Relationships Between Spatial Metrics and Forest Landscape Beauty Across Viewing Distance Zones: Implications for Forest Management in Ino Town, Japan

Abstract

To develop targeted forest management strategies, management staff must understand the statistical relationships between forest aesthetic values and landscape metrics across specified distance ranges. However, as the existing studies based on distance-zone theory have failed to isolate the impacts of landscape features in different zones, their practical applicability to forest management is limited. The present study aims to clarify the different effects of landscape elements on the modeling of forest scenic beauty. To this end, the relevant features are divided into near (0–400 m), middle (400 m–2.5 km), and far (2.5–5 km) zones. A regression analysis stratified by viewing zones confirmed the dominant role of the near zone and revealed different influences of individual landscape elements across the viewing zones. The landscape patterns identified through a cluster analysis, together with pattern-specific regression models, further clarified different explanatory powers of the landscape elements under different conditions, highlighting the elements that should be prioritized to enhance aesthetic value. These findings refine the existing theories and clarify how landscape elements influence aesthetic value across different viewing zones, highlighting the importance of distance-specific landscape element management.

1. Introduction

Human health and well-being are shaped by a wide range of landscape-related factors in rural, natural, semi-natural, and urban environments, including accessibility, biodiversity, opportunities for recreation, and environmental quality [1,2,3]. Among these factors, the aesthetic qualities of forest landscapes represent a particularly important dimension, as they are perceived to provide visual pleasure and are closely associated with stress reduction, psychological restoration, and improved self-reported well-being [4,5,6,7,8]. Given the impacts of land-use/cover changes on aesthetic value [9], decision-makers predicting the effects of forest management must understand the consequences of landscape elements and establish land-use policies that enhance or maintain the original aesthetic quality [10,11].

To identify the crucial elements and assess the impacts of proposed forest policies, many studies have established relationships between objective landscape elements and aesthetic values, acquiring subjective perceptions of forest-landscape aesthetic values through questionnaires or interviews [12,13,14,15], through online big data such as social media posts [16,17,18,19,20], and more recently, through machine learning techniques [21]. These subjective perceptions have been linked to several objective spatial metrics, including composition metrics such as water and vegetation [22,23], and configuration metrics such as landscape diversity [24] and visual scale [25,26]. Most of these studies collectively assess the landscape elements within a fixed viewing distance rather than across different distance ranges. However, in practical applications, understanding the effects of landscape metrics across specified distance ranges is necessary for developing targeted forest management strategies.

The visual distance from a given viewpoint influences human perception of natural environments. This dependence has been recognized since the early development of landscape theory. Originating from painting and pictorial arts, landscape is frequently divided into foreground, middle-ground, and background zones [27]. The foreground is the zone of maximum detail perception, including the surface patterns of bark and the juxtaposition of tree trunks. The middle-ground view provides a broader spatial context, revealing patterns such as the connections of hills into ranges or the visual simplification of vegetative surfaces into textures. The background view is visually simplified, dominated by elements such as skylines and ridge lines against other land surfaces [27]. Schirpke et al. [28] related scenic beauty in mountain regions to different zones. After calculating the distance-weighting factors, they found that the near zone exerts the strongest impact. Building on these findings, many studies have calculated landscape elements using distance–buffer theory, providing more precise models of scenic beauty. For example, Wu et al. [29] demonstrated that the distance d from the viewpoint influences the relationship between landscape features and visual quality. Accordingly, they weighted each landscape element by 1/d. Schirpke et al. [28] adjusted the spatial resolution of digital elevation model (DEM) data to 20 m × 20 m data in the near zone, 100 m × 100 m in the middle zone, and 1 km × 1 km in the far zone. This model has been refined and applied in subsequent studies [9,30,31]. Despite these advances, previous studies have not isolated the landscape metrics across different viewing distances, which is necessary for assessing the distinct effects of these metrics. Therefore, the existing approaches provide limited actionable insights into land-use planning in forest policy.

To address these research gaps, we investigated the effects of landscape elements within different viewing zones on human perceptions of forest landscapes and propose targeted forest-management strategies for different landscape patterns. We hypothesized that the same elements may exert different, or even opposite, effects at varying viewing distances and in varying landscape patterns. The objectives were met through the following processes: (1) mapping the perceived scenic beauty of the forest within a forested region in Japan, (2) measuring the spatial metrics derived from land-cover data in each view distance zone, (3) modeling the perceived forest scenic beauty using regression analysis with the spatial metrics as independent variables, and (4) classifying the perceived landscape patterns using cluster analysis and modeling the scenic beauty of each pattern.

2. Methods

2.1. Study Area

This study was conducted in Ino Town, Kochi, Japan (Figure 1), where the areal coverage exceeds 470 km^2, and the population is 21,374 (2020 Population Census). Ino Town is predominantly forested (~423 km²; ~90%), with farmland and urbanized areas concentrated in the southern region. To assess the scenic beauty in Ino Town, viewpoints were systematically sampled along road centerlines (provided by the Base Information Map in Japan, 2022) at 100-m intervals. The landscape metrics were calculated at 14,891 viewpoints, excluding one viewpoint due to overlap between the road centerline buffer and building data.

2.2. Data Description

2.2.1. DEM and Digital Surface Model Data

The DEM data (0.5-m spatial resolution), representing the regional topography with a 5-km buffer surrounding the study area, were obtained from the Forestry Agency of Japan (2018). A digital surface model (DSM) raster map was created for modeling the heights of the landscape elements. The height of the forested land cover was accurately represented by a digital canopy height model with a spatial resolution of 0.5 m. As precise building-height data were unavailable, a height of 3 m (the nominal height of the single-story structures that dominate buildings in Ino Town) was assigned to both residential and retail building categories. All other landscape elements were assigned a height of zero. To ensure consistency with the spatial resolution of the land-cover map, both the DEM and DSM data were resampled at 5-m resolution.

2.2.2. Land-Cover Data

The land-cover data supporting the eye-level scenic beauty assessments were compiled from multiple datasets. Based on the classification scheme in [32] (

Table A1. Sources of land cover data.

Land-Cover Categories		Data Sources
1	Forest	JAXA high-resolution land-use and land-cover map of Japan; Height values are based on the Digital Canopy Height Model published by the Forestry Agency of Japan
2	Green spaces	Google map
3	Arable	JAXA high-resolution land-use and land-cover map of Japan
4	Grassland	JAXA high-resolution land-use and land-cover map of Japan
5	Water	JAXA high-resolution land-use and land-cover map of Japan
6	Artificial	JAXA high-resolution land-use and land-cover map of Japan
7	Residential buildings	Basic Information Map published by the Geospatial Information Authority of Japan
8	Retail buildings	Basic Information Map published by the Geospatial Information Authority of Japan
9	Roads	Digital National Land Information published by the Geospatial Information Authority of Japan

), the land cover was classified into nine categories. The forest, grassland, arable land, water, and artificial-surface categories were extracted from the land-use and land-cover map of Japan, which was developed using Advanced Land Observing Satellite imagery and published by the Japanese Aerospace Exploration Agency (JAXA) in 2022. Forest types were excluded from this study for two reasons. The first is related to distance band theory. At the analyzed viewing distances (400 m–5 km), specific forest types become visually indistinguishable, and the aesthetics are dictated mainly by the presence/absence and form of the forests relative to other landscape elements. The second reason refers to the methodological foundation established in this study. Distance band theory has not been previously applied to Japanese forest landscapes, necessitating the establishment of fundamental relationships between landscape elements before proceeding to a detailed forest type analysis.

Building polygons were extracted from the Basic Information Map (2020). Ordinary buildings and robust structures were categorized as residential buildings and retail buildings, respectively. Road centerlines were obtained from the Basic Information Map (2020) and buffered by 5 m prior to rasterization to derive the road category, thereby preventing the loss of narrow linear features during resampling. Green space polygons were manually digitized based on their corresponding park areas identified through Google Maps. All raster and vector datasets were standardized to a spatial resolution of 5 m for consistency (Figure 2).

2.3. Scenic Beauty Value Based on an Existing Model

Many previous studies have examined general preferences for natural landscapes [1,33,34]. Common theories, such as prospect-refuge theory, have been shown to apply across different cultural contexts [35]. Based on these findings and adopting a previous aesthetic model [32], we obtained a large volume of sample data while accounting for the influence of viewing distance on human visual perception. Sahraoui et al. [32] modeled aesthetic value using landscape features derived from eye-level views at specified viewpoints simulated by processing 2D spatial information with PixScape 1.2.7 software (see https://thema.univ-fcomte.fr/productions/software/pixscape/en.html (accessed on 28 January 2026)). The regression model was constructed using human perception data obtained from a photographic survey involving a heterogeneous group of participants (referred to as Group 3 in [32]; R² = 0.700).

As described in previous methods, the DEM and land-cover datasets were imported into PixScape to calculate the eye-level landscape metrics by which the scenic beauty values were calculated. Unlike the previous study, the vertical viewing angle was set between −46° and 67°, corresponding to the natural head-and-eye fixed vertical field of view [36], cited in [37]. The horizontal field of view was set to 360°, capturing a full panoramic perspective. The virtual eye height was set at 1.5 m above ground level. Figure 3 shows a representative eye-level image generated by PixScape at a selected viewpoint (Figure 3a) and the corresponding Google Street View image of the same location (Figure 3b). A comparison of the images confirms the accuracy of the visual simulation. At any eye-level viewpoint, the model predicts a scenic beauty score as follows:

VALUE = 0.0184 + 0.0201 × Forest + 0.0019 × SKYLINE + 0.0212 × Grassland + 0.0001 × DISTave + 0.2294 ×
Water + 0.0106 × GreenSpaces + 0.0180 × Roads − 0.3476 × RetailBuild,

(1)

where Forest, Grassland, Water, Greenspaces, Road, and RetailBuild represent the composition metrics of their respective land-cover types, indicating the visible proportion of each type from the specified viewpoint. SKYLINE, a configuration metric that captures the shape of the horizon, was calculated as the ratio of the total length of the visible skyline to the width of the field of view. DISTave, representing the average line-of-sight length, quantifies the degree of openness or depth of the visible landscape. Scenic beauty (VALUE) is computed as a weighted linear combination of the above eye-level visual variables with fixed coefficients. Because VALUE is constructed from these inputs, subsequent regression analyses are not intended to re-estimate or validate the coefficients embedded in the scenic beauty formula. Instead, we use the derived VALUE as a consistent response scale to examine how landscape–beauty relationships differ across viewing distance zones and landscape-pattern contexts.

2.4. Viewing Distance Zones

According to previous studies, the boundaries of environmental factors such as weather, daylight, and atmospheric conditions must be flexibly designated in each viewing zone. Litton [27] defined the foreground as the area extending from 0 to a distance between 1/4 and 1/2 mile (approximately 402–805 m), the middle-ground as the area from 1/4–1/2 mile to 3–5 miles (approximately 4828–8047 m), and the background as beyond 3–5 miles. Japanese landscape theory has incorporated these boundary designations and adapted their ranges to different vegetation characteristics. Specifically, Japanese landscape theory defines the near zone (also called the individual-tree domain) as the area within 340–460 m, where individual trees and plant details are clearly perceptible. The middle zone (or texture domain) covers 340–460 m to 2.1–2.8 km, where the overall textures and structures of vegetation become more apparent. The far zone (or topographic field) includes landscapes beyond 2.1–2.8 km and is characterized by broad topographic forms and simplified visual elements [38].

To separately calculate the spatial metrics in the three defined view-distance zones, the land-cover data within different viewshed distances were extracted using the Viewshed tool in ArcPy 3.2 within ArcGIS Pro 3.2. Based on Japanese landscape theory [38], the viewshed was segmented to facilitate the division of landscape characteristics into the near zone (within 400 m), the middle zone (400 m–2.5 km), and the far zone (2.5–5 km) (see Figure 4). The number of visible land-use grid cells (at 5 m × 5 m resolution) within the defined distance ranges was calculated at each road sampling point, providing a detailed assessment of scenic beauty across the different spatial zones.

2.5. Spatial Metrics

Landscape metrics were calculated using only land-cover grids within the visible area delineated for each viewpoint and distance zone, ensuring that only visible landscape elements were included in the analysis. The landscape metrics were the composition metrics, which quantify the areas of individual land-cover elements, and the configuration metrics, which characterize the spatial pattern or arrangement of landscape elements [39]. These metrics were selected because they are related to scenic beauty in existing studies [32,39,40] and are extractable from available data. Ultimately, the scenic beauty calculation involved nine composition metrics representing the visible land-cover types, which are aligned with land-cover (Table A1). Using PyLandStats 3.0.0 [41], an open-source Python library that computes metrics based on FRAGSTATS definitions (the aerial-level variable calculations are detailed in [42]), five configuration metrics were calculated in Python 3.12 and used separately to characterize spatial patterns and to support the clustering analysis. Among the configuration metrics, the Shannon diversity index [43] and patch density represent the landscape diversity degree, edge density represents the complexity of spatial organization, and the contagion index [44] quantifies the degree of landscape aggregation. Visual scale indicators, which influence landscape preference [25], were included in the eye-level analysis. The visual scale is represented by viewshed size, which measures the number of grid cells visible from the viewpoint. As spatial configuration patterns are inherently designed to quantify the spatial patterns at broader scales, only the nine composition metrics and viewshed size were applied in the near zone; all metrics were included in the middle and far zones.

2.6. Statistical Analyses

Using the scenic beauty values as the dependent variables and the spatial landscape metrics from different view distance zones as the independent variables, the regression analyses evaluated the effectiveness of the scenic beauty predictions and the effects of spatial metrics across distance zones. To avoid overparameterization and reduce multicollinearity among candidate predictors, regression models were specified using an AIC-based stepwise selection procedure (see Supplementary Materials). Because the data are geographically referenced, residual spatial autocorrelation was assessed using Moran’s I based on a Queen contiguity spatial weight matrix, and spatial error models (SEM) were estimated as a robustness check. Results of the spatial diagnostics and SEM analyses are reported in the Supplementary Tables.

Rather than focusing solely on scenic value, we determined the perceived landscape patterns and their ideal combination through a clustering analysis using the K-means algorithm, selected for its conceptual simplicity and practical effectiveness. The K-means algorithm classified landscape patterns based on eye-level metrics that influence scenic beauty values, thereby identifying typical landscape patterns. These pattern types were subsequently used for pattern-specific regression analyses. Specifically, regression models were estimated separately for each landscape pattern, using landscape metrics from different viewing distance zones as explanatory variables and applying the same AIC-based stepwise procedure and spatial robustness checks (see Supplementary Materials). This approach enabled a comparison of how relationships between landscape elements and scenic beauty vary across viewing distance zones under different landscape pattern contexts. Implications for forest management, particularly the configuration and composition of forest-land cover within each viewing distance zone from a given viewpoint, were elucidated through this approach.

3. Results

3.1. Effects of Viewing Distance on Scenic Beauty Values

3.1.1. Scenic Beauty Mapping

Based on the model of Sahraoui et al. [32], the scenic beauty values were calculated using the eye-level landscape metrics generated by Pixscape software. The scenic beauty values derived from the model represent a relative index of perceived aesthetic quality, rather than absolute aesthetic scores. The descriptive statistics of the eye-level landscape metrics and scenic beauty values are listed in

Table A2. Descriptive statistics of eye-level landscape metrics and scenic beauty values (n = 14,891).

	Min.	Max.	Mean	SD	Median
Scenic beauty value	−0.286	0.109	0.041	0.007	0.041
Forest	0.000	0.638	0.336	0.163	0.380
Green spaces	0.000	0.213	0.000	0.004	0.000
Grassland	0.000	0.391	0.018	0.045	0.000
Water	0.000	0.141	0.000	0.003	0.000
Retail buildings	0.000	0.886	0.001	0.013	0.000
Roads	0.000	0.980	0.594	0.094	0.587
Skyline	1.000	1.976	1.388	0.122	1.374
DISTave (m)	3.980	697.525	24.510	36.091	13.208

. As shown in Table A2, the values vary within a limited numerical range in the study area, reflecting comparative differences in perceived scenic beauty among viewpoints. For visualization purposes only, the values were grouped into quartiles to facilitate interpretation of relative differences across viewpoints, rather than the absolute aesthetic thresholds (Figure 5). Under this quartile classification, the areas with high aesthetic values were primarily concentrated in the northern high-altitude regions, along the main roads of the central zone, and near the water bodies in the southern part of the study area (Figure 5). In contrast, urbanized areas in the south were predominantly associated with lower scenic beauty values.

3.1.2. Regression Analysis Across Distance Zones

Using the scenic beauty values as the dependent variable, we constructed regression models based on the corresponding landscape elements in each visual range. The descriptive statistics of landscape metrics are listed in

Table A3. Descriptive statistics of spatial metrics across near, middle, and far distance zones (n = 14,891).

	Min.	Max.	Mean	SD	Median
Near zone
Forest	0	99.94	78	24.13	85.79
Green spaces	0	46.51	0.06	0.98	0
Arable	0	78.11	2.28	7.18	0
Grassland	0	96.21	5.53	11.73	0
Water	0	86.41	0.87	5.35	0
Artificial	0	60.17	3.49	7.69	0
Residential buildings	0	86.01	2.25	6.88	0
Retail buildings	0	28.45	0.14	0.83	0
Roads	0.06	41.07	7.38	8.13	3.83
Middle zone
Forest	0	100	74.42	40.47	98.44
Green spaces	0	15.17	0.01	0.25	0
Arable	0	45.85	0.36	1.64	0
Grassland	0	100	1.94	5.89	0
Water	0	49.57	0.24	1.67	0
Artificial	0	24.2	0.37	1.44	0
Residential buildings	0	43.51	0.37	1.62	0
Retail buildings	0	20.63	0.04	0.31	0
Roads	0	14.58	0.31	0.84	0
Patch density	0.16	218.74	17.76	24.89	7.51
Shannon diversity index	0.54	0.93	0.59	0.06	0.57
Edge density	8.92	186.58	26.17	23.01	16.79
Contagion index	58.56	87.39	73.34	8.39	74.31
Viewshed size	0	112,803	9525.21	13,811.18	3755
Far zone
Forest	0	100	49.27	48.88	57.74
Green spaces	0	2.94	0	0.03	0
Arable	0	30.82	0.11	1.07	0
Grassland	0	99.87	1.08	5.29	0
Water	0	29.63	0.03	0.56	0
Artificial	0	81.11	0.2	2.54	0
Residential buildings	0	57.07	0.11	1.17	0
Retail buildings	0	22.31	0.03	0.42	0
Roads	0	26.67	0.14	0.77	0
Patch density	0.03	114.51	4.09	9.92	0.07
Shannon diversity index	0.67	1.01	0.69	0.03	0.68
Edge density	5.69	89.37	9.59	8.27	5.97
Contagion index	50.32	84.28	61.81	11.79	66.7
Viewshed size	0	521,026	9982.13	25,799.27	14

. The R² values of the models differed in different visual ranges (

Table 1. Viewing-zone–specific regression models of forest scenic beauty value, including coefficients and R² statistics.

Variables	Near Zone		Middle Zone		Far Zone
Constant	0.823	***	0.828	***	0.825	***
Composition metrics
Forest	0.005	**	0.012	***	0.004	***
Green spaces	−0.017	**
Grassland	0.042	***	0.046	***	0.018	***
Artificial land	−0.025	***	0.010	*
Residential buildings	−0.016	***
Retail buildings	−0.132	***	−0.043	***
Roads	0.003	**
Configuration metrics
Viewshed size	0.011	***	0.044	***
Patch density 1	-		0.031	***	0.124	***
Shannon diversity index 1	-		0.019	**	0.110	***
Edge density 1	-		−0.052	***	−0.102	***
Contagion index 1	-		−0.028	***	−0.012	***
R2	0.264		0.168		0.173
Adjusted R2	0.264		0.167		0.173

Note: * p < 0.05; ** p < 0.01; *** p < 0.001; ¹ metrics included only in the middle and far zones.

). The explanatory power of the model was modest overall, with the highest value observed in the near zone (R² = 0.264), followed by the far zone (R² = 0.173) and the middle zone (R² = 0.168). The influences of the landscape elements also differed across different visual ranges. Specifically, increasing the viewing distance decreased the numbers and strengths of the significant composition elements (in the far zone, only forest and grassland remained significantly correlated with scenic beauty) and generally increased the influence of the configuration elements. Although residual spatial autocorrelation was detected (

Table 2. Global Moran’s I for residuals of viewing-zone-specific models.

Model (Residuals)	Moran’s I	Expected I	z-Score	p-Value
Near zone	0.238	−0.00007	49.557	<0.001
Middle zone	0.325	−0.00007	67.650	<0.001
Far zone	0.340	−0.00007	70.726	<0.001

Note: n = 14,891 for all models; Moran’s I was computed for residuals from the corresponding ordinary least squares models using Queen contiguity weights (row-standardized).

), supplementary SEM estimates yielded consistent coefficient patterns and did not change the conclusions (Table S2).

3.2. Landscape Pattern Analysis

3.2.1. Clustering Analysis

The cluster analysis utilized the eye-level metrics used in the scenic beauty calculation (Section 2.3). K-means clustering of the sample viewpoints yielded four distinct cluster groups (as determined by the elbow method).

Table 3. Welch’s one-way ANOVA results for scenic beauty and eye-level metrics across landscape pattern clusters.

	Cluster 1	Cluster 2	Cluster 3	Cluster 4	F	η2	p
	n = 5989	n = 1193	n= 4955	n = 2754
Scenic beauty value	0.042	0.047	0.042	0.036	985.416	0.166	***
Forest	0.484	0.158	0.342	0.078	40,781.766	0.892	***
Green spaces	0.000	0.000	0.000	0.001	18.170	0.004	***
Grassland	0.002	0.149	0.009	0.012	14,274.740	0.742	***
Water	0.000	0.001	0.000	0.001	40.049	0.008	***
Retail buildings	0.000	0.000	0.000	0.005	130.604	0.026	***
Roads	0.513	0.663	0.632	0.671	5455.392	0.524	***
Skyline	1.406	1.341	1.370	1.401	152.900	0.030	***
DISTave (m)	15.066	73.096	21.479	29.454	1078.687	0.179	***

Note: *** p < 0.001; η² denotes effect size.

summarizes the ANOVA results of the eye-level features and scenic beauty values in each cluster. Although ANOVA results indicated statistical significance for most variables, effect size estimates (η²) revealed substantial heterogeneity: some variables (e.g., forest, grassland, and roads) exhibited large effect sizes, whereas others (e.g., green spaces and water) showed very small effects, indicating limited substantive differences despite statistical significance. Representative images and the spatial distributions of the four clusters are shown in Figure 6 and Figure 7, respectively.

Cluster 1, characterized by a high proportion of forests and roads and a narrow field of view, is mainly distributed along forest roads and represents the most common landscape type. Cluster 1 yielded a moderate scenic beauty value (0.042). Cluster 2, dominated by natural landscapes with open views, yielded the highest scenic beauty value but the smallest number of samples. This cluster is primarily distributed in elevated areas, relatively open forest roads, and waterfront zones in the southern region. Cluster 3, characterized by a moderately open view and a relatively high proportion of nearby forest elements, is scattered mainly across the transitional zones between urban and forested areas. The scenic beauty level of Cluster 3 is moderate. Cluster 4, characterized by fewer natural elements and a greater presence of artificial features, is concentrated in southern urban areas and other developed zones. This cluster yielded the lowest scenic beauty value.

3.2.2. Regression Analysis by Landscape Pattern

Through regression analyses of each cluster, this subsection relates the landscape elements across different visual ranges and assesses their influences on scenic beauty within different landscape types. As shown in

Table 4. Pattern-specific regression models using landscape elements in different distance zones as the explanatory variables.

Variables	Cluster 1		Cluster 2		Cluster 3		Cluster 4
Constant	0.833	***	0.820	***	0.837	***	0.823	***
Near zone
Forest	−0.006	***			−0.010	***	−0.010	**
Green spaces					−0.040	***
Arable	−0.008	***			−0.012	***
Grassland	0.102	*	0.020	***
Artificial	−0.008	***	−0.020	***	−0.013	***	−0.021	***
Residential buildings	−0.009	**					−0.016	***
Retail buildings							−0.174	***
Roads	−0.002	***	0.012	*
Viewshed size	0.009	***			0.013	***
Middle zone
Forest			0.009	**
Green spaces			−0.124	***
Arable			0.020	*
Grassland			0.018	**	0.006	***
Water			−0.022	**	−0.013	***
Artificial	−0.017	***
Residential buildings					−0.012	***
Retail buildings					−0.072	***	−0.044	*
Roads	0.006	**			−0.002	***
Patch density					0.012	**
Shannon diversity index	0.008	**	−0.057	***
Edge density	0.013	***					0.043	***
Contagion index			−0.008	**	−0.002	*
Viewshed size	−0.013	**	0.101	***
Far zone
Forest	0.001	*	0.007	***
Water	−0.006	**	−0.052	**
Grassland			0.013	*
Residential buildings	−0.017	*
Retail buildings	0.024	*
Patch density	−0.031	***			0.058	***
Shannon diversity index	0.022	**	0.102	***	−0.036	***
Edge density	0.013	*			−0.036	***
Contagion index			−0.013	***
Viewshed size			−0.056	**	0.089	***	0.106	***
n	5989		1193		4955		2754
R2	0.280		0.753		0.433		0.249
Adjusted R2	0.277		0.749		0.431		0.246
Sig.	<0.001		<0.001		<0.001		<0.001

Note: * p < 0.05; ** p < 0.01; *** p < 0.001.

, the explanatory power of the models varied among the clusters, being highest in Cluster 2 (R² = 0.753), followed by Cluster 3. The explanatory powers were lowest in Clusters 1 and 4 (both with R² values below 0.3).

We further examined the differences among the regression coefficients of landscape elements in different visual ranges of the clusters. A consistent trend was observed in the near-view range: tall elements such as forests and buildings were negatively correlated with scenic beauty, whereas flat natural elements, particularly grassland, were strongly positively correlated with scenic beauty. This pattern aligns with the positive relationship between scenic beauty and viewshed size. In the near zone, positive and negative associations between the natural elements and scenic beauty were observed in Clusters 2 and 3, respectively. In this range, configuration diversity tended to enhance scenic beauty in Clusters 1 and 4 (middle zone) and in Clusters 1 and 2 (far zone), while showing negative effects in Cluster 2 (middle zone) and Cluster 3 (far zone). In the far zone, the composition elements exerted their main effects in Clusters 1 and 2. High diversity of the configuration elements in the far zone was positively correlated with scenic beauty in Clusters 1 and 2, but higher diversity and complexity tended to lower the scenic beauty values in Cluster 3. Finally, viewshed size and scenic beauty were negatively related in Cluster 2 but positively related in Clusters 3 and 4. Residual spatial autocorrelation was detected in the cluster-specific models (Table S3); however, supplementary SEM analyses yielded similar coefficient patterns and did not change the cluster-based interpretations (Table S4).

4. Discussion

Introducing the concept of view distance within landscape theory, this study aimed to predict forest scenic beauty values from spatial metrics across the near, middle, and far-distance zones. Through this approach, we can gain a more nuanced understanding of how landscape elements and their arrangement across different view distances affect the perceived aesthetic qualities of forest landscapes. Moreover, understanding the differential effects of spatial elements across landscape patterns enables the development of targeted strategies for specific spatial contexts, thereby supporting improved forest management decisions.

4.1. Effects of Spatial Metrics Across Distance Zones

This study reports the first examination of landscape-element effects on aesthetic value across different viewing distances. In the spatial metric calculations, the distance boundaries (400 m, 2.5 km, and 5 km) were decided based on the existing literature in Japan.

The regression models in each visual range confirmed that the influence of landscape elements depends on scale. Although all three models exhibited relatively low R² values, diagnostic checks suggest no major departures from the assumptions of linear regression (see Supplementary Materials). Furthermore, because each model was fitted within a single viewing distance zone and included only zone-specific predictors, modest R² values are considered acceptable for the comparative purpose of this analysis. The strongest effect (R² = 0.264) was observed in the near zone, consistent with previous studies [28,45]. However, unlike prior research reporting a stronger mid-view influence, we report a slightly stronger effect of the far view (R² = 0.173) than of the mid-view (R² = 0.167), likely imposed by differences in the defined viewing distance ranges. In particular, the mid- and far-view ranges were farther from the observer in prior works (middle zone: 1.5–10 km; far zone: 10–50 km) than in the present study. Moreover, whereas earlier studies observed pronounced differences among the three ranges and a relatively weak far-view effect, our findings suggest more comparable influences of the near-, mid-, and far-view ranges. This discrepancy may stem from methodological differences. Specifically, previous work estimated the weight of each visual range from photographs, removing one or more ranges at each time [28,45]. This approach focuses solely on the visible features, overlooking openness and other non-visible attributes that can also shape aesthetic perception.

Furthermore, as revealed in the distance-specific regression analyses, the number of composition variables exhibiting significant effects decreased with increasing viewing distance. This trend was particularly evident for elements occupying smaller proportions of the visual field, such as residential buildings. Previous studies have similarly stated that increasing the distance decreases the discernibility of details [46] and diminishes the index related to visual features [29]. Investigating the influences of elements across different distance zones, we found that overall, artificial and natural elements tended to exert negative and positive effects, respectively. The influence of the same element also differed across different visual ranges; for instance, artificial land exerted negative and positive effects in the near and middle-view zones, respectively, possibly because artificial land provides an open visual field at mid-range. In contrast, the influences of the configuration variables increased with increasing distance. Metrics such as patch density and Shannon diversity, which are known to enhance scenic beauty [25,40,47], proved to be more influential in the far zone than in the other zones, suggesting that spatial structure and landscape heterogeneity more saliently shape scenic beauty as the visual distance increases.

4.2. Implications for Forest Management Based on Landscape Patterns

To understand how elements influence predictions of scenic beauty and to develop forest management policies, a consideration of the landscape patterns might be essential. Therefore, the final part of this study explored the relationships between landscape metrics and scenic beauty value in specific landscape patterns.

The K-means cluster analysis broadly divided the eye-level landscape patterns into natural-oriented (Clusters 1 and 2) and artificial-oriented (Clusters 3 and 4) patterns. The regression analysis results revealed varying explanatory powers in different landscape clusters, which may be attributed to the different influences of the landscape elements on aesthetic perception at multiple spatial levels. The explanatory power was highest in Cluster 2 (R² = 0.753), likely owing to the broad viewing range with unobstructed near and middle views and extensive forest cover in the far view. The explanatory contribution of multi-scale landscape factors can be strengthened within the spatial composition of Cluster 2. In contrast, the explanatory power of the landscape elements was below 0.3 in Clusters 1 and 4, possibly because local obstructions in the near and middle zones limit the visual impact of far-view elements, thereby reducing the model’s capacity to explain aesthetic evaluations.

Based on the regression results of each landscape type, we propose differentiated strategies to enhance the aesthetic quality of forest landscapes. In Cluster 1, which mainly represents roads through the forest, trees should be removed from the near or middle zones to expand the view in areas of predominantly natural middle or distant areas. In Cluster 2, characterized by open natural scenery, enhancing the naturalness of the middle view and the heterogeneity of the far view is apparently more effective than merely increasing the openness. In Clusters 3 and 4, where artificial structures constitute major visual barriers and expanding the openness may be difficult, partial screening of buildings through strategic placement of near- and middle-view vegetation is a potentially practical and aesthetically pleasing solution.

Note that our cluster analysis does not classify forest types. Aesthetic differences are primarily driven by the openness, topographic context, and spatial relationships of forests with non-forest elements, rather than by the forest composition itself. The identified significant zone-specific and pattern-specific effects of landscape elements (such as grassland) provide practical insights for the management of landscapes dominated by forests, irrespective of specific forest types.

4.3. Limitations and Future Work

Several limitations of this study must be mentioned. First, the scenic beauty values were predicted by a regression model developed in previous research and were not backed by subjective surveys. That is, scenic beauty scores were computed using an established perception-based scoring model that is itself a linear combination of several eye-level visual variables (e.g., forest, grassland, water, roads, retail buildings, skyline, DISTave, and green spaces) with fixed coefficients. Therefore, statistical associations between scenic beauty scores and landscape variables that overlap conceptually or operationally with these inputs should not be interpreted as independent validation of those specific components. In this study, our inferences focus on comparative patterns (i.e., how relationships vary across viewing distance zones and landscape-pattern contexts) under a consistent scoring framework, rather than on the absolute effect sizes of individual predictors. Future work could examine coefficient sensitivity and alternative parameterizations of the scenic beauty model. In addition, as the evaluations of eye-level metrics and scenic beauty were derived solely from the information of element proportions in the view, they cannot capture the detailed differences among observations of landscape elements in real-life scenarios. The employed regression formula was originally developed using images from urban fringe areas in Europe and may not be fully applicable to forest areas in Japan. Therefore, a perceptual survey of landscape preferences using photographs from the study area is warranted in the future.

Second, while the modest explanatory power of the zone-specific models is consistent with the analytical design aimed at isolating viewing-distance-specific effects, incorporating additional relevant variables may help better capture the complexity of the relationships between landscape elements and scenic beauty perception across viewing distances. Spatially explicit regression approaches (e.g., geographically weighted regression) could also be explored to examine continuous spatial heterogeneity in these relationships, thereby complementing the present zone- and pattern-based framework.

Moreover, the topological data in the DEM were limited to Kochi Prefecture. As data were missing in the northern area of the 5-km buffer of Ino Town, the open area values were artificially raised at viewpoints close to the northern border. Finally, the present study established fundamental distance-zone relationships in forest-dominated landscapes; the effects of specific forest types (coniferous, deciduous, mixed) and structural characteristics (age, density, species diversity) on aesthetic perception within each distance zone should be investigated in the future.

5. Conclusions

In summary, this study addressed a research gap in forest landscape perception by examining distance- and pattern-dependent variations in how spatial landscape elements relate to perceived scenic beauty. The results extend previous understanding of landscape effects—often expressed in simplified terms (e.g., uniformly positive effects of natural elements and negative effects of artificial features)—by demonstrating that these effects vary and can even be reversed across viewing distance zones and eye-level landscape patterns.

From a practical perspective, these findings can help forest managers and landscape planners prioritize context-specific interventions to enhance aesthetic value rather than applying uniform management measures, thereby supporting more efficient use of limited resources. Notably, although this study was conducted in Japan, the eye-level features were derived from objective spatial datasets, the study area encompasses diverse forest-related landscape environments, and the results are reported across landscape patterns. Therefore, the proposed analytical workflow can be applied in other regions where comparable spatial data are available, and the resulting insights may be informative for scenic beauty discussions in other spatial (e.g., rural or peri-urban) and socio-cultural contexts.