Evaluation of Ecological Environment Quality Using an Improved Remote Sensing Ecological Index Model

Abstract

The Remote Sensing Ecological Index (RSEI) model is widely used for large-scale, rapid Ecological Environment Quality (EEQ) assessment. However, both the RSEI and its improved models have limitations in explaining the EEQ with only two-dimensional (2D) factors, resulting in inaccurate evaluation results. Incorporating more comprehensive, three-dimensional (3D) ecological information poses challenges for maintaining stability in large-scale monitoring, using traditional weighting methods like the Principal Component Analysis (PCA). This study introduces an Improved Remote Sensing Ecological Index (IRSEI) model that integrates 2D (normalized difference vegetation factor, normalized difference built-up and soil factor, heat factor, wetness, difference factor for air quality) and 3D (comprehensive vegetation factor) ecological factors for enhanced EEQ monitoring. The model employs a combined subjective–objective weighting approach, utilizing principal components and hierarchical analysis under minimum entropy theory. A comparative analysis of IRSEI and RSEI in Miyun, a representative study area, reveals a strong correlation and consistent monitoring trends. By incorporating air quality and 3D ecological factors, IRSEI provides a more accurate and detailed EEQ assessment, better aligning with ground truth observations from Google Earth satellite imagery.

1. Introduction

The continuous development of society and increasing urbanization have greatly impacted the ecological space in urban areas [1], and many regional and global ecological problems are becoming more prominent, such as vegetation degradation [2], urban heat islands [3], and air pollution [4]. Therefore, the efficient, comprehensive, and accurate monitoring and assessment of regional ecological environments is crucial. EEQ monitoring using remote sensing data (imagery) has gradually become a common and efficient research method [5]. Studies [6,7,8,9] have primarily used a single 2D factor to evaluate the ecological environment status. However, the ecological environment is composed of multiple elements, and the assessment of a single ecological factor can only explain the ecological characteristics in one direction and cannot comprehensively reflect the ecological environment’s status [10]. In recent years, researchers have explored the EEQ assessment by coupling several individual ecological indices, which can be divided into three main categories: the ecological index (EI) model; the remote sensing ecological index (RSEI) model; and other ecological environment models [11].

The EI index model, which aggregates the biological abundance, vegetation coverage, water network density, land degradation, and environmental quality, is the most widely used ecological environment evaluation index by the Ministry of Ecology and Environmental Protection of China [12]. It has been widely applied for regional EEQ assessment in China [13,14,15]. Based on the EI model, Ouyang et al. [16] subdivided the six ecological factors into 20 sub-indicators, and the weights were calculated using the hierarchical analysis method, making the evaluation results more precise. Li et al. [17] selected the ecological factors needed in the inverse EI model based on the GEE platform, such as MODIS data, Landsat data, and night-time lighting data, and introduced a humidity factor to generate a new ecological index (NEI) model. However, the EI model relies on annual statistical or land use and land cover data, which are difficult to obtain, and the results must be divided along the administrative boundaries, which goes against natural ecological laws. Moreover, the weights and evaluation units are difficult to obtain and not universal.

The widely used RSEI model was first proposed by Xu [18] and it combined four factors including dryness, greenness, heat, and moisture with the principal component analysis (PCA) method. The RSEI model can reflect the pressure of human activities on the environment, the response to climate change, and changes in environmental conditions [19,20,21,22]. In recent years, research on improving RSEI has mainly been reflected in the selection of variable factors and the stable solution of the model. For the former, the ecological factors deriving from remote sensing imagery are usually added or substituted. For example, the mine-specific eco-environment index model was constructed by adding the atmospheric environment and mining scale [23]. The modified remotely sensed ecological index was constructed by introducing the landscape diversity index to evaluate oasis ecology [24], and the comprehensive salinity index and the remotely sensed water network density estimation model were constructed to evaluate the ecological quality of the northwest arid zone [25]. For the latter, the entropy weighting method was used instead of PCA to construct the RSEI model that considers all ecological factors [26]. Studies have proven that the first principal component of PCA selected for the RSEI model cannot fully characterize ecological features; therefore, the first three principal components of PCA are used [27]. Moreover, the nonlinear kernel principal component method has also been used to construct nonlinear remote sensing ecological indices [28]. The RSEI model is conducive to quantitatively evaluating EEQ changes at the regional scale, and its reliability and credibility have been verified. Furthermore, its EEQ results can be visualized, scaled, and compared at different spatiotemporal scales. However, the construction of an RSEI model is complex and subjective, and the EEQ is prone to be the opposite when using PCA. In particular, it is only suitable for a small number of calculations and is not suitable for long-term sequence and batch operations [29]. Most importantly, in large-scale EEQ assessment, the RSEI and its improved models only utilize two-dimensional ecological factors and have not considered vertical three-dimensional information [30].

Other ecological models include the pressure-state-response (PSR) model [31,32], the InVEST model [33], and so on. These ecological evaluation models have made important contributions to EEQ evaluation [34,35,36,37]. Ashraf et al. [38] researched the development of the Spatial Ecosystem Health Index (SEHI), including indicator weights and selection using the remotely sensed Pressure-State-Response (PSR) framework, the Hierarchical Analysis Method (AHP), and the Principal Component Analysis (PCA) in 1990, 2003, 2013, and 2021, respectively. Based on the land use and land cover (LULC) data in 2000, 2010, and 2020, Li et al. [39] first predicted the trends and results in 2030. Then, the habitat quality was evaluated from 2000 to 2030 using the InVEST model, which showed that habitat quality was highly correlated with land use change. However, most of these models rely computationally on hard-to-access data such as LULC, and these models often focus on specific ecological services or stressors, which may lead to the neglect of other ecosystem functions and services. Moreover, ecosystems are multidimensional representations of the real world, and the 2D expression of these models may not fully account for these complex interactions.

The objectives of this study were as follows: (1) to integrate 2D and 3D ecological factors for EEQ evaluation, enabling a transition from 2D to 3D ecological monitoring across large regions, (2) to incorporate air quality as a factor in the proposed IRSEI model for more comprehensive ecological environment monitoring, and (3) to employ a combined subjective–objective weighting model to assess ecological quality, enhancing the stability and rationality of EEQ results.

2. Materials

2.1. Study Area

This study selects a well famous ecological support district, named Miyun (Figure 1), located in northeast Beijing (40°13′N–40°48′N, 116°40′E–117°30′E). It is surrounded by mountains and rolling peaks in the east, north, and west. The ancient Great Wall stretches over the mountains. In the middle is the rippling Miyun Reservoir, and in the southwest is the flooded alluvial plain. It falls under the warm temperate monsoon continental semi-moist and semi-arid climate, governed by the high-pressure systems of Siberia and Mongolia in winter and influenced by continental low pressure and Pacific Ocean high pressure in summer. It experiences four distinct seasons with noticeable variations in dryness, wetness, coldness, and warmth. The average temperature in January is −3.5 °C, while in July it is 26.8 °C, with an annual precipitation of around 577 mm. As the most important water source in Beijing, it has the largest forest, the best wetland resources, and the richest biodiversity. In addition, the functional zoning distribution from A1 to A5 can carry different functions [40].

2.2. Data and Pre-Processing

The remote sensing data used in this study are 2D multispectral remote sensing images and 3D spaceborne laser scanning point clouds.

(1)

Multispectral remote-sensing images (2D)

These used remote sensing imagery are Landsat8 OLI imagery, obtained from the United States Geological Survey (USGS) [41]. Owing to the vegetation growth condition, we selected the Landsat8 OLI image on 18 September 2019, with a spatial resolution of 30 m × 30 m and cloud content of 0.01%. The pre-processing is accomplished by Google Earth Engine (GEE) including radiometric calibration, atmospheric correction, stitching, and cropping, and these images can be further used to calculate 2D ecological factors.

(2)

Laser-scanning point cloud (3D)

The three-dimensional (3D) indicators, generated from the global ecosystem dynamics investigation (GEDI) and the Ice, Cloud, and Land Elevation Satellite-2 (ICESat-2), can provide more detailed and accurate results, as well as new ecological insights; therefore, they can be effectively used to improve ecological models for large-scale ecosystem investigations. The adopted 3D indices used in the proposed method are calculated from the 3D products (Guo et al. 2021). These 3D data were combined with inverse vegetation canopy height data using neural network-guided interpolation [42]. The resulting products have no saturation effect in areas with a high forest canopy and can be used to reflect vegetation vertical structure for regional ecological assessment.

3. Methods

The proposed methodological framework (Figure 2) encompasses three key components: ecological factor calculations (Section 3.1), model construction (Section 3.2), and model validation (Section 3.3). The 2D and 3D ecological factors were initially calculated from the Landsat8 OLI image, ICESat-2, and GEDI data. Subsequently, the IRSEI model for EEQ assessment was constructed through a combined subjective–objective approach (PCA-AHP). Finally, the performance of the newly proposed IRSEI model was evaluated.

3.1. Ecological Factors Calculation

The IRSEI model incorporates vegetation, humidity, aridity, heat, and air quality as ecological factors, while the original RSEI model considers vegetation, humidity, aridity, and heat factors. When these ecological factors are calculated, a normalization process should be performed.

(1)

Vegetation factor

The vegetation factor serves as a crucial indicator in assessing the quality status of the regional ecological environment. Here, a new Comprehensive Vegetation Index (CVI) was introduced, which incorporated 2D planar features (NDVI) and 3D vertical distribution information (CVHI). As the significant differences between the NDVI and CVHI, the non-dimensional treatment is adopted, and the final formula is expressed as follows:

$$CVI=w_1\cdot NDVI+w_2\cdot CVHI$$

(1)

where NDVI is the normalized difference vegetation index. The CVHI [42] is the vegetation canopy height index, which is the 3D canopy height data acquired from the GEDI and ICESat-2. The linear model, where the coefficients w₁ and w₂ are set to 0.5 based on prior experience, was used. Specifically, the three ecological factors were normalized to scale their values between 0 and 1.

(2)

Wetness factor

The land surface moisture reflects the moisture content of water bodies, soil, and vegetation, and is closely related to ecology [22]. According to previous studies [43,44], the wetness can be obtained by inverting the bands and can be calculated by

$$\begin{array}{ll}WET(OLI)=& 0.1511\times {\rho }_{blue}+0.1973\times {\rho }_{Green}+0.3283\times {\rho }_{\mathrm{Re}d}+\\ & 0.3407\times {\rho }_{NIR}-0.7117\times {\rho }_{SWIR1}-0.4559\times {\rho }_{SWIR2}\end{array}$$

(2)

where WET (OLI) is applied to Landsat OLI images. The variables ${\rho }_{blue}$, ${\rho }_{Green}$, ${\rho }_{\mathrm{Re}d}$, ${\rho }_{NIR}$, ${\rho }_{SWIR1}$, and ${\rho }_{SWIR2}$ are the reflectances of bands 2, 3, 4, 5, 6, and 7 of the OLI images, respectively.

(3)

Dryness factor

The dryness factor responds to the dryness of the ground surface, which influences some ecological phenomena. The dryness factor is expressed by the bare soil index (SI), but in the regional environment, part of the built land also causes the surface to “dry out”. Therefore, SI and the building index (IBI) were chosen to express the dryness factor (NDBSI) with the following formulas [22]:

$$NDBSI={\displaystyle \frac{SI+IBI}{2}}$$

(3)

$$SI={\displaystyle \frac{\left({\rho }_{SWIR1}+{\rho }_{\mathrm{Re}d}\right)-\left({\rho }_{NIR}+{\rho }_{blue}\right)}{\left({\rho }_{SWIR1}+{\rho }_{\mathrm{Re}d}\right)+\left({\rho }_{NIR}+{\rho }_{blue}\right)}}$$

(4)

$$IBI={\displaystyle \frac{2{\rho }_{SWIR1}/\left({\rho }_{SWIR1}+{\rho }_{NIR}\right)-\left[{\rho }_{NIR}/\left({\rho }_{NIR}+{\rho }_{\mathrm{Re}d}\right)+{\rho }_{Green}/\left({\rho }_{Green}+{\rho }_{SWIR1}\right)\right]}{2{\rho }_{SWIR1}/\left({\rho }_{SWIR1}+{\rho }_{NIR}\right)+\left[{\rho }_{NIR}/\left({\rho }_{NIR}+{\rho }_{\mathrm{Re}d}\right)+{\rho }_{Green}/\left({\rho }_{Green}+{\rho }_{SWIR1}\right)\right]}}$$

(5)

where ${\rho }_{blue}$, ${\rho }_{Green}$, ${\rho }_{\mathrm{Re}d}$, ${\rho }_{NIR}$, and ${\rho }_{SWIR1}$ refer to the reflectance of the 1st, 2nd, 3rd, 4th, 5th, and each band of the TM image, and the reflectance of the 2nd, 3rd, 4th, 5th, and 6th bands of the OLI image, respectively.

(4)

Heat factor

Land Surface Temperature (LST) is the most direct indicator of heat and is frequently employed in ecological quality assessments [25]. The LST, serving as the heat factor in this study, is calculated as follows:

$$L=gain\times DN+bias$$

(6)

$$T=K_2/\ln \left(K_1/L+1\right)$$

(7)

$$LST=T/\left[1+\left(\lambda \cdot T/\rho \right)\ln \epsilon \right]$$

(8)

where L is the temperature value of the thermal infrared band; gain is the gain value; DN is the gray value of the pixel; bias is the bias value; T is the temperature value at the sensor; K₁ = 774.89 W/(m²·sr·µm); K₂ = 1321.08 k; $\lambda$ is the central wavelength of the thermal infrared band; ρ is 1.438 × 10⁻² K; and ε is the specific emissivity of ground objects estimated by the Sobrino model.

(5)

Difference factor for air quality

The particulate matter, a large proportion of dust, is the chief pollutant of the atmosphere pollution in Beijing. According to the PM2.5 of particles, the reflectivity of the red band increases, and the reflectivity of the near-infrared band decreases [23]. Thus, the difference index (DI) was constructed to characterize the air quality:

$$DI={\rho }_{red}-{\rho }_{nir}$$

(9)

where ρ_red and ρ_nir are the surface reflectance in the near-infrared and red bands, respectively.

3.2. Model Construction

Two ecological models, that are the proposed IRSEI and common RSEI model, were constructed for EEQ evaluation with the following extracted ecological factors: NDVI, CVI, WET, NDBSI, LST, and DI.

(1)

IRSEI model

The proposed IRSEI model, which combines 2D and 3D ecological factors is organized as follows:

$$IRSEI=f\left(CVI,WET,NDBSI,LST,DI\right)={\displaystyle \sum _{i=1}^n\left(W_i\times K_i\right)}$$

(10)

where n denotes the number of extracted ecological factors; W_i and K_i are the weight and index values of the i-th ecological factor, respectively.

To achieve the final IRSEI model, the PCA approach is frequently used to estimate the weights, which can downscale the regional ecological environment with multiple factors. However, the impact of different evaluation factors on actual environmental problems is ignored. In the study of Xu et al. [45], it was shown that the first component of PCA has obvious ecological benefits, while the other components do not have clear ecological meanings. Therefore, when there is a situation where the contribution rate of the first component is low, the single use of PCA analysis will lose a large amount of data [46]. In addition, AHP is a method used for subjectively determining weights, enabling the quantification of expert a priori knowledge, and facilitating a comprehensive evaluation of multiple indicators. However, the results of AHP tend to be overly subjective and may overlook the information inherent in the data itself [47].

Therefore, a hybrid method/model (PCA-AHP) is proposed for the new IRSEI, which can fix the weights (W_i) by combining objective analysis (PCA) and subjective analysis (hierarchical analysis; AHP).

Specifically, for the weight of each ecological factor (W_i), we first assumed that W_1i and W_2i are subjective and objective weights, which can be calculated using the PCA and AHP methods, respectively. Using the minimum information entropy and Lagrange’s mean theorem, the final weight (W_i) can be expressed as:

$$\min F={\displaystyle \sum _{i=1}^n{W}_i\left(\ln W_i-\ln W_{1i}\right)}+{\displaystyle \sum _{i=1}^n{W}_i\left(\ln W_i-\ln W_{2i}\right)}$$

(11)

$$s\cdot t{\displaystyle \sum _{i=1}^n{W}_i=1(W_i>0,i=1,2,\dots n)}$$

(12)

Also, using the Lagrangian median theorem, it is known that:

$$W_i={\displaystyle \frac{\sqrt{W_{1i}{W}_{2i}}}{{\displaystyle {\sum }_{i=1}^n\sqrt{W_{1i}{W}_{2i}}}}}(i=1,2,\dots n)$$

(13)

where Wi is the weight of the PCA–AHP hybrid model and i is the i-th factor.

(2)

RSEI model

The common RSEI model is constructed with 2D ecological factors, which are given by

$$RSEI=f\left(NDVI,WET,NDBSI,LST\right)$$

(14)

To obtain the RSEI model, the PCA method, which can effectively remove redundant information, is utilized, and the first component of PCA (PC₁) is selected for RSEI [18,45,48]. Finally, the model can be expressed as:

$$RSEI=1-P{C}_1\left(NDVI,WET,NDBSI,LST\right)$$

(15)

3.3. Model Validation

To evaluate the performance of the two models, field measurements and the commonly used average degree of correlation were used. Directly assessing model accuracy using extensive field measurements was challenging [49]. Thus, high-resolution remote sensing images from Google Earth (GE) were commonly used to validate EEQ accuracy [50,51,52]. To this end, we could randomly select checkpoints over the experimental areas, and the EEQ values of each selected sample were then compared to corresponding ecological performance as visually assessed from the GE images.

Table 1. The classification and description of EEQ.

Grade	Values	Ecological Performance
Excellent	[1, 0.8)	High-vegetation cover, good natural conditions, and stable ecosystems.
Good	[0.8, 0.6)	Good natural conditions and good vegetation cover for human life.
Moderate	[0.6, 0.4)	Vegetation cover is medium and more suitable for human life.
Fair	[0.4, 0.2)	Vegetation cover is poor and arid and there are limiting factors for human life.
Poor	[0.2, 0]	Low vegetation cover, harsh conditions, and restrictions on human life

, based on References [22,25,26], outlines the EEQ and referenced performance criteria.

The average degree of correlation, that is the internal plausibility of the selected ecological factors, was calculated for model validation, which indicates the overall representativeness of the IRSEI and the correlation among the various indicators [23]. The average degree of correlation can be expressed as follows:

$$\overline{R}={\displaystyle \sum _{i=1}^n\left(\frac{\left|R_i\right|}{n}\right)}$$

(16)

where n is the number of ecological factors, and R_i is the correlation coefficient between any two ecological indicators.

4. Results

4.1. Results of Calculated Factors

The calculated factors (NDVI, WET, CVI, NDBSI, DI, and LST) are visualized in Figure 3. Their respective maximum values are 0.863, 0.726, 0.616, 0.449, 0.563, and 0.402.

As seen in Figure 3, ecological factors (NDVI, CVI, and WET) that have a positive impact on EEQ have smaller values in areas with greater human disturbance (e.g., urban areas), and larger values in places with better ecology (e.g., forests), while the opposite is true for the negative ecological factors. The recently introduced 3D ecological factor, CVI, and the conventional 2D ecological factor, NDVI, share a similar overall trend but exhibit distinctions in detail. CVI integrates the 3D parameters of vegetation, facilitating a more distinct separation of vegetation compared to traditional NDVI.

4.2. Results of the Two Constructed Models

To obtain the final models, the PCA method was used to generate the RSEI model (4 factors), whereas both PCA and AHP methods were adopted for the IRSEI model (5 factors). The results for each principal component calculated by PCA are listed in

Table 2. Contributions of the model’s ecological factors by PCA method.

Model	Factors	PC1	PC2	PC3	PC4	PC5
IRSEI (Proposed)	CVI	0.703	0.609	0.257	0.260	0.019
	WET	0.382	−0.616	0.152	0.212	0.638
	NDBSI	−0.491	0.463	0.235	−0.041	0.698
	LST	−0.288	0.005	−0.165	0.938	−0.096
	DI	−0.188	−0.188	0.910	0.071	−0.310
	EV 1	0.032	0.008	0.005	0.002	0
	pEV (%) 2	68.169	17.266	10.257	3.286	1.023
RSEI (Original)	NDVI	0.337	−0.384	0.550	0.661
	WET	0.572	0.575	−0.429	0.399
	NDBSI	−0.671	−0.057	−0.385	0.630
	LST	−0.330	0.721	0.604	0.083
	EV 1	0.021	0.002	0.001	0
	pEV (%) 2	83.25	9.56	6.00	1.150

¹ EV is the eigenvalue; ² pEV is the contribution of the eigenvalue; PC_i is the i-th component of PCA.

.

As can be seen from Table 2 that only the first principal component (PC₁) has real ecological significance and integrates the main information (68.619%, 83.25%) for both the RSEI and IRSEI models, regardless of whether 4 or 5 ecological factors are considered. In PC₁, the ecological factors that are beneficial to the EEQ show positive values, while detrimental ones are negative. This illustrates its stability for the first principal component PC₁. In addition, among other principal components (PC_2–5), their values are either positive or negative, making it difficult to distinguish the actual ecological meaning. Therefore, it is not suitable for us to adopt PC_2–5 for EEQ evaluation except the PC_1. Moreover, in the PC₁ component of the IRSEI model, CVI has the largest value (0.703), which indicates that vegetation plays a decisive role in the evaluation of ecological environment quality. This result is consistent with similar studies [11,53,54]. Furthermore, the proportion of PC₁ in the IRSEI model reached 68.169%, while other principal components showed both positive or negative values, making it difficult to distinguish their actual ecological significance.

Thus, it is not suitable to solely use PCA to generate the IRSEI model and assess the EEQ, as it results in the loss of 31.831% of the information. To overcome this key issue, the AHP analysis method, incorporating expert prior knowledge, was employed to refine the proposed IRSEI model; the details are listed in

Table 3. The weights for the IRSEI model while using only the AHP method.

	Ecological Factors					Weight	CI	RI	CR
	CVI	WET	NDBSI	LST	DI
CVI	1	3	2	4	3	0.398	0.023	1.12	0.021
WET	1/3	1	1/2	2	2	0.160
NDBSI	1/2	2	1	3	2	0.242
LST	1/4	1/2	1/3	1	1/2	0.079
DI	1/3	1/2	1/2	2	1	0.122

CI is the consistency index; RI can be obtained by looking up the table. CR is the consistency ratio.

. It is generally believed that a CR less than 0.1 indicates compliance with the consistency test, indicating that the constructed model is reliable [55].

As can be seen from Table 3, the consistency ratio (CR) was 0.021, which was less than 0.1; thus, the results obtained have passed the consistency test, indicating that the calculated weights are reliable. Furthermore, by combining the PCA–AHP method, the weights (W_i) for the IRSEI model can be further fixed based on the minimum entropy. In conjunction with the five EEQ evaluation grades outlined in Table 1, the weights for the proposed IRSEI model can be determined using Equation (11) to Equation (13), as listed in

Table 4. The final weighs (Wi) of the IRSEI model using the PCA–AHP method.

Factors	CVI	WET	NDBSI	LST	DI
Weights (Wi)	0.372	0.173	0.242	0.107	0.106

.

4.3. Results of Model Validation

As outlined in the model validation process (Section 3.3), 50 random samples were selected. These points were evenly distributed across the study area (Figure 4).

The accuracy of the evaluation results was verified using Google Earth images. The EEQ grades determined by the RSEI and IRSEI models were matched with the reference grades, and the details are illustrated in

Table 5. Comparison of EEQ grades derived from the RSEI and IRSEI models.

Model	The Number of EEQ Grades That Matched the Reference Grades	Overall Accuracy (OA)
RSEI	34/44	77.27%
IRSEI	37/44	84.09%

, where the overall accuracy for the RSEI and IRSEI models is 77.27% and 84.09%, respectively.

A visual comparison of EEQ accuracy between the RSEI and IRSEI models, based on Google Earth imagery, is presented in Figure 5. The IRSEI model demonstrates a higher level of detail compared to the RSEI model.

To further evaluate the internal performance of the IRSEI model, the correlation coefficients between the various factors, as well as the coefficients between each calculated factor and the IRSEI model, are shown in Figure 6 and Figure 7, respectively.

As can be seen from the figures, CVI and WET exhibit positive correlations with the actual values, while the remaining factors display negative correlations. Moreover, the proposed IRSEI model demonstrates a strong average correlation of 0.755, indicating its effectiveness in integrating key ecological factor information and outperforming individual factors in explaining EEQ.

4.4. Results of Model Application in Miyun

To evaluate the effectiveness and reliability of the proposed IRSEI model, the Miyun region was selected as a case study area. The model’s performance was assessed through EEQ evaluation (Section 4.4.1) and consistency analysis within the context of territorial spatial planning (Section 4.4.2).

4.4.1. Evaluation of Eco-Environmental Quality (EEQ)

With the acquired RSEI and IRSEI models, the results of EEQ can be mapped into five grades according to the criteria outlined in Section 3.3, as illustrated in Figure 8.

As can be seen from Figure 8, the spatial distribution of EEQ by the two models is roughly the same. Regions with low EEQ values (poor) are concentrated in the southwest and northeast, while areas with good EEQ are predominantly located in the mountainous northwest and southeast. These findings align with existing knowledge of the region [56,57]. A graphical representation of the EEQ evaluation results is provided in Figure 9.

As shown in Figure 9, the general trends of urban ecological monitoring were similar, but there were differences in ecological levels. Notably, the most significant differences occur between the good and moderate categories. Despite identical grading standards, the IRSEI model can offer a more distinct EEQ classification. To further examine these model-specific EEQ discrepancies, an area transfer matrix (

Table 6. EEQ change the transfer matrix from RSEI to IRSEI.

EEQ Grade		IRSEI
		Poor	Fair	Moderate	Good	Excellent
R S E I	Poor	4.41	0.076
	Fair	6.271	98.748	3.311
	Moderate		121.076	319.154	63.118
	Good			338.877	754.036	162.845
	Excellent			0.157	73.867	132.396

) was constructed.

It can be seen from Table 6 that 62.97% of the EEQ evaluation results (diagonal line in the table) are the same, indicating that the evaluation results of IRSEI and RSEI are the same. In addition, 11.04% of the IRSEI areas (above the diagonal) are better than the RSEI areas, while 25.99% (below the diagonal) have deteriorated ecological grades. This shows that IRSEI is more sensitive to EEQ than RSEI. A visual map of the EEQ change transfer matrix is illustrated in Figure 10.

Figure 10 illustrates that regions experiencing EEQ improvement under the IRSEI model are primarily located in the mountainous northwest (40°30′N–40°48′N, 116°40′E–117°00′E) and southeast (40°20′N–40°30′N, 117°00′E–117°15′E), characterized by abundant vegetation and superior EEQ. Conversely, areas exhibiting EEQ deterioration are concentrated in the urbanizing southwest and northeast, where rapid development and increased human impact have adversely affected the ecological environment.

4.4.2. Consistency Assessment in Territorial Spatial Planning

To verify the reliability of the proposed model, areas of EEQ results acquired by the IRSEI and RSEI models are spatially segmented and counted, as listed in

Table 7. Area of EEQ grade in different ecological functional zones.

Zones	Models	Poor	Fair	Moderate	Good	Excellent
A1	IRSEI	6.651	113.56	171.783	69.435	3.071
	RSEI	2.941	59.13	145.638	142.868	13.924
A2	IRSEI	0.345	10.203	61.789	264.264	185.36
	RSEI	0.147	4.921	48.485	394.54	73.868
A3	IRSEI	0.059	3.852	26.536	26.337	6.107
	RSEI	0.024	0.964	13.511	41.489	6.899
A4	IRSEI	2.624	65.119	203.305	240.184	32.975
	RSEI	1.061	31.507	160.15	316.071	35.422
A5	IRSEI	0.695	25.348	196.223	288.886	72.348
	RSEI	0.23	10.706	131.989	360.203	80.373

A1–A5 are the five functional zones from the Territorial Spatial Planning (2017–2035) of Miyun.

.

As seen in Table 7, the EEQ grades from the IRSEI and RSEI display the same trend in different ecological functional zones (A1–A5), aligning well with the actual situation of the territorial spatial planning data in Miyun. The ecological function zone with the best ecological environment quality (EEQ) is A2 because it has high-vegetation coverage and the richest biomass, while the poorest EEQ is the comprehensive development core zone (A1) due to the most serious human interference. In addition, the zones A3, A4, and A5, designated as ecological protection areas by the government, demonstrate relatively good EEQ. Furthermore, the EEQ statistical results obtained by IRSEI and RSEI models are visualized in Figure 11.

As can be seen from Figure 11, the statistical results across different ecological functional zones reveal significant differences in EEQ. In the ecological function zone A2, the mean value of EEQ acquired from the proposed IRSEI model is 0.739, while RSEI is 0.713. In the poorest zone, A1, the mean values of EEQ calculated by the IRSEI and RSEI are 0.473 and 0.558, respectively. These results indicate that the EEQ achieved by the IRSEI is more effective and accurate than the RSEI model.

5. Discussion

5.1. Advantages of 3D Factors

In previous studies, 2D greenness indices are always created for the RSEI models [54]. These indices are the Normalized Difference Vegetation Index (NDVI) [58], Enhanced Vegetation Index (EVI) [23], Leaf Area Index (LAI) [59], Fractional Vegetation Cover (FVC) [60], and others. Different from these existing EEQ evaluation models that only use 2D ecological factors, the proposed IRSEI model introduces a new 3D ecological factor (CVI) in addition to 2D factors. The CVI factor can obtain vertical structural parameters of vegetation. Compared to the 2D NDVI factor, the 3D CVI factor that integrates vegetation canopy can more clearly distinguish between meadow and forest by values. Thus, higher CVI values were observed in forested areas, while lower values were found in urban and low-vegetation areas, as shown in Figure 3a,b. This is consistent with studies that tested in small areas [30]. Liu et al. [54] employed the method of multiplying the height of the vegetation canopy by the vegetation cover to calculate the three-dimensional greenness rate. Nonetheless, the three-dimensional greenness index is exclusively applicable to forested areas and lacks applicability to non-forested areas. Its contribution to evaluating the quality of the ecological environment is rather limited. Other 3D factors predominantly concentrate on small-area studies [30,61], utilizing methods such as airborne radar or ground-based scanners to acquire three-dimensional parameters of vegetation.

To intuitively understand the advantage of 3D ecological factors, three key areas (A1–A3) are selected and scaled to compare between 2D and 3D ecological factors, as illustrated in Figure 12.

As can be seen from Figure 12, the 3D CVI factor in the scaled areas (A1–A3) has a good discrimination ability for vegetation with different canopy heights and can distinguish woodlands, urban green spaces, and cultivated land with high-vegetation canopies to a large extent. The incorporation of 3D data into CVI effectively addresses the issue of saturation of remote sensing information when ecological factors are inverted through the sole use of optical remote sensing data [62]. This is consistent with the real-world observations, indicating that vegetation canopy height can effectively address the shortcomings of 2D EEQ factors, offering a more comprehensive representation of surface vegetation conditions. Therefore, the IRSEI model can transform EEQ evaluation from 2D to 3D and improve EEQ accuracy by introducing 3D factors.

5.2. Comparison of RSEI and IRSEI Models

The PCA has been widely used to determine the weights of EEQ models [63]. It is an adaptive method that may extract noise as a principal component [64], leading to unreliable results for some of the principal components. As listed in Table 2, the contributions of PC₁ to the RSEI model was 83.25%, while it was 68.169% to the proposed IRSEI model. In addition, the positive and negative values from PC₂ to PC₅ cannot explain EEQ. Therefore, it is difficult to accurately characterize the regional EEQ using only PC₁, which contributes less [65,66,67].

To address this issue, a PCA–AHP hybrid weighting method was introduced, which can couple the weights of each factor based on the minimum entropy theory [68]. It avoids the subjective results inherent to the AHP-only approach and the instability of PC₁ contribution caused by only the PCA method [46], as listed in Table 4. Therefore, the PCA–AHP method emerges as a reliable approach for EEQ assessment.

Furthermore, to compare these two models, we have visualized the relationships between each EEQ model and its factors, as illustrated in Figure 13.

As illustrated in Figure 13, the IRSEI and RSEI models exhibit consistent patterns in the influence of key ecological factors on EEQ. Specifically, NDVI and WET positively contribute to RSEI, while NDBSI and LST negatively impact it. These findings align with previous studies [22,24,69]. Furthermore, the effect of CVI on the new IRSEI model was found to be greater than that of NDVI on the RSEI model, indicating that the vegetation factor (CVI), with the inclusion of 3D data, had a stronger explanatory power on the ecological environment [54]. The distribution of each ecological factor under different EEQ models was calculated (Figure 14).

It can be seen from Figure 14 that there is a significant difference in distribution between the two models. NDVI is mainly concentrated in the region of high values (0.6–1). The main reason is that NDVI exhibits low sensitivity in high-vegetation areas is low, which leads to saturation. This saturation effect makes it challenging to accurately differentiate between various vegetation types or conditions within these areas [54,62]. In contrast, CVI demonstrates a more balanced distribution with aggregations at both extremes. The reason for this is mainly that, after adding 3D data, CVI is more sensitive to vegetation compared to NDVI [30,54].

As seen in Figure 15, EEQ quantitatively reflected by the IRSEI model and remote sensing images is more consistent. For the monitoring of high-density contiguous built-up areas and high-vegetation coverage areas, the IRSEI model is more consistent with actual ecological environment conditions than RSEI. In Figure 15(B1), for contiguous built-up areas, RSEI indicates that the ecological quality of these urban green spaces is too high. Contrastingly, IRSEI, by adding 3D factors and increasing air quality factors, can more accurately monitor the ecological environment and avoid a higher EEQ setting. Consistent with the findings of Mao and Wang [70], who conducted a dynamic change analysis of ecological vulnerability in Miyun, the EEQ in the northwest forested area is significantly higher than in built-up areas. As seen in Figure 15(B2,B3), the integration of the 3D factor accentuates the disparity between RSEI and IRSEI in forest areas with high-vegetation cover. As expected, forest areas with high-vegetation cover exhibit superior EEQ compared to those with lower cover [71]. The IRSEI model effectively captures this relationship, demonstrating its ability to accurately represent ecological phenomena.

6. Conclusions

In this research, a novel IRSEI model for large-scale EEQ assessment is proposed. It introduces air quality and 3D vegetation factors into the IRSEI framework and utilizes the PCA–AHP approach to determine optimal factor weights. Comprehensive experiments were conducted in an ecological support district. The results indicated that the proposed IRSEI model outperformed the existing RSEI model, and can achieve results that are more consistent with real-world EEQ conditions. The following conclusions were obtained from the experiment: (1) The difference index for the air quality factor can capture the changes in particle concentration, which is the chief pollutant of atmosphere pollution in Miyun, while the 3D vegetation factor can express the 3D canopy height. Consequently, the IRSEI model delivers more detailed EEQ assessments. (2) The PCA–AHP approach successfully balances objective and subjective information among the input ecological factors, thereby enhancing model stability and weight reliability.

In the future, more 3D ecological factors (e.g., building heights, and DEM), natural and human factors (e.g., population density, road network density), and time series data need to be taken into consideration.