Optimizing Urban Forest Landscape for Better Perceptions of Positive Emotions

Abstract

Interacting with urban spaces that are green and blue is believed to promote mental well-being and positive emotions. Therefore, there is an incentive to strategically design urban forest landscapes in a given space to evoke more positive emotion. In this study, we conducted a pilot study in Northeast China with 24 parks from 11 cities across 3 provinces. The subjects of the study are the visitors and a total of 1145 photos and selfies were collected from open micro-twitters in Sino Weibo (~50 individuals per park). Facial expressions of happy and sad emotions were recognized and rated as percent scores by FireFACE v1.0. Demographically, male adolescents smiled more than male visitors in other age groups and female teens. Females expressed more positive emotions than males according to their positive response index (PRI; happy-sad). Multivariate linear regression indicated positive contribution of green space to happy scores (estimate of 0.0040) and a stronger negative contribution of blue area to sad scores (estimate of −0.1392). Therefore, an urban forest landscape can be optimized by mapping green- and blue-spaces to predict spatial distributions of positive emotions. Male teens are recommended more as frequent visitors than people in other age ranges.

1. Introduction

Mental health has been a global public health challenge [1]. There is a wide consensus that interacting with nature can have immense positive effects on mental health [2]. Research shows that people who are more connected with nature feel positive emotions more frequently [3]. Natural environments are made up of green and blue spaces, and both play a direct or indirect role in health and well-being [4]. Green space includes all types of planted landscape surfaces [5]. Blue space can be an individual or a group of freshwater ecosystems such as sea, watershed, river, lake, and coast [6]. Both green and blue spaces improve emotional perception [6,7,8,9]. Resident environments usually contain both green and blue spaces. However, the effects of interacting with a landscape with both spaces on emotional perception is not well known.

Urban green and blue spaces are useful public places that promote health and well-being [10]. Urban forests are the main type of urban green spaces, and are widely perceived as a nature-based solution that builds on tree-based urban ecosystems to address multiple challenges [11]. The landscape structure (area and patchiness) of urban forests is important for well-being since it is associated with the occurrence of public events that impair health [12,13]. People may have a perception towards the changes in size and type of an urban forest park, which would further influence the responses of human health and well-being in association with this perception [14]. A sustainable city needs strategies to optimize urban forests’ landscape structure for maintaining ecosystem services for human health and well-being [15]. Adapting practical plans, however, is difficult due to the insufficient knowledge to support optimizing urban forest structures for emotional perception.

Landscape metrics are a practical instrument to assess urban forest structure [16]. Landforms are framed by a structure with horizontal and vertical metrics which are characterized by area and height, respectively. Spatial metrics can be defined by eight categories wherein land area accounts for two classes with involvements in at least 40% of all metric variables [17]. The size of an area of green land has a strong impact on the perception of an experience in nature, even as early as childhood [18]. A large urban forest park is a common requirement by adults [19]. For them, a large and open field can help reduce mental stress [20]. Height of aerial objects on landscape surfaces is the major metric that accounts for visual perception. In urban forest landscapes with continuous spreading of canopies, adults tended to prefer tree views with expanded canopy but a short trunk [21,22]. However, children like tall but thin canopies [23]. As urban forests increase, the number of people that interact with green spaces will also increase. Therefore, emotional perceptions toward different green space landscapes will become more diverse in accordance with varied demographics.

Studies on the emotional benefits of green spaces mostly used self-reported scores on surveys, which are typically limited to a small geographical scale, intrusive, time-consuming, and labor intensive [24]. Current data for analyzing emotional perception also mainly derive from self-reported scores [25,26]. However, queries suffer due to subjective bias, emotion-restrained habit, fatigue error, and lack of validation [27,28,29]. Scholars of recent studies still continue using this methodology, wherein results are still limited by geographical scale and number of subjects [30,31]. Instead, facial expressions are a more direct presentation of perceived emotion by a green space user who experienced environmental variation [32]. New frontiers of forest landscape have been established by employing facial expressions as a new metric to assess emotional responses to different green spaces at local [28,29,32,33], regional [27,34], and national scales [35]. These together suggest a potential to continue using facial expression scores as a meter to gauge emotional responses to varied landscape structures. Facial data will be useful as a dependent variable reference for optimizing landscape structures at regional scales. To the best of our knowledge, however, using this innovative approach in urban green space planning has been rare.

In this study, we assessed the emotional responses of visitors to urban forest parks in Northeast China. Facial expressions were collected to build a dataset for evaluating emotional perception. Spatial landscape metrics were quantified using geographic information system (GIS)-based calculation to facilitate estimates at regional scale to the highest accuracy. Our objective was to detect relationships between spatial metrics of urban forest parks and emotional responses of visitors to supply statistical bases for optimizing urban forest landscape. We hypothesized that: (i) positive emotions will be perceived by visitors with the same demographic characteristics; (ii) contact with both green and blue spaces benefits positive emotions; and (iii) positive emotions are associated with landscapes with larger areas of green and blue spaces.

2. Materials and Methods

2.1. Study Sites

This study was conducted in urban forest parks of cities in Northeast China. A total of 24 parks were chosen as sites for data collection, which were located in 10 cities across 3 provinces in Northeast China (

Table 1. Coordinates of urban forest parks located in cities of provinces at Northeast China.

Province	City	Park Name	Longitude	Latitude
Heilongjiang	Harbin	Dingxiang park	126°34′	45°45′
Heilongjiang	Harbin	Jinlongshan park	127°25′	45°31′
Heilongjiang	Harbin	Sidalin park	126°37′	45°47′
Heilongjiang	Harbin	Sun-isle park	126°36′	45°48′
Heilongjiang	Mudanjiang	Jingbohu reserve	128°57′	44°04′
Heilongjiang	Mudanjiang	North Hill park	129°36′	44°37′
Heilongjiang	Tsitsihar	Hecheng park	123°57′	47°18′
Heilongjiang	Tsitsihar	Longsha park	123°57′	47°21′
Heilongjiang	Tsitsihar	Zhalong reserve	124°14′	47°12′
Jilin	Changchun	Jingyue Pool park	125°28′	43°47′
Jilin	Changchun	Nanhu park	125°18′	43°51′
Jilin	Changchun	North Lake park	125°22′	43°59′
Jilin	Changchun	Water Culture park	125°21′	43°51′
Jilin	Changchun	World Sculpture park	125°20′	43°49′
Jilin	Jilin	Longtan Mt. park	126°37′	43°53′
Liaoning	Anshan	Qianshan Scenic spot	123°08′	41°01′
Liaoning	Dalian	Baiyu Mt. park	121°15′	38°49′
Liaoning	Dalian	Labor park	121°38′	38°55′
Liaoning	Dalian	Tiger Shore park	121°41′	38°52′
Liaoning	Dandong	Jinjiang Mt. park	124°23′	40°8′
Liaoning	Fushun	Yueya Isle park	123°50′	41°52′
Liaoning	Jinzhou	East Lake Forest park	121°10′	41°6′
Liaoning	Shenyang	Qipan Mt. Scenic spot	123°39′	41°56′
Liaoning	Shenyang	South Lake park	123°25′	41°46′

). The regions of Northeast China include a geographical range across mild temperate and cold temperate climate zones. It has a monsoon climate of medium latitudes with four seasons. Local summer is warm with frequent rainfalls and winter is chilling and dry. Annual precipitation ranges between 300 mm to 1000 mm along a geographical gradient from northwest to northeast. The total land area of Northeast China is 1.62 million square kilometers with a total population of 0.12 billion (data from 2010) [36]. Together, the host cities of the parks in this study accounted for a total area of 0.25 million square kilometers.

2.2. Data Collection

Visitors of parks in the study were chosen as subjects. Their intended facial expressions were sourced from posts on Sina Weibo, a micro-blog platform similar to Twitter, to assess emotional responses. Using photos as sources of facial expressions for evaluating emotional responses to urban forest experiences have been carried out several times [27,34,35,37]. We acknowledge that intended facial expressions do not fully represent spontaneous emotions in response to experiences in an urban forest park. However, we controlled our source of data to prevent any technical errors that would occur when subjects are conscious of being photographed. In addition, we assumed that subjects who took photos and posted them were unaware about the usage of these photos in this study [38]. Ethical statement for human studies has been reviewed and approved by the committee of ethic board of Environment and Resources College of Dalian Nationalities University (ES-ERC-2021-001).

To warrant a basic requirement for the subject number in parks and minimize differences between observations, initial number of subjects per park was demanded to be more than 50 individuals. An available subject is a person who posted his/her selfie(s) or photo(s) to Sina Weibo with a geographical record at one of the urban forest parks in this study [27,35]. A photo needs to meet all of the requirements below to pass screening to be used:

(i)

Only photos of a person with typical characteristics of an oriental face;

(ii)

All sense organs (eyebrows, eyes, nose, mouth, and ears) have to be fully visible;

(iii)

Facial photos cannot be over decorated on makeup or jewelry;

(iv)

The subject of the photo has to be outside according to the background scenery.

Information such as gender and age were manually evaluated by the same technicians throughout the whole process of screening to reduce bias in perception between individuals. Gender was categorized into female and male according to visual perceptions of sexual characteristics. Age was categorized into five ranges: toddler (1–5 years-old), adolescent (6–19 years-old), youth (20–25 years-old), middle-aged (35–50 years-old), and senior (over 60 years-old) [39,40].

2.3. Imagery Processing

Photos with more than one person were cropped to only have one person. Selfies and individual photos were rotated to a horizontal baseline across the cheek and analyzed by FireFACE software (1.0 version). Two basic facial expressions, happy and sad, were recognized and rated by the instrument. These two emotional expressions were used since they have the highest accuracy among all facial expressions [39]. We used a new variable to assess the net presentation of positive emotion (i.e., positive response index (PRI)), that is evaluated as happy score minus sad score [27,29,32,35]. Finally, a total of 1145 photos were derived from Weibo across the year of 2020. Photos passed the screening criteria for facial analysis and obtained recognition with rated score output (

Table 2. Number of gender and age variations for visitors in urban forest parks at Northeast China.

Park Name	Gender		Age Category					Sum
	Female	Male	Toddler	Adolescent	Youth	Middle	Senior
Dingxiang park	46	3	3	0	46	0	0	49
Jinlong Mt. park	22	13	0	4	19	12	0	35
Sidalin park	35	10	0	0	42	2	1	45
Sun-isle park	41	8	4	4	37	4	0	49
Jingbo Pool reserve	39	10	5	0	41	3	0	49
North Hill park	36	11	6	0	40	1	0	47
Hecheng park	44	4	0	0	45	2	1	48
Longsha park	39	11	3	2	43	2	0	50
Zhalong reserve	40	3	0	0	40	0	3	43
Jingyue Pool park	43	3	0	0	41	4	1	46
Nanhu park	43	7	0	0	50	0	0	50
North Lake park	43	5	7	0	37	1	3	48
Water Culture park	49	3	3	0	46	1	2	52
World Sculpture park	40	9	0	0	45	4	0	49
Longtan Mt. park	31	19	0	0	42	6	2	50
Qianshan Scenic spot	34	13	0	0	44	3	0	47
Baiyu Mt. park	27	21	1	4	37	2	4	48
Labor park	48	1	0	1	42	6	0	49
Tiger Shore park	39	11	0	0	46	4	0	50
Jinjiang Mt. park	43	6	0	0	43	6	0	49
Yueya Isle park	40	6	0	0	45	1	0	46
East Lake Forest park	44	5	0	0	48	1	0	49
Qipan Mt. Scenic spot	38	9	0	1	38	6	2	47
South Lake park	47	3	0	0	44	5	1	50
Total	951	194	32	16	1001	76	20	1145

).

2.4. Data of Landscape Metrics

In this study, we employed spatial metrics of area and height of urban forest patches in landscapes. Landsat 8 OLI images at a resolution of 15 m were used as the source of landscape metrics. Landsat images were downloaded from USGS data center [41]. Landsat images with cloud-cover over 10% were excluded to ensure the maximum disclosure of land areas. Landsat 8 imageries were downloaded from a database in two periods of 22 June–14 November 2019 and 10 May–7 August 2020 to meet the requirements of existed imagery data for specific coordinates with cloud-cover lower than 10%. Imageries were firstly calibrated to correct radiometric errors and geometric distortions [42]. The man-computer interactive interpretation was used to outline the geographic ranges of park area, green-land, and blue-land classifications using ArcGIS 10.2 software. The largeness of a park is not equal to the sum of green-land and blue-land areas due to the presence of bare soils and buildings. Satellite layers in Baidu Maps were used as references of geographical range for outlining. All urban forest parks were projected to generate shape files with coordinates of WGS-1984-UTM-Zone-49N, WGS-1984-UTM-Zone-51N, and WGS-1984-UTM-Zone-52N. Land areas for green and blue spaces were given by calculating the geometry of the digital field in the area property of the generated shape file.

Digital elevation model (DEM) data were downloaded as ASTER GDEM 30M data map products from NASA EarthData [43]. DEM maps were fused by outlined shape files with classified green-land and blue-land layers in the same coordinates as mentioned above. Average elevation for a region was calculated using the function of partition statistics of records from all 15 m-spacing grids in ArcGIS 10.2 software. To estimate the height of mapped surface features in green- and blue-spaces, another digital surface model (DSM) data was superimposed onto the DEM layer, again using the same coordinates. DSM data were downloaded as AW3D30 DSM data map products from Japan Aerospace Exploration Agency [44]. The feature height can be calculated as the difference between heights given by the DSM layer (the greatest height on surface) and that by the DEM layer (the ground elevation) [45]. Since we have outlined urban forest parks in landscape patches as green- and blue-lands, the feature height can be the height of a tree, a shrub, a crowd of reeds, or a vegetative plant on a lawn. Impervious surface and constructions were excluded from the Landsat 8 map layer to avoid recognizing the surface feature as a building.

Generally, our parks had an area of green and blue spaces up to 128.1 and 4.5 km², respectively, with the total average area of an intact park at 170.9 km². Elevation ranged between 11.0 m and 516.5 m with an average tree height of 9.8 m.

2.5. Statistical Analysis

All statistics were evaluated using SPSS v23.0 software (IBM, Armonk, NY, USA). Data were log-transformed to enable analysis using general linear models (GLMs), but they were transformed back when means and standard errors were presented. One-way analysis of variance (ANOVA) was used to detect differences in facial expression scores (happy, sad, PRI) among urban forest parks. Another two-way ANOVA was used to detect the interactive effects of gender and age on facial expression scores. These tests will reveal the source of variations in dependent data in accordance with different locations and demographics. When significant effects were indicated by ANOVA (α = 0.05), log-transformed results were compared using Duncan tests (p < 0.05). Simple correlations, such as the relationship between facial expression scores and variables for landscape structure or topography, were detected using a Pearson model. Thereafter, to detect relationships between dependent variables and facial expressions, multivariate linear regression (MLR) was used to estimate the combined contributions of the independent variables of landscape metrics (area and height) and topography (coordinate and elevation) to happy, sad, and PRI scores. A forward selection of estimates was employed at the significant level of 0.05 for entry of data. When significant estimates were predicted by MLR model, optimized facial expression scores were predicted by combined estimates that passed significant level. Spatial distributions were mapped using ArcGIS 10.2 software for landscape metrics that accounted for significant estimates and predicted facial expression scores. The former described the geographical attributes that drive emotional responses, and the latter depicted the optimized spatial distribution of emotions.

3. Results

3.1. Differences of Facial Expressions among Urban Forest Parks

Different facial expressions among urban forest parks are shown in

Table 3. Difference of facial expression scores for visitors in urban forest parks at Northeast China.

Park	Happy Score (%)	SE 1	Comparison	Sad Score (%)	SE	Comparison	PRI 2 (%)	SE	Comparison
Dingxiang park	51.63	33.94	ABCDE 3	7.06	9.41	D	44.57	39.79	ABCD
Jinlong Mt. park	60.69	31.70	A	12.94	14.78	ABCD	47.76	42.71	AB
Sidalin park	43.15	35.06	ABCDEFGHI	14.20	14.95	ABC	28.95	45.27	ABCDEFG
Sun-isle park	41.51	32.37	ABCDEF	8.58	9.34	ABCD	32.93	37.68	ABCDEFG
Jingbo Pool reserve	60.28	36.45	ABCD	11.17	16.96	D	49.11	48.93	A
North Hill park	24.17	28.27	HI	12.18	11.06	AB	11.98	34.70	G
Hecheng park	39.52	30.33	ABCDEFGH	13.33	14.08	AB	26.18	40.33	ABCDEFG
Longsha park	42.91	35.09	ABCDEFGH	9.04	10.06	BCD	33.87	41.18	ABCDEFG
Zhalong reserve	49.84	34.41	ABCDE	11.11	10.59	ABCD	38.73	42.42	ABCDEF
Jingyue Pool park	50.57	34.06	ABCDEF	8.84	12.39	BCD	41.73	42.08	ABCDE
Nanhu park	31.92	31.76	GHI	14.30	13.05	AB	17.62	39.63	EFG
North Lake park	54.43	32.49	ABC	10.41	15.01	BCD	44.02	42.43	ABCD
Water Culture park	34.38	31.37	EFGHI	13.24	13.59	AB	21.14	40.33	CDEFG
World Sculpture park	33.51	32.67	FGHI	18.67	14.84	A	14.84	43.32	FG
Longtan Mt. park	44.37	36.14	BCDEFGHI	11.39	12.41	ABCD	32.98	44.53	ABCDEFG
Qianshan Scenic spot	25.33	29.39	I	15.41	12.07	A	9.92	37.22	G
Baiyu Mt. park	57.37	32.67	AB	12.41	14.87	ABCD	44.96	44.30	ABC
Labor park	47.98	36.55	ABCDEFGH	8.38	9.90	BCD	39.60	43.15	ABCDEF
Tiger Shore park	44.18	34.21	ABCDEFGH	15.52	16.08	AB	28.66	46.17	ABCDEFG
Jinjiang Mt. park	36.31	32.90	DEFGHI	16.75	15.90	AB	19.56	44.43	DEFG
Yueya Isle park	36.72	32.35	BCDEFGHI	13.60	13.42	AB	23.12	40.69	BCDEFG
East Lake Forest park	35.40	31.61	CDEFGHI	11.86	11.21	ABCD	23.54	38.62	BCDEFG
Qipan Mt. Scenic spot	56.13	34.23	ABC	6.68	8.03	CD	49.45	39.99	A
South Lake park	44.26	32.99	ABCDEFGH	9.50	10.96	BCD	34.76	40.68	ABCDEFG
F value 4	3.06			2.62			2.70
p value	<0.0001			<0.0001			<0.0001

Note: ¹ SE, standard error; ² PRI, positive response index; ³ Different capital letters in a column stand for significant difference of scores for a type of facial expressions; ⁴ F and p values are coefficients to evaluate analysis of variance.

. Difference in the location of the park resulted in significant differences in facial expression scores for happy and sad emotions. However, the difference did not follow any geographical patterns along longitude or latitude gradients (Table 1 and Table 3).

Differences in happy scores were not due to the parks’ differing locations (Figure 1A). For example, happy score was higher for visitors in Jinlong Mt. Park, a park in the northern part (45°31′ N, 127°25′ E), compared to that for others either in adjacent northern parts, such as North Hill Park (44°37′ N, 129°36′ E) and Longtan Mt. Park (43°53′ N, 126°37′ E), or in southern parts such as Jinjiang Mt. Park (40°08′ N, 124°23′ E) and Qianshan Scenic Spot (41°01′ N, 123°08′ E) (Table 1 and Table 3). In addition, happy scores for visitors in Baiyu Mt. Park, which is located in the most southern part (38°49′ N, 121°15′ E), were not statistically different compared to those in parks located over 45° latitudes.

Sad scores were alternatively high and low along geographical gradients (Figure 1B). For example, sad scores were higher for visitors in Qianshan Scenic Spot (41°01′ N, 123°08′ E) and World Sculpture Park (43°49′ N, 125°20′ E) compared to that in adjacent parks such as South Lake Park (41°46′ N, 123°25′ E) and Jingyue Pool Park (43°47′ N, 125°28′ E), respectively (Table 1 and Table 3). Parks in northern parts had relatively lower sad scores for local visitors, such as those in the Jingbohu Reserve (44°04′ N, 128°57′ E) and Dingxiang Park (45°45′ N, 126°34′ E). However, their difference from some local parks were not significant, e.g., compared to those in Sun-isle Park (45°48′ N, 126°36′ E) and Zhalong Reserve (47°12′ N, 124°14′ E).

Accordingly, PRI also showed alternatively high and low scores along geographical gradients (Figure 1C). For example, PRI was found to be higher in the northern part of Hecheng Park (47°18′ N, 123°57′ E) and southern part of Qianshan Scenic Spot (41°01′ N, 123°08′ E) (Table 1 and Table 3) compared to that in the Jingbohu Reserve (44°04′ N, 128°57′ E) and in East Lake Forest Park (41°06′ N, 121°10′ E), respectively.

3.2. Differences of Facial Expressions among Demographics

Happy score was not statistically different between different age groups for female visitors, but it was significantly different for male visitors (Figure 2A). Gender and age had an interactive effect on happy scores. Male adolescents had higher happy scores compared to female adolescents and youths and male toddlers, youths, middle-aged, and seniors. Females had higher happy scores (45.64 ± 29.01%) compared to males (32.44 ± 30.74%). Adolescents had higher happy scores (68.95 ± 35.04%) relative to youths (41.26 ± 28.70%).

Gender and age also had a significant effect on sad scores (Figure 2B). Male adolescents had the lowest sad score compared to most other types of visitors except for female toddlers. Male seniors’ sad scores were 123.61% higher than that of female seniors. Females had lower sad scores (11.35 ± 11.00%) than males (14.75 ± 13.05%). Adolescents had lower sad scores (5.08 ± 7.81%) relative to youths (12.09 ± 11.09%) and seniors (18.05 ± 19.10%).

Adolescents and seniors had contrasting PRI between genders (Figure 2C). Compared to female adolescents, male adolescents’ PRI is higher by 299.55%. In contrast, male seniors’ PRI was 90.59% lower compared to female seniors’ PRI. Females had higher PRI (34.29 ± 36.44%) compared to males (17.68 ± 38.88%). Happy scores showed a general decreasing trend of PRI with the increase of age, except adolescents (63.87 ± 40.30%) which was significantly higher than youths (29.17 ± 36.04%).

3.3. Correlation between Facial Expression Scores and Landscape Metrics

Happy scores had a negative relationship with sad scores (

Table 4. Correlations between facial expression scores and landscape metrics for structure and topography in urban forest parks at Northeast China.

	Correlation	Happy 1	Sad 1	Longitude	Latitude	GreenA 2	WaterA 3	ParkA 4	Elevation	FeatH 5
Happy	R	1	−0.511286	−0.02211	0.05369	0.41902 7	0.34927	0.42615	0.13418	0.05388
	P	-	0.0107	0.9183	0.8032	0.0415	0.0944	0.0379	0.5319	0.8025
Sad	R	-	1	−0.03516	−0.17126	0.02447	−0.46039	0.01905	0.09637	0.07808
	P	-	-	0.8705	0.4236	0.9097	0.0236	0.9296	0.6542	0.7169
Longitude	R	-	-	1	0.61595	0.40595	−0.00262	0.39883	0.74154	0.03224
	P	-	-	-	0.0014	0.049	0.9903	0.0535	<0.0001	0.8811
Latitude	R	-	-	-	1	0.1535	0.05897	0.1593	0.39415	0.37988
	P	-	-	-	-	0.4739	0.7843	0.4572	0.0567	0.0671
GreenA	R	-	-	-	-	1	0.04993	0.99927	0.77121	0.10851
	P	-	-	-	-	-	0.8168	<0.0001	<0.0001	0.6138
WaterA	R	-	-	-	-	-	1	0.05894	0.10726	0.26935
	P	-	-	-	-	-	-	0.7844	0.6179	0.2031
ParkA	R	-	-	-	-	-	-	1	0.75911	0.11704
	P	-	-	-	-	-	-	-	<0.0001	0.586
Elevation	R	-	-	-	-	-	-	-	1	0.23982
	P	-	-	-	-	-	-	-	-	0.259
FeatH	R	-	-	-	-	-	-	-	-	1
	P	-	-	-	-	-	-	-	-	-

Note: ¹ happy and sad scores are transformed as logarithm; ² GreenA, green land area; ³ BlueA, aquatic area; ⁴ ParkA, intact area of an urban forest park; ⁵ FeatH, height of surface feature; ⁶ values in white font and black background indicate a significantly negative correlation; ⁷ values in italic font and light gray background indicate a significantly negative correlation.

). Among all landscape metric variables, green spaces and intact parks are positively related to happy scores. In contrast, aquatic spaces had a negative relationship with sad scores. Latitude had no relationship with any other variables. However, longitude had positive relationships with latitude, green land area, and elevation. In addition, green land area also had a relationship with intact park area and elevation. The relationship between intact park area and elevation was also positive.

3.4. Multivariate Linear Regression of Landscape Metrics to Facial Expression Scores

Among all landscape metrics, only green land area was estimated to have a minor positive contribution to happy scores (

Table 5. Multivariate linear regression of landscape metrics to facial expression scores for visitors in urban forest parks in Northeast China.

Dependent	Variable	Parameter	SE 1	Type II Sum of Squares	F Value	p Value
Happy 2	Intercept	5.3449	0.0735	591.6274	5283.91	<0.0001
	Green land area	0.0040	0.0018	0.5466	4.88	0.0379
Sad 2	Intercept	3.6123	0.0788	255.8518	2103.10	<0.0001
	Aquatic area	−0.1392	0.0572	0.7199	5.92	0.0236
PRI 3	Intercept	27.5659	2.4732	13,391.0000	124.23	<0.0001
	Aquatic area	3.7204	1.7062	512.5047	4.75	0.0407
	Green land area	0.1254	0.0556	548.5958	5.09	0.0349

Note: ¹ SE, standard error for estimated parameter; ² happy and sad scores are transformed as logarithm; ³ PRI, positive response index.

). Most curve-fit confidences were concentrated when green land area was less than 30 km² (Figure 3A). In contrast, aquatic area had a strong negative contribution to sad scores. It was more confident to fit curves for sad scores when aquatic area was less than 1.3 km² (Figure 3B). Both green land area and aquatic area were estimated to have strong positive relationships with PRI. According to absolute value estimates, aquatic areas’ negative contribution to sad scores was more powerful compared to green land areas’ positive contribution to happy scores. However, green land area and aquatic area had an even more powerful contribution to PRI than to happy and sad scores (Table 5).

3.5. Optimizing Urban Forest Landscape

Since green land area and aquatic area were estimated to have significant contributions to facial expression scores (Table 5), these two landscape-metrics can be used to optimize landscape for improved emotions. Green land areas were generally greater in northern parts of the study area than in the southern parts (Figure 4A). Spatial interpolation indicated that green land areas tended to be larger in Jinlongshan Park (45°31′ N, 127°25′ E) of Harbin City and the Jingbohu Reserve (44°04′ N, 128°57′ E) of Mudanjiang City at Heilongjiang Province (Table 1). However, aquatic area tended to be larger in the central part of study area (Figure 4B). Aquatic area tended to be greater in Qianshan Scenic Spot (41°01′ N, 123°08′ E) of Anshan City at Liaoning Province and Jingyue Pool Park (43°47′ N, 125°28′ E) of Changchun City at Jilin Province (Table 1). Therefore, it is predicted that experiences in urban forest parks from these regions will benefit emotions by promoting positive effects and countering negative effects.

According to regression models (Table 5), green land area and aquatic area were inputted as independents. Thereafter, predicted facial expression scores can be regressed against landscape metrics and their spatial distribution patterns reflect the optimized proposals.

Optimized proposals are shown in Figure 5. Happy scores were higher in northern parts of the study area which accord with the distribution pattern of green land area (Figure 5A). Regions with higher regressed happy scores had lower sad scores (Figure 5B). In addition, southern parts of the study area facilitated lower sad scores. PRI showed a general increasing trend along the geographical gradient from south to north (Figure 5C).

4. Discussion

In this study, we did not find any gradual changes in facial expression scores along geographic gradients. Thus, the emotions of the people we focused on were not affected by geographical gradient change. It was found that facial expressions reflect emotions that are perceived when experiencing changing environments [32,34]. In our study area, regional climates in forest ecosystems are prone to gradually change along a geographical gradient [46]. To our knowledge, however, there is few evidence that supports that regional climate in Northeast China can drive emotional responses. Liu et al. [34] reported that people’s facial expressions were driven by a daily minimum temperature gradient. The study area therein was much larger than the Northeast China region. In Northeast China, the facial expressions of urban forest experiencers were found to show responses to gradients of time [28] and urbanization levels [27]. None of these variables were considered as independent variables in this study. All of our facial expression scores were driven by variations in demographics and landscape pattern areas. Neither of these variables change along any geographical gradients.

Male adolescents were the happiest group among all male visitors. They also looked happier than female adolescents. Therefore, less sad emotions were expressed by male adolescents compared to other groups of visitors. This is in accordance with findings that happy scores had a negative relationship with sad scores. Photos of unaware pedestrians did not show any differences between male adolescents and other groups of visitors in urban forest parks [29,39]. However, an enormous survey interviewing 200,000 teenagers from 42 European countries also suggested that male adolescents looked happier than female adolescents [47]. The difference in the frequency of happy expressions between males and females may be the result of differing gender characteristics. Girls are more self-conscious so they look more serious than boys [48]. Since our photos were sourced from online posts, adolescents were more aware that they were being photographed than toddlers. Hence, expressions were more expressive on adolescent faces [49]. The decrease in the frequency of happy faces as age increased was found, both in our study for males and in Liu et al. [34]. Compared to adolescents, a smile is associated with a stronger social motivation for adults [49]. When adults are aware that they are being photographed, they tend to be more serious to hide immediate emotions, but young children perceived it as an interesting interaction with others. All these can be attributed to the significant effect of age on emotional perception.

We found that females expressed more positive emotions than males. Females were also found to post happier faces when photos were intended [27,34]. However, photos of people who were unaware showed null [29] or even inverse results [28]. Corroborating our results, gender was also found to impact emotional expressions and females tended to smile more than males [50,51]. The internet is another world where people present an intended version of themselves. Males tend to accentuate objects and achievements, which does not need too many smiles [52,53]. However, females focused more on familial or interpersonal relations, which motivates expressing positive emotions. Facial expressions of people in urban forest parks, who are unaware that they are being photographed, reflect real-time emotions that are not influenced by social presentation and the gender effect vanished. In summary, we do not accept our first hypothesis since emotional presentation through facial expressions varied by both gender and age.

All of our facial expressions can be regressed against landscape metrics of land area, without any significant estimates found for topography (latitude, longitude, and elevation) and feature height. We can accept our second hypothesis since positive emotions were promoted by green land area and negative emotions were reduced by aquatic area. It has been known that interacting with blue space can evoke positive emotions and our results present a further explanation; the larger the blue space area, the less that sad emotions are perceived. The decrease of sad emotions helps to increase happy emotions since happy and sad scores have a negative relationship. The estimated contribution from the size of aquatic lands was greater than that from the size of green lands. Therefore, the positive emotions evoked from interacting with nature was the result of a small promotion of positive emotions by green spaces and a greater reduction of negative emotions by blue spaces. Our results concur with those found in Ireland and Guyana [54,55]. However, our study revealed more details for the comparison of emotional response to these two types of nature. Our results suggest that, in urban forest parks of Northeast China, people may be not so happy when interacting with blue spaces compared to interacting with forests, but there is a strong motivation to hide their negative emotion.

After using significant estimates from areas of green and blue spaces for optimization, we found an increasing, expressed positive emotion gradient along the latitude gradient. However, using our original happy scores and PRI data, no obvious gradient can be mapped. This is due to the fact that people in some southern parks posted positive expressions that had high scores. These parks were not linked with high scores of both happy and sad expressions since their areas had extremely small green and blue spaces. As the difference between happy score and sad score, PRI was lowered by the regression model in these southern parks since predicted happy scores suffered a greater decline (a range of ~17%) than sad scores (~12%). Therefore, we can also accept our third hypothesis.

5. Conclusions

In this study, we established a dataset of a total of 1145 photos from Sina Weibo on the faces of visitors of urban forest parks of Northeast China. Generally, we found that positive emotions being evoked from interacting with nature was associated with landscape metrics of green and blue space areas. The largeness of the green space positively contributed to the expression of happy emotions, but patch size of blue space had a stronger contribution as s suppression of negative emotions. With a purpose to optimize the landscape for expressing more positive emotions, areas of green and blue spaces were used as independent variables for regressing happy and sad scores. Therefore, net emotion expression, calculated by subtracting sad scores from happy scores, can be mapped as an increase along the latitudinal gradient from south to north. Accordingly, we recommend frequently visiting urban forest parks in Heilongjiang Province to perceive more positive emotions. Male teens are more recommended to visit these parks as they have a higher tendency to smile compared to other types of people.

Our study can be referred to for urban greening and landscape planning divisions for designing a landscape of regional urban forest parks with a need to predict potential distribution of emotional response. Future studies can continue our design and expand the scale to a larger scale. More work is also suggested to detect optimizing projects for landscapes in other regions of the world.