Virtual nature, real relief: how exposure to virtual natural environments reduces anxiety, stress, and depression in healthy adults

Abstract

Stress, anxiety, and depression represent significant challenges to global public health. Exposure to virtual natural environments, as a convenient and scalable intervention, has shown uncertain effects on healthy adults. This systematic review and meta-analysis aims to comprehensively evaluate the impact of exposure to virtual natural environments on stress, anxiety, and depression in healthy adults. A total of 24 studies were included after searching five databases. The results indicate that exposure to virtual natural environments effectively reduces anxiety levels (SMD = 0.82, p < 0.001, large effect), stress levels (SMD = 0.577, p = 0.003, moderate effect), and depression levels (SMD = 0.621, p < 0.001, moderate effect) in healthy adults. These findings suggest that exposure to virtual natural environments has a positive impact on mental health and can serve as a viable alternative when direct access to natural environments is not feasible.

Introduction

According to the statistical reports from Gallup, the stress levels of adults have remained at a high level in recent years, with minor fluctuations but still significantly higher than those a decade ago^1,2. The increasing focus on stress can be attributed, in part, to its significant correlation with depression and anxiety. Studies have confirmed that stress is an important trigger for depression and anxiety³, and the onset of these conditions can often be traced back to stressful life events^4–7. Data from the World Health Organization in 2022 indicated that the impact of anxiety and depression is far greater than expected worldwide. The prevalence of mental disorders among working-age adults remains high, resulting in substantial economic losses^8,9. Clearly, stress, anxiety, and depression have become a major public health challenge globally, urgently requiring effective strategies to address them.

Nature, as a non-pharmacological intervention for alleviating anxiety, stress, and depression^10–12, offers benefits across the entire life span and among diverse occupational groups^10,13–15. Moreover, its high inclusivity provides a variety of engagement pathways, such as exposure to natural environments^16,17, walking in nature^18,19, and engaging in physical exercise within natural settings^20,21, thereby facilitating multidimensional connections between individuals and nature. However, rapid urbanization has constrained opportunities for nature contact in both time and space^22–24. In contrast, advancements in digitalization and technological innovation have created new possibilities for overcoming these limitations²⁵. Virtual reality (VR) technology, in particular, serves as an ideal medium. It transcends temporal and spatial barriers, eliminates physical constraints, and allows individuals to immerse themselves in the beauty of nature^26,27. Additionally, VR is highly controllable and ecologically valid, capable of simulating realistic natural environments²⁸.

Indeed, there is a growing body of research on the impact of virtual nature exposure on mental health, encompassing both clinical and healthy populations. Studies have been conducted on individuals with depression²⁹, anxiety³⁰, dementia³¹, and breast cancer³², as well as on working professionals³³ and college students^34,35. The methods of virtual nature exposure employed in these studies are diverse, ranging from traditional 2D screens³⁶, 360° projections³³, to head-mounted VR displays³⁷. Additionally, the intervention processes involve various factors, such as multisensory stimulation (visual, auditory, and tactile)³⁸, different sound level settings (high vs. low volume)³⁹, and distinct virtual nature environments (e.g., Wood, Waterfall with trees, Pool with plants)⁴⁰. The effects of virtual nature exposure are multidimensional and significant. Psychologically, it effectively alleviates depressive symptoms, reduces anxiety, and mitigates accumulated stress^41,42. Physiologically, it promotes relaxation by influencing heart rate variability and regulating blood pressure^41,42. Moreover, the visual stimulation provided by virtual nature environments holds unique value. Research indicates that this visual input not only enhances attention restoration but also alleviates pain perception to some extent, reducing physical discomfort through positive sensory experiences⁴².

However, while the diversity of populations, methods, and variables in the research on virtual nature exposure enriches and enlivens this field, it also introduces a degree of ambiguity. Particularly for healthy adults, the effects of virtual nature exposure appear to lack a consistent conclusion. For instance, Yu et al.⁴³ demonstrated that exposure to a virtual forest environment could improve participants’ anxiety levels, whereas Brivio et al.³³ did not observe significant improvements. Although numerous systematic reviews and meta-analyses have been conducted on this topic^44,45, these studies primarily focus on clinical populations. The specificity of the included groups introduces potential psychological and physiological confounders, thereby limiting the generalizability of the findings. For non-clinical populations, existing mini-reviews⁴² and systematic reviews⁴¹ provide a broad overview of the effects of virtual nature exposure. However, they lack a clear quantification of these effects, offering only generalized introductions rather than precise assessments.

Given the detrimental and pervasive nature of stress, anxiety, and depression among healthy adults, the urgent need for nature exposure, and the limited accessibility to natural environments, coupled with the convenience of virtual natural environment exposure and the uncertainty regarding its effectiveness, this systematic review and meta-analysis aims to systematically investigate the effects of virtual natural environments on anxiety, stress, and depression levels in healthy adults. Outcome measures include standardized scales used to assess levels of anxiety, stress, and depression.

Results

Study selection

The flowchart of the study selection process is shown in Fig. 1. The initial search yielded a total of 30,367 studies. After automatic removal of duplicates using Endnote software (Version X9; Thomson ResearchSoft, USA) and manual screening, 6449 duplicates were excluded. In the subsequent preliminary review based on titles and abstracts, 23,762 studies were excluded for reasons such as irrelevance to the research topic, registration protocols, reviews, theses, patents, books, conference papers, and experimental protocols. Consequently, 156 studies were assessed for eligibility. Of these, 132 studies were excluded due to reasons such as long-term interventions, lack of required indicators, absence of pretests, failure to obtain experimental data, data non-compliance, incompatible environments, incompatible interventions, protocols, non-healthy populations and non-adult populations. Ultimately, 24 studies were included in this review. All retrieved records and the reasons for their inclusion and exclusion are detailed in the Supplementary Table 2 for reference.

Fig. 1. Visualization of the study selection process.

A total of 30,367 studies were identified during the search phase. After excluding 6449 duplicates through both automated and manual screening, 23,762 studies were excluded based on initial review of titles and abstracts. Following full-text review, an additional 132 studies were excluded. Ultimately, 24 studies were included in the analysis.

Study characteristics

This systematic review and meta-analysis ultimately included 24 studies for further statistical analysis. All participants in the included studies were healthy adults (as reported in the individual studies), encompassing college students, faculty and staff, healthy adults without specified occupations, and ward staff. All participants underwent interventions involving exposure to virtual natural environments, which included three types of non-real nature exposure methods: 2D, 360°, and VR head-mounted display. The specific characteristics of the included studies are detailed in Table 1.

Table 1.

Characteristics of included studies

Study	Sample (sex)	Group	Age stage	Details of interventions for the virtual nature exposure
Yu 201843	30 (male:13, female: 17)	Adults	20–29 years	Equipment: HTC Vive VR system (head-mounted display) developed by HTC and Valve Corporation. Virtual Nature Environment: The video captured natural elements of the Wandanda forest park, including twin waterfalls, maple tree trail, pine grove, cypress, fir forest observatory, and the Cingshuei River.
Yu 202047	34 (male:6, female 28)	Adults	58.76 ± 8.36 years	Equipment: Samsung Gear VR head-mounted display and a Samsung Galaxy S8 smartphone with iPhone earbuds. Virtual Nature Environment: Videos and soundtracks primarily contained natural elements such as waterfalls, creeks, broad-leaved forests, ferns, and bird songs, manmade structures (e.g., roads, fences, tourists) were not present.
Yang 201434	30 (male)	College Students	22–25 years	Equipment: VRP 11.0 (3D Interactive Virtual Platform) with NVIDIA 3D glasses. Virtual Nature Environment: A picturesque ocean island surrounded by azure seawater, featuring a lookout tower, pavilion, benches, clear weather with sunshine reflecting off the sea, cloudless sky, and sounds of waves, seagulls, and birds. Tourists were depicted resting or strolling on the island.
Yahaya 202354	8 (male: 4, female:3, not to disclose gender:1)	Students: 4 Full-time employees: 4	16–30 years old: 62%, 31–45 years old: 38%	Equipment: Virtual environments created using Visual Studio in the Unity engine. The application was developed for desktop VR and does not require a VR headset. Virtual Nature Environment: Terrain included oak trees, ferns, mosses, mountains, and waterfalls.
Williams 202355	14 (male:11, female:3)	Ward Staff	Not reported	Equipment: Wireless Oculus Go head-mounted display with built-in speakers and a handheld controller. Virtual Nature Environment: Virtual environments included beach scenes, underwater scenes, swimming with dolphins, and mountain landscapes.
Wang 201940	96 (male:33, female:63)	College Students	24.03 ± 5.29 years	Equipment: Not reported Virtual nature environment: It includes Structure (S), Wood (W), Wood with bench (WB), Wood with platform and bench (WPB), Platform with trees (PT), Waterfall with trees (WT), Pool with plants (PP). S: Wooden pavilion, building, wooden platform, gravel paving, plants, tables and chairs W: Plants, soil WB: Plants, wooden benches close to nature WPB: Plants, wooden platforms, wooden tables and benches, wood paving PT: Plants, wooden platforms, wooden planks, trees WT: Plants, pool water, waterfall, rocks PP: Pool water, rocks, railing, fountain head, plants
Vaquero-Blasco 202148	23 (male: 8, female: 14, non-binary: 1)	Adults	22.65 ± 5.48 years	Equipment: VR head-mounted display Virtual Nature Environment: Participants freely choose from four environments: beach, cascade cave, aurora borealis, and space.
Takayama 202249	25 (male: 12, female: 13)	Adults	36.1 ± 8.13 years	Equipment: Five projectors (FP-Z5000, FUJIFILM Holdings Corp., Tokyo, Japan) projected video images taken in Urahoro-cho, Hokkaido, onto the three south–northwest walls and ceiling of the experimental room. Speakers (ICS-15, ONKYO Home Entertainment Corp., Osaka, Japan, one being a woofer) installed at six locations in the room emitted environmental sounds linked to the video images. Virtual Nature Environment: Forest environment of Urahoro-cho.
Suseno 202336	62 (male: 33, female: 29) VR group: 25 2D group: 28	Adults	VR group: 20.7 ± 1.72 years 2D group: 19.8 ± 1.31 years	Equipment: Virtual reality environment was shown on the HMD HTC VIVE Pro Eye. 2D video environment was shown with a 22-inch flat-screen 1 m away. Virtual Nature Environment: Nokia Bay was used as the simulated natural environment for both the virtual reality and the 2D video.
Reese 202250	25 (sex not specified)	Adults	24.2 ± 3.7 years	Equipment: Oculus Rift head-mounted display with two-hand controlling device and respective sensors. Virtual Nature Environment: The VR environment featured a forest landscape, including trails, a wooden cabin, and stairs. Birdsong was played in the VR environment, and participants heard footsteps while moving.
Liu 202356	156 (male:48, female:108)	College Students	19.8 ± 1.18 years	Equipment: Pico Neo 3 VR headset for 360° video presentation. Virtual Nature Environment: Travel VR video of Cuba; specific details not provided.
Lahti 202046	129 (sex not specified)	Adults	51.8 ± 16.8 years	Equipment: Samsung Gear VR headset and Samsung Galaxy S7 mobile phone (attached to the virtual headset) for the MelloVR application. Virtual Nature Environment: Beach, waterfall, underwater, space float, paddling.
Kim 202137	74 (male: 37, female: 37)	Adults	19–59 years	Equipment: Samsung Gear VR with head-mounted display device featuring separate screens for each eye, integrated head tracking, and stereo earphones. Virtual Nature Environment: Participants viewed a VR video of immersive natural scenes while walking on a trekking course with famous scenery and a relaxing soundtrack.
Hsieh 202339	45 (male: 16, female: 29) no-decibel group: 15 low-decibel group: 15 high-decibel group: 15	College Students	19–23 years	Equipment: VR equipment not reported. Virtual Nature Environment: Natural forest and water scenery.
Hong 201957	40 (male: 23, female: 17)	Adults	24.4 ± 2.8 years	Equipment: VR headset (HTC VIVE PRO, HTC, Taiwan). Virtual Nature Environment: The forest physiognomy in the VR forest video is a mixed forest.
Chen 202351	173 (male: 85, female: 75, other: 2, no response: 11)	Adults	37 ± 17 years	Equipment: Immersive CAVE utilizing two rear projectors for 2K-4K-8K 3D data sources displayed on a curved panoramic screen. Virtual Nature Environment: Natural environments were captured from nearby Boston parks.
Brivio 202033	VR Group: 20 (male: 8, female: 12) 360 group: 20 (male: 11, female 9)	College students and faculty	VR Group: 26.15 ± 5.12 years 360 group: 26.40 ± 6.69 years	Equipment: Virtual environment designed using Unity software (Unity Technologies 2015). The 360° video was created using two Kodak PixPro SP360 4 K cameras and Kodak PixPro 360 Stitch software, with similar features to the virtual environment. Both environments were visualized through a VR i7 head-mounted display with an LG Nexus smartphone (Android 4.4.2 KitKat, 4.95″ 1920 × 1080 Full HD Display). Virtual Nature Environment: Included trees, flowers, water, houses, bridges, and natural sounds.
Bodet-Contentin 202258	88 (male: 17, female 71)	Intensive unit caregivers	< 30 years: 25, 30–40 years: 43, > 40 years: 20	Equipment: Healthy Mind VR program (Healthy Mind, Paris, France) and an Oculus GO VR device (Oculus VR, San Francisco, CA, USA) with an audio headset (Bose, Framingham, MA, United States). Virtual Nature Environment: Three natural virtual environments were available for selection: garden, forest, or mountain.
Bielinis 202035	42 (male: 19, female: 23)	College Students	26.24 ± 6.23 years	Equipment: High-resolution 55” LCD monitor (MultiSync P554, NEC, Tokyo, Japan) with dimensions of 71.4 × 124.4 cm² (1920 × 1080 resolution). Virtual Nature Environment: Scots pine (Pinus sylvestris), coniferous forests dominated by Scots pine and Norwegian spruce, and a path made of wooden beams.
Annerstedt 201352	20 (male) VR with sound group: 10 VR no sound group: 10	Adults	VR with sound group: 28.2 ± 10.3 years VR no sound: 26.7 ± 3.4 years	Equipment: CAVE™ system with three rear-projected walls (4 m × 3 m) and a floor projection (EON Development Inc.). Virtual Nature Environment: Featured twittering birds and a babbling brook.
Adhyaru 202259	39 (male: 7, female: 32)	Adults	36.61 ± 10.26 years	Equipment: Oculus Go with Single Fast-Switch LCD 2560 × 1440 screen. Virtual Nature Environment: Participants interacted with the environment to varying degrees; some navigated through the scene and adapted the environment (e.g., planting “virtual trees” or changing the weather), while others remained stationary, looking around and listening to calming sounds (soft music, animal sounds).
Ma 202560	49 (male: 10, female: 39)	College Students	22.06 ± 4.56 years	Equipment: PICOS VR headset. Virtual Nature Environment: The video consisted of two nature-based environments, though specific details were not reported.
Pratviel 202461	36 (male: 27, female: 9)	Sport students	19.2 ± 1.3 years	Equipment: HTC Vive Pro headset (HTC America, Inc., Seattle, WA, USA). Virtual Nature Environment: Developed using Unity (Unity Technologies, San Francisco, CA, USA), the environment represents an island floating in a sea of clouds, featuring natural elements such as rocks, grass, and trees, a large vista, and dimmed lighting.
Reese 202153	64 (male: 17, female: 47)	Adults	23 ± 3.87 years	Equipment: OculusRift head-mounted-display with its two-hand controlling device and sensors. Virtual nature environment: coastal nature environment.

Risk of bias and quality of evidence

Overall, 1 study⁴⁶ (4%) was classified as having a low risk of bias, 13 studies^{33–36,40,43,47–53} (54%) were classified as having an unclear risk of bias, and 10 studies^{37,39,54–61} (42%) were classified as having a high risk of bias. The risk of bias is detailed in Fig. 2. In the domain of randomization, 2 studies explicitly mentioned the use of allocation concealment measures, leading to most studies being rated as “unclear risk of bias” or “high risk of bias” in this domain. In the domain of deviations from intended interventions, 4 studies were rated as “high risk of bias” due to open-label design, which led to blinding failure, subjective outcome assessment bias, and inappropriate analysis methods. In the domain of missing outcome data, 1 study was rated as “high risk of bias” due to a high dropout rate, with the reasons for dropout potentially related to the true values of the outcomes.

Fig. 2. Overall risk of bias presented as percentage of each risk of bias item across all included studies.

The figure includes six domains of outcome reporting assessment. Green indicates low risk of bias, yellow represents unclear risk of bias, and red denotes high risk of bias.

After assessing the quality of evidence using the GRADE framework, the evidence quality for all outcomes was judged to be very low (Table 2). This result was primarily influenced by the high risk of bias in some studies. Additionally, heterogeneity among studies also contributed to the reduced overall credibility of the evidence.

Table 2.

GRADE Summary of Evidence

Outcome	Number of studies	Study design	Risk of bias	Inconsistency	Indirectness	Imprecision	Confidence rating
Anxiety Levels	21	Randomised and non-randomised studies					Very low
Stress Levels	6	Randomised and non-randomised studies					Very low
Depression Levels	9	Randomised and non-randomised studies					Very low

Risk of bias: very serious; serious; not serious, study design includes both randomised controlled trials (RCTs) and non-randomised studies (NRS). RCTs were initially rated as “high” quality and NRS as “low”, with overall certainty rated based on the body of evidence following GRADE criteria.

Publication bias

Visual inspection of the funnel plots suggested potential publication bias for anxiety, stress, and depression levels, although this could not be quantified directly (Fig. 3). Therefore, Egger’s test was employed to quantify the risk of publication bias. The results of Egger’s test indicated no significant publication bias for anxiety levels (t = 1.79, p = 0.082 > 0.05), stress levels (t = 1.78, p = 0.135 > 0.05), and depression levels (t = 1.24, p = 0.251 > 0.05).

Fig. 3. Visualization of publication bias, and Egger’s test.

This figure includes assessments of publication bias and results of Egger’s Test for the outcomes of anxiety, stress, and depression. a Funnel plot for Anxiety. b Egger’s Test for Anxiety. c Funnel plot for Stress. d Egger’s Test for Stress. e Funnel plot for Depression. f Egger’s Test for Depression.

Meta analysis of anxiety levels

This systematic review and meta-analysis included a total of 21 studies, comprising 37 comparison groups and 1471 participants, to assess the impact of exposure to virtual natural environments on anxiety levels (Fig. 4). The meta-analysis revealed that exposure to virtual natural environments significantly reduced anxiety levels in healthy adults (SMD = 0.82, 95% CI [0.639, 1.001], p < 0.001), with a large effect size. However, the results exhibited substantial heterogeneity (I² = 80.2%, p < 0.001).

Fig. 4. Visual representation of the effect of virtual natural environment exposure on anxiety levels.

This forest plot illustrates the effect of exposure to virtual natural environments on anxiety levels. Each dot represents the effect size of an individual study or the overall effect size. The error bars indicate the lower and upper limits of the 95% confidence interval (CI). The mountain plot enhances the visualization effect. The color gradient of the plot indicates the magnitude of the effect: a deeper blue signifies a larger effect, while a deeper yellow signifies a smaller effect.

Meta analysis of stress levels

This systematic review and meta-analysis included a total of 6 studies, comprising 7 comparison groups and 250 participants, to assess the impact of exposure to virtual natural environments on stress levels (Fig. 5). The meta-analysis revealed that exposure to virtual natural environments significantly reduced stress levels in healthy adults (SMD = 0.577, 95% CI [0.137, 1.016], p = 0.003), with a moderate effect size. However, the results exhibited substantial heterogeneity (I² = 81.0%, p < 0.001).

Fig. 5. Visual representation of the effect of virtual natural environment exposure on stress levels.

This forest plot illustrates the effect of exposure to virtual natural environments on stress levels. Each dot represents the effect size of an individual study or the overall effect size. The error bars indicate the lower and upper limits of the 95% confidence interval (CI). The mountain plot enhances the visualization effect. The color gradient of the plot indicates the magnitude of the effect: a deeper blue signifies a larger effect, while a deeper yellow signifies a smaller effect.

Meta analysis of depression levels

This systematic review and meta-analysis included a total of 9 studies, comprising 10 comparison groups and 425 participants, to assess the impact of exposure to virtual natural environments on depression levels (Fig. 6). The meta-analysis revealed that exposure to virtual natural environments significantly reduced depression levels in healthy adults (SMD = 0.621, 95% CI [0.298, 0.943], p < 0.001), with a moderate effect size. However, the results exhibited substantial heterogeneity (I² = 79.6%, p < 0.001).

Fig. 6. Visual representation of the effect of virtual natural environment exposure on depression levels.

This forest plot illustrates the effect of exposure to virtual natural environments on depression levels. Each dot represents the effect size of an individual study or the overall effect size. The error bars indicate the lower and upper limits of the 95% confidence interval (CI). The mountain plot enhances the visualization effect. The color gradient of the plot indicates the magnitude of the effect: a deeper blue signifies a larger effect, while a deeper yellow signifies a smaller effect.

Subgroup analysis

The results of the subgroup analysis (Table 3) indicated a significant interaction between subgroups for the categorical variable “Age groups” (p = 0.004), suggesting that age groups were a primary source of heterogeneity. In contrast, no significant interactions were observed for “Exposure duration” (p = 0.319) or “Exposure environments” (p = 0.937). Specifically, an exposure duration of 10–15 min was associated with a large effect size (SMD = 1.11, 95% CI [0.72, 1.51], p < 0.001). Among the different age groups, individuals under 20 years old, those aged 20–24.99 years, and those aged 25–30 years all exhibited large effects (SMD > 0.80, p < 0.05), whereas individuals over 30 years old had a smaller effect size (SMD = 0.46, 95% CI [0.35, 0.58], p < 0.001). In the virtual natural environment exposure scenario, both real-video content (SMD = 0.83, 95% CI [0.62, 1.04], p < 0.001) and virtual-video content (SMD = 0.81, 95% CI [0.46, 1.16], p < 0.001) demonstrated substantial effects. The detailed results of the subgroup analysis have been added to Supplementary Table 3 for reference.

Table 3.

Subgroup Analysis Results

Variable	Group Data Size	Effect Size (SMD)	95% Confidence Interval	Subgroup p-value	Interaction p-value
Exposure duration
less than 5 min	5	0.70	0.03, 1.38	0.040	0.319
5–9.9 min	19	0.71	0.49, 0.92	<0.001
10–15 min	11	1.11	0.72, 1.51	<0.001
more than 15 min	2	0.67	0.21, 1.12	0.004
Age groups
under 20 years old	3	1.06	0.26, 1.85	0.009	0.004
20–24.99 years old	18	0.93	0.67, 1.18	<0.001
25–30 years old	6	1.07	0.27, 1.87	0.009
over 30 years old	10	0.46	0.35, 0.58	<0.001
Exposure environments
Virtual	12	0.81	0.46, 1.16	<0.001	0.937
Real	25	0.83	0.62, 1.04	<0.001

Meta-regression

To explore the moderating effects of continuous variables on the anxiety-alleviating effects of natural interventions, we conducted univariate random-effects meta-regression analyses (Fig. 7). The results indicated that exposure duration was positively correlated with the intervention effect on anxiety levels (β = 0.0322, p = 0.045), accounting for 12.1% of the heterogeneity between studies. Conversely, participant age was negatively correlated with the intervention effect on anxiety levels (β = –0.0208, p = 0.039), explaining 13.5% of the heterogeneity. No significant moderating effects were observed for other covariates, such as exposure environment. The detailed results of the meta-regression analysis have been added to Supplementary Table 4 for reference.

Fig. 7. Visualization of meta-regression analysis.

This figure provides a visual representation of the meta-regression analysis, illustrating the relationship between the effect sizes of virtual natural environment exposure and potential moderator variables. a Exposure Duration. b Age Groups. c Exposure Environments.

Sensitivity analysis

Sensitivity analyses revealed that despite the high heterogeneity in anxiety levels, the exclusion of any single study did not significantly alter the overall effect size or lead to a shift in the summary effect (i.e., rendering the results non-significant or reversing the direction of the effect). Similarly, for stress and depression levels, which also exhibited high heterogeneity, the exclusion of specific studies (Vaquero-Blasco 2021⁴⁸ and Yang 2014³⁴) reduced heterogeneity to 0%. The overall effect sizes for stress (SMD = 0.29, 95% CI [0.10, 0.47]) and depression (SMD = 0.44, 95% CI [0.30, 0.58]) remained robust and unchanged after these exclusions. These findings indicate that although high heterogeneity was observed in anxiety, stress, and depression levels, the overall results of this systematic review and meta-analysis remained stable and were not significantly influenced by the heterogeneity (Figs. 8–10). The detailed statistical results of the sensitivity analyses are provided in the supplementary Table 5 for reference.

Fig. 9. Visual representation of sensitivity analysis results for stress levels.

This figure illustrates the results of the sensitivity analysis for stress levels using the leave-one-out method. a Each dot corresponds to the heterogeneity after excluding a particular study. The nodes are color-coded with a gradient: deeper blue indicates higher heterogeneity, while deeper yellow indicates lower heterogeneity. b Each dot corresponds to the effect size after excluding a particular study. The nodes are color-coded with a gradient: deeper blue indicates a higher effect size, while deeper yellow indicates a lower effect size. Both (a, b) should be viewed together to observe how the overall heterogeneity and overall effect size change after the removal of certain studies.

Fig. 8. Visual representation of sensitivity analysis results for anxiety levels.

This figure illustrates the results of the sensitivity analysis for anxiety levels using the leave-one-out method. a Each dot corresponds to the heterogeneity after excluding a particular study. The nodes are color-coded with a gradient: deeper blue indicates higher heterogeneity, while deeper yellow indicates lower heterogeneity. b Each dot corresponds to the effect size after excluding a particular study. The nodes are color-coded with a gradient: deeper blue indicates a higher effect size, while deeper yellow indicates a lower effect size. Both (a, b) should be viewed together to observe how the overall heterogeneity and overall effect size change after the removal of certain studies.

Fig. 10. Visual representation of sensitivity analysis results for depression levels.

This figure illustrates the results of the sensitivity analysis for depression levels using the leave-one-out method. a Each dot corresponds to the heterogeneity after excluding a particular study. The nodes are color-coded with a gradient: deeper blue indicates higher heterogeneity, while deeper yellow indicates lower heterogeneity. b Each dot corresponds to the effect size after excluding a particular study. The nodes are color-coded with a gradient: deeper blue indicates a higher effect size, while deeper yellow indicates a lower effect size. Both (a, b) should be viewed together to observe how the overall heterogeneity and overall effect size change after the removal of certain studies.

Sensitivity analyses were also conducted for studies deemed to have a high risk of bias according to the risk of bias assessment. The results indicated that the exclusion of these high-risk studies did not reverse the direction of the effect sizes for anxiety (SMD = 0.83, 95% CI [0.51, 1.06]), stress (SMD = 0.73, 95% CI [−0.13, 1.60]), or depression (SMD = 0.63, 95% CI [0.19, 1.07]). These findings further demonstrate that even with the inclusion of studies with a high risk of bias, the overall effect sizes remained robust, thereby reinforcing the reliability of the overall conclusions. Detailed data for the sensitivity analysis results after excluding studies with a high risk of bias are provided in the supplementary Table 6 for reference.

Discussion

The results of this meta-analysis demonstrate that exposure to virtual natural environments plays a significant role in regulating mental health and is an effective means of improving anxiety levels (SMD = 0.82, p < 0.001, large effect), stress levels (SMD = 0.577, p = 0.003, moderate effect), and depression levels (SMD = 0.621, p < 0.001, moderate effect) in healthy adults. Specifically, for anxiety levels, an exposure duration of 10–15 min is recommended (SMD = 1.11, p < 0.001, large effect). Both real-video content (SMD = 0.83, p < 0.001, large effect) and virtual-video content (SMD = 0.81, p < 0.001, large effect) in virtual natural environments are effective. However, for individuals over 30 years old, the effect is smaller (SMD = 0.46, p < 0.001, small effect). These findings suggest that when access to real natural environments is not feasible for improving mental health, brief exposure to virtual nature environments can serve as a temporary alternative. This is particularly relevant for populations such as office workers and students who have limited opportunities to engage with real natural settings. Although virtual environments cannot fully replicate the experience of real nature, their convenience and accessibility make them a practical option within urban contexts.

The significant positive impact of natural environments on mental health promotion is a widely recognized and scientifically validated conclusion^18,62,63. Natural environments offer diverse forms of exposure, ranging from basic contact with nature (such as forests, parks, and beaches)^16,17, to walking in natural settings^18,19, and engaging in physical exercise within these environments^20,21. These varied forms of engagement cater to the diverse needs of individuals seeking to improve their mental health. The high effectiveness of nature in promoting mental health is attributed to its ability to increase the distance between individuals and stressors, as well as to reduce the perception of these stressors⁶². However, with the continuous advancement of urbanization, individuals face increasing constraints in terms of both available time and accessible space, making it difficult to directly or indirectly escape these stressors^22–24.

Against this backdrop, rapid technological advancements have provided a virtual reality-based indirect approach to address this challenge⁶⁴. Reese et al.⁵⁰ demonstrated that virtual natural environments may serve as an effective therapeutic intervention for improving mental health. In their study involving 52 participants, a brief virtual forest walk was shown to enhance subjective well-being⁵⁰. Wen et al.⁶⁵ further expanded on this concept through a systematic review and meta-analysis of 15 studies across various age groups and clinical conditions. Their findings indicated that immersion in virtual natural environments effectively reduced anxiety, stress, and depression levels. However, the inclusion of a diverse range of ages and clinical conditions in Wen et al.'s study⁶⁵ limits the interpretability of their results and the generalizability of virtual natural environments as a therapeutic tool. This systematic review and meta-analysis corroborate Wen et al.'s⁶⁵ findings regarding the efficacy of virtual natural environments in reducing anxiety, stress, and depression. However, by focusing specifically on healthy adults, this review mitigates the risk of overestimating effects due to low baseline levels resulting from clinical conditions.

The efficacy of virtual natural environment exposure in reducing anxiety, stress, and depression can be explained through two well-established theories: the Attention Restoration Theory (ART) and the Stress Recovery Theory (SRT)^53,66,67. According to ART, natural environments are more effective than urban environments in providing physiological, emotional, and attentional restoration. Immersion in natural settings allows for the recovery of directed attention, which is often depleted in urban environments^68–70. Unlike urban environments, natural environments can capture involuntary attention without taxing the attentional capacity^59,71. SRT posits that exposure to natural environments enhances positive affect^53,72,73, accelerates stress recovery, reduces negative emotions, and promotes positive mood states, primarily through the activation of the parasympathetic nervous system^35,68,69. These theories are equally applicable to virtual natural environments, explaining their effectiveness in promoting mental health^53,66,67.

Additionally, the NN50 (number of interval differences of successive normal-to-normal intervals >50 ms) is closely related to parasympathetic nervous system activity. Higher stress levels are associated with a lower percentage of NN50^74,75. Monitoring the heart rate variability index NN50 can thus elucidate the effectiveness of virtual natural environments⁷⁶. For instance, Kim et al.³⁷ reported that exposure to virtual natural environments increased NN50 in participants, indicating that the parasympathetic nervous system can be effectively activated through virtual natural environments to reduce stress levels³⁷.

When exploring virtual natural environments, they are often compared with real natural environments, forming an interesting and valuable research variable. Browning et al.⁷⁷ and Reese et al.⁵⁰ provided comparability for this variable. Browning et al.⁷⁷ compared the effects of exposure to real and virtual natural environments in a sample of healthy college students, finding that both environments had similar positive impacts on positive affect. Reese et al.⁵⁰ also obtained similar results, showing that both real and virtual natural environments effectively increased positive affect, decreased negative affect, and reduced stress levels. The form of virtual exposure, as one of the variables, also affects intervention outcomes to some extent. Brivio et al.³³ compared two exposure forms—virtual reality and 360° panorama technology—and found that virtual reality had a better effect on reducing anxiety levels. Suseno et al.³⁶ compared virtual reality with 2D video and similarly found that virtual reality was more effective in improving anxiety levels. The virtual exposure scenario is another key variable affecting intervention outcomes. Yu et al.⁴³ compared virtual natural environments with virtual urban environments and found that virtual natural environments had a better intervention effect on reducing anxiety and depression levels compared to virtual urban environments. Therefore, virtual natural environments appear to be a more effective choice for improving mental health. However, virtual natural environments also contain many potential and interesting variables. For example, Hsieh et al.³⁹ introduced auditory elements into virtual natural environments and found that sound acts as a “catalyst” for improving mental health. Low-decibel water sounds combined with virtual natural environments can better enhance positive affect, while high-decibel sounds have a more pronounced effect on the resting sympathetic nervous system. Song et al.³⁸ further introduced tactile, olfactory, and gustatory sensations into virtual natural environments and found that taste is an important sense for enhancing psychological recovery. The interactions between touch and taste, as well as smell and taste, significantly enhanced the psychological recovery effect.

In the realm of mental health research, a multitude of variables interweave to shape the rich tapestry of the field. These multivariate elements not only carve out diverse pathways for mental health research, allowing investigators to explore mental health phenomena from various dimensions, but also provide a rational explanation for heterogeneity in meta-analyses. Specifically, for anxiety levels, subgroup analysis results indicated that “Age groups” was a primary source of heterogeneity (p = 0.004). Meta-regression results revealed that exposure duration was positively correlated with the intervention effect on anxiety levels, while participant age groups were negatively correlated with the intervention effect, accounting for 12.1% and 13.5% of the heterogeneity between studies, respectively. Sensitivity analysis results showed two key findings: first, studies by Vaquero–Blasco et al.⁴⁸ and Yang et al. (2014)³⁴ were sources of single-study inconsistency for stress and depression levels; second, simulated exclusion of studies with high risk of bias did not reverse the original results, indicating that the meta-analysis findings remain robust despite retaining these high-risk studies. However, this finding was not a reason for us to exclude any particular study. On the contrary, we chose to actively retain studies with higher heterogeneity because we acknowledge the existence of high heterogeneity and also permit the presence of multiple variables. Instead, what we need to explain and demonstrate is that, despite retaining these studies with higher heterogeneity, our meta-analysis results remain robust. In fact, it is the richness of these underlying variables that enables different studies to focus on unique areas, thereby presenting a diverse array of research priorities in the aggregate. This diversity not only enriches the content of mental health research but also legitimizes the presence of heterogeneity in meta-analyses.

In this systematic review and meta-analysis, several limitations were identified. The rigor of the included studies varied, with some studies falling short of optimal standards in experimental design and statistical analysis, thereby affecting the credibility of the results. Additionally, the high heterogeneity observed in the meta-analysis reduced the overall quality of evidence. Given the complexity of subjects in environmental psychology research, such variability is somewhat expected; however, caution is required when interpreting the results to avoid overgeneralization. Moreover, the presence of numerous potential variables, such as intervention duration and exposure scenarios, coupled with the limited number of included studies, precluded an in-depth exploration of the relationships between these variables and their interactive effects on mental health. This limitation restricts a comprehensive understanding of the field. Finally, the inclusion criteria permitted the inclusion of studies without controls. These uncontrolled studies, which cannot fully eliminate biases potentially introduced by confounding factors (such as time effects), may slightly weaken the causal inference strength of the meta-analysis results.

To address these limitations, future research can be expanded in several targeted directions. First, specific research questions should be further refined. For example, studies could investigate whether the addition of natural sounds enhances the stress-alleviating effects of visual virtual reality exposure in student populations, or examine the differential impacts of various virtual natural scenes (such as forests and oceans) on anxiety levels among office workers. Second, in terms of study design, factorial randomized controlled trials and other more systematic approaches are recommended. By manipulating independent variables such as “presence or absence of natural sounds” and “types of virtual scenes,” researchers can precisely measure their main effects and interactions on intervention outcomes. Meanwhile, strict control of sample size and representativeness, along with the use of randomization and blinding, can reduce bias and enhance the strength of causal inferences. In addition, the research dimensions should be broadened. On one hand, multivariate relationships should be explored in depth, such as the dose-response effects of different exposure durations and frequencies on mental health improvements. On the other hand, comparisons should be made between the psychological intervention effects of virtual and real-world natural environments, as well as the impact of interaction patterns between humans and virtual natural environments (such as the degree of interactive manipulation) on intervention outcomes. These efforts will provide more precise evidence for optimizing virtual natural intervention programs. Lastly, future research should equally emphasize psychological and physiological outcomes, integrating both to jointly support the validity of the conclusions.

In conclusion, exposure to virtual natural environments has a positive impact on mental health, effectively alleviating anxiety, stress, and depression levels. Therefore, when direct access to natural environments is not feasible, virtual natural environments can serve as a viable temporary alternative for improving mental health. However, given the lack of physiological indicators explored in the current studies, we recommend a cautious approach when considering virtual natural environments as a direct substitute. Further research incorporating physiological measures is warranted to fully understand the mechanisms and long-term effects of virtual natural environments on mental health.

Methods

The study has been registered in the International Prospective Register of Systematic Reviews (PROSPERO: CRD420251034979).

Literature search

This systematic review and meta-analysis employed a comprehensive and systematic approach to literature retrieval, covering five major authoritative databases: Web of Science (all databases), PubMed, ProQuest, Embase, and the Cochrane Central Register of Controlled Trials (CENTRAL). The search timeframe spanned from the inception of each database to March 6, 2025. The final search strategy was developed based on high-quality systematic reviews and meta-analyses previously published in the field¹⁸. Boolean logic operators (“AND”/“OR”) were used to combine subject headings and free-text terms to construct the search queries. To ensure transparency of the methodology, the standardized search process for the Web of Science database is illustrated as an example (see Table 4). The complete search strategy and parameter settings are provided in the Supplementary Table 1 for reference.

Table 4.

Comprehensive search strategy and methodological details

Database	Search query (Boolean operators: AND, OR)	Search Time Span
Web of Science (all database), PubMed, ProQuest, Embase, Central	#1 AB=(stress OR anxiety OR depression OR psychological stress OR emotional distress OR mental health OR Depression Anxiety Stress Scales OR DASS OR Self-rating Depression Scale OR SDS OR Hamilton Depression Rating Scale OR HAM-D OR Hospital Anxiety and Depression Scale OR HADS OR Self-rating Anxiety Scale OR SAS OR Hamilton Anxiety Rating Scale OR HAMA OR Beck Anxiety Inventory OR BAI OR State-Trait Anxiety Inventory OR STAI OR Profile of Mood States OR POMS OR visual analogue scales OR VAS OR Standard Stress Scale OR SSS OR Beck Depression Inventory-II OR BDI-II) #2 AB=(Virtual OR VR OR virtual reality OR virtual environment OR simulated environment OR immersive OR immersive virtual) #3 AB=(Outdoor OR environment OR Green OR outside OR environment OR forest OR woodland OR park OR greenspace OR open space OR green gym OR nature OR natural environment OR mountain OR garden OR wood OR wilderness OR countryside OR landscape OR beach OR sand) #1 AND #2 AND #3	Databases searched from inception to March 6, 2025.

Eligibility criteria

The inclusion and exclusion criteria for this systematic review and meta-analysis were strictly defined according to the PICOS principle. The detailed criteria for inclusion and exclusion of studies are summarized in Table 5.

Table 5.

Inclusion and Exclusion Criteria Based on PICOS

Category	Inclusion criteria	Exclusion Criteria
Population	- Adults: Defined as individuals aged 18–60 years, consistent with the World Health Organization’s age classification)85,86, The specific age range for each study was based on the reported mean age. - Healthy individuals: Participants without psychological or physiological disorders.	- Children (<18 years) and older adults (>60 years): Limited quantitative studies; inconsistent metrics; potential for increased heterogeneity and bias. - Unhealthy individuals: Physical/psychological conditions and comorbidities may interfere with effect assessment and reduce causal inference accuracy.
Intervention	- Virtual natural environments: Exposure to natural settings delivered via virtual reality, augmented reality, 360-degree videos, or 2D videos.	- Non-virtual environments: Exposure to real natural environments. - Non-natural virtual environments: Exposure to virtual urban or virtual office environments.
Comparator	- This study focuses solely on the effects of virtual natural environments on anxiety, stress, and depression. No control group is required.	/
Outcome	- Studies that utilized valid scales to accurately measure anxiety, stress, and depression.	- Studies using scales that do not accurately measure anxiety, stress, and depression. - Studies lacking accessible data (e.g., mean ± standard deviation for pre- and post-intervention measurements). - Studies for which data could not be obtained despite communication with the corresponding authors up to the time of publication. - Lack of pretest.
Study design	- Randomized controlled trials. - Randomized crossover trials. - Controlled trials. - Crossover trials.	- Case reports. - Animal studies. - Reviews. - Protocols. - Patents. - Long-term intervention.

Data extraction

To ensure the accuracy of data extraction, three co-authors (L.C., R.Y., and J.Y.) participated in the data extraction process. An independent researcher (L.C.) systematically entered the basic characteristics of the included studies into a standardized data collection template (Microsoft Excel 2019, Microsoft Inc., USA), which included information such as the first author, baseline/post-intervention data (mean ± standard deviation), demographic characteristics of the participants, intervention implementation parameters, duration of intervention, and frequency of exposure. After the initial data extraction, a second co-author (R.Y.) independently reviewed all extracted content. In cases where discrepancies or ambiguities were identified, a third co-author (J.Y.) provided final confirmation. For numerical data presented in graphical form, the GetData Graph Digitizer (Version 2.20; GetData, USA) was used to reverse-engineer the data from the graphs. When pre- and post-intervention data were not available, the corresponding authors were contacted via email or ResearchGate to request supplementary data. Studies were excluded if the required data remained inaccessible before manuscript submission.

Quality assessment of evidence

The quality of evidence in this systematic review and meta-analysis was assessed using the Grading of Recommendations Assessment, Development and Evaluation (GRADE) approach. The standardized evidence evaluation process was implemented via the GRADEpro GDT platform (available at https://www.gradepro.org). The assessment of evidence quality was structured around five core domains: risk of bias, inconsistency, indirectness, imprecision, and publication bias. The credibility of evidence for each outcome measure was initially evaluated independently by two reviewers (L.C. and R.Y). The final rating was determined according to the four-tier classification system established by Guyatt et al.⁷⁸ (high, moderate, low, and very low quality of evidence). Any discrepancies during the evaluation process were resolved by a third reviewer (J.Y.).

Risk of bias assessment

The risk of bias in this study was assessed using the Rob 2.0 framework (Revised Cochrane Risk of Bias Tool for Randomized Trials), which evaluates domains such as random sequence generation, allocation concealment, blinding, missing outcome data, and selective outcome reporting⁷⁹. The methodological reliability of the studies was categorized into three levels:

1. Low risk of bias: All domains were rated as low risk.

2. High risk of bias: Any domain was rated as high risk.

3. Some concerns: Studies that did not meet the criteria for either low or high risk of bias.

The assessment was independently conducted by two co-authors (L.C. and R.Y.). Any disagreements were resolved by a third co-author (J.Y), who provided the final arbitration.

Synthesis methods

All data integration and analyses were conducted using Stata 15 software (StataCorp. 2017. Stata Statistical Software: Release 15. College Station, TX: StataCorp LLC), while forest plots and sensitivity analysis results were visualized using the R programming language(version 4.3.3; R Foundation for Statistical Computing, Vienna, Austria). Given the cross-cultural scale differences in key outcome measurements among the included studies (including version iterations and country-specific norms), the standardized mean difference (SMD) was used as the core effect size metric, along with its 95% confidence interval (95% CI). The effect size of SMD can be interpreted as follows: trivial (SMD < 0.20), small (0.20 ≤ SMD < 0.50), medium (0.50 ≤ SMD < 0.80), and large (SMD ≥ 0.80)⁸⁰. Heterogeneity was assessed using the I² statistic, with the following thresholds: I² < 25% indicating negligible heterogeneity, 25%–75% representing moderate heterogeneity, and >75% reflecting substantial heterogeneity⁸¹. All statistical inferences were based on two-tailed tests, with p < 0.05 considered statistically significant. Although the Cochrane Handbook does not provide universal guidance on model selection, this study prioritized the random-effects model for the following reasons: (1) significant heterogeneity in baseline population characteristics among the included studies (including age distribution, educational level, differences in virtual reality exposure environments, and history of virtual reality exposure); (2) variations in virtual environment construction techniques across different platforms in the intervention protocols; and (3) the fixed-effects model’s insensitivity to potential heterogeneity⁸². Publication bias was visually assessed using funnel plot symmetry and quantitatively evaluated using Egger’s test. The trim-and-fill method was employed to account for missing studies.

When the meta-analysis revealed substantial heterogeneity, we further conducted subgroup analyses and meta-regression analyses to explore potential covariates that might influence the intervention effects. Subgroup analyses for categorical variables were stratified based on the following characteristics: Exposure environments (Real vs. Virtual), participant age groups (under 20 years old, 20–24.99 years old, 25–30 years old, over 30 years old), and virtual natural environment exposure duration (less than 5 min, 5–9.9 min, 10–15 min, more than 15 min). A p < 0.01 in the interaction test between subgroups was considered statistically significant⁸³. For continuous covariates, if the number of relevant studies exceeded 10, we employed a random—effects meta—regression model to examine whether the intervention effects were influenced by factors such as mean age and exposure duration⁸². A p < 0.05 in the regression analysis was deemed statistically significant⁸⁴. Additionally, to investigate potential biases from outliers, we used a leave-one-out method to assess the impact of individual studies on the overall effect and heterogeneity⁸². For studies classified as “high risk of bias” in the risk of bias assessment, we retained them to avoid selection bias arising from subjective exclusion⁸². Concurrently, we simulated the overall results after excluding these high-risk studies to quantify the impact of risk of bias on the final findings⁸².

Author contributions

Chen Lunxin: Writing—review and editing, Writing—original draft, Supervision, Software, Project administration, Methodology, Investigation, Data curation. Yan Ruixiang: Writing—original draft, Software, Methodology. Jialiang Yu: Validation, Project administration, Conceptualization.

Data availability

The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.

Code availability

The code used in the current study is available from the corresponding author on reasonable request.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41746-025-02057-4.