on improving the quality of activities and edge interactions in areas dominated by artificial surfaces while expanding and optimizing green spaces with community-based, accessible micro-parks.

5.4. Study Limitations and Future Directions

Despite significant findings, this study has limitations. The complexity of geographic and urban systems means UV driving factors often interact rather than act independently [62]. Our approach may not fully capture these interactions. Incorporating the Geographical Detector model could better explore these synergies. Secondly, UV’s dependency on LEF/LSF might vary across different spatial scales, which the current study may not have fully accounted for. This could affect the accuracy of policy recommendations. Addressing these relationships through multi-scale analysis would enhance understanding.

Additionally, our model currently depends on limited data sources. Enhancing accuracy and detail would require integrating additional data, such as Points of Interest (POI) and age demographics, to gain a better understanding of activity patterns across different groups. This is because people with different age, gender, and occupational profiles have distinct behavioral patterns. Thus, continuous improvements in modeling and data integration are needed for future research.

6. Conclusions

This study constructs a comprehensive assessment framework for UV and land functions, delving into their complex relationship and emphasizing the significant role of land development in shaping urban vitality heterogeneity. The main findings are as follows:

UV Spatial Variation: The spatial distribution of UV in Luohe City shows localized changes under different scenarios (weekdays and weekends). The old town consistently has high UV levels, while the peninsula’s UV increases on weekends. High UV areas are mainly in built-up regions, with non-built-up areas showing much lower UV values. UV Spatial Autocorrelation: UV distribution shows significant spatial autocorrelation across all scenarios with clear clustering. High (low) UV areas exert a positive (negative) influence on surrounding areas. In particular, the HMV indicator reveals that Luohe lacks persistent UV growth centers to stimulate development in adjacent areas.

Functional Coupling: LSF and LEF distributions differ notably between built-up and non-built-up areas, with both primarily found in built-up regions. Their distribution within urban areas is unsynchronized, suggesting a lack of full spatial coupling between social and ecological functions. Functional Dependence: Spatial analysis shows UV mainly relies on LSF supply. LEF boosts UV significantly only when LSF is sufficient. This underscores the importance of integrating LSF in urban planning as a foundational element, with LEF serving as a supplementary factor to further enhance UV under specific conditions.

Key Drivers: MGWR analysis identifies ASR and LERNCI as key drivers of UV under different scenarios, but their influence varies across scenarios. Temporal Dynamics: On weekends, the attractiveness of LEF, such as GSR and VQ, increases significantly, reflecting a heightened demand for ecological amenities during residents’ leisure activities. This emphasizes not only the temporal dynamics of residents’ interactions with urban areas but also the need for flexible, multifunctional mixed land use.

The insights presented here can inform urban development strategies in cities in China and globally, promoting sustainable urban environments that satisfy the social and ecological needs of their populations. By aligning land development policies with the intrinsic drivers of UV identified in this study, cities can enhance their livability, resilience, and overall vitality.