Potential drivers of urban green space availability in Latin American cities

Urban green space (UGS) encompasses natural and human-designed vegetated areas such as parks, yards, gardens, forests and green belts that provide ecosystem and cultural services affecting human well-being. While UGS can reduce heat, air pollution and support social and physical activity, poorly maintained UGS can also pose disservices. UGS is intertwined with cities' natural (topography, soils, climate), built (urban form, land-use) and socioeconomic contexts, which together shape its availability and quality. Research in Latin America (LA) remains comparatively limited and concentrated in a few cities, despite the region’s high urbanization, biodiversity and diverse climates. LA’s rapid, often unplanned urban growth and widespread informal settlements may have reduced green space and exacerbated inequities in access and quality. This study asks: how much UGS is available across LA cities, how is it spatially configured, and what is the relative importance of climate versus modifiable built and socioeconomic factors in explaining city-level UGS? The purpose is to fill regional knowledge gaps, benchmark UGS patterns across 371 large cities in 11 countries, and assess potential drivers to inform policy and equitable greening strategies.

Prior studies from the Global North and China document links between UGS and urbanization, density and socioeconomic status, but findings on density are mixed (negative, positive or null associations). Reviews highlight that wealthier places often have more UGS, though evidence in LA is inconsistent, with recent work showing negative relationships between greenness and socioeconomic status in some cases. LA research output is growing but remains concentrated in Brazil, Argentina and Chile and in a few large cities. Rapid urbanization, insufficient planning and informal settlement growth have often displaced green areas, while some cities (for example, Curitiba) pursued greening strategies. The climatic diversity of LA (tropical, temperate, arid) may modulate UGS benefits and disservices relative to evidence from cooler, wealthier contexts. Global-scale studies suggest precipitation and temperature are key determinants of UGS quantity, raising questions about how much city-level policy can offset climatic constraints. Equity literature indicates socioeconomic gradients in UGS access and use, but regional generalizations remain uncertain due to limited, non-representative samples.

Design and sample: City-level cross-sectional analysis of 371 large Latin American cities (>100,000 residents) in 11 countries from the SALURBAL project. Unit of analysis was each city’s urban extent (contiguous built-up area within administrative boundaries) delineated from the Global Urban Footprint (12 m) via spatial clustering of built-up pixels (2017). Population for urban extent was estimated from built-up area shares across sub-city units. UGS metrics (2017 unless noted):

• Greenness (NDVI): MODIS Terra MOD13Q1.006 (250 m, 16-day). Annual maximum NDVI computed per pixel; city-level median of annual maxima; water masked.
• Green space land-cover map: Sentinel-2 TOA median composite (2017) per city; supervised classification (clouds/water removed) produced 10 m binary green space map (grass, shrubs, forests, farmland). Validation in 11 cities showed overall accuracy 0.76–0.96 depending on climate zone. Green patches defined using Moore neighborhood; resampled to 30 m for landscape metrics.
• Configuration metrics: percent green space area; green patch density; clumpiness (aggregation, range −1 to 1); mean nearest-neighbor distance between patches (isolation). Per capita green space = total green area divided by city population.
• Urban parks (2020): Defined as publicly accessible, vegetated, delimited urban open spaces. Identified via web mining of Google-categorized “park” locations filtered by expert-curated key terms; cleaned to exclude non-parks (zoos, golf courses, cemeteries, green walls). Verified greenness via NDVI within 900 m buffer and delimited park boundaries using OpenStreetMap street networks (excluded locations without surrounding streets). Validated against municipal data in 77 cities with high agreement. Metrics: number of parks per 1,000 residents; park area per 1,000 residents. City characteristics:
• Natural: climate zone (tropical, temperate, arid; Köppen–Geiger), topography (flat vs hilly, slope >5°), coastal vs inland (boundary within 1 km of coastline).
• Built: city area (km²), population, population density, street intersection density (intersections per km², OSM 2017).
• Socioeconomic: GDP per capita (PPP, constant 2011 intl USD; 2015 subnational estimates), educational attainment (% aged ≥25 with secondary+), unemployment rate (% of labor force ≥15). Years vary by country (2005–2018, per national censuses and projections). Statistical analysis:
• Descriptive comparisons of UGS metrics across tertiles of built/socioeconomic variables and categories of natural variables; ANOVA with Tukey’s HSD for multiple comparisons.
• Random forest regressions (non-parametric ensembles) to assess relative variable importance and partial dependence for quantity metrics (NDVI, % green space, green space area per capita) and park metrics (parks per capita, park area per capita). Hyperparameters tuned via grid search (mtry optimization). Configuration metrics were not modeled with RF due to low explained variance (≤26%). Variable importance measured by permutation decrease in accuracy.
• Sensitivity analyses: adjusted multivariable linear regressions of each UGS metric on natural (climate, terrain, coastal), built (population, population density, intersection density) and socioeconomic (GDP per capita, education, unemployment) covariates, modeled categorically (tertiles or natural groupings). City area excluded due to multicollinearity (r=0.96 with population). Models re-estimated with random intercepts for country and with continuous predictors and interaction terms (climate × built/socioeconomic).

• Regional benchmarks: Across 371 cities, NDVI ranged 0.10–0.80 (mean 0.53); percent green space 4–72% (mean 41%); mean green space area per 1,000 residents 123,220 m². Mean park area per 1,000 residents was 2,142 m², far lower than total green space per capita.
• Climate zone variability: Arid cities exhibited lower NDVI, lower percent and per capita green space than tropical or temperate cities (arid coverage ~12–14% lower on average), but had higher urban park availability. Arid cities’ green space was more fragmented (higher patch density), more isolated and less aggregated.
• Topography/coastal: Hillier cities had higher NDVI, greater percent green space, and less fragmentation; flatter cities had more parks per capita. Coastal cities generally had lower NDVI and % green space than inland counterparts.
• Built environment: City area negatively associated with NDVI. Green space per capita decreased with higher population and higher population density; higher density cities had a smaller share of green space area and fewer parks per capita. Higher intersection density associated with lower NDVI, lower % green space and lower green space per capita, but with higher green patch density, greater isolation, and more parks per capita.
• Socioeconomic environment: Socioeconomically advantaged cities tended to have higher NDVI and % green space and less fragmented, more aggregated green space. NDVI showed a U-shaped pattern with GDP per capita and unemployment: cities at both low and high ends had higher NDVI than mid-level cities. Higher GDP per capita was associated with greater parks per capita.
• Random forest performance and importance: RF explained 65% of NDVI variance (RMSE=0.08), 41% of % green space (RMSE=9%), and 63% of green space per capita (RMSE=42,352 m² per 1,000 persons). Climate zone contributed 54%, 43%, and 28% of total variable importance for NDVI, % green space, and green space per capita, respectively. For park metrics, climate was much less influential (17% for parks per capita; 8% for park area per capita); education, population density, and intersection density were the most important drivers of park availability.

The study demonstrates that climate zone is the dominant driver of city-level vegetation-based UGS quantity in Latin America, implying fundamental climatic constraints on greening potential and on the ecosystem services UGS can provide. Arid cities, despite having less vegetated green space and more fragmented configurations, showed higher urban park availability, suggesting deliberate policy responses to create recreational spaces where natural vegetation is sparse or less suitable for recreation. Built environment pressures—particularly higher population and intersection densities—were associated with lower UGS quantity, highlighting trade-offs between compact urban form and green space availability. Socioeconomic gradients generally favored greener and more cohesive UGS in advantaged cities, although nonlinear (U-shaped) relationships with NDVI indicate that both underdevelopment and high investment can yield higher greenness for different reasons (undeveloped areas vs intentional park investments). The relative importance results clarify that while climate limits overall vegetative greenness, urban planning levers (education proxies, density, connectivity) are more influential for park provision, which may better reflect policy choices and demand. These findings address the research questions by quantifying UGS heterogeneity, linking it to natural/built/socioeconomic environments, and separating climatic constraints from modifiable urban characteristics, thereby informing realistic, context-sensitive greening strategies.

Environmental context, especially climate zone, largely predetermines how much vegetated green space a Latin American city can sustain, whereas the distribution and availability of urban parks appear more policy-responsive. Large regional disparities imply different capacities for greening as a heat-mitigation and health-promotion strategy; arid cities may face the greatest challenges for tree-based cooling and may need complementary approaches. Given unequal UGS distribution, ensuring equitable access and quality—particularly in disadvantaged and informal areas—should accompany expansion efforts. Methodologically, combining multiple UGS metrics is essential, as vegetation-based indicators and park-based measures capture distinct ecological and social health pathways. Future research should: (1) analyze within-city patterns, especially formal vs informal neighborhoods; (2) incorporate vegetation types and quality to assess ecosystem services/disservices; (3) track temporal dynamics of parks and UGS; and (4) refine causal understanding of nonlinear socioeconomic relationships and interactions with climate and urban form.

• UGS measures do not capture specific vegetation types or quality, limiting inference on particular ecosystem (dis)services.
• Temporal misalignment: UGS data (2017) and parks (2020) versus socioeconomic data from varying years (2005–2018). NDVI trends 2005–2017 were slow, mitigating some concerns for greenness metrics, but park–socioeconomic misalignment remains a limitation.
• Spatial mismatch: UGS and built/natural metrics were computed for urban extents, whereas socioeconomic variables used administrative city boundaries.
• Park data may contain geographic bias inherent to web-mined/geotagged sources; validation was at city level, but finer-scale biases may persist.
• Findings are city-level and may not generalize to within-city scales where inequalities and accessibility are best assessed.
• Random forest explained limited variance for configuration metrics (≤26%), so drivers of configuration were not analyzed via RF.