Evaluation of Perceptions Using Facial Expression Scores on Ecological Service Value of Blue and Green Spaces in 61 Parks in Guizhou

Abstract

This study selected 61 parks in Guizhou province as research points and collected 3282 facial expression photos of park visitors in 2021 on the Sina Weibo platform. FireFACE v1.0 software was used to analyze the facial expressions of the visitors and evaluate their emotional perception of the landscape structure and ecosystem service value (ESV) of different landscape types of blue–green spaces. Research shows that the average ESV of green spaces in parks is USD 6.452 million per year, while the average ESV of blue spaces is USD 3.4816 million per year. The ESV of the blue–green space in the park shows no geographical gradient changes, while the happiness score in facial expressions is negatively correlated with latitude. Compared to blue spaces, green spaces can better awaken positive emotions among visitors. The ESV performance of different types of green spaces is as follows: TheroponcedrymV > GrasslandV > Shrubland V. The landscape structure and ESV of the blue–green space in the park can be perceived by visitors, and GreenV and vegetation height are considered the main driving factors for awakening positive emotions among visitors. In Guizhou, when the park area decreases, people are more likely to experience sadness. Regressions indicated that by increasing the green space area of the park and strengthening the hydrological regulation function of the blue–green space, people can achieve a more peaceful mood. Overall, people perceive more positive sentiments with high ESV in blue–green spaces of Karst parks but low ESV in shrubland.

1. Introduction

The term ecosystem services (ESs) refers to all benefits derived directly or indirectly from ecosystems by human beings, including provisioning, regulating, supporting, and cultural services [1,2]. All these types of services act as a guarantee for sustainable social development [3], the enhancement of human–nature interaction [4], the mitigation of climate change [5], and an increase in the perception of quality life [6]. Under the background of dually economic globalization and urbanization sprawl, excessive anthropogenic activities have resulted in the destruction of ecosystem structure and the degeneration of ESs [7]. These negative influences have threated human survival and the ecosystem environment, which is attracting a lot of attention to the responses of ESs [8]. ES value (ESV) can be used to assess the magnitude that humans can enjoy ESs, hence creating a scientific evaluation of regional ESV benefits in terms of policymaking for environmental management and economic development [9]. A scientific system for ESV evaluation is a necessary reference for exploiting the ESs provided by natural resources.

In 1997, Costanza et al. evaluated ESV in terms of monetary values for global ecosystems, which activated a hotspot of relevant studies on natural capital and placed them as classical references to which further studies show follow up and depend on [10]. Thereafter, Costanza et al. (2014) further optimized the evaluation system of ESV [11], and further studies on ESV were commenced to rapidly accumulate this in Western countries [12,13]. Currently, the assessment of ESV has been extended to the systematic employment of various parameters from gauging equivalent factors, such as, but not limited to, carbon tax, shadow engineering, travel cost, etc. [14,15]. However, multiple parameters result in a disputation that ESV assesses by different methodologies that may generate trade-offs between each other for a given region [16,17]. This was caused at least partly by insufficient detections about humans’ perceptions towards ESV in the region, and some parameters were irrelevant to users’ awareness about regional ecosystem sustainability [18,19]. No matter what type of ES was focused on, human beings are always relevant users; hence, ESV has to be assessed with the dimension of relevance in terms of human perception.

People are paying increasing attention to public health as a response to possible worries about the consequences of fast economic acceleration. The COVID-19 pandemic resulted in attention being concentrated on mental health and perceptional outcomes [20,21]. An experience of nature is taken as a nature-based solution to improve mental concentration and promote attention recovery [22]. The stress recovery theory (SRT) proclaimed that a touch of nature can alleviate mental stress and promote stress recovery [23]. Green and blue spaces (GBSs) account for the majority of frequent visits to nature in a city, which function as ecological infrastructures that provide ESs to promote public health and well-being [24,25,26,27]. The psychological and mental benefits obtained from experiences with GBSs can further result in preventing anxiety [28], depression [25], chronic disease [26], and the impact of air pollution [29,30]. These effects were mainly detected for visitors exposed to GBSs in parks, communities, and streets in built-up regions of the host city [31]. Compared to urban infrastructures, parks located a remote distance from cities’ downtowns are more attractive to visitors; hence, GBSs in parks located in a wild environment may be able to elucidate stronger enhancement than those placed under neighboring conditions [32,33]. The effects of GBSs in parks on human health and well-being can also be evaluated as a part of ecosystem cultural services (ECSs) [27,34]. Regarding a vast geographical range of GBSs located in parks with highly varied conditions and natural doses, ESV should also be within a large variation depending on the specific locations of interest [35,36]. For a specific given ecosystem type, a large knowledge gap is apparent in the current literature about the detection of ESV in ECSs for GBSs in local parks.

Ecosystem services and their values both result from the dependence on host ecosystem structures that can be analyzed using landscape metrics [37,38]. For example, ecosystem degradation has resulted in shrinking the areas of pastures, marshlands, and surface-waters and reduced the ESs of habitat integrality for the preservation of species in Niemodlin Forests [38]. Light and moderate ecosystem degradations were found to reduce ESs in carbon (C) storage, nutrient supply, and water retention [39]. However, degraded forests had a different landscape, and thus aesthetic scenery, that was more attractive for visitors with enhanced CES [33]. The Karst ecosystem is suffering from ecological degradation due to rocky desertification; some regions of southwestern China are characterized as some of the largest Karst landforms globally [40,41]. The literature documents highly heterogeneous distribution patterns of eight ESs in the Karst ecosystem of Guizhou [42], which was formed by contrasting driving forces between urbanization and restoration [43]. Different types of ESs varied in response to the magnitude of rocky desertification, which mainly reshaped ESs in soil retention and water yield but did not cause distinct differences in C storage and crop production [44]. These regulation ESs, plus provision ESs, were formed through changes in ecological factors following the growing area of stony land conversion, but ESs in urban sprawl bundle were shaped with a higher reliability on socioeconomic factors [42]. However, findings about the ESV in the Chinese Karst ecosystem did not fully concur with ESs on Karst lands with varied levels of rocky desertification. Lu et al. (2022) reported that fully decertified lands were evaluated to have a higher ESV than those subjected to partly rocky lands [45]. High elevation (~1100 m) resulted in a higher ESV in provision ESs compared to lower altitude (~800 m) [45], but the ESV for ECSs increased, with rising elevation dominating the topographical index [46]. The evaluation instrument in these studies varied with low matching comparations and, furthermore, the rationality for contrasting ESV needs to be judged by a third-party parameter that has reasonable scientific meaning. A better understanding of perceived health and well-being in the parks of the Karst regions will benefit strategies that improve their ESV in this function. Thus, little is known about this effect on humans’ mental well-being in the Karst regions.

Hauman beings are the major receivers of ESs, and their perceptions about ESV are a fatal reliance on which ESs are satisfied. The perceptions of ESV were evaluated for local users and stakeholders of natural resources [47,48,49]. Current studies mostly employed self-reported data that were collected from questionnaire surveys or face-to-face interviews. Although theses conventional methodologies accounted for the major body of current studies, the results are still being queried for existing human errors [22] and a lack of validation [50]. The development of big data utilization contributes to the employment of facial expression scores as a gauge measuring perceptions of people in an experience with nature [51,52,53]. Inspired by this novel methodology, it was successfully used to assess ESV in wetlands at the provincial scale [27]. This suggests a probability of using this novel methodology for evaluating ESV in an altitude-determined landscape.

In this study, a total of 61 parks were selected as study plots where facial photos of visitors were obtained from a social network service (SNS) platform. Facial expression scores were recognized and rated for evaluating the perceptions of ESV and detecting landscape metrics that can evoke valued ESs perceived by people. The Karst ecosystem was chosen as the objective of this study, and our findings will be useful for managing a local ecosystem to provide highly valued services for users. Our study generates a contribution to the field of ESV evaluation in a special ecosystem type of the Karst regions in Guizhou. The novelty of our study is shown by employing emotional scores as an assessment of the perceptions towards experiences in the GBSs of the parks in Guizhou.

2. Materials and Methods

2.1. Study Area

Guizhou province was chosen as the study area as its 92.5% realm is covered by mountains and hills, contributing towards making it one of the top three largest Karst landforms [54]. Guizhou has a total population of 38.56 million in a total land area of 17.62 × 10⁴ km². Guizhou is located in the southwestern part of China, where local biomes are subjected to a subtropical monsoon climate. The annual temperature is about 15 °C, and averaged rainfall is 1100–1300 mm. In total, nine municipal regions (six prefecture-level cities and three autonomous prefectures) are within Guizhou, containing 61 parks (Figure 1). The overall terrain spans from low elevation in the west to high elevation in the east, with an average elevation of 1100 m.

2.2. Landscape Structure

Landscape metrics were analyzed using ArcGIS 10.8 version (Eris Branch Inc., Shanghai, China). The horizontal plane of the landscape structure can be characterized as alternatively linked green (vegetation) and blue (lakes, wetlands, river, etc.) patches, and the vertical plane was mainly characterized by the gradient change in green patches. Land use data were adapted from a dataset published in 2021 (resolution: 1 m × 1 m), on which the areas of 61 parks were outlined by human–computer interactions. Landscape-projected coordinates were adapted with systems of WGS_1984_UTM_Zone_48N and WGS_1984_UTM_Zone_49N. Elevation was determined by a digital elevation model (DEM) in ASTER GDEM 30M [55]. The height of the highest surface feature was evaluated using the digital surface model (DSM) from AW3D30 DSM [56]. Vegetation height (VegH) was calculated as the difference in the DSM by subtracting DEM [57].

2.3. Facial Photos Collection and Treatment

Sina Weibo is the biggest SNS platform in China, which is also termed ‘Chinese Twitter’. Sina Weibo has been repeatedly used as a source of data about facial photos to undergo further analysis [52,58]. In this study, Sina Weibo was also employed as a source of facial photos. We stated that all photos that were about to be documented for analysis had to be taken within places of objective parks, where only those with backgrounds in the open air could be targeted. Indoor photos were excluded from our photo pool. All photos that we decided to use in this study had to have three characteristics:

(1)

All the people in the photos had to have the typical facial characteristics of East Asian races;

(2)

The facial organs of the five senses had to be fully exposed without any covers or shades;

(3)

No excessive digital decoration could be employed on the faces.

Some photos contained more than one faces; hence, they had to be divided into smaller ones containing single faces per shot [59]. To increase the rate of the successful recognition of the facial expressions, all photos were rotated using WPS software (Kingsoft Office, Beijing, China) to make the nose vertical to horizontal line of the host cheek. The photos were further cropped to leave facial areas accounting for over 66.7% of area in a whole photo. The treated photos were analyzed for happy, sad, and neutral emotion scores by FireFACE v1.0 software (Zhilunpudao Inc., Changchun, China). This software also calculated the rates of the positive response index (PRI) through subtracting the happy score from the sad score. This set of operations used to analyze happy, sad, and neutral expression scores using the suggested software has been validated by Guan et al. [60] and tested by Wei et al. [61]. Finally, a total of 3282 photos were successfully recognized by the software and rated to emotional scores.

2.4. Economic Evaluation of Ecosystem Services

According to previous studies [10,15], local ESs were divided into 4 class-I types and 11 class-II types. A basic unit was equaled to one seventh of food production from a unique area of local farmland [12]. Regarding the difference in the price of yearly commodities that may affect the price of the ESV, the prices of crop foods produced on the farmlands of Guizhou in 2021 were taken as the standard reference. Different prefectures had varied socioeconomic states, meaning that pricing based on local market price may result in the underestimation of ESV [62]. To further increase the estimate accuracy, the ESV estimate was modified on the basis of the coefficients in the nexus of spatial heterogeneity, socioeconomic regulation, and resource scarcity [63]. The per unit area coefficients for the ESV estimate are listed in

Table 1. Coefficients of ESV estimate per unit area in Guizhou province (USD hm⁻²).

Ecosystem Service Value		Green Spaces			Blue Spaces
First Service Category	Second Service Category	Theropencedrymion	Shrubland	Grassland	Water
Provisioning	Food production	68.68	42.09	51.69	177.23
	Raw material production	157.29	95.26	76.06	50.95
	Water resources supply	81.97	48.74	42.09	1836.55
Regulating	Gas regulation	520.61	312.37	267.32	170.58
	Climate regulation	1557.41	937.11	706.71	507.32
	Environmental purification	440.86	283.57	233.35	1229.54
	Hydrological regulation	777.60	742.15	517.66	22,650.04
Support	Soil conservation	633.60	381.05	325.66	206.03
	Maintain nutrient cycle	48.74	28.80	25.11	15.51
	Biodiversity	576.00	347.81	296.12	564.92
Cultural	Aesthetic landscape	252.55	152.86	130.71	418.71

. Therein, the ESV was categorized to be the first and second rank. In the first service category, four major functions were involved, namely provisioning, regulating, support, and cultural. In the second service category, every first service category was further divided into three to four specific functions.

The area ratio of blue space to green space (BAGA) was found to be a critical factor that can be perceived as a determinant of ESV [19]. This parameter was considered for further analysis plus the ratio of ESV in blue space to green space (BVGV). Detailed calculations are as follows:

(1)

A basic unit of ESV (E; unit of CNY ha⁻¹) is calculated as follows:

$$E=\frac{1}{7}\times {\sum }_{i=1}^n\frac{m_i{\times p}_i\times q_i}{\mathrm{M}}$$

(1)

where E equals the sum of all (since i = 1 up to n) the crop species, mainly rice, wheat, corn, soybean, potato, etc.; p_i is the price of i crop species (CNY ton⁻¹); q_i is the crop yield per unit area for species i (ton ha⁻¹); and m_i is the farming area of ith species. Data were obtained from yearbooks and published data from the statistical bureau. Finally, averaged regional crop food production was priced to be USD 235.76 ha⁻¹ after being modified through eliminating the spatiotemporal differences.

(2)

Net primary production coefficients

Net primary production (NPP; unit of kg ha⁻¹ yr⁻¹) is a basic gauge of the provision service in an ecosystem [64]. The Thornthwaite Memorial model was used to estimate the ESV coefficients [65]:

$$\mathrm{N}\mathrm{P}\mathrm{P}=3000\times [1-e^{-0.0009695\times (\mathrm{v}-20)}]$$

(2)

$$\mathrm{V}=\frac{1.05\times r}{\sqrt{1+{\left(\frac{1.05\times r}{\mathrm{L}}\right)}^2}}$$

(3)

$$\mathrm{L}=300+25\times \mathrm{T}+0.05\times {\mathrm{T}}^3$$

(4)

$$\mathrm{H}\mathrm{e}\mathrm{t}\mathrm{e}=\frac{{\mathrm{N}\mathrm{P}\mathrm{P}}^{\prime }}{{\mathrm{N}\mathrm{P}\mathrm{P}}^{″}}$$

(5)

where V is the actual evapotranspiration (mm); r is annual rainfall (mm); L is the averaged amount of evaporation (mm); T is annual temperature (°C); Hete is the spatial heterogeneous coefficient; NPP′ is NPP in the objective study region, and NPP″ is NPP at the national scale.

(3)

Socioeconomic coefficients

Regarding variation in regional household incomes, people also varied in terms of their willingness to pay for ESV [66]. The socioeconomic regulation coefficient (D) is calculated as follows:

$$\mathrm{D}=\mathrm{R}\times \mathrm{W}$$

(6)

where R is the willingness to pay for ESV, and W is the ability to afford to make the payment. Furthermore, R can be calculated as follows:

$$\mathrm{R}=\frac{{\mathrm{Z}}_{\mathrm{m}}^{\prime }}{{\mathrm{Z}}_{\mathrm{g}}^{\prime }}$$

(7)

$$\mathrm{Z}=\frac{1}{1+{\mathrm{e}}^{3-\frac{1}{{\mathrm{E}}_{\mathrm{n}}^{\prime }}}}\times \frac{\mathrm{h}}{\mathrm{H}}$$

(8)

$${\mathrm{E}}_{\mathrm{n}}^{\prime }={\mathrm{E}}_{\mathrm{a}}\times {\mathrm{P}}_{\mathrm{a}}+{\mathrm{E}}_{\mathrm{b}}\times {\mathrm{P}}_{\mathrm{b}}$$

(9)

where Z is the coefficient of the social development assessment; Z′_m and Z′_g are the development levels in the study area and at the national scale, respectively; E′_n is the Engel coefficient; E_a and E_b are the Engel coefficients for the prefecture and rural households, respectively; P_a and P_b are prefecture and rural populations, accounting for those in total, respectively; and h and H are coefficients assessing the urbanization levels within the study area and at the national scale, respectively.

$$\mathrm{W}=\frac{{\mathrm{G}\mathrm{D}\mathrm{P}}^{\prime }}{{\mathrm{G}\mathrm{D}\mathrm{P}}^{″}}$$

(10)

where GDP′ and GDP″ are the gross domestic product in the study area and at the national scale, respectively.

(4)

Scarce resource coefficients

Regional population has a trade-off with the holding capacity of natural resources; hence, a dense population will impact ESV [67]. The scarce resource coefficient (F) is calculated as follows:

$$\mathrm{F}=\frac{\mathrm{l}\mathrm{n}\mathrm{g}}{\mathrm{l}\mathrm{n}\mathrm{G}}$$

(11)

where g is the population density in the study area, and G is the average population density at the national scale.

(5)

Regional ESV estimation

Region ESV (CNY) can be estimated according to the following equation:

$$\mathrm{E}\mathrm{S}\mathrm{V}=\sum _{i=1}^n({\mathrm{A}}_i\times \mathrm{N}\times \mathrm{D}\times \mathrm{F})$$

(12)

where n is the type of landscape, including a mixed broadleaf–conifer forest, coppice, meadows, waterbodies, etc. The coefficients of N, D, and F were estimated to be 1.45, 0.60, and 1.08, respectively.

2.5. Mapping and Statistical Analysis

In our study, a park comprises green and blue spaces, impervious land (parking lot, road, playground, etc.), bare soils, farmland, etc. Therefore, the area of the park is larger than any of the metrics therein. The landscape structure was characterized as the metrics of green space area (GreenA), blue space area (BlueA), park area (ParkA), elevation, VegH, and BAGA prices of their services (GreenV, BlueV, ParkV, and BVGV). The exposed sentiments of the visitors were characterized as the facial expression scores for happy, sad, and neutral emotions and the PRI. All these data were mapped in the study area and further used as the baselines to interpolate their distributions in ArcGIS.

All data were analyzed using SPSS v.12.0 software (IBM China-Branch, Shanghai, China). The data failed to follow the normal distributions; hence, they were transformed by ranking when being analyzed. Furthermore, the landscape metrics and their service prices were transformed to logarithms to eliminate abnormally extreme observations that caused heterogeneous variance. An analysis of variance (ANOVA) was used to detect locational variation in parks on the facial expression scores, landscape metrics, and ESV. When significant effects were indicated, the results were compared using Duncan’s test, with critical significance at 0.05. General linear models (GLMs) were used to detect relationships between geographical coordinates (latitude or longitude) and dependent variables. Multivariate linear regression (MLR) was used to detect multiple sources of independent drivers contributing to the dependent variables.

3. Results

3.1. Description of Dataset

As is shown in

Table 2. Description and difference examination of data.

Parameters	N	Max	Min	Mean	sd	CV	F Value	p Value
	Facial Expression Scores
PRI (%)	3282	58.11	−9.58	29.54	13.75	0.47	4.97	<0.0001
Happy (%)	1522	98.73	70.83	87.26	5.89	0.07	8.42	<0.0001
Sad (%)	222	86.39	0.00	50.61	24.02	0.47	0.91	0.507
Neutral (%)	1538	86.08	62.11	76.71	4.61	0.06	9.18	<0.0001
	Landscape metrics
Elevation (m)	61	2481.73	393.56	1172.93	429.30	0.37	9.14	<0.0001
VegH (m) 1	61	17.38	2.64	10.53	3.09	0.29	2.83	0.011
ParkA (km2) 2	61	452.10	0.03	18.75	66.43	3.54	0.69	0.696
TheropencedrymionA (km2) 3	61	230.56	0.01	9.94	37.25	3.75	0.76	0.638
ShrublandA (km2) 4	61	0.03	0.00	0.00	0.00	7.48	1.35	0.241
GrasslandA (km2) 5	61	142.09	0.00	4.94	20.16	4.08	0.55	0.817
BlueA (km2) 6	61	43.92	0.00	1.25	6.30	5.03	0.57	0.801
GreenA (km2) 7	61	372.65	0.01	14.88	55.96	3.76	0.66	0.726
	Ecosystem service value (thousand USD yr−1)
TheropencedrymionV 8	61	117,941.17	5.25	5083.71	19,053.22	3.75	0.76	0.638
ShrublandV 9	61	10.96	0.00	0.19	1.40	7.48	1.51	0.178
GrasslandV 10	61	37,972.80	1.22	1321.34	5388.12	4.08	0.55	0.817
BlueV 11	61	122,231.64	0.00	3481.66	17,526.56	5.03	0.44	0.893
GreenV 12	61	155,913.98	6.47	6405.24	23,976.97	3.74	0.70	0.691
ParkV 13	61	157,528.69	6.47	9886.90	29,476.40	2.98	0.58	0.789
ProvisioningV 14	61	9633.08	0.39	648.35	1940.01	2.99	0.56	0.803
RegulatingV 15	61	110,163.23	4.17	7201.69	21,693.89	3.01	0.55	0.813
SupportV 16	61	38,249.99	1.59	1668.84	5891.48	3.53	0.73	0.666
CulturalV 17	61	7704.48	0.32	368.01	1206.48	3.28	0.68	0.708

Note: ¹ VegH, vegetation height; ² ParkA, park area; ³ TheropencedrymionA, theropencedrymion area; ⁴ ShrublandA, shrubland area; ⁵ GrasslandA, grassland area; ⁶ BlueA, blue space area; ⁷ GreenA, green space area; ⁸ TheropencedrymionV, theropencedrymion ecosystem service value; ⁹ ShrublandV, coppice ecosystem service value; ¹⁰ GrasslandV, grassland ecosystem service value; ¹¹ BlueV, blue space ecosystem service value; ¹² GreenV, green space ecosystem service value; ¹³ ParkA, park ecosystem service value; ¹⁴ ProvisioningV, provisioning value; ¹⁵ RegulatingV, regulating value; ¹⁶ SupportV, support value; ¹⁷ CulturalV, cultural service value.

, the visitors showed distinct differences, as the happy score accounted for 46% of the total emotion while the sad score accounted for only 7%. As a result, the PRI was generally higher than zero, suggesting that visitors can obtain positive emotions. The neutral score accounted for 47%, which was comparable with the happy score.

The locational variation among cities did not result in any further effects on the landscape metrics and ESV. Overall, GreenA was 14.88 km², which was larger than BlueA (1.25 km²). Green space was mainly dominated by broadleaf–conifer forests. The averaged ESV in green space per park was evaluated to be 640.52 × 10⁴ USD yr⁻¹, which was higher than that in blue space per park (348.17 × 10⁴ USD yr⁻¹). In green space, the ESV for different vegetation types ranged in an order of broadleaf–conifer forest > meadow > coppice.

3.2. Spatial Distributions of Landscape Metrics

In eastern regions of Bijie and the central area of Tongren, ParkA and GreenA were larger than those in other regions (Figure 2D,F). The distribution pattern of GreenA concurred with that in theropencedrymionA, but it showed distinct distributions with ShrublandA and GrasslandA (Figure 2A–D). The theropencedrymionA distribution dominated that of GreenA. Large areas of BlueA were concentrated in the west of Guizhou, especially in western regions of Bijie and southwestern regions of Guiyang.

3.3. Spatial Distributions of Ecosystem Services Values

The ecosystem service value of the green space was higher in eastern Bijie and central Tongren, which nearly overlapped that in TheropencedrymionV (Figure 3A,D). Both ShrublandV and GrasslandV showed even levels in most parts in the study area, except for high levels distributed in western corners around the central area of Bijie (Figure 3B,C). GreenV generally followed the same pattern as that for TheropencedrymionV and GrasslandV, all with high values distributed in the unique region of Bijie (Figure 3A,C,D). BlueV was extremely higher in western Bijie and moderately higher in Guiyang than that in most of the other regions (Figure 3E). Overall, ParkV was even in most of the regions of the study area, except for higher levels in the western and eastern regions of Tongren, central region of Tongren, and Qiannan (Figure 3F).

3.4. Spatial Distributions of Facial Expression Scores

The happy score was generally higher in southwestern regions than in northeastern regions, which showed a grossly contrasting pattern with the PRI (Figure 4A,D). High values of the PRI were found to be distributed in regions of Zunyi, Tongren, and Qiandongnan, which resulted from subtracting the moderate happy score from the low sad score. Although the happy score was high in eastern Bijie and Qianxinan, sad and neutral scores were also high in these regions, which together resulted in lower levels of PRI scores (Figure 4B–D).

3.5. Parameter Changes along Topographical Gradients

According to Figure 5, along the increasing gradient of longitude, VegH showed an increasing trend, and the elevations of the parks declined (Figure 5A,E). None of the landscape metrics showed any relationship with latitude for parks, which resulted from high variations in landscape metrics at low and high latitudinal parks.

Among all the facial expression scores, only the happy score showed a negative relationship with the latitude gradient, which was not accompanied with any significant relationships with PRI scores (Figure 6). Neither the sad score nor neutral score showed significant relationships with topographic gradients due to massed dots in moderate ranges.

None of the ESVs showed any relationships with topographic gradients (Figure 7).

3.6. Relationships between Landscape Metrics and Their ESV and Facial Expression Scores

According to

Table 3. Pearson correlation between park facial expression scores and landscape parameters.

Landscape Metrics	Facial Expression Scores
	PRI (Rank)	Happy (Rank)	Sad (Rank)	Neutral (Rank)
Elevation	0.010	−0.004	−0.036 *	0.015
VegH	0.060 **	0.067 **	−0.040 *	−0.056 **
ParkA	0.125 **	0.134 **	−0.082 **	−0.121 **
TheropencedrymionA	0.120 **	0.134 **	−0.069 **	−0.126 **
ShrublandA	−0.018	−0.011	0.021	<0.001
GrasslandA	0.109 **	0.115 **	−0.076 **	−0.101 **
BlueA	0.036 *	0.039 *	−0.023	−0.041 *
GreenA	0.125 **	0.136 **	−0.078 **	−0.125 **
BAGA	−0.040 *	−0.048 **	0.021	0.042 *
TheropencedrymionV	0.124 **	0.138 **	−0.072 **	−0.130 **
ShrublandV	−0.007	−0.001	0.011	−0.007
GrasslandV	0.111 **	0.118 **	−0.078 **	−0.103 **
BlueV	0.044 *	0.044 *	−0.039 *	−0.035 *
GreenV	0.128 **	0.140 **	−0.079 **	−0.130 **
ParkV	0.112 **	0.119 **	−0.074 **	−0.109 **
BVGV	−0.044 *	−0.054 **	0.020	0.052 **
ProvisioningV	0.111 **	0.118 **	−0.074 **	−0.107 **
RegulatingV	0.110 **	0.117 **	−0.074 **	−0.105 **
SupportV	0.122 **	0.133 **	−0.077 **	−0.123 **
CulturalV	0.118 **	0.128 **	−0.076 **	−0.118 **

Note: ** at level 0.01 (two-tailed) means that the correlation was significant. * at level 0.05 (two-tailed) means that the correlation was significant.

, several landscape metrics (TheropencedrymionA, GrasslandA, BlueA, GreenA, and ParkA), plus nearly all types of ESV, had positive relationships with the PRI and happy scores. Meanwhile, these landscape metrics and ESV showed negative relationships with the sad and neutral scores. Elevation and VegH in parks had negative relationships with sad scores, while VegH had a positive relationship with the happy score.

Almost all types of ESV had positive relationships with the happy and PRI scores, while they had negative relationships with the sad and neutral scores. Both BAGA and BVGV had negative relationships with the happy and PRI scores. In contrast, these two parameters had positive relationships with the neutral score.

3.7. Multivariate Linear Regression of Facial Expression Scores against Landscape Parameters

According to

Table 4. Multivariate linear regression of facial expression scores of visitors against landscape structure metrics and ecosystem service value in parks.

Dependent Variables	Independent Variables	Estimate	SE 1	F Value	p Value
PRI (Rank)	Intercept	27.542	3.662	7.522	<0.0001
	GreenV 2	2.95	0.431	6.843	<0.0001
	VegH 3	7.471	3.531	2.116	0.034
Happy (Rank)	Intercept	19.236	1.17	16.435	<0.0001
	GreenV	4.532	0.74	6.126	<0.0001
	Hydrological regulation 4	−1.727	0.725	−2.383	0.017
Sad (Rank)	Intercept	17.518	0.352	49.724	<0.0001
	ParkA 5	−1.212	0.319	−3.794	<0.0001
Neutral (Rank)	Intercept	40.128	4.266	9.406	<0.0001
	TheropencedrymionV 6	−7.421	1.87	−3.968	<0.0001
	Hydrological regulation	5.323	1.211	4.396	<0.0001
	ParkA	−12.069	3.292	−3.666	<0.0001
	GreenA 7	11.326	3.701	3.06	0.002

Note: ¹ SE, standard error; ² GreenV, green space ecosystem service value; ³ VegH, vegetation height; ⁴ Hydrological regulation, hydrological regulation value; ⁵ ParkA, park area; ⁶ TheropencedrymionV, theropencedrymion ecosystem service value; ⁷ GreenA, green space area.

, both GreenV and VegH made positive contributions to the PRI. GreenV also made a positive contribution to the happy score, but hydrological regulation made a negative contribution. ParkA made negative contributions to the sad and neutral scores. In addition, TheropencedrymionV also made a negative contribution to the neutral score, but hydrological regulation and GreenA made positive contributions to the neutral score.

4. Discussion

4.1. Evaluation of Ecosystem Service in Guizhou

Parks are public infrastructures that provide recreation and entertainment for human users, wherein green and blue spaces account for the major services of recovery and enjoyment [68]. In Guizhou, the ESV in green space was evaluated to be 640.52 × 10⁴ USD yr⁻¹, and that in blue space was 348.16 × 10⁴ USD yr⁻¹, which were totaled to be 988.70 × 10⁴ USD yr⁻¹. This value accounts for 1.57% of the price of total ESs in the Karst ecosystem in Guizhou [45]. In detail, the ESV in the green and blue spaces of Guizhou included the prices of the services in regulation (720.17 × 10⁴ USD yr⁻¹), support (166.88 × 10⁴ USD yr⁻¹), provision (64.84 × 10⁴ USD yr⁻¹), and culture (36.80 × 10⁴ USD yr⁻¹). Furthermore, the cultural service value (CSV) in green space was estimated to be 31.56 × 10⁴ USD yr⁻¹, which is lower than that in the parks of Jiangxi (94.13 × 10⁴ USD yr⁻¹) [19]. This was attributed to the variation in the coefficients. The CSV in Jiangxi was estimated on the basis of the global monetary coefficients [10,62], which was better matched for regional estimation in developed countries, but not suitably used in developing countries if no regional fixation was employed. In addition, the study in Jiangxi did not specify detailed species for greeneries in green space, which resulted in the involvement of coppice and grassland and an overrated CSV in Jiangxi. Given that landscape area is a key metric that determines the magnitude of CSV, the total area of green and blue spaces in Guizhou (16 km²) was larger by 1.7 times than that in Jiangxi (9.63 km²); hence, it was reasonable for Guizhou to have a higher CSV than Jiangxi.

Guizhou was grossly estimated to have their ESV distributed higher in the east than in the west [69,70]. Our results further revealed that the ESV was low in the central part and southwestern area of Guizhou and high in the surrounding regions around the provincial outline. Most lands in Guizhou were covered by the Karst landform with imbalanced development levels, which together resulted in the heterogeneous pattern of the regional ESV [71]. The central and southern parts of Guizhou were subjected to larger areas of stony landforms, which largely limited the ESV and led to a comprehensive lower trend in the west than in the east.

Land use/land cover (LULC) have close relationships with ecosystem function and landscape structure [72]. Generally, ecosystem function is strengthened with larger gross biomass accumulation. Therefore, forests should have a higher ability to provide ecosystem services than coppices and meadows as the per-area production in forests is higher than that for shrub and grass [73]. In the Karst ecosystem, Lu et al. (2022) also revealed that forests contributed more to local ESV, followed by farmland, coppices, and grasslands [45]. In our study, ESV was estimated to be the highest in broadleaf–conifer forests, followed by grassland and coppices. A higher ESV in grassland than in coppice can be explained by two aspects. Firstly, the LULC product properties affect the ESV estimate in terms of the resolution, gross precision, and kappa coefficient. In this study, LULC was employed with a 1 m resolution, precision of 73.61%, and kappa coefficient of 0.6595 [74]. This dataset has been proven to be reliable for estimates at the national scale, but its use at the provincial level varies in different regions. One of leading reasons for this may result from the fact that it is easy to confuse the estimate of the ESV on coppice with that on grassland due to similar surface feature characteristics. Hence, a practical solution is to estimate ESV in a provincial area with larger areas as references in neighboring provinces so as to fix the estimate precision in objective areas [75]. Secondly, our objective surfaces were parks, which had a nature to harbor a larger area of grasslands than coppice. That may be another reason why grassland had a higher ESV than coppice as the ESV is an area-dependent parameter.

It was interesting to find no differences in the landscape metric area and ESV between cities, and neither had any relationships with topography. The Karst landform accounted for 73.6% of the whole area in Guizhou. In Karst regions, the ESV in parks could be afforded mainly by the built environment, which was planned with similar budgets among the cities in Guizhou. Otherwise, elevation did not show continuous changes along the topographical gradients; hence, there were no strong drivers that could form ESV distribution along any of the topographical gradients. In Guizhou, the ESV may also change with the increase in elevation in a pattern of “V” [45]. In our study, this responsive pattern of ESV only occurred in Bijie following a gradient order of east > west > southwest.

4.2. Emotional Perception of Ecosystem Service Value and Landscape Structure

In this study, the happy score had a negative relationship with latitude, but the PRI had no relationship with topography. These together suggest that people in regions at low latitudes tended to show more smiles, but more complicated emotions were perceived in larger areas. The change in position did not cause emotional responses, as people experience emotions in response to a particular experience in a landscape, which was within a position that cannot be replaced. This is why the high VegH in large green space can be perceived by visitors who experienced positive emotions. It was surprising that elevation had a negative relationship with the sadness score. This may be caused by season or weather conditions, as warm temperature or sunny days can promote people to perceive more positive emotions. It was found that a range of 17.5–22.3 °C can be an optimization that triggers positive emotions in urban wetland parks [76]. In parks with high elevation, experiences in blue space in summer and autumn can evoke positive emotions, but extreme weather, over moist weather, and blowing velocity together evoke sorrowful emotions [24]. The average temperature in Guizhou was usually 15–16 °C. Under solar radiation, the temperature in Zunyi, Bijie, and Tongren was higher than that in most of the other regions, especially in Liupanshui and Guiyang. The happy score was found to be distributed generally higher in the west and lower in the east, which was mostly accounted for by high happy scores in Bijie. Local weather was accompanied by more sunny days in Bijie, which may be the reason why more smiles could be seen locally. Similarly, the high PRI scores found in Zunyi and Tongren may have resulted from seasonal impacts. Although VegH made a larger contribution to the PRI than ESV in green and blue spaces, none of the landscape metrics made any contributions to the happy score. Therefore, it is easier to disclose positive emotions by perceiving ESV than in green and blue spaces in Guizhou, which concurs with the findings of Zheng et al. (2023) [19].

Guizhou is typically featured as an “ecological fragile region” with a “low economic outcome” [71]. To decide on a strategy, the Guiyang government put forth a project of “a city of thousand parks”, which resulted in a total of 1025 parks built already totaling an area of 45.78 km² up to the end of 2022. In this study, the average area of the parks in Guiyang was only about a quarter of that in the whole of Guizhou. Relatively smaller parks tended to induce more sad emotions. The parks in Guiyang had relatively larger areas of waterbodies, which elicited peaceful perceptions that alleviated sadness. This is irrelevant to the landform or elevation of the parks.

Although our study was conducted in Guizhou and our objective landform was in the Karst regions, the methodology to detect perceptions about ESV using facial expression scores was not limited to the area of this study. Our study provided new data about facial expression scores with experiences of ESV in different locations of parks mostly distributed in remote rural areas. This methodology was used in Jiangxi for wetland parks [19] and Guangdong to detect degraded forest ecosystem services [33]. However, more studies employing this methodology have appeared to detect ecosystem services in urban parks of built-up regions of cities [52,59,61,77]. These together suggest a probability that our findings can be verified in a wider geographical range, including urban areas in Karst regions and more types of landforms.

4.3. Limits of This Study

The data pool used in this study included 3282 photos of visitors in 61 parks of Guizhou, which means there were about 54 photos per park or 54 technical replicates of emotional data. The general number of visitors per park in similar studies was about 20 individuals per park. Hence, the power of our data pool was more reliable than that in most other studies. However, there were only 10–15 visitors’ photos found in five parks, much lower than the expected number of 20. Although these photos accounted for 2% of the number of the total, this still resulted in a risk of the uneven distribution of the data pool. Further work should screen parks to eliminate this issue. In addition, it is hard to fully indicate one’s emotions just according to his/her instant gesture in a photo. Further work is required to conduct a crossover bioassay to verify the extent to which volunteers can show emotions in accordance with their true perceptions.

Secondly, our study only tested one-year data, which failed to reflect any time-dependent variations in ESV. This may cause biases in landform changes across years for some special parks. The time-dependent variable can also reflect population emotions subjected to yearly changes. These were not covered in our study but suggested to be considered in the future.

Finally, we did not consider landscape differences among wetlands, mangroves, or rivers. This may result in the underestimation of ESV in blue spaces, as some small-area but special types of blue spaces may have been excluded from our data pool.

5. Conclusions

This study estimated that the average ESV of green spaces and blue spaces per park in Guizhou province is USD 6.4052 million per year and USD 3.4816 million per year, respectively. The overall spatial distribution of ESV in the blue–green space of the park is low in the central region, with high values concentrated in the surrounding areas, and there is no significant linear change in ESV on the geographical gradient. In green spaces, the ESV performance of different landscape types is as follows: coniferous and broad-leaved mixed forests > grasslands > shrubs. In addition, this study found that the landscape structure and ESV of the blue–green space in the park can be perceived by tourists and expressed through facial emotions. GreenV and vegetation height were considered to be the main driving factors for arousing positive emotions among visitors, while the hydrological regulation function of blue–green spaces may inhibit park visitors from smiling but promote people’s inner peaceful perceptions. Taking Guizhou as an example, future work should be conducted to construct large green spaces with more tall plants so as to induce positive sentiments. However, the theropencedrymion area can be encouraged if not too many indifferent emotions are expected. Overall, visitors can perceive ESV in the park and receive more smiles when experiencing blue–green spaces. Our methodology assessing emotional scores to detect perceptions towards ESV in local GBSs can refer to other Karst regions and other land types. Our findings regarding the relationships between ESVs and facial expression scores need to be verified in more types of lands across the world.