The ‘3-30-300 rule’ for urban nature exposes acute canopy deficits in 8 global cities

doi:10.21203/rs.3.rs-3960404/v1

Download PDF

Article

The ‘3-30-300 rule’ for urban nature exposes acute canopy deficits in 8 global cities

https://doi.org/10.21203/rs.3.rs-3960404/v1

This work is licensed under a CC BY 4.0 License

Journal Publication

published 19 Nov, 2024

Read the published version in Nature Communications →

Version 1

posted

You are reading this latest preprint version

The ’3-30-300 rule’ is a recently proposed metric which sets minimum standards for access to nature in cities for human wellbeing. It specifies homes, schools and workplaces should have a view of 3 trees, be located in a neighbourhood with over 30% tree canopy cover and be within 300m walk of a park. This metric is an important progression for assessing urban nature because it is easy to understand, highly local, and sets a pass/fail benchmark for green infrastructure. Using a global dataset of over 2.5 million buildings in eight cities, we show that most buildings fail the ’3-30-300’ rule due to inadequate tree canopy. The ‘3’ standard was met more often, while ‘300’ was patchy. Further analysis indicates that existing trees are too small for adequate canopy cover. Cities must invest in improving planting conditions to support tree growth and enhance governance to reduce premature removals and excessive pruning.

Scientific community and society/Geography

Earth and environmental sciences/Environmental sciences

Urban forests and green spaces are increasingly recognised as important urban infrastructure, with a range of studies identifying benefits including localised cooling ^1,2, flood risk reduction ^3,4, mental and physical health ⁵ and the support of urban biodiversity ^6,7. In a bid to improve access to nature in cities, strategies seeking to expand urban forests and open spaces in cities have become common ^8–10. These strategies commonly use metrics to track key strategic considerations such as tree canopy cover, physical access to nature and tree diversity. The metrics tend to either serve as high-level targets for action (targets for canopy cover, or annual planting counts such as those under the United Nations ‘Trees in Cities Challenge’¹¹), or as tools to monitor the progress of set plans using metrics such as species diversity and tree diameter¹².

In contrast to the considerable array of emerging metrics that serve as targets or management tools (see Supplementary 1), metrics that benchmark the adequacy of green infrastructure at a very local level (e.g. individual neighbourhoods) in terms of human needs are rare, particularly in urban forestry ¹². The metrics that do exist are often highly spatially aggregated, such as municipal canopy cover, or per-capita green open space ¹³.

A new metric proposed by Konijnendijk (2022) shows promise as a ‘measure of adequacy’ for urban green infrastructure. The ‘3-30-300’ rule specifies that every dwelling, educational facility and workplace should have a view to 3 trees, 30% canopy in the neighbourhood, and a walk of less than 300 metres to the nearest park. The local aspect of greening is important; a municipality with treeless streetscapes and a few large, well-forested parks may score well on aggregated metrics of canopy and per-capita greenery, but will be exposed as inadequate by the highly local ‘3’ and ‘30’ requirements. Given the well-documented heterogeneity and inequality of green infrastructure distribution in cities^14–16, this is a significant progression.

Unlike many previous metrics, the 3-30-300 rule is not just a measurement, nor is it aggregated: it is a ‘pass/fail’ benchmark that is highly local in focus. Metrics of adequacy play important roles in many parts of society, for example when traffic projections are used to prioritise highway spending¹⁷, school tests determine university access¹⁸, and Michelin stars determine the prestige and profitability of restaurants¹⁹. These powerful metrics are not simply measurements, but normative benchmarks, where achievements are deemed to pass or fail. In addition, the 3-30-300 measure has intuitive appeal and clarity; Konijnendijk (2022) notes that a similarly intuitive heuristic for species diversity (Santamour’s 10-20-30 rule) has become well-established in urban forestry circles²⁰.

The intentionally accessible, engaging design of the 3-30-300 rule seems to have worked. While it was first conceived in 2021 and only published in peer-reviewed literature in 2022 ^21,22, the metric has been promoted by the IUCN ²³ and has seen quite rapid diffusion, with at least six cities in Europe, the US and Canada adopting the measure, with more examples arising in international organisations, regional governments and NGOs²⁴. Academic studies applying the 3-30-300 rule are still rare. Nieuwenhuijsen et al. (2022) show that full achievement of the metric’s tests was associated with fewer psychologist and psychiatrist visits in a sample of Barcelona residents, while Battisti et al. (2024) used the rule to support spatial prioritisation of nature-based solutions for heatwave risk management in Turin. Browning et al. (2024) offer detailed advice to guide calculation of the metric. To date, it is not clear how cities worldwide currently perform with respect to the rule, and which of the different components of the rule are closer to fulfillment.

Here, we apply the 3-30-300 rule to a global dataset of eight cities, many of which are leaders in urban forestry and green space development: New York, Singapore, Seattle, Denver, Amsterdam, Buenos Aires and the inner-city municipalities of Melbourne and Sydney. In total we calculate achievement of the rule for over 2.5 million individual buildings.

Our first research question focuses on city performance. Does our selection of cities meet this new benchmark? Where cities fall short, how badly do they fall short? Our second question concerns the relationship between the component tests of the 3-30-300 rule. Are the results related? For example, do buildings with short walks to parks tend to also have good neighbourhood canopy, or views to trees? Could one of the three tests be redundant? We examine these correlations, particularly in light of the very different results we observe within individual cities in relation to each of 3-30-300 tests.

Third, we focus on Seattle and New York to understand why cities that seemingly have abundant views to trees (as established by the ‘3’ test) still can fall badly short on the ‘30’ canopy test. Detailed datasets for these cities enable us to plot both the density of planting, and the median size of trees planted near buildings in each city.

A fourth cross-cutting strand of enquiry examines how feasible it is to calculate the 3-30-300 rule for individual buildings at the city scale; practicality is vital to ongoing large-scale use of urban green space indicators, particularly in resource-constrained municipal teams²⁵. We yield insights in this area by approaching the analysis from a practitioner perspective – working with resources plausibly within reach of a municipal urban forestry/open space officer, approaching the calculation using existing data, and only ‘no-code’ software for data processing (i.e. using graphic user interfaces), recognising that many municipal urban forestry teams are primarily trained in arboriculture, ecology and landscape architecture rather than computer science. We present much of these reflective insights via Methods, and in our Discussion rather than as direct Results.

Our paper offers both an insight into the (grave) state of green infrastructure provision in major cities around the world, and also into the practicality of the metric that proposes to measure this. We conclude with a discussion of the prospects for improving the index, and indeed for future benchmarks for urban nature.

Overall performance: do cities meet the benchmark?

Cities had strongly different results for each of the three tests. Figures 1a and 1b map the achievement levels of each city for each test, highlighting the ‘30’ rule as a particularly weak point for most cities, and the ‘300’ rule as very patchy and heterogenous. Results for ‘3’ were often more positive. Interestingly, in New York, Melbourne and Seattle, almost every building had a view to three or more trees, but this did not correspond to impressive rates of achievement of the ‘30’ rule.

Overall achievement of all three standards within the rule was very poor in every city, with Figure 2 (Panel A) demonstrating that only Seattle had low rates of outright failure and modest rates of success; even this result can at best be considered a serious shortfall, with over 80% of the building stock failing the overall test. Given the diversity of methods, currency and quality of the open data that was assembled for this study, is important that these scores are not used to make comparisons between cities; that is neither the intent nor design of this research. It is unambiguous, however, that no city came close to meeting this basic minimum standard for urban green infrastructure.

Figures 1a, 1b and Figure 2 (panel A) show the results of a very simple binary test, with no indication of how close cities were to reaching the standard. To a degree, a strict standard is appropriate given it is argued to be a minimum standard for human wellbeing²¹. However, the binary nature of the test can obscure good progress, as is evident when examining Figure 2b, where cities with very low overall achievement of 3-30-300 test (e.g. Sydney, Denver, Singapore) are demonstrated to have significant proportions of their buildings in neighbourhoods with over 20% tree canopy cover – not quite enough to pass the test this metric poses, but close. Similarly, large proportions (approx. ~70%) of Amsterdam and Melbourne’s buildings are within 500m of a park (Figure 2c). In almost all cities, the ‘three’ test produced a more binary result – most cities had relatively few buildings with views of just one or two trees, and had large proportions with views to three trees and/or none.

Supplementary 2 offers examples of how 3-30-300 results can be mapped to visualise both outright achievement of the rule and, using a scoring out of ten, also visualise the extent of partial achievement to better support spatial prioritisations. Supplementary 3 shows how changes in the definition of key components of the test – such as the spatial definition of ‘neighbourhood’ – can lead to variances in results.

Correlations: does passing one test make it likely a building will pass other tests?

It may seem intuitive that having visible trees or better access to parks should lead to higher canopy results. However, our analysis of correlations between the metrics using the Spearman correlation coefficient showed very weak relationships; performance in one test does not reliably predict results in any other test (Figure 3). While p-values were significant, this is due to the substantial sample size and they are irrelevant in practice.

In neighbourhoods which provide good views to trees, why are canopy metrics not met?

Figure 1b shows that Seattle and New York performed excellently in the ‘3’ test, but had very different rates of achievement of the ‘30’ test (45% and 1% respectively). Both cities offer point data showing the location and canopy area of every individual tree in the city. This enabled us to calculate the median canopy area of all trees near buildings (i.e. within 30m, as per the ‘3’ test). We sought to infer whether the difference between the 3 and 30 results in and between each city was reflective of differences in tree size, and/or planting density. As the 30 test was conducted at neighbourhood scale (represented by census tracts in US cities) this was the unit of aggregation for which we sought medians and planting density.

Figure 4 demonstrates that tree size was a major difference between the New York and Seattle datasets. The median size of trees near buildings in Seattle was much higher than in New York. However, planting density (trees/ha) was significantly higher in New York, where the vast majority of median tree canopies were smaller than an average parking space.

It is important to recognise that these are asynchronous results collected using methods with different levels of precision and therefore this must not be understood as a fair comparison of each city, but rather a comparison of two urban forests seeking to achieve a goal by different means. In this case is clear that a lower density of large trees around buildings outperformed a denser planting of small trees in this comparison. It is also clear that both populations were not sufficiently large or dense to perform well in the ‘30’ test; a population with trees as large as Seattle’s but planted as densely as those in New York may have achieved much higher success. It is less clear that the predominance of very small trees in New York reflects immaturity of trees or simply constrained growth. We also cannot be certain that the larger size of Seattle’s median tree necessarily reflects an absence of these constraints.

This study has identified some broad patterns in the achievement of the 3-30-300 rule, while also demonstrating a range of challenges and uncertainties associated with its calculation. We discuss our overarching findings, then consider future prospects for the development of this indicator, noting a number of areas in which future indices might build on 3-30-300. A detailed discussion of our technical insights is included in a ‘Methodological Frontiers’ section, at Supplementary 4.

The index isn’t too tough; our cities are too grey (and our trees are too small)

The 3-30-300 rule, calculated at a city scale, does offer a clear and intuitive benchmark for green infrastructure provision in cities. It is somewhat demanding to calculate, both computationally and in terms of input data, but we have demonstrated that this is possible, and that the component measures are all relevant and independently valuable.

The chief insight we gleaned by applying this benchmark within eight cites is that full achievement of the 3-30-300 rule is very rare. The limitations imposed by the use of open data and relatively simple methods introduce the likelihood of some inaccuracy, but the consistency of this finding across eight diverse cities indicates that even with some margin of error, provision of green infrastructure for human wellbeing is inadequate in the cities we studied. We note that Battisti et al. (2024) found similarly poor results in Turin, and Nieuwenhuijsen et al. (2022) recorded low results in Barcelona, with both studies identifying particularly low achievement of the ‘30’ rule. This suggests that the typical built form we are familiar with in cities does not sufficiently support healthy tree canopy growth, nor allocate sufficient open space. It may be tempting to interpret this finding as an indication that the 3-30-300 hurdles are too high. We urge against this interpretation, particularly in the case of the ‘30’ test for canopy, for reasons we explain below.

It is concerning that every city in our study has a large majority of buildings that fail the ‘30’ test, given that the selection of 30% neighbourhood canopy is proposed as a bare minimum²¹, and studies have indicated that cover over 40% may be required to substantially lower daytime air temperatures ²⁶. This is a significant shortfall, with every city except Singapore and Seattle having fewer than 20% of buildings passing the ‘30’ test. By contrast, in half of the cities studied, 25-40% of buildings were in neighbourhoods with very little canopy (0-10%). The acute need for canopy as heatwave protection infrastructure is already evident in cities around the world; 2023 was the planet’s hottest year on record, and 25% of the global population faced dangerous levels of extreme heat ²⁷. If adaptation action is not taken, an increase in heat deaths of 370% is anticipated by mid-century, accompanied with a 50% increase in heat-related labour loss²⁸. As these extreme heat events become more common, it is urgent that cities reach safe levels of canopy cover.

As a number of studies contend that a lack of space makes achievement of 30% canopy infeasible in urban areas^21,24,29, we consider this a call for departures from the status quo of streetscape design, which often assume existing allocations of space to be permanent and unchangeable. We encourage a move beyond this notion; where sufficient space is not vacant, reallocation of street space may be needed. 30% canopy may indeed be infeasible if planting can be rejected on the basis that space is currently allocated to parking or roadways, or is deemed ‘unplantable’ due to generous clearance requirements associated with utilities or vehicle sightlines³⁰. This demands problem-solving, not an implicit capitulation to the tacit norm that vehicles and services hold greater rights to public space than green infrastructure elements. Our position aligns with the suggestion by Nieuwenhuijsen et al. (2022) that we ‘dig up asphalt’ to respond to the identified low achievement of 3-30-300.

A reconsideration of the priority of green infrastructure in relation to grey infrastructure is especially important given the finding that the ‘3’ result does not predict higher canopy; this indicates that there are many buildings that have good access to trees, but not to adequate canopy, because trees are too young and/or unhealthy to achieve good cover. This suggests that tree planting to date may not be providing sufficient space for successful tree establishment and growth, and that tree protection policies are not strong enough to avoid removal or excessive pruning of maturing trees, resulting in a predominance of younger and/or heavily pruned trees.

Patchy results for the ‘300’ test indicate that our sample of cities have deficiencies in green space provision, though the extent of this deficiency may be subject to some argument, depending on the precise configuration of the test; we consider these points in our discussion of methodological frontiers (Supplementary 4). Nevertheless, in many cities this result should serve as a prompt to expand programs that create urban parks.

Prospects for better benchmarks for urban nature

The 3-30-300 index’s simplicity and ‘rhyme’ is one of its strengths, but it does limit the consideration of a few important aspects of urban green infrastructure planning. We note a few particular points that might be addressed either by future revisions of the rule, or by combining it with other analytic tools (for example, as demonstrated by Battisti et al. (2024) in their use of both 3-30-300 and a heat vulnerability index to examine conditions in Turin).

We consider the most critical area of opportunity for metrics like 3-30-300 is in integrating considerations beyond just human wellbeing. While engaging, the existing metric is not necessarily a predictor of ecosystem service benefits or biodiversity benefits. Cooling, stormwater interception and habitat provision are examples of important ecosystem services that we might guess are higher in areas that achieve high 3-30-300 scores, but it is not clear that 3-30-300 achievement necessarily means optimal provision of these ecosystem services. For example, a street with a single exotic tree species planted in paved tree plots might plausibly meet the 3-30-300 rule, but will offer very little support in terms of stormwater management or biodiversity, due to low rates of understorey planting, stormwater infiltration, and the low habitat value of the trees to native species. More direct modelling of services like shading is now possible at the city scale, as demonstrated on recent projects such as Cornell’s ‘tree folio’ project ³², as is ecosystem connectivity ³³. This means that a revised (or additional) benchmark could reasonably provide more explicit consideration to ensuring benefits such as cooling, stormwater interception and everyday experiences of nature.

3-30-300 could also better incentivise the delivery of quality green infrastructure. The 3-30-300 rule does consider quality to some degree using size as a proxy (requiring ‘reasonably established’ trees in the ‘3’ test, and parks above a certain size in the ‘300’ test). However, the link between these proxies and the quality of parks and trees is quite tenuous. Recent studies have demonstrated how the quality of parks³⁴ and the health of trees³⁵ can be measured more precisely at the city scale.

Views to nature may also be somewhat underestimated by the ‘3’ rule. Pleasant natural views including green roofs, meadows, climbing plants and seascapes are all discounted by this test. Studies have demonstrated that non-tree green infrastructure including green roofs ³⁶ and green facades ³⁷ can offer psychological benefits. An approach to measurement of views that treats green elements more equably may be appropriate than a tree count; a number of studies have quantified visible urban greenery (not just trees) using computer vision (Labib et al., 2020, Browning et al. 2024).

Another area of potential is a grading that captures the 3-30-300 concept without demanding a sharp binary, whereby 29% canopy is just as much a failure as 1%. The examples of a notional grading demonstrated in Supplementary 2 ranged between 0 and 10, but this could be refined. Firstly, a graded scoring system might weight the components of 3-30-300, rather than scoring them equally (for example, canopy may be more important than tree views). Scorings that range from 0-10 might inadvertently position full 3-30-300 achievement as an aspirational maximum score, rather than a bare minimum for human wellbeing. Accordingly, gradings could be revised to reward buildings for exceeding the standard, and assign penalties for missing the standard. This could result a scoring that ranges from -5 to +5, with achievement of 3-30-300 as the ‘0’ point.

One final promising prospect is that the index could be calculated in a more sophisticated way at the national or global scale. Calculating 3-30-300 results for every building in a city is not straightforward, and it demands quite comprehensive data. While this may make 3-30-300 analysis difficult to access for many local governments, a number of major software firms own large, comprehensive spatial datasets, and have considerable capacity to access and process remote-sensed data (e.g. Google’s Earth Engine, or Microsoft’s Bing Maps API). The input data required to complete the 3-30-300 analysis demonstrated in this paper likely exists in these large corporate datasets, or could be derived from processing of 3D data and aerial images. With sufficient data, an analysis could consider individual dwellings (rather than buildings), facilitating the use of 3-30-300 results in real estate markets, or for benchmarking cities against each other (e.g. in the style of The Economist’s Global Liveability Index)³⁸ with relative methodological consistency.

Calculating the 3-30-300 index for multiple cities at the resolution of single buildings was undertaken in five steps. First, suitable open data was assembled, which also served to determine which cities were included in the study. Second, a generic process was prepared by which the data could be analysed to determine achievement of the 3-30-300 benchmark. Third, this process was tailored to each city’s particular input data. Fourth, recognising that full achievement of the metric is rare, we experimented with scoring systems to offer a clearer sense of how close each building might be to achieving the standard. Fourth, we examined correlations between our results for each of the tests (3,30 and 300) to understand how independent they are and therefore how necessary each rule is to include in the metric. Fifth, we determined the distribution of the age and planting density of trees in Seattle and New York to understand why the ‘3’ and ‘30’ tests seem so unrelated in the cities in our study.

This section describes each of these five steps, concluding with a discussion of the data and processing limitations encountered.

1. Identifying cities with appropriate data

We used ‘opentrees.org’ as an initial point of departure for assembling open data for each city, as it aggregates and maps freely-accessible tree inventory datasets from dozens of cities around the world. By browsing the map, candidate cities were identified, aiming to ensure representation of most continents; however, the key determinant of which cities formed part of this study was the completeness of their data. We aimed to secure data for at least one, and up to three, cities per continent. To be included in the study, the following datasets needed to be freely available for download in easily-used formats (e.g. csv, shapefile, geojson):

For the ‘3’ rule:

Tree inventory data (i.e. point data showing tree location)
Building footprints (i.e. the outline of shape of the structure, as seen from above)

For the ‘30’ rule:

Tree canopy data showing tree cover (not just green cover from NDVI)
‘Neighbourhoods’ – often census tracts were used as a substitute

For the ‘300’ rule:

Building footprints (same as for the 3 trees rule)
Roads (centreline data, not polygons)
Parks

While a few cities had adequate open data (e.g. Seattle, Melbourne), most cities we considered had one or two important datasets missing; useful canopy data in particular was scarce, though tree inventories and even building footprints led to the rejection of some candidates. Figure 5 summarises our search process.

Easily-accessed data was prioritised, with data for some large US cities proving relatively simple to find. Australian data was less comprehensive, with successful searches limited to only the small central-city municipalities of Melbourne and Sydney (each being a small fraction of the wider city); even in these relatively well-resourced state capitals, some augmentation of local data with state datasets was necessary. Assembly of data to ensure inclusion of a city from Asia, South America and Europe required significantly more effort. Open datasets in many European cities proved particularly difficult to navigate. This was firstly due to unique data structures (e.g. instead of a single citywide tree inventory, three unlinked datasets titled ‘trees in parks’, ‘trees in flower beds’, ‘trees in streets’ might exist, perhaps each broken into sets of 4 due to file size). Secondly, data was often stored in the local language and only searchable in that language; even with the aid of digital translation and careful assembly of these disparate datasets, datasets were usually found to be incomplete. By accessing a combination of municipal, national and EU data, the necessary components were assembled for Amsterdam, though some clipping and file type conversion was required to deliver useable datasets. Similar creative searching was required to prepare the components for an Asian and South American city.

At the completion of this search, data was assembled for New York, Seattle, Denver, Melbourne, Sydney, Amsterdam, Buenos Aires and Singapore. Regrettably, no suitable data for cities in Africa could be identified by our process, nor for a city in mainland Asia or the Middle East. Supplementary 5 details the individual datasets used. Generally, municipal open data was the main source for each city, supplemented by data either provided by higher levels of government (e.g. national or state level canopy data) or from large international datasets (e.g. an EU dataset of building footprints, roads and parks from OpenStreetMap, or large datasets of building footprints generated by Microsoft and Google generated by machine learning tools). In the case of New York, a very detailed tree inventory dataset was used arising from a recent study ³⁹. In Singapore and Buenos Aires, we relied on averaging ‘green view index’ data (calculated by MIT by processing images from Google Streetview) within neighbourhoods as a substitute for canopy cover ^40,41. In all cases, datasets were trusted to be reasonably error-free; we did not systematically validate each input dataset beyond checking their coverage and completeness.

This study’s use of open data means it has not produced a perfect ‘apples-with-apples’ comparison between cities. Calculations, input data and processing methods are broadly similar, but vary between cities. The 3-30-300 index is a heuristic, and our intent is to examine its application at large spatial scales; the somewhat different components used in each city do not detract fundamentally from the broad insights that this study aims to derive. However, different datasets may introduce a bias towards over- or under-estimation of achievement of the metric, and the specific dataset traits that served as limitations to this study are discussed at the end of Methods.

2. Development of generic analytical workflow

The work of Browning et al. (2024) was drawn on to establish our approach to each of the three tests that form the ‘3-30-300’ rule; the experts involved in that study assessed the suitability of a range of options for the calculation of each criterion. The methods employed in this paper generally follow this expert guidance, though deviations were necessary in a few instances. In all cases, geospatial software with a graphic user interface (GUI) was used; while the analysis started out using ArcMap 10.6.1 and QGIS 3.32, the analysis was moved to FME Workbench 2021 software due to its relative stability and efficiency of processing ⁴². The description that follows is generic and can be executed in any spatial software package, or using a myriad of coding approaches.

a. Testing whether three trees are in view of each building

This step of the analysis used only building footprints and tree inventory data. The workflow used in FME Workbench is shown at Supplementary 6, as is a figure showing the product of the analysis, with ‘lines of sight’ between trees and buildings demonstrated.

To determine whether each building was likely to have a view of three or more trees, each building footprint polygon was converted into a set of vertices (or points) at five metre intervals. This served to represent potential window locations on each building façade, assuming that most unobstructed facades on buildings will have one or more windows at reasonable intervals, and that obstructed facades could be identified in subsequent analytical steps. Second, a ‘nearest neighbour’ test was used to draw lines between each ‘window’ vertex and each tree location. The closest three trees within 30m of each window vertex were identified by this mechanism. The resulting lines represent sightlines from a notional window location at every 5m along the building façade. Two clean-up steps were required to derive meaningful insights: first, the ‘sightlines’ were intersected with a building layer, and any sightline that crossed a building was deleted, because buildings between a given tree and window represent visual obstructions. Second, as the conversion of building footprints into notional ‘window’ vertices created multiple sightlines from each building to each tree, the risk of double counting needed to be managed. Therefore, the sightlines were sorted by length, and then checked for duplicates to eliminate all but the shortest sightline to each tree for a given building. A simple statistical summary counting the number of trees per property was then conducted, and joined back to the building polygon layer. An estimated visible tree count per building was thus derived.

b. Testing whether each neighbourhood has 30% canopy cover

This was a relatively simple step of the analysis, using only the polygon layers representing canopy cover and neighbourhood boundaries (or census tract boundaries). We calculated the area of each neighbourhood in this study, then the canopy polygons were erased or clipped from the neighbourhood layer. The resulting clipped area – representing land that does not fall under canopy – was calculated by area again. The difference for each neighbourhood was computed, and this canopy cover was expressed as a simple percentage of the original neighbourhood area.

c. Testing whether each property is within a 300m walk of a park

This is a standard application of network analysis tools, available in many spatial software packages, and is recommended by Browning et al (2024). The analysis requires use of building, parks and road centreline layers. A routing algorithm is used to calculate and record the shortest distance between each property and the closest park; this is not conceptually complex but usually requires some setup, and is relatively computationally demanding at the city scale. The many intermediate steps in this analysis are shown Supplementary 6 as an annotated export directly from the GUI-based processing software used (FME Workbench).

In the case of the software used in this study, the most important tool used was called ‘ShortestPath’ and required two inputs: a road network, and a ‘from-to’ line essentially linking each property to its closest park (as the crow flies). The tool calculates the shortest path along the network to link the start of the line to its end. This meant that setup involved deriving ‘from-to’ lines using Nearest Neighbour analysis, which in turn required all origins and destinations to be single vertices (points) instead of polygons. Therefore, the analysis converted each building into a centroid, and each park into a set of points at 2m intervals. Nearest neighbour analysis used the building centroid as a base, and identified only the nearest park point, thereby avoiding duplication. With each building centroid now equipped with a set of coordinate attributes for its nearest park, generation of the ‘from-to’ line became possible. The ‘from-to’ line and the road layer were then fed into the ‘shortest path’ analysis for routing. This effectively measured a walking route along the road network for each from-to pair. However, testing a prototype of the analysis, we found that the tool calculates distances between intersections, sometimes routing a walker to an intersection some distance from their actual park destination. Therefore, a ‘chopper’ tool was created to break every road into 10m segments, ensuring that routing calculated routes that start and finish as close as possible to the relevant park and building points. Even with this improvement added, the shortest path tool struggled at times to find routes between buildings and parks where one end of the trip was that were slightly distant from a road (as shown at Figure X) so we added a ‘snapping distance’ to the shortest path calculator which automatically brought origin and destination vertices onto the network at a maximum distance of 100m. This introduces a level of approximation to estimates of walking distance, but avoids producing a high proportion of ‘path not found’ responses. Even with the snapping tool active, a portion of legitimate paths were not solved by the tool, and on inspection these were often in locations where buildings closely abutted parks with no interceding road – an example of this problem is included at Supplementary 6.

This posed a risk of underestimating the accessibility of parks to these properties, because recording the ‘300’ test result for these properties as ‘path not found’ would fail to recognise that these are often places that have very convenient park access. Accordingly, we added a final step which analysed any ‘path not found’ properties and determined whether they were in a crow-flies distance of under 50m. We replaced any ‘path not found’ attributes for these buildings with the measured distance, thereby recognising these locations as likely ‘pass’ results for the 300 test.

3. Tailoring the generic analysis to each city

New York was used to iteratively develop a prototype of the analysis. Being a large city with over 1 million buildings and over 6 million trees, standard desktop computers quickly proved inadequate for the required analysis on the city as a whole. To reduce processing demands, the city was broken into its component boroughs. That enabled the ‘3’ and ‘300’ analysis, but the spatial intersection of millions of tree canopy polygons with the city’s census districts proved excessively demanding (processing crashed frequently, and forecast processing times often exceeded 1 week). The tree inventory data for the city included an ‘area’ field representing canopy area for each tree, which enabled a much simpler analysis involving the summing of this area based on census districts. Otherwise, the analysis followed the generic workflow.

Melbourne, Sydney, Seattle and Denver proceeded smoothly through the generic analysis with minimal modification, though Seattle, Melbourne and Sydney had some inputs from larger datasets which required the data be clipped to the city’s boundaries. The small municipalities of central Melbourne and Sydney processed in a matter of hours, whereas the full extent of Denver and Seattle took some days, with network analysis and canopy analysis both proving quite computationally demanding.

Amsterdam and Singapore both did not offer comprehensive Open Data layers for their road networks, which required extraction and processing of OpenStreetMap (OSM) data, as well as an unsuccessful attempt to exploit national-level data provided by Microsoft offering machine-learned road networks ⁴³. While promising, these files proved unsuitable, as they were very large, unusually formatted (.tsv files) and required extraction using coding approaches (as distinct from the GUI-based approach employed in this paper). Network analyses with OSM data required a level of additional processing to ensure road segments were linked, but even after multiple runs and inspections continued to produce anomalous results at unacceptable rates, and ultimately network analysis approaches were abandoned in these cities, and crow-flies distances between parks and buildings were calculated instead simply by buffering each building by 300m and testing whether a park intersected each buffer. We note the work of Jafari et al. (2022), which offers detailed instructions on network setup using OSM data in the R programming language for city-scale analysis of walk distances. As this required coding it fell outside the scope of this study.

Buenos Aires and Singapore both do not provide canopy datasets, but were analysed as part of the ‘Treepedia’ study conducted by the MIT Senseable City Lab ^41,45. This free dataset quantifies canopy at street level at observation points at regular intervals in city streets using computer vision. While park canopy is excluded from this study, it gives a good sense of streetscape canopy cover, and with simple averaging within neighbourhood polygons it serves as a proxy for polygon-based canopy analysis in these cities. Amsterdam also did not have an immediately available canopy dataset, but the Dutch government offers one for the entire country; this was in a raster format so processing was required to ensure raster values were appropriately averaged within neighbourhood polygons in Amsterdam. Sydney’s canopy came pre-measured, as part of a dataset provided by the state government of New South Wales ⁴⁶. This data was at a sub-neighbourhood scale (‘modified mesh blocks’, often as small as just one side of the road on a city block) so this required aggregation to a larger unit , the ‘SA1’, which approximates a census district or neighbourhood, and usually includes 200-800 residents⁴⁷.

Buenos Aires and Singapore also required careful clipping of country-scale building datasets as part of data setup.

4. Quantifying internal correlations between 3,30 and 300 tests

The Spearman correlation coefficient was used to assess the strength and direction of relationships between the variables of interest. The Spearman correlation is particularly suitable for analysing non-normally distributed data and does not assume linearity. The correlation coefficients were calculated for the pairs of variables: canopy cover vs. tree views, tree views vs. distance to parks, and canopy cover vs. distance to parks. The magnitude of the coefficient reflects the strength of the association, and the sign reflects the direction of the effect, i.e. with values closer to +1 indicating strong positive relationship and -1 indicating a stronger negative relationship.

5. Examining New York and Seattle to understand differences between ‘3’ and ‘30’ results

Tree point data for New York and Seattle was of a standard that enabled analysis of tree populations within viewing distance of buildings used for the ‘3’ test (30m as defined in this study). By subsetting each dataset spatially, we isolated trees in this range, then for each census district we calculated the median area of each tree’s canopy, and the planting density per hectare of trees. This enabled us to visualise the distribution of tree sizes and platning density within the subset of each tree population that sits within 30m of buildings.

6. Limitations

Two key potential areas of limitation should be noted in this study’s methods, especially given that we seek to interrogate the difficulty and viability of calculating this heuristic metric within municipal teams. The first limitation relates to a set of flaws in the input data, which were inevitably incomplete and/or asynchronous given their diverse sources. The second key area of limitation is in the approaches we took to processing the data, which have introduced margins of error through their inherent assumptions and imperfect tools. Supplementary 7 summarises the issues observed through the course of this study, and offers a detailed description of the limitations in processing and data.

Data Availability Statement

The large quantities of data generated by this study are available on request.

Competing interests

The authors declare no competing interests in relation to the work above.

Ossola, A. et al. Small vegetated patches greatly reduce urban surface temperature during a summer heatwave in Adelaide, Australia. Landsc. Urban Plan. 209, 104046 (2021).
Stevens, H. R., Graham, P. L., Beggs, P. J. & Ossola, A. Associations between violent crime inside and outside, air temperature, urban heat island magnitude and urban green space. Int. J. Biometeorol. (2024) doi:10.1007/s00484-023-02613-1.
Haghighatafshar, S. et al. Efficiency of blue-green stormwater retrofits for flood mitigation – Conclusions drawn from a case study in Malmö, Sweden. J. Environ. Manage. 207, 60–69 (2018).
Szota, C. et al. Street tree stormwater control measures can reduce runoff but may not benefit established trees. Landsc. Urban Plan. 182, 144–155 (2019).
Astell-Burt, T. & Feng, X. Association of Urban Green Space with Mental Health and General Health among Adults in Australia. JAMA Network Open vol. 2 (2019).
Soanes, K. & Lentini, P. E. When cities are the last chance for saving species. Front. Ecol. Environ. 17, 225–231 (2019).
Kirk, H. et al. Building biodiversity into the urban fabric: A case study in applying Biodiversity Sensitive Urban Design (BSUD). Urban For. Urban Green. 62, 127176 (2021).
Brink, E. et al. Cascades of green: A review of ecosystem-based adaptation in urban areas. Glob. Environ. Chang. 36, 111–123 (2016).
Uittenbroek, C. J. From Policy Document to Implementation: Organizational Routines as Possible Barriers to Mainstreaming Climate Adaptation. J. Environ. Policy Plan. 18, 161–176 (2016).
Frantzeskaki, N. Seven lessons for planning nature-based solutions in cities. Environ. Sci. Policy 93, 101–111 (2019).
United Nations Economic Commission for Europe. Trees in Cities Challenge. https://sdgs.un.org/partnerships/trees-cities-challenge (2019).
Barron, S., Sheppard, S. R. J. & Condon, P. M. Urban forest indicators for planning and designing future forests. Forests 7, 1–17 (2016).
Jafrin, M. & Beza, B. B. Developing an open space standard in a densely populated city: A case study of Chittagong city. Infrastructures 3, 1–25 (2018).
Foster, A., Dunham, I. M. & Bukowska, A. An environmental justice analysis of urban tree canopy distribution and change. J. Urban Aff. 00, 1–16 (2022).
Schwarz, K. et al. Trees grow on money: Urban tree canopy cover and environmental justice. PLoS One 10, 1–17 (2015).
Astell-Burt, T., Feng, X., Mavoa, S., Badland, H. M. & Giles-Corti, B. Do low-income neighbourhoods have the least green space? A cross-sectional study of Australia’s most populous cities. BMC Public Health 14, 19–21 (2014).
Jafino, B. A., Kwakkel, J. & Verbraeck, A. Transport network criticality metrics: a comparative analysis and a guideline for selection. Transp. Rev. 40, 241–264 (2020).
Ingram, J., Elliott, V., Morin, C., Randhawa, A. & Brown, C. Playing the system: incentives to ‘game’ and educational ethics in school examination entry policies in England. Oxford Rev. Educ. 44, 545–562 (2018).
Daries, N., Moreno-Gené, J. & Cristobal-Fransi, E. Michelin stars shine brightly, but are they profitable? Int. J. Gastron. Food Sci. 24, (2021).
Galle, N. J. et al. Mapping the diversity of street tree inventories across eight cities internationally using open data. Urban For. Urban Green. 61, (2021).
Konijnendijk, C. C. Evidence-based guidelines for greener, healthier, more resilient neighbourhoods: Introducing the 3–30–300 rule. J. For. Res. (2022) doi:10.1007/s11676-022-01523-z.
Nieuwenhuijsen, M. J. et al. The evaluation of the 3-30-300 green space rule and mental health. Environ. Res. 215, (2022).
IUCN. Promoting health and wellbeing through urban forests – Introducing the 3-30-300 rule. https://iucnurbanalliance.org/promoting-health-and-wellbeing-through-urban-forests-introducing-the-3-30-300-rule/ (2021).
Browning, M. H. E. M. et al. Measuring the 3-30-300 rule to help cities meet nature access thresholds. Sci. Total Environ. 907, (2024).
Carmen, R. et al. Keep it real: Selecting realistic sets of urban green space indicators. Environ. Res. Lett. 15, (2020).
Ziter, C. D., Pedersen, E. J., Kucharik, C. J. & Turner, M. G. Scale-dependent interactions between tree canopy cover and impervious surfaces reduce daytime urban heat during summer. Proc. Natl. Acad. Sci. U. S. A. 116, 7575–7580 (2019).
Nature. Earth just had its hottest year on record — climate change is to blame. https://www.nature.com/articles/d41586-023-03523-3 (2023).
Romanello, M. et al. The 2023 report of the Lancet Countdown on health and climate change: the imperative for a health-centred response in a world facing irreversible harms. Lancet 402, 2346–2394 (2023).
Vuckovic, M., Schmidt, J. & Cornel, D. Nature-Based Solutions Analytical assessment of street-level tree canopy in Austrian cities : Identification of re-naturalisation potentials of the urban fabric. Nature-Based Solut. 3, 100062 (2023).
Croeser, T. et al. Finding space for nature in cities: the considerable potential of redundant car parking. npj Urban Sustain. 2, (2022).
Battisti, L. et al. Spatializing Urban Forests as Nature-based Solutions : a methodological proposal. Cities 144, 104629 (2024).
Cornell DAS Lab. Treefolio: NYC Tree Dashboard. https://www.treefolio.org/ (2023).
Kirk, H. et al. Ecological connectivity as a planning tool for the conservation of wildlife in cities. MethodsX 10, 101989 (2023).
Wu, W., Tan, W., Wang, R. & Chen, W. Y. From quantity to quality: Effects of urban greenness on life satisfaction and social inequality. Landsc. Urban Plan. 238, 104843 (2023).
Chi, D., Degerickx, J., Yu, K. & Somers, B. Urban tree health classification across tree species by combining airborne laser scanning and imaging spectroscopy. Remote Sens. 12, (2020).
Lee, K. E., Williams, K. J. H., Sargent, L. D., Williams, N. S. G. & Johnson, K. A. 40-second green roof views sustain attention: The role of micro-breaks in attention restoration. J. Environ. Psychol. 42, 182–189 (2015).
Elsadek, M., Liu, B. & Lian, Z. Green façades: Their contribution to stress recovery and well-being in high-density cities. Urban For. Urban Green. 46, 126446 (2019).
McArthur, J. & Robin, E. Victims of their own (definition of) success: Urban discourse and expert knowledge production in the Liveable City. Urban Stud. 56, 1711–1728 (2019).
Lin, J. et al. Relationships between urbanization, tree morphology, and carbon density: An integration of remote sensing, allometric models, and field survey. Urban For. Urban Green. 76, 127725 (2022).
Li, X. & Ratti, C. Mapping the spatial distribution of shade provision of street trees in Boston using Google Street View panoramas. Urban For. Urban Green. 31, 109–119 (2018).
Li, X. et al. Assessing street-level urban greenery using Google Street View and a modified green view index. Urban For. Urban Green. 14, 675–685 (2015).
Safe Software. About FME Workbench. https://docs.safe.com/fme/html/FME-Form-Documentation/FME-Form/Workbench/About FME Workbench.htm (2021).
Microsoft. GitHub - microsoft/RoadDetections: Road detections from Microsoft Maps aerial imagery. GitHub https://github.com/microsoft/RoadDetections (2022).
Jafari, A., Both, A., Singh, D., Gunn, L. & Giles-Corti, B. Building the road network for city-scale active transport simulation models. Simul. Model. Pract. Theory 114, 102398 (2022).
Seiferling, I., Naik, N., Ratti, C. & Proulx, R. Green streets − Quantifying and mapping urban trees with street-level imagery and computer vision. Landsc. Urban Plan. 165, 93–101 (2017).
New South Wales Government. Greater Sydney Region Tree Canopy to Modified Mesh Block 2019 | Dataset | SEED. https://datasets.seed.nsw.gov.au/dataset/greater-sydney-urban-vegetation-cover-to-modified-mesh-block-2019 (2020).
Australian Bureau of Statistics. Statistical Area Level 1. https://www.abs.gov.au/statistics/standards/australian-statistical-geography-standard-asgs-edition-3/jul2021-jun2026/main-structure-and-greater-capital-city-statistical-areas/statistical-area-level-1 (2021).

There is NO Competing Interest.

330300Supplementaries.docx

Download PDF

Journal Publication

published 19 Nov, 2024

Read the published version in Nature Communications →

Version 1

posted

You are reading this latest preprint version

The ‘3-30-300 rule’ for urban nature exposes acute canopy deficits in 8 global cities

Status:

Journal Publication

Version 1

Abstract

Figures

Main

Results

Overall performance: do cities meet the benchmark?

Correlations: does passing one test make it likely a building will pass other tests?

In neighbourhoods which provide good views to trees, why are canopy metrics not met?

Discussion

The index isn’t too tough; our cities are too grey (and our trees are too small)

Prospects for better benchmarks for urban nature

Methods

1. Identifying cities with appropriate data

2. Development of generic analytical workflow

a. Testing whether three trees are in view of each building

b. Testing whether each neighbourhood has 30% canopy cover

c. Testing whether each property is within a 300m walk of a park

3. Tailoring the generic analysis to each city

4. Quantifying internal correlations between 3,30 and 300 tests

5. Examining New York and Seattle to understand differences between ‘3’ and ‘30’ results

6. Limitations

Declarations

References

Additional Declarations

Supplementary Files

Status:

Journal Publication

Version 1