Exploring Aesthetic Preference for Agricultural Landscapes in Hangzhou Plain: A Visual Choice Experiment from Two Perspectives

Abstract

The aesthetic value of agricultural landscapes is gaining importance as rural tourism burgeons during urbanization. To ascertain key elements influencing the visual appeal of agricultural landscapes, this research employed a visual choice experiment in Hangzhou Plain during the spring flowering period to assess public preferences for four landscape attributes in ground and aerial perspectives. The mixed logit model was utilized to evaluate the respondents’ average preference, while the latent class logit model helped in identifying distinct preference groups. The research revealed that participants exhibited different preferences between the two perspectives. The diversity within public preferences was highlighted, with respondents favoring oilseed rape-dominated landscapes with a single agricultural land cover proportion in ground perspective while favoring diverse landscapes in aerial perspective. Gender, education level, landscape familiarity, connection to agriculture, and membership in relevant organizations significantly shape individual preferences. These results can help refine multi-objective policy targeting by incorporating aesthetic value perspective in agricultural landscapes.

1. Introduction

Agriculture has expanded from a pure food supply to environmental protection, aesthetics, and education [1,2,3,4]. Agricultural multifunctionality, the ability of agriculture to simultaneously provide a range of ecosystem services, has increasingly become a key element in transforming and modernizing traditional agriculture [5]. The cultural ecosystem services of agricultural landscapes have received increasing attention in recent years [6,7,8,9]. Among these services, the aesthetic service is considered critical [10]. Consequently, enhancing the visual quality of agricultural landscapes is increasingly becoming a primary driver for their development [11]. However, the assessment of aesthetic services is particularly challenging, as human perception is often highly subjective [12]. While landscape aesthetic values can be partially related to biophysical and ecological characteristics of landscapes, such as landscape pattern [13,14,15,16] and biodiversity [17,18], human preferences are crucial for their valuation. Therefore, evaluating human preferences is of significant importance [9,19,20,21,22,23].

While the majority of studies on agricultural landscape preferences have been carried out in European countries [24], studies in the context of China are also of great significance. China, developing based on agriculture, has enacted policies to secure grain supply, such as carrying out rural land consolidation [25] and large-scale construction of high-standard farmland. Despite considerable advancements in high-standard farmland development and land consolidation, these projects often focus on capacity enhancement of food supply, neglecting the aesthetic and ecological aspects of agricultural landscapes [26]. This excessive pursuit of production efficiency has led to the homogenization of agricultural landscapes [27], yet its aesthetic impact remains far from clear. Existing literature focusing on Europe has already yielded contradictory conclusions regarding the visual appeal of such landscapes: some studies indicate that people prefer landscapes with diverse agricultural land cover [9,28] and smaller field sizes [29], while others suggest that the public sometimes favors large-scale, continuous, and striking landscapes, such as expansive rice paddies and the sea of flowers and forests [7,30,31,32]. This disparity highlights the regional diversity in landscape preferences. Therefore, what preferences people hold for farmland landscapes in the context of China is also a question worthy of exploration.

Meanwhile, with the rapid urbanization since China’s reform and opening up policy, the rural population has been flocking to the cities. To adapt to the economic transformation and meet the consumption demands of urban residents, at the end of the twentieth century, villages in the developed coastal regions of China began to develop tertiary industries represented by agricultural picking and sightseeing tours, meeting the leisure and recreational needs of urban residents [33]. In 2017, China put forward the “Rural Revitalization” strategy to harmonize the contradiction between the imbalance of urban and rural development. Leisure agriculture and rural tourism have developed rapidly under the dual role of government and market [34,35]. The aesthetic value of agricultural landscapes has been increasingly emphasized due to its close relationship with the tourism industry [11,19]. Importantly, such value is not inherent in the landscape itself, but emerges from people’s perceptions of and interactions with landscape attributes [36,37]. Public preferences therefore play a central role in the evaluation of agricultural landscape aesthetics.

Methods for assessing landscape preferences can be divided into revealed preference and stated preference. The revealed preference method is typically determined indirectly by observing behavior or analyzing documents, while the stated preference method directly asks participants to evaluate specific landscape scenes or attributes through questionnaires or experiments, revealing their underlying preferences and values [12,23]. In the stated preference method, participants are often shown a series of landscape images. Different images are often rated using a Likert scale or by selecting preferred or rejected options from a set of images in methods of using such images for preference assessment. These stated preferences are then used to infer which landscape attributes are relevant for assessing landscape aesthetic values [38]. However, these methods typically do not adequately take the influence of the interactions of multiple landscape attributes on visual landscape preference into account [39,40]. In contrast, choice experiments (CE), which aim to estimate attribute utility based on individual responses to multiple combinations of decision attributes [41], are more conducive to understanding the benefits of each attribute combination, and are increasingly applied to agricultural landscape planning and management, as well as to the public’s visual preferences for different landscape scenarios under different political administrations [8,9,28,40,42,43,44].

Notably, the visual attributes of agricultural landscapes exhibit dynamic changes with shifts in observation perspectives. It is worth noting that the ground and aerial perspectives differ not only in observation height but also represent two fundamentally distinct modes of landscape perception [45]. Ground-based observers are typically situated within or at the edge of farmland, where their aesthetic experience is multisensory—beyond vision, it integrates wind sounds, bird songs, crop odors, soil moisture, and other environmental inputs. From the perspective of cognitive psychology, this immersive experience emphasizes the observer’s on-site interaction with the landscape [46]. In contrast, the aerial perspective offers a highly visual, detached mode of observation. A physical and psychological distance exists between the observer and the landscape, making aesthetic judgments more dependent on holistic visual composition, texture contrast, and spatial pattern.

While existing studies have primarily focused on ground perspectives [8,9,13,40,43], there is a noticeable oversight in understanding preferences for agricultural landscapes from aerial perspectives. In the meantime, low-altitude tourism as a form of tourism utilizing air transportation and aircraft in low-altitude airspace, is gaining increasing favor among consumers due to its striking visual appeal [47,48]. The growth of rural tourism and leisure agriculture has promoted the rapid development of low-altitude tourism, which provides a new perspective for landscape observation and highlights the research gap. Therefore, comparing the similarities and differences in public preferences in the two perspectives helps elucidate the trade-off mechanism between “multi-sensory experiences” and “pure visual experiences” in landscape aesthetics, thereby providing a theoretical foundation for multi-perspective tourism and landscape planning.

It is important to acknowledge that agricultural landscapes are highly dynamic across seasons [49]. This study focused on the spring flowering period (April), when wheat and oilseed rape created a colorful mosaic of green and yellow. During the rest of the year, the landscape exhibited highly homogeneous features such as bare soil after harvest and uniformly green rice fields. Therefore, the findings are specific to this narrow temporal window. Tourist perceptions and preferences may differ substantially in other seasons, and generalization to the full year requires further seasonal comparisons.

Therefore, this study takes the Hangzhou Plain agricultural landscape in spring flowering period as an example, and empirically analyzes the aesthetic value of the landscape from two perspectives, ground and aerial, through CEs. It aims to address three questions: (1) What are the differences in the public’s aesthetic preferences for landscape attributes under different perspectives? (2) To what extent do preferences vary across individuals within each perspective? (3) How do participants’ socio-cultural characteristics influence these preferences?

2. Materials and Methods

2.1. Study Area

The study was conducted in the plain area in the northeastern part of Hangzhou City (29°11′–30°33′ N, 118°21′–120°30′ E), the capital of Zhejiang Province, China. The specific sampling area is located in farmland in Qianxi Village, Jingshan Town, Yuhang District, featuring over 1.33 km² of contiguous fertile land. Hangzhou encompasses ten districts, two counties, and one county-level city, covering an area of 16,852 km². Hangzhou’s farmland covers 2356 km², accounting for 14% of the city’s total area. The lowland farmland of Hangzhou is characterized by a flat terrain and dense river network, interspersed with villages, rivers, and countless lakes and ponds (Figure 1). Hangzhou’s renowned natural and cultural landscapes attracted approximately 192.80 million domestic tourists and 114.02 million international overnight visitors in 2024 [50]. Notably, Hangzhou’s rural tourism sector has already welcomed over 74.40 million visitors in 2024 [51]. As a central component of the rural landscape, the agricultural landscape plays an essential role in rural tourism.

2.2. Choice Experiment Design

2.2.1. Choice Experiments

The fundamental principle of choice experiments is that rational individuals select the option with the greatest utility from a set of alternatives, each defined by a unique combination of attributes. By varying these attribute combinations and observing responses, it is possible to evaluate the relative strength of preference for different attributes [52].

The choice experiment model has three advantages over multivariate model-based studies. First, it forces participants to make trade-offs by selecting their preferred landscape from a set of images. Second, this model offers a robust quantitative analytical framework for analyzing the factors that influence participants’ choices. Third, it accommodates the variability in individual preferences when evaluating landscape aesthetics [8]. Consequently, within the field of landscape preference research, choice experiments are an effective tool for investigating how landscape attributes contribute to aesthetic value.

2.2.2. Landscape Attribute Table Construction

Landscape attributes refer to the physical characteristics of a specific landscape [53,54]. This study focuses on agricultural landscape attributes that are influenced by agricultural policies, thus attributes relevant to agricultural management were selected for CE. The selection of four agricultural landscape attributes was finalized through multiple consultations with local experts in landscape architecture, environmental science, ecology, and agronomy, as well as validation and confirmation by other stakeholders (

Table 1. Description of the landscape attribute.

Attributes	Level	Description	Representation of Ground Perspective	Representation of Aerial Perspective	Hypothesized Effect on Aesthetic Preference	Source
Proportion of agricultural land cover	0	single cover	Only one crop	Only one crop	/ 1	[8,9,28,53]
	1	medium	Proportion of minor crops less than 30%	Proportion of minor crops less than 30%	+
	2	high	Proportion of minor crops greater than 30% less than 50%	Proportion of minor crops greater than 30% less than 50%	+
Major types of crops	0	wheat	Major crop grown is wheat	Major crop grown is wheat	/	Outline of the 14th Five-Year Plan (2021–2025) for National Economic and Social Development and Vision 2035 of the People’s Republic of China.
	1	oilseed rape	Major crop grown is oilseed rape	Major crop grown is oilseed rape	+
Linear elements	0	low	No visible linear elements	No visible linear elements	/	The National Plan for the Construction of High-standard Farmland (2021–2030); Standard for planning and design of ecological engineering for land consolidation and rehabilitation; [8,9,27,28,40,53,55]
	1	medium	Linear vegetation, including windbreaks and vegetative cover roadside in the background	Linear vegetation in the middle ground	+
	2	high	Linear vegetation, including windbreaks and vegetative cover roadside both in background and foreground	Linear vegetation in the fore-, middle and background	+
Point elements	0	low	No visible point element	No visible point element	/	[8,9,27,28,40,53]
	1	medium	4 shrubs	3 bushes and 1 pond	+
	2	high	6 shrubs, 1 tree	4 bushes and 3 ponds	+

¹ /: Using level 0 as the baseline to hypothesize the impact of levels 1 and 2; +: Lead to higher preference compared with level 0.

). With the emergence of leisure agriculture and aerial tourism, this study examines these agricultural landscape attributes from both ground and aerial perspectives. Due to the shift in perspective, these attributes exhibit different visual representations.

(1)

Proportion of agricultural land cover. This study took the Hangzhou-Jiaxing-Huzhou Plain as the main research area. The Hangjiahu Plain is aesthetically valued for its spring cultivation of both wheat and oilseed rape. During spring, wheat and oilseed rape are cultivated in varying proportions across the same farmland, their contrasting colors and patchy spatial arrangement create a visually diverse mosaic. The more balanced the proportion of the two crops, the higher the chromatic heterogeneity. Thus, the proportion of agricultural land cover is determined by the extent of two different crops cultivation in the farmland. This visual landscape attribute is categorized into three levels: single cover, medium proportion, and high proportion.

(2)

Major types of crops. The Outline of the 14th Five-Year Plan (2021–2025) for National Economic and Social Development, along with the Long-Range Objectives through 2035, advocates improving the arable land fallow system and implementing crop rotation, to strengthen the production of main food crops, while expanding winter oilseed rape cultivation in winter farmlands, and enhancing oilseed supply security. Recognizing the actual planting situation of two crops in the Hangjiahu Plain and the duration of the research, the study primarily selected wheat (Triticum aestivum L.) and oilseed rape (Brassica napus L.) as the main crops, defining large areas planted mainly with oilseed rape (with little or no wheat) as ‘oilseed rape’, and those planted mainly with wheat (with little or no oilseed rape) as ‘wheat’.

(3)

Prevalence of linear elements. The prevalence of linear elements in agricultural landscapes is marked by elements such as ridges, ditches, hedgerows, and field margins [54]. These elements serve as vital habitats and offer ecosystem services, functioning as ecological infrastructure for various species [55]. The presence or absence of vegetation cover along each field margin characterizes this attribute. It is categorized into three levels of prevalence: low, medium, and high.

(4)

Prevalence of point elements. This attribute represents the visual landscape characterized by patches of non-agricultural habitat within farmland, including individual trees or clusters of trees, shrubs, grasses, and ponds. The levels of this visual landscape attribute are categorized as low, medium, and high levels of point elements.

2.2.3. Landscape Images Selection

In 2019, the Zhejiang Provincial Department of Agriculture and Rural Affairs, in cooperation with the Zhejiang Provincial Department of Culture and Tourism, selected 100 “most beautiful countryside” in Zhejiang Province. These farmlands, usually larger than 33.33 ha, have moderate size, density, well-maintained roads and ditches, orderly and pristine crop layouts, and an aesthetically appealing environment.

In this study, base images were all taken on 21 April 2023, when wheat and oilseed rape were in the flowering period, on farmland in Qianxi Village, Jingshan Town, Yuhang District, which was selected as one of Hangzhou’s most beautiful countryside, using Canon 5D4 camera (Canon Inc., Tokyo, Japan) and DJI AIR3S drone (DJI, Shenzhen, China). The base images include both ground and aerial perspectives, and were taken from exactly the same location and angle. Two images from them were ultimately selected for further study. The ground perspective includes villages, flat farmland, and roads, while the aerial perspective shows a dense river network interspersed with farmland and villages, as well as villages, flat farmland, rivers, and roads in the middle and background.

2.2.4. Choice Sets Design

The Ngene software package (version 1.4) was used to create an orthogonal design with 18 choice sets, each with three alternatives. Since studies have shown that longer surveys reduce response rates and may affect the quality of the data collected due to participant fatigue, to maintain the validity of the questionnaire and to avoid participant fatigue, the split questionnaire design (SQD) was employed in this study, dividing the sets into two groups of nine and randomly distributing among the participants [56,57].

Based on the selected sets, a visualization approach was employed to depict the four landscape attributes and their respective levels against a representative background image (Figure 2). Unlike many studies that rely on generic photographs [13,58,59], this study used a typical view of the landscape as a constant base image for all landscapes in the study area. Adobe Photoshop 2026 was used to modify the levels of various landscape attributes on this base image, thereby creating multiple scenarios. This method has the advantage of controlling for other factors that may influence visual preferences, such as weather conditions, photographic perspective, and composition.

2.2.5. Socio-Cultural Explanatory Variables

Previous research has shown that demographic and cultural characteristics significantly influence landscape preferences, with considerable variation among individuals [60]. Common demographic characteristics considered in landscape preference studies include gender, income, age, place of residence, and education level. In particular, participants’ places of residence, their connections to agriculture, and familiarity with agricultural landscapes are important considerations in these studies [53]. Living environment, whether urban or rural, significantly influences landscape preferences [61]. Individuals develop a sense of place and identity through environmental interactions and experiences [62], which influences their agricultural landscape preferences. Education level also strongly influences landscape preferences, with individuals with higher education tending to prefer cultural landscapes [63,64]. In addition, landscape familiarity [44,65] and agricultural connections influence preferences, with those associated with agriculture often favoring less natural and more intensively managed agricultural landscapes [19,65]. Based on the literature reviewed, a standardized set of explanatory variables was established (

Table 2. Classification of participants’ socio-cultural explanatory variables.

Variable	Values
Gender	0: male	1: female
Education level	0: low education (senior high school degree or below)	1: high education (junior college/university degree or above)
Rural/urban place of residence	0: rural	1: urban
Landscape familiarity	0: not familiar	1: familiar
Connection to agriculture	0: not related to farming	1: related to farming
Related organizations	0: no	1: yes

).

2.2.6. Questionnaire Design and Survey

Based on the above considerations, the questionnaire was designed (Appendix A). The questionnaire comprised three sections. Section 1 briefly introduced the research background and explained the task. Section 2 assessed landscape preferences in ground perspective and aerial perspective, respectively. Each question was followed by three images displaying different combinations of landscape attributes, totaling nine questions per section. Participants were not provided with explanations of landscape attributes, ensuring their decisions were based solely on visual perception. Section 3 collected personal information (Table 2). A pre-test was conducted to ensure the comprehensibility, readability and authenticity of the visualization after the questionnaire was designed. Authenticity assessment of the images yielded an average score of 4.27/5 from respondents, indicating that these processed images are deemed capable of accurately reproducing authentic agricultural landscapes.

Data collection was conducted in January 2025 using the online sampling service “SoJump” [66]. As rural tourism in Hangzhou primarily attracts visitors from nearby areas and Zhejiang residents constitute a key source market, the demographic characteristics of the study sample are considered consistent with those of Zhejiang Province residents. To ensure representativeness, the sample demographics were matched with data from the “Zhejiang Provincial Statistical Yearbook” [67]. Of the 233 questionnaires received, 223 were considered valid after excluding incomplete responses or questionnaires from outside Zhejiang Province. To assess the adequacy of sample size, we conducted a statistical power analysis. With sample of 223 respondents, the study demonstrated an 80% power at the α = 0.05 level to detect moderate to significant preference differences (z ≥ 2.8), and over 60% power for smaller but statistically significant effects (z ≥ 2.0) [68]. This indicated that the sample size possessed sufficient statistical power so that genuine significant preference patterns across different landscape attributes and viewing perspectives could be identified.

Given the primarily online format of the survey, there was a slight overrepresentation of participants aged 35–59 and those with higher education levels compared to the general demographics of Zhejiang Province. Also, our sample may underrepresent non-internet users and thus carry self-selection bias. Nonetheless, post-quota adjustments ensured that the sample’s gender, age, education level, and place of residence demographics closely mirrored those of Zhejiang Province (Appendix B).

2.3. Model Specification and Statistical Analysis

To evaluate participants’ preferences for agricultural landscape attributes, this study adopted a mixed logit model (MLM). MLM, as opposed to the more basic multinomial logit (MNL) model, more realistically assumes that utility parameters vary among individual participants [69]. In our MLM, each attribute level was analyzed using dummy coding, with the lowest level of all attributes serving as the reference category in the model.

Given the large number of attribute levels involved in the study, setting all attribute coefficients to random distributions would lead to convergence difficulties and parameter instability during model estimation. To address this, the study adopts a two-stage approach: only assign coefficients of key attribute levels to normally distributed random parameters while ensuring model convergence, and retaining fixed parameters for other attributes. This approach achieves robust model performance while effectively capturing participants’ preference differences for key attributes.

The degree of consensus among participants and the variability in their preferences were analyzed using a latent class logit model (LCM) approach. This model is based on the principle that observed responses can be classified into distinct homogenous preference groups, called classes. It posits that heterogeneity among participants can be effectively represented within a finite number of these classes [70]. The LCM is an extension of the MNL model that focuses primarily on incorporating latent classes. The LCM efficiently classifies individuals into a limited number of groups characterized by strikingly similar β-vectors. Within this model, the probability that a given individual n, who was a member of class q, chose alternative i from a given choice set t was specified by:

$$P_{ni}|\left(class=q\right)={\displaystyle \frac{\exp \left({x^{\prime }}_{nit}{\beta }_q\right)}{{{\displaystyle \sum }}_{j=1}^j\exp \left({x^{\prime }}_{nit}{\beta }_q\right)}}$$

(1)

It is important to note that this probability depends on the individual’s membership in class q. The assignment of individuals to a specific class was probabilistic and was determined in the model based on their choice of behavior. Parameters for both the class selection model and the choice probability model were estimated using a standard maximum likelihood procedure [71].

The probability of class membership of an individual n was modeled based on its specific characteristics, denoted as ${Z^{\prime }}_n$, and the relevant class conditional parameter vector ${\gamma }_q$.

$$H_{nq}={\displaystyle \frac{\exp \left({Z^{\prime }}_n{\gamma }_q\right)}{{{\displaystyle \sum }}_{q=1}^Q\exp \left({Z^{\prime }}_n{\gamma }_q\right)}}$$

(2)

Combining the class membership probabilities with the probabilities from the choice model yielded the aggregate log-likelihood function for an individual n.

$$S_n={\displaystyle {\displaystyle \sum }_{n=1}^N}\ln \left\{{\displaystyle {\displaystyle \sum }_{q=1}^Q}{H}_{nq}{\displaystyle {\displaystyle \prod }_{t=1}^T}{\displaystyle {\displaystyle \prod }_{j=1}^J}{\left[{\displaystyle \frac{\exp \left({x^{\prime }}_{njt}{\beta }_q\right)}{{{\displaystyle \sum }}_{j=1}^J\exp \left({x^{\prime }}_{njt}{\beta }_q\right)}}\right]}^{{\gamma }_{njt}}\right\}$$

(3)

where Q represents the number of classes, T denotes the number of choice sets presented to an individual, and J signifies the number of choices within each choice set. The variable ${\gamma }_{njt}$ was set to 1 if individual n selected the j-th alternative in the t-th choice set, and it was 0 in all other cases. To determine the number of classes for the latent class model, the Bayesian Information Criterion (BIC) is employed because of its widespread use in latent class modeling.

In our final analysis, the relationship between socio-cultural explanatory variables and landscape attributes was examined by interacting dummy-coded socio-cultural variables with landscape attributes, which were then estimated using the MNL model. All model estimations for this study were performed using Stata software (version 18.0).

3. Results

3.1. Preference for Landscape Attributes

The estimation results of MLM showed that the participants exhibited higher preferences for most landscape attributes, with the notable exceptions being the medium and high proportions of agricultural land cover and linear elements in the ground perspective, as well as the medium proportion of agricultural land cover and the high level of point elements in the aerial perspective, which received lower evaluations (

Table 3. Mixed logit estimations for ground perspective and aerial perspective; higher coefficients correspond to higher preference.

Attributes	Level	Ground Perspective		Aerial Perspective
		Coeff. (p-Value)	Coeff. Std. (p-Value)	Coeff. (p-Value)	Coeff. Std. (p-Value)
Proportion of agricultural land cover	medium	−0.120 *	0.0696	−0.0413	0.0765
	high	−0.178 *	0.0912	0.0731	0.0977
Major types of crops	oilseed rape	0.136 **	0.0612	0.172 **	0.0691
Linear elements	medium	−0.0293	0.0749	0.0205	0.0850
	high	−0.236 ***	0.0768	0.0770	0.0936
Point elements	medium	0.0734	0.0740	0.0590	0.0783
	high	0.0317	0.0703	−0.303 *	0.155

*** p-value < 0.01; ** p-value < 0.05; * p-value < 0.1.

). Other estimated coefficients were statistically significant, except for the medium level of linear elements and point elements in the ground perspective, as well as the proportion of agricultural land cover, linear elements and medium level of point elements in the aerial perspective, which did not meet the critical significance level of 0.1.

Participants exhibited similar partial preferences in both perspectives, as depicted in Figure 3. The highest preference was for oilseed rape as the major crop, with more positive attitudes towards medium level of point elements, and more negative attitudes towards the proportion of agricultural land cover. The preferences also revealed heterogeneity between the perspectives: a lower preference for linear elements in the ground perspective and a higher, albeit non-significant, preference for high level of linear elements in the aerial perspective. Additionally, participants in the ground perspective showed the lowest preference for a high level of linear elements, while those in the aerial perspective displayed the lowest preference for a high level of point elements. In general, the utility coefficients for the landscape attributes were relatively small, with the exception of the value of −0.303 for the high level of point elements in the aerial perspective.

To assess the robustness of the mixed logit estimates, we also estimated a conditional logit model using the same data. The results reported the same direction for all attribute levels as the mixed logit model (Appendix C.1). Both models had BIC values of 6884.518 in ground perspective and 6885.853 in aerial perspective, indicating comparable model fit quality. This consistency supported the robustness of our main findings.

3.2. Analysis of the Extent of Agreement Among Participants

This study utilized the LCM to assess the level of agreement among participants. The BIC was used to determine the model class. According to the results shown in Appendix C.2 and Appendix C.3, though the LCM for the ground perspective did not have the lowest BIC value with three model classes, some categories exhibited extremely low participation rates with more model classes, indicating weak representativeness and structural instability. To prevent over-segmentation as well as enhancing interpretability of the model, the number of classes for the ground perspective was determined to be three, and the number for the aerial perspective was determined four.

Participants were divided into three potential categories in the ground perspective, with which in the two largest classes (Classes 2 and 3) exhibiting contrasting preferences (

Table 4. Coefficients of the latent class models for ground view and aerial view.

Ground Perspective		Class1		Class2		Class3		-
Latent class probability		0.109 ***		0.297 ***		0.594 ***		-
Attributes	level	Coefficient	Standard error	Coefficient	Standard error	Coefficient	Standard error	-	-
Proportion of agricultural land cover	medium	−0.419	0.577	−0.133	0.0841	0.165	0.321	-	-
	high	−0.805	0.646	−0.203 **	0.0928	0.214	0.369	-	-
Major types of crops	oilseed rape	2.100 ***	0.702	0.230 ***	0.0837	−1.080 ***	0.346	-	-
Linear elements	medium	−0.137	0.531	−0.0428	0.0947	−0.165	0.433	-	-
	high	−27.561	124.728	−0.0846	0.0936	−0.950 ***	0.364	-	-
Point elements	medium	0.310	0.725	0.193 **	0.0866	−1.071 **	0.478	-	-
	high	1.164 *	0.629	0.147 *	0.0765	−1.256 *	0.704	-	-
Aerial perspective		Class1		Class2		Class3		Class4
Latent class probability		0.102 ***		0.098 ***		0.585 ***		0.215 ***
Attributes	level	Coefficient	Standard error	Coefficient	Standard error	Coefficient	Standard error	Coefficient	Standard error
Proportion of agricultural land cover	medium	−0.154	0.230	1.654 *	0.910	−13.870	94.423	0.301	0.269
	high	−0.250	0.372	2.409 **	1.192	0.811	0.831	0.606	0.395
Major types of crops	oilseed rape	0.315 **	0.125	−0.551	0.400	−1.032	0.929	−0.104	0.405
Linear elements	medium	0.240	0.188	−2.181 ***	0.810	0.395	1.000	−0.309	0.723
	high	0.271 *	0.157	0.959 **	0.380	14.940	94.420	−0.636	0.827
Point elements	medium	−0.0224	0.152	1.829 **	0.799	−13.118	94.420	0.188	0.254
	high	−0.208 *	0.110	1.284 **	0.608	1.035	1.063	−0.239	0.268

*** p-value < 0.01; ** p-value < 0.05; * p-value < 0.1.

), as visually depicted in Figure 4. Class 3, comprising 59.4% of the participants, showed a statistically significant aversion to landscapes dominated by oilseed rape, point elements and high level of linear elements in the ground perspective. Utility coefficients for landscape attributes lacking statistical significance suggested an indifference towards the landscape attributes. Conversely, Class 2, representing 29.7% of the participants, showed a more positive inclination towards landscapes dominated by oilseed rape and point elements. The remaining class, Class 1, representing 10.9% of the participants, also showed stronger preference for rapeseed-dominated landscape and high level of point elements.

In the aerial perspective, the two largest groups (Classes 3 and 4) of participants demonstrated contrasting preferences for proportion of agricultural land cover, linear elements and point elements. However, the insignificant utility coefficients of the landscape attributes meant that their preferences for the four landscape attributes were not obvious. The two smaller categories (Classes 1 and 2) showed more distinct preferences contrarily, as depicted in Figure 4. Class 2, with 9.8% of participants, showed preferences for point elements and high level of linear elements, while showing lower preferences for medium level of linear elements. Conversely, Class 1, comprising 10.2% of the participants in the aerial perspective, showed no clear preference for proportion of agricultural land cover, while exhibiting strong fondness for positively evaluated landscapes dominated by oilseed rape. High level of linear elements also received higher evaluations, while high level of point elements received lower.

3.3. Effects of Socio-Cultural Explanatory Variables on Preferences

A forward stepwise MNL with interactions was used to derive the model results (

Table 5. The effects of individual socio-cultural explanatory variables on preferences for landscape attributes.

Socio-Cultural Explanatory Variables	Attribute	Level	Model Coefficient (Ground Perspective)	Model Coefficient (Aerial Perspective)
Gender	Proportion of agricultural land cover	medium	n.s. 1	n.s.
		high	−0.126 *	n.s.
	Major types of crops	oilseed rape	−0.152 **	n.s.
	Linear elements	medium	n.s.	−0.217 **
		high	n.s.	n.s.
	Point elements	medium	n.s.	0.144 *
		high	n.s.	n.s.
Education level	Proportion of agricultural land cover	medium	n.s.	n.s.
		high	n.s.	n.s.
	Major types of crops	oilseed rape	0.173 **	0.160 **
	Linear elements	medium	n.s.	n.s.
		high	−0.271 *	n.s.
	Point elements	medium	n.s.	n.s.
		high	n.s.	−0.407 ***
Frequency of visits	Proportion of agricultural land cover	medium	n.s.	n.s.
		high	n.s.	n.s.
	Major types of crops	oilseed rape	n.s.	0.208 *
	Linear elements	medium	−0.128 *	−0.175 **
		high	0.203 **	n.s.
	Point elements	medium	n.s.	n.s.
		high	n.s.	n.s.
Connection to agriculture	Proportion of agricultural land cover	medium	n.s.	n.s.
		high	n.s.	n.s.
	Major types of crops	oilseed rape	0.120 *	0.0974 *
	Linear elements	medium	n.s.	0.208 *
		high	n.s.	n.s.
	Point elements	medium	0.176 **	n.s.
		high	n.s.	n.s.
Related organizations	Proportion of agricultural land cover	medium	n.s.	−0.184 **
		high	0.175 **	n.s.
	Major types of crops	oilseed rape	n.s.	n.s.
	Linear elements	medium	n.s.	−0.165 *
		high	0.173 *	0.256 ***
	Point elements	medium	n.s.	−0.186 **
		high	n.s.	n.s.

¹ n.s. = non-significant interaction that was excluded from the model. *** p-value < 0.01; ** p-value < 0.05; * p-value < 0.1.

), with nonsignificant interactions removed. The table indicated that socio-cultural explanatory variables had different effects on landscape attribute preference.

In contrast to women, men showed a lower preference for high proportion of agricultural land cover and landscapes dominated by oilseed rape in the ground perspective. In the aerial perspective, they demonstrated lower preference for medium level of linear elements and higher preference for medium level of point elements.

Educational level also predominantly influences landscape attribute preference in both perspectives. Participants with higher education showed preference for the landscapes dominated by oilseed rape in both perspectives. Contrarily, they showed opposite preferences for high level of linear elements in the ground perspective and high level of point elements in aerial perspectives.

Compared to participants with less connection to agriculture, those with closer connection showed a stronger preference for oilseed rape as the major crop. In the ground perspective, participants familiar with agricultural landscapes showed a greater preference for medium level of point elements, and in the aerial perspective, they preferred medium level of linear elements. Similarly, the group of participants with non-agricultural backgrounds who visited agricultural landscapes more frequently showed stronger preference for high level of linear elements in ground perspective and oilseed rape as the major crop in aerial perspective.

Participants who were affiliated with agricultural and environmental organizations showed a stronger preference for high level of linear elements in both the ground and aerial perspectives, while showing less preferences for medium levels of linear and point elements in aerial perspective (Table 5). Additionally, these members showed stronger preference for a high proportion of agricultural land cover in the ground perspective.

4. Discussion

4.1. Landscape Attribute Preferences

The results of the MLM revealed that participants demonstrated modest negative preferences toward the proportion of agricultural land use in ground perspective. Previous studies have also demonstrated markedly divergent outcomes across different regions. This appeared to align with the conclusions of those studies indicating that people prefer large-scale, continuous, and striking landscapes [7,30,31,32]. Since those studies reached opposite conclusions [9,28,29], it also highlighted the geographical variability of landscape preferences.

An important finding of this study is that participants exhibited markedly different preferences in aerial perspective compared to ground perspective, showing a stronger preference for diverse landscapes with high proportion of agricultural land cover. Such difference may stem from variations in public emotional experiences from different perspectives. As previously mentioned, the ground perspective offers an immersive experience that places visitors right at the edge of a farmland, experiencing a range of sensations including auditory, olfactory, and thermal perceptions. In such settings, the sense of order, security, and familiarity in the landscape becomes particularly crucial—hence the public generally favors views with expansive, clearly visible landscapes [72,73]. In contrast, aerial perspective presents a top-down panoramic view, where the participants observe the farmland from a detached vantage point, with sensory input highly simplified to color, texture, shape, and spatial distribution. In such settings, complex and diverse topographical features and colors seem to evoke greater visual pleasure [74]. However, results in this study derived from static photographs cannot directly provide multisensory inputs of real-world agricultural landscapes, and thus may differ from those obtained in actual environmental settings.

In both the ground and aerial perspectives, participants showed strong preferences for agricultural landscapes where oilseed rape is the primary crop, suggesting that bright colors are more appealing. This finding is consistent with previous research [7,8,75,76,77].

Meanwhile, our results provide no statistically significant evidence that point elements positively influence aesthetic preferences in ground perspective. In aerial perspective, high level of point elements received a significantly negative coefficient, while the medium level was not significant. This suggests that excessive point elements may be perceived as clutter from above, rather than being visually attractive. Similarly, studies have suggested that in terms of agricultural production, point elements in landscapes could impede mechanized operations, thereby affecting productivity. From a visual perspective, elements such as trees or ponds could also disrupt the uniformity of farmland [32]. However, other studies have indicated that point elements significantly improve the visual quality of agricultural landscapes [28]. Whether point elements are viewed positively [9,28,40] or negatively [32], they have a significant impact on agricultural landscapes. Nevertheless, there is currently a research gap on point elements in agricultural landscapes [53], which highlights the need for further research in this area.

Furthermore, participants exhibited negative reactions to linear elements in ground perspective, indicating a preference for concrete ditches over ecological ditches. Considering agricultural production, ecological ditches and ridges tend to increase maintenance costs for farmers compared to their concrete counterparts. In terms of visual landscape aesthetics, concrete ditches and ridges without vegetation present a more orderly appearance. These factors may underlie the prevalent negative preference for linear elements of high-level field margin. In line with this, multiple studies have emphasized that windbreak forests can enhance the visual diversity and attractiveness of agricultural landscapes [27,78,79,80,81,82,83].

Existing research indicates that seasonality influences factors such as color and shape in agricultural landscape features [8,84]. This study focused primarily on landscapes with spring oilseed rape and wheat rotations. Future research should examine seasonal variation in agricultural landscapes to more accurately assess preferences for agricultural landscape features in landscape image-based survey studies.

4.2. Homogeneity and Heterogeneity of Landscape Preferences

Since MLM’s reliance on average utility coefficients, which tend to marginalize the preferences of specific demographic segments, LCM approaches categorize preferences in a way that recognizes the diversity among different groups, thereby providing more detailed insights for policy formulation. Our study delineated primary representative categories under both methodologies, focusing on distinct category characteristics. Interestingly, these populations exhibited contrasting preferences.

The preferences of class 2 in ground perspective which comprised 29.7% of participants and class 1 in aerial perspective comprising 10.2% of participants indicated a preference for large and intensive agricultural landscapes, as they demonstrated negative preferences for most landscape attributes. More specifically, they prefer uniformity, scale, and openness in agricultural landscapes, aligning with the aesthetic factor of “vastness” highlighted in landscape preference studies [85,86].

Meanwhile, different preferences for landscape attributes were observed between class 3 (59.4%) in ground perspective and class 2 (9.8%) in aerial perspective. The utility coefficient analysis of class 3 in ground perspective showed a preference for open yet colorful landscapes. Conversely, class 2 in aerial perspective demonstrated a positive preference for the combination of high agricultural land cover, medium levels of point elements, and medium to high levels of linear elements. This aligns with the findings of several previous studies on agricultural landscape preferences that certain combination can enhance landscape attractiveness [9,28,40,44,59]. Notably, preferences of these participants for high-level linear elements were significantly higher than those for medium-level elements, indicating a stronger inclination toward landscape features such as windbreak forests and ecological ditches with unified foreground and background.

The LCM results indicated no significant preference among the classes 3 and 4 participants in the aerial perspective. One possible explanation is that the plethora of alternatives in the experimental design may have resulted in reduced participant attention during the survey. Future studies should consider adopting an efficient design to rationalize the number of alternatives [52]. The inclusion of low-quality questionnaires, possibly due to insufficiently rigorous screening criteria, may have contributed to this phenomenon.

Landscape perceptions exhibit considerable variation between individuals and social groups, which has been confirmed by previous LCM research in agricultural landscapes [8,28,40,42]. These findings highlight the importance of considering the diverse needs of different populations in landscape management, particularly in the context of cultural ecosystem services (CES) in agricultural settings.

4.3. Influence of Socio-Cultural Characteristics

Respondents possessing higher education levels exhibited a marked disfavor towards several landscape attributes except for the type of major crop. One possible explanation related to a bias in higher education favoring agricultural scale and modernization, leading to a preference for large-scale farmland and neatly constructed concrete ditches and roads [87]. Research indicated that exposure to natural landscapes can reduce stress, enhance mood [88], and improve long-term health and well-being relative to urban environments [89]. Our study revealed that urban residents enjoying visiting agricultural landscapes had increased preferences for certain landscape attributes. This trend may be due to urbanites’ limited daily exposure to natural environments, leading to an increased desire for more natural agricultural landscapes [8,90].

Those with ties to agriculture displayed more preference for medium levels of linear and point elements in both perspectives. One possible explanation is that familiarity with specific landscapes during childhood shapes landscape preferences [62]. Additionally, individuals associated with agriculture showed a preference for landscapes dominated by oilseed rape in both perspectives, possibly due to its visually appealing bright colors [8] and higher economic value as an oilseed crop [91,92].

Similarly, individuals associated with agricultural and environmental organizations displayed specific preferences. With reference to results that were not statistically significant, it was speculated that they have a positive preference for higher levels of these attributes despite the aversion to the medium levels of these elements. In stark contrast, previous studies have found that people closely connected to agriculture prefer fewer natural landscapes [65], more intensively managed landscapes [63], or high-intensity land uses [19], regarding excessive linear and point elements as hindrances to mechanized operations. Participants in this study exhibited a significant preference for diverse landscapes, probably indicating that they place greater emphasis on the positive effects derived from elements such as crop rotation management, ecological ditches, and windbreak forests.

4.4. Relevance for Planning and Policy Making

To enhance the efficiency of agricultural land use and increase agricultural intensification and mechanization, the Chinese government has implemented land consolidation and high-standard farmland construction. Concurrently, to ensure oilseed supply, there has been a government push for oilseed and main food crop rotation. The MLM results suggest public satisfaction with landscapes shaped by these policies. First, public preference for landscapes with oilseed rape, observed in both perspectives, indicates that oil-main food crop rotation may be further promoted. Meanwhile, agricultural tourism planning may prioritize oilseed rape landscapes as the main competitive feature. Second, in the aerial perspective, there is a marked preference for linear elements with a medium-level of field margins, suggesting that the construction of windbreaks could be promoted to increase coverage. Third, given the public’s slight preference for unified, large-scale, and open agricultural landscapes in the ground perspective, agritourism planning could prioritize the preservation of visual openness along ground tourist routes. However, the small utility coefficients indicated that the formulation of policy recommendations requires further careful consideration. Future research with larger sample sizes and field validation is needed before such measures are implemented.

However, it is not sufficient to consider policy solely in terms of aesthetics. In recent decades, intensified agriculture, marked by escalated pesticide use and large-scale mechanization, has led to increased food production and global environmental degradation affecting climate, land, water, and biodiversity [93,94]. To mitigate the negative environmental impacts of agriculture, many countries have implemented policies, most notably the Agricultural Environmental Scheme (AES) in Europe, which has shown significant effectiveness and is relevant to China. Research shows that field margins contribute to natural pest control [95], pollination [96], water purification [97], soil and water conservation [83], and carbon sequestration [98]. However, there is a dearth of studies on cultural services [99,100]. Results in this study indicated that high levels of linear elements may fail to effectively attract public interest in ground perspective, but may have high aesthetic value in aerial perspective. Consequently, China can consider adopting elements of the European AES, especially in field margin management, to enhance the aesthetic value of linear elements and the multiple ecosystem services of farmland. Similarly, research has demonstrated that agricultural diversification enhances a variety of ecosystem services without affecting yields [101]. In this study, class 3 (ground perspective) and class 2 (aerial perspective) exhibited a preference for a medium to high proportion of agricultural land cover. Consequently, it may be beneficial to increase agricultural land cover in selected areas during crop rotation to promote landscape heterogeneity and amplify multiple ecosystem services.

It is noteworthy that the implementation of landscape aesthetic management ultimately rests on farmers’ decisions. Crop diversification, preservation of linear or point-like landscape features, and adjustment of rotation schemes may conflict with farmers’ economic interests and agricultural priorities. Without adequate incentives, farmers may be reluctant to alter their existing farming practices. Therefore, future research may examine farmers’ willingness to adapt and develop compensation mechanisms that balance landscape aesthetic objectives with agricultural livelihood needs.

5. Conclusions

This study employed a stated choice experiment with digitally calibrated images in Hangzhou Plain during the spring flowering period to assess public preferences for landscape attributes associated with farmland management. The MLM results, which reflect average public preferences, revealed that participants favored oilseed rape-dominated landscapes with a single agricultural land cover proportion in ground perspective, whereas in aerial perspective, they preferred diverse landscapes featuring a high proportion of agricultural land cover and abundant linear and point elements. The LCM results identified different public groups with varying preferences. The two groups with the most significant preferences in the two perspectives had contrasting preferences. Socio-cultural factors such as gender, education level, landscape familiarity, connection to agriculture, and membership in relevant organizations significantly influenced these preferences. The results suggested that the landscape shaped by current policies may align with the average preferences of the public. Promoting oilseed main food crop rotation and increasing the coverage of windbreaks on farmland can be prioritized in landscape planning, and priority can be given to landscapes dominated by oilseed rape and those with open vistas for ground agricultural tourism routes. Also, China could propose strategies for maintaining buffer zones around field margins to balance ecological and aesthetic values amidst agricultural intensification.