It is not always greener on the other side: Greenery perception across demographics and personalities in multiple cities

This study analyzes the discrepancies between objective and subjective urban greenery perceptions across five countries using street view imagery and a survey of 1,000 participants, revealing that while demographics and personality have little influence, an individual's geographic location is a primary factor shaping how they perceive green spaces.

The Gist

Imagine you are standing on a busy street corner. You look around and think, "Wow, this place feels very green and refreshing." But if you were to take a robot camera, count every single leaf and patch of grass, and calculate a "Green Score," the robot might say, "Actually, this street is only 20% green."

This paper is about that gap between what we feel and what the numbers say. The authors call it: "It is not always greener on the other side."

Here is the story of their research, broken down into simple concepts.

🌳 The Big Question: Do We See Green the Same Way?

For years, city planners have used two main tools to measure how "green" a city is:

1. The Robot Eye (Objective): They use street-view cameras and computer algorithms to count pixels of green. This gives a hard number called the Green View Index (GVI). It's like a nutrition label for a street.
2. The Human Eye (Subjective): They ask real people, "How green does this street feel?" This is a feeling, not a number.

The researchers wanted to know: Do these two things match? And if they don't, why? Is it because of your age? Your personality? Or maybe where you live?

🧪 The Experiment: A Global Taste Test

To find out, they didn't just look at one city. They set up a massive "taste test" involving:

• 1,000 People from 5 different countries (USA, Singapore, Chile, Nigeria, and the Netherlands).
• 400 Street Photos from 5 different cities (San Francisco, Singapore, Abuja, Amsterdam, Santiago).
• The Task: Participants were shown two photos side-by-side and asked, "Which one looks greener?"

They also asked the participants deep questions about their lives (age, income, personality traits) to see if those factors changed their answers.

🔍 The Surprising Findings

1. The "Feel" is Usually Higher than the "Fact"

Just like a movie trailer makes a film look more exciting than the movie itself, people almost always rated streets as greener than the robots did.

• The Analogy: Imagine a salad with a tiny bit of lettuce on top of a mountain of cheese. The robot counts the lettuce and says, "This is 5% green." But you, looking at the fresh, crisp lettuce, think, "This is a very green salad!"
• The Result: The human feeling of greenness consistently overestimated the actual amount of vegetation.

2. Your Personality Doesn't Matter (Much)

The researchers wondered: Do extroverts see more green? Do older people see less?

• The Verdict: No. Surprisingly, your age, gender, income, or personality type (like being nervous or outgoing) had almost zero effect on how green you thought a street was.
• The Analogy: It's like how everyone agrees that a loud siren is loud, regardless of whether you are a rock star or a librarian. Greenness perception seems to be a universal human reaction, not a personal one.

3. Where You Live is the "Secret Sauce"

If personality doesn't matter, what does? Your home address.

• The Finding: Where you live shapes your "green meter." A person from a very green city (like Amsterdam) might look at a street in a dry city (like Abuja) and think, "That's pretty green!" But a person from a very dry city might look at the same street and think, "That's just a few trees."
• The Analogy: Think of your brain as a camera with a custom filter. If you grew up in a desert, your filter is set to "low green," so even a small patch of grass looks huge. If you grew up in a rainforest, your filter is set to "high green," so you need a whole forest to feel impressed.
• The Takeaway: You can't just ask "people" what they think; you have to ask local people. A global survey might miss the mark because everyone's "green filter" is different.

4. It's About the Arrangement, Not Just the Distance

The team also looked at where the green was.

• The Myth: "Greenery right next to me feels greener than greenery far away."
• The Reality: Not necessarily. People liked greenery that was spread out evenly across the whole view (like a sprinkler system) more than greenery that was just a big clump right in front of them.
• The Analogy: Imagine a room with one giant potted plant in the corner vs. a room with small plants on every shelf. The second room feels "greener" overall, even if the total amount of plant is the same. The pattern matters more than the proximity.

🏗️ What Does This Mean for Our Cities?

The authors suggest that city planners need to stop relying only on the "Robot Eye" (the GVI numbers).

1. Don't just count leaves: If you want a city to feel green, you can't just plant trees in a few spots. You need to spread the greenery out so it feels like a natural part of the whole street.
2. Ask the Locals: When designing a park or a street, don't just use a global model. Ask the people who actually live there what they feel. Their "green filter" is the one that matters most.
3. The "Not Always Greener" Lesson: Just because a city has a high "Green Score" on a map doesn't mean people walking there will feel refreshed. And sometimes, a street with very little green can feel surprisingly green if it's arranged just right.

🎯 The Bottom Line

Greenery isn't just about biology; it's about psychology and context. We don't just see trees; we see our memories, our culture, and how the green is arranged around us. To make cities truly green, we need to design for the human heart, not just the computer algorithm.

Technical Summary

1. Problem Statement

Quantifying urban greenery is critical for climate mitigation, public health, and urban planning. Traditionally, this has been done through objective metrics (e.g., Green View Index or GVI derived from Street View Imagery) and subjective metrics (human perception surveys). However, a significant discrepancy often exists between the two: a location may be objectively "green" (high vegetation pixel count) but perceived as less green by humans, or vice versa.

Current research lacks a comprehensive, multi-city understanding of:

• The magnitude and nature of the gap between objective measurements and subjective perception.
• Whether demographic factors (age, gender, income) or personality traits significantly influence this perception.
• How the spatial arrangement of greenery (distribution vs. proximity) affects human perception compared to simple vegetation quantity.

2. Methodology

Dataset: SPECS

The study utilizes the Street Perception Evaluation Considering Socioeconomics (SPECS) dataset, a large-scale global survey.

• Participants: 1,000 individuals from 5 countries (Chile, Netherlands, Nigeria, Singapore, USA).
• Demographics: Detailed data collected on gender, age, income, education, race, and Big Five personality traits (extraversion, agreeableness, conscientiousness, neuroticism, openness).
• Imagery: 400 Street View Images (SVI) selected from 5 cities (Santiago, Amsterdam, Abuja, Singapore, San Francisco), with 80 images per city.
• Task: Participants performed pairwise comparisons (e.g., "Which place looks greener?") to generate perception scores.

Objective Metrics (Machine-Derived)

1. Green View Index (GVI): Calculated as the ratio of vegetation pixels to total pixels using panoptic semantic segmentation (ZenSVI package, trained on Mapillary Vistas). This captures trees, grass, and green terrain.
2. Spatial Entropy: A measure of the distribution (evenness) of greenery within the image. Calculated using Shannon entropy on sliding windows to determine if vegetation is concentrated or dispersed.
3. Greenery Proximity: Estimated using Depth Anything V2 (monocular depth estimation) to calculate the mean distance of vegetation pixels from the viewer.

Subjective Metrics

• Q Scores: Perception scores derived from pairwise comparisons using the Strength of Schedule (SOS) method, normalized to a [0, 1] range to match GVI. Scores were calculated based on specific image-participant location pairs.

Analytical Approach

1. Correlation & Bias Analysis: Pearson correlation and Bland-Altman plots to compare Q scores vs. GVI.
2. Statistical Testing: Wilcoxon signed-rank tests to identify systematic biases; Mann-Whitney U tests to compare spatial metrics between high-perception and low-perception image groups.
3. Predictive Modeling: A Random Forest model was trained to predict green perception Q scores.
   • Inputs: Demographics, personality, image features (GVI, entropy, depth, Sky View Index), and other perceptual indicators (safety, beauty, etc.).
   • Feature Importance: Conditional permutation importance was used to handle multicollinearity and identify the most influential variables.

3. Key Results

A. Objective vs. Subjective Discrepancy

• Moderate Correlation: There is a significant but moderate positive correlation ($R \approx 0.64$) between GVI and perceived greenery.
• Systematic Bias: Subjective scores consistently exceed objective GVI values. Humans tend to overestimate greenery compared to pixel-based counts.
• Context Matters: The correlation strength varies significantly by location. For example, US participants rating Singapore images showed the weakest correlation ($R=0.19$), while Chilean participants rating Amsterdam images showed the strongest ($R=0.70$).

B. Demographic and Personality Factors

• Negligible Influence: Standard demographic factors (gender, age, income, education, race) and most personality traits do not significantly influence greenery perception.
• The "Location" Effect: The city of residence of the participant is the only demographic factor with a significant impact. Cultural and environmental familiarity shapes the perceptual baseline.
• Generalizability: Because demographics matter little, global models using diverse samples are feasible, but they must account for the specific location of the evaluator.

C. Spatial Arrangement vs. Proximity

• Distribution is Key: Images perceived as "very green" (top 25% Q scores) had significantly higher spatial entropy. This means greenery perceived as high quality is dispersed throughout the scene rather than concentrated in one spot.
• Proximity is Secondary: The distance of greenery from the viewer (proximity) showed inconsistent effects. In some contexts, closer greenery increased scores; in others, distant greenery (e.g., continuous corridors) was preferred.
• Gestalt Principle: Perception appears to follow holistic principles where the overall spatial arrangement matters more than the immediate proximity of specific elements.

D. Predictive Modeling

• Top Predictors: The Random Forest model identified GVI as the strongest predictor, followed by Spatial Entropy and Sky View Index.
• Irrelevant Predictors: Demographics and personality traits had negligible feature importance.
• Model Performance: The model achieved an $R^2$ of 0.49, indicating that while objective metrics explain a large portion of perception, a significant portion remains unexplained by current features (likely due to cultural/contextual nuances).

4. Key Contributions

1. Global Scale Validation: First large-scale study (5 cities, 1,000 participants) to systematically compare objective and subjective greenery metrics across diverse cultural and urban contexts.
2. Demographic Insight: Challenged the assumption that demographics drive perception, finding instead that place of residence (cultural/environmental context) is the primary demographic driver.
3. Spatial Arrangement Metric: Introduced and validated Spatial Entropy as a critical factor in greenery perception, demonstrating that how greenery is distributed is as important as how much exists.
4. Bias Quantification: Quantified the systematic overestimation of greenery by humans compared to machine vision, providing a baseline for calibrating urban planning tools.

5. Significance and Implications

• Urban Planning: Relying solely on GVI or NDVI is insufficient for human-centric design. Planners must consider the spatial arrangement (dispersal) of greenery, not just total coverage.
• Policy Design: Green interventions should prioritize creating continuous, dispersed green corridors rather than isolated pockets of vegetation to maximize perceived greenness.
• Research Methodology: Future studies on urban perception should prioritize collecting data from local residents rather than global crowds, as local context heavily influences the baseline of what is considered "green."
• AI & GeoAI: Machine learning models for urban sensing should incorporate spatial distribution metrics (entropy) and be calibrated against local subjective baselines rather than global averages.

In conclusion, the paper argues that "it is not always greener on the other side" because perception is deeply rooted in local context and spatial configuration, not just the raw quantity of vegetation.