QGIS · Estonian 15-minute accessibility data

Regreening Prioritisation Matrix

Finding the empty parcels where new biodiversity reaches the most children and families — translating the ..._W15 walk-accessibility attributes into a defensible priority score and a do-this-first matrix.

For everyoneFor technical reviewers

This document works on two levels. Part 1 explains, in plain language, what we are doing and why — for decision-makers and the public. Part 2 sets out the data, formula, software and limitations so an independent GIS analyst or academic reviewer can check the method.

In one minute

We are finding the empty plots of land where planting new, biodiverse green space would do the most good — above all for children and families.

For every location in the city we ask four simple questions: do people live here, are children nearby, is nature already missing, and is it a lively walkable place? The answers become a single 0–100 priority score and a clear map of where to act first.

The approach rests on published health research and a transparent, repeatable method that anyone is free to audit.

Why these criteria — the evidence base

Biodiversity → child health

Adding forest floor, turf & planting to daycare yards measurably improved children's immune regulation within 28 days. Contact with biodiversity, not just "green", is the active ingredient.

Roslund et al. 2020, Science Advances

Reach where people are

GIS multi-criteria siting consistently finds population density the single strongest factor (weight ≈0.36) for where new green space delivers most benefit.

Frontiers Public Health 2024; Env. Systems Research 2024

Target the deficit (equity)

The 3-30-300 rule: everyone within 300 m of quality green space. <15% of people in 862 European cities comply — so prioritise the under-served, not the already-green.

Konijnendijk 2023; Nature Communications 2026

Empty lots work

A Philadelphia randomised trial: greening vacant lots cut residents' depression by up to 69% in poorer areas, and lowered stress/heart rate — cheap and scalable.

South/Branas et al. 2018, JAMA Network Open

The scoring model — regreen_score (0–100)

Household demand · DUs_Sum25%

Children reach · schools, daycare, play20%

Biodiversity deficit · inverted35%

Walkable context · daily services20%

score = 100 × (
0.25 · demand_norm
+ 0.20 · children_norm
+ 0.35 · (1 − green_norm) ← deficit
+ 0.20 · urban_norm
)

each term min-max normalised across all parcels

Why these weights — the academic argument. The weights are not arbitrary; they encode a hierarchy that the evidence supports, while the exact magnitudes remain an explicit, testable calibration.

Biodiversity deficit (0.35, the largest). The programme's purpose is to add biodiversity where people currently lack it, so deprivation carries the heaviest weight. Two lines of evidence justify this: the health case — direct contact with environmental biodiversity measurably improves children's immune regulation [1, 2] — and the equity case — under the 3-30-300 standard, fewer than 15% of residents across 862 European cities have adequate green access [4, 5]. Targeting the gap is the single strongest lever.

Household demand (0.25) and children reach (0.20). GIS multi-criteria siting studies consistently find population the dominant determinant of where new green space delivers benefit (the leading factor, with weights near 0.36 in AHP analyses) [6, 7]; demand is weighted second. Children are a distinct, high-value subset of that population — the health evidence is child-specific — but a complement to, not a replacement for, the population term. Together the two “people” terms (0.45) slightly exceed the deficit, so the model balances where the need is against how many people are reached: need takes the largest single weight, people the largest combined weight.

Walkable context (0.20). Green space is used and cared for best when embedded in active, walkable neighbourhoods; the 15-minute-city accessibility literature motivates rewarding integration into daily life, set equal to children as a supporting rather than leading factor.

Status of the weights. The literature robustly supports the ranking and direction of the four terms; the precise values (35/25/20/20) are an expert judgement by the research team, offered as a transparent hypothesis to be stress-tested with a sensitivity analysis or a formal AHP elicitation.

How it works — the four things we look at

Every location gets a score out of 100, built from four simple questions. The share next to each one is how much it counts towards the final score.

25%

Do many people live here?

▲ more homes = higher priority

A new green space helps more people when more people live next to it. We simply count the homes in each area, so that investment lands where it benefits the most residents.

Example: a dense residential block counts for far more than an empty industrial strip.

What we count: number of dwellings (individual flats — one block holds many), scored relative to the busiest area.

20%

Are children nearby?

▲ more children = higher priority

The health benefits of contact with nature are strongest for children. We look at how many kindergartens, crèches, schools, playgrounds and children's activities sit within a 15-minute walk — a good signal of where families with young children actually are.

Example: a block located between two kindergartens and a school scores higher.

What we count: kindergartens, childcare, schools, playgrounds, children's clubs, children's shops.

35%

Is nature already missing? — counts the most

▼ less nearby nature = higher priority

This is the heart of the programme, so it carries the most weight. If people can already walk to a park, trail, swimming spot or outdoor sports ground, the area is well served — so it scores lower for new greening. Where there is nothing green within a 15-minute walk, it scores higher. This is how we steer investment towards the people who are missing out today, rather than adding more green where there is plenty already.

Example: a housing estate with no walkable green space scores high; a neighbourhood next to a forest park scores low.

What we count (and want to be LOW): walking routes, dog-walking areas, nature destinations, swimming and bathing spots, parks, outdoor sport.

20%

Is it a real, lively neighbourhood?

▲ more daily life = higher priority

We favour places that are already part of everyday life — where people walk to the shop, the pharmacy, the doctor, the post office, the library or a bus stop. Greening here completes a living neighbourhood instead of landing in the middle of nowhere.

Example: a high-street corner beats a remote field with nothing around it.

What we count: grocery, pharmacy, GP, eatery, post office, library, public transport, daily-goods shop.

Putting it together. Each location is scored on all four questions, then the answers are blended using the shares above into a single 0–100 priority score. A worked example: an empty lot beside two schools, in a dense block, with no park within a 15-minute walk but next to shops and a bus stop, scores very high — lots of children and residents, no nature today, clearly a lived-in place. An empty lot far from any homes and already next to a forest scores low, because the few people there are already well served.

How a score is actually worked out. For each question we add up what is reachable within a 15-minute walk (for “homes,” how many dwellings the area contains). There is no fixed target — each area is scored relative to the busiest area in the city: the top area earns the full share, everyone else a proportion. “Homes” are individual dwellings (flats), not buildings. The four shares are then added into the 0–100 score.

Scoring wheel — the formula, as a wheel

Here the wheel is the score. Each wedge is one weighted term in the regreening formula, and the size of the wedge is its weight — so the biggest slice, the biodiversity deficit, is what moves the score most. The outer ring lists the attributes that feed each term. ▲ means more of it raises priority; ▼ means the term is inverted, so low outdoor-nature access is what scores high. Everything else in the dataset (health, culture, civic, retail) is mapped in the reference table below but does not enter the formula.

Biodiversity deficit · 35% ▼ Household demand · 25% ▲ Children reach · 20% ▲ Walkable context · 20% ▲

Wedge size = weight in the formula · ▲ raises priority · ▼ inverted (less access = higher priority)

Fair-Share (academic) regreening index mapped across Tallinn — This wheel, mapped onto Tallinn. The Fair-Share index applied to every parcel in the city — priority rising where nature is missing but people and children are present.⤢ Click to enlarge

Full attribute reference — every attribute, scored or not

Category	Dataset attributes (Estonian original in brackets)	Role in regreening score
Children & learningBenefit driver ▲	Kindergarten (Lasteaed) · Childcare (Lastehoid) · Basic school (Põhikool) · Upper-secondary (Gümnaasium) · Higher ed (Kõrgkool) · Vocational (Kutseõppeasutus) · Continuing ed (Täienduskoolitus) · Playground (Mänguväljak) · Kids' hobby & playroom (Huvi, Mängutuba, Huvikeskus) · Children's shop (Pood-Lapsed) · underlying demand: residents DUs_Count / DUs_Sum	Children reach ↑ + demand (DUs)
Health & careBenefit driver ▲	Pharmacy (Apteek) · GP (Perearst) · Specialist (Eriarst) · Dentist (Hambaarst) · Ambulance (Kiirabi) · Care home (Hooldekodu) · Day centre (KOV-Päevakeskus) · Health service (Teenus-Tervis)	Context ↑ vulnerable-group benefit
Nature & recreationBiodiversity gap ▼	Walking routes (Liikumine-Jalutamine) · Dog walking (Koduloomad) · Nature destination (Sihtkoht) · Swimming (Suplus) · Theme park (Teemapark) · Outdoor sport (Välisport: staadion, tennis, ballcourt, pool)	Biodiversity deficit ↓ scored inverted — the core
Everyday & foodLived-in context ▲	Grocery (Toidupood) · Market (Turg) · Daily goods (Tarbekaubad) · Eatery (Toidukoht) · Bar / café / pub (Toidukoht-Bar) · Post (Post)	Context ↑
Movement & accessLived-in context ▲	Transit terminal (ÜT-terminal) · Bike parking & rental (Rattaparkla, Rattarent) · Car rental (Autorent) · Fuel (Kütus)	Context ↑ (transit)
Community & cultureLived-in context ▲	Local government (KOV) · County centre (Maakonnakeskus) · Church (Kirik) · Culture centre (Kultuurikeskus) · Public library (Raamatukogu) · Cinema (Kino) · Museum (Muuseum) · Theatre (Teater) · Gallery (Galerii) · Hall & stage (Saal, Lava) · Nightclub (Ööklubi) · Police (Politsei) · Fire/rescue (Päästekomando) · Civil shelter (Varjumiskoht) · Cemetery (Kalmistu) · Public toilet & waste (WC, Konteinerid)	Context ↑ (library)
Retail & servicesLow relevance	Clothing & textile (Riided, Tekstiil) · Electronics (Tehnika) · Furnishings (Sisustus) · Books/stationery/music (Raamatud, Kirjatarbed, Muusika) · Plants (Taimed) · Hardware (Ehitus) · Sport & bikes (Sport, Rattad) · Hobby/pets/vehicles/optics (Hobi, Koduloomad, Mootorsõidukid, Prillid) · Beauty & care (Pood-Ilu, Teenus-Ilu, Massaaž) · Gym & indoor sport (Sisesport, Jõusaal, Ujula) · Home services (Kodu, Laenutus, Print, Autopesu) · Hotel (Majutus)	—

Note: the role labels above describe the whole category. Only the specific attributes in the scored baskets actually enter the formula — e.g. within Health & care only pharmacy and GP feed walkable-context, while ambulance, dentist, specialist and care home are not scored. Car rental, fuel and transit sit under movement/services and are not part of the biodiversity (nature) term.

The prioritisation matrix — add greenery, or address services first?

The four colours match the index map. As the greening case weakens (deep green → orange), the binding constraint shifts from missing nature to missing everyday services. The top two classes say “add greenery”; the bottom two say “greenery helps, but service access is the real gap.”

Greening priority →

DEEP GREEN · HIGHEST INDEX

1 · Add greenery — top priority

Dense, child-rich areas that lack nearby nature but already have everyday services within reach. New biodiverse green here reaches the most people and completes the neighbourhood.

→ Plant first.

LIGHT GREEN · HIGH INDEX

2 · Add greenery — strong case

A clear greening gap with plenty of people to benefit. The case for new nature is strong; act here alongside the top tier.

→ Plant / pipeline.

LIGHT YELLOW · LOWER INDEX

3 · Greenery helps — but services are the gap

These areas score lower mainly because everyday services are sparse — not because nature isn’t wanted. Greenery still adds value, but improving access to shops, care and transit matters as much.

→ Pair greening with services.

ORANGE · LOWEST INDEX

4 · Services first — greenery is secondary

The dominant shortfall is missing everyday services (and often few residents). Greening is a welcome co-benefit, but the primary intervention is basic amenities and access.

→ Address services; green opportunistically.

Deep green = add greenery · Orange = service access is the main gap

◀ greenery is the answerservices are the main issue ▶

The feasibility axis — your next layers

is_empty

Vacant / transformable

The precondition. Only empty (or convertible) parcels can actually be regreened — gates everything.

is_public

Ownership

Public land = act directly. Private = strategic target needing negotiation.

GRD_500m / GRD_1km

Surrounding density

Already in your data. Dense surroundings = more people reached per intervention; very low density caps impact.

Workshop variants — Team 1 and Team 2

During the workshop, two teams proposed their own regreening priorities. Each is rebuilt below as a wheel using only the attributes in our dataset, with equal weight per category (20% each) and a presence-based reading (more of a category within a 15-minute walk → higher priority). Where the data cannot represent a category, it is flagged in grey.

Team 1

Team 2

Team category	Mapped to our data	Status
T1 · Streets (home streets)	Dwellings (DUs) as a residential proxy	~ Proxy — no street layer
T1 · Education	Kindergarten, childcare, schools, higher & vocational	✓ Full
T1 · Shops — food (parking lots)	Grocery, market, eatery, bar	~ Partial — parking lots ✗ no data
T1 · Services	Pharmacy, post, bank, fuel, beauty, home, WC… (Teenus*)	✓ Full
T1 · Dense housing	Dwellings (DUs_Sum)	✓ Full (same signal as “streets”)
T2 · Home (kodu)	Dwellings (DUs)	✓ Full
T2 · Teenused – TOIT (food)	Grocery, market, eatery, bar	✓ Full
T2 · Liikumine (movement)	Walking, dog-walking, nature destinations, swimming, sport	✓ Full *
T2 · Huvid (interests)	Hobby, hobby centre, culture (cinema, museum, theatre…)	✓ Full
T2 · Haridus (education)	Kindergarten, childcare, schools, higher & vocational	✓ Full

What doesn't map — read before using.
Team 1: “parking lots” have no equivalent attribute and are dropped; “streets / home streets” has no street-network layer, so it is proxied by dwelling presence — which is also what “dense housing” uses, so those two categories collapse onto the same signal (dwellings effectively counts twice). Separating them would need a street layer and a parking layer.
Team 2: all five categories map. One caveat: “liikumine” is scored here as presence (more movement options → higher), the opposite of our main method, where missing nearby nature is the priority. If Team 2 meant “regreen where outdoor movement is lacking,” that term should be inverted.
Both scores use equal weights and presence-based normalisation — they can be rank-weighted or inverted on request. Produced by Regreening_team_scores.py as team1_score and team2_score.

Three narratives — and what each says about the city

Each wheel is really a different story about what a good city is and where green belongs. None is more correct than another — they optimise for different values. Here is what each narrative sees, and what changes when you follow it.

The Fair-Share city

our research-based (“academic”) narrative

The storyA fair city gives everyone — above all children — nature on their doorstep, so plant where nature is missing but families are present.

Reads the city bywhat is absent: green deficit, child density, equity.

Lights updense, child-rich neighbourhoods with no park within a 15-minute walk — however quiet.

Strengthtackles health inequality; grounded in published evidence.

Blind spotcan pick low-visibility places; leans on a nature-access proxy, not measured vegetation.

The Active city

Team 1's narrative

The storyGreen should reinforce the parts of the city that are already busy and built-up — homes, streets, schools, shops and services.

Reads the city byits built fabric and footfall: density and amenities.

Lights updense, central, well-serviced blocks where many people pass every day.

Strengthhigh visibility, many daily users, builds on existing investment.

Blind spotadds green where some may already exist; can widen the gap for under-served areas (and streets/parking aren't in the data).

The Everyday-life city

Team 2's narrative

The storyGreen should support daily wellbeing — home, food, movement, hobbies and learning all within reach.

Reads the city byquality of everyday life: services, recreation, culture, education.

Lights upcomplete, lived-in neighbourhoods that already have amenities and leisure.

Strengthholistic wellbeing; supports active, sociable lifestyles.

Blind spotrewards already-equipped areas; treats existing recreation as a plus, so may overlook places that lack it.

How alike are the three? — the numbers

We scored all 5,966 parcels on each of the three indices and compared them directly. Two questions decide the strategy: how strongly do the indices agree (correlation across every parcel), and how much do their shortlists overlap (the top 20% each index would actually fund). The answer is not a clean three-way split.

Team 1 ↔ Team 2 — near-identical

Across every parcel the two team indices correlate at r = 0.96; their top-20% shortlists overlap by 88%. Statistically they are one narrative under two names.

Pearson r 0.96 · 1,053 of 1,194 shared

Fair-Share ↔ teams — related, not opposed

Fair-Share is not the teams' opposite: it correlates r = 0.83 with Team 1 and r = 0.68 with Team 2. All three lean on the same dense, populated city — they share most of their priorities.

Pearson r 0.83 / 0.68

The decisive 19%

Where they part is the tail: 222 parcels — about a fifth of the Fair-Share shortlist — are sites neither team picks. They are quieter, under-served places that rank well below the teams' funding cut-off (≈ 30 / 100 on the team indices).

222 of 1,194 · team score ≈ 30

The real finding — it is not three narratives, nor two, but a shared core plus one fault line. Of the 1,194 parcels in any index's top-20% shortlist, 831 (about 70%) are chosen by all three — a large agreed core to start on whatever you believe. The two team indices are effectively the same map (r = 0.96). The one genuine choice is at the margin: the Fair-Share index redirects 222 sites (19% of its shortlist) to dense, child-heavy, green-deprived neighbourhoods that the team lens ranks well below its funding cut-off, because those places have less footfall and amenity to register. The decision is not “which of three stories,” but: do you also fund the 222 the equity lens reveals?

What changes if you follow each narrative

If you follow…	Fair-Share (academic)	Active city (Team 1)	Everyday-life (Team 2)
Shortlist overlap with the other two	shares 831-parcel core; adds 222 unique sites	88% identical to Team 2; 80% inside Fair-Share	88% identical to Team 1; 70% inside Fair-Share
Green goes to…	quieter, under-served, family-heavy areas missing nature	dense, central, high-traffic blocks	complete neighbourhoods with services & leisure
Who gains most	children & residents currently without nature	the many people already in busy areas	residents enjoying daily amenities
Headline win	fairness & child-health equity	visibility & footfall	everyday quality of life
Main risk	some chosen sites are less prominent	green-poor areas stay green-poor	same — under-served areas overlooked
City it builds	a more equal green city	a more concentrated green city	a more liveable green city

What the choice costs and buys. All three indices already agree on 831 parcels (~70% of every shortlist) — start there whatever you believe. Because Team 1 and Team 2 are near-identical (r = 0.96), choosing “a team narrative” is a single option, not two. The one real lever is the 222 sites (19% of the Fair-Share shortlist) that the team lens cannot see: fund them and you close gaps in dense, child-heavy, green-deprived neighbourhoods; skip them and greening concentrates in the already-thriving core. Per-parcel scores are in acessiblity tallinn.csv (join by fid; columns regreen_score, team1_score, team2_score).

Part 2 · for technical review

Method, data and limitations

This part documents the methodology in full so a GIS analyst or academic reviewer can reproduce and critique it. It mirrors the standalone technical report (Regreening_Methodology_Report.docx) and the delivered PyQGIS scripts.

Method at a glance

1 · Inputs (grid + accessibility)→ 2 · Group into 4 terms→ 3 · Normalise 0–1→ 4 · Weight & invert→ 5 · Score 0–100→ 6 · Quantile tiers→ 7 · Diagnose drivers→ 8 · Feasibility (next)

Input data and coordinate system

The layer is a polygon grid of 125 m × 125 m cells (cell side derived from the coordinate spacing in the data), carrying INSPIRE/EEA-style identifiers (e.g. GRD_ID “125mN6611375E595375”). Coordinates (E ≈ 595,000; N ≈ 6,611,000) are consistent with the Estonian national projected CRS L-EST97 / EPSG:3301 — to be confirmed in the layer properties. A projected, metric CRS is appropriate because the analysis is grid-based and the accessibility metric is distance-defined.

Each POI-…_W15 field is a continuous “within-15-minute-walk” accessibility value (sample values ≈ 0.0006–7.5), i.e. a distance-decayed accessibility surface rather than a raw count of reachable places; the model treats these as interval-scaled access intensities. Dwelling units (DUs_Sum) provide the demand measure. Two data-quality features shaped the implementation: several numeric fields are stored as text with “-” / blank placeholders, and field names contain Estonian diacritics (ä, õ, ü, š) and parentheses.

Normalisation

Each of the four terms is reduced to one value per cell (a single field, or the sum of its member attributes) and min-max normalised to [0, 1]. Because the inputs are non-negative and their minimum is effectively zero, this simplifies to division by the layer maximum:

x̂ = x / max(x) ( max taken across all cells in the dataset )

Observed maxima in the working dataset were: dwellings 774, children 83.9, biodiversity-access 22.4, walkable-context 114.9. Because normalisation is relative to the dataset maximum, scores are comparable only within this extent and will shift if the study area or its maximum changes.

Weighting and the composite score

Weights sum to 1.0 and encode the evidence in Part 1: the biodiversity deficit carries the largest weight (0.35) as the core of a biodiversity-led, equity-driven intervention; household demand (0.25) reflects the dominance of population in green-space siting studies; children reach (0.20) operationalises the child-health objective; walkable context (0.20) keeps interventions in active neighbourhoods. The two “people” terms together (0.45) make reaching residents and children the largest combined emphasis. Weights are expert-set and should be treated as a hypothesis to be calibrated.

regreen_score = 100 × ( 0.25·D̂ + 0.20·K̂ + 0.35·(1 − Ĝ) + 0.20·Û )

where D̂, K̂, Ĝ, Û are the normalised demand, children, biodiversity-access and context terms. The biodiversity term is inverted as (1 − Ĝ) so that low current access scores high. The model is linear-additive and fully compensatory (a strong term can offset a weak one).

Definitions & scoring detail — reviewer clarifications

“Homes / demand” = dwelling units, not buildings. DUs are individual dwellings (flats); one apartment block holds many. The demand term uses DUs_Sum, the dwellings within each 125 m cell.
Catchment. The children, biodiversity and walkable-context terms use the _W15 fields — destinations reachable within a 15-minute walk. Demand (DUs) is the cell's own dwelling count, not a walking catchment.
How each term becomes 0–1. Member attributes are summed per cell, then divided by the highest value found anywhere in the dataset (min-max, min ≈ 0). A term earns its full share only at the single highest cell; every other cell gets a proportion. The scale is relative to this dataset, not an absolute target (see limitations).
What the biodiversity term contains — and excludes. Only outdoor-nature/recreation access: walking routes, dog-walking, nature destinations, swimming/bathing, theme park, outdoor sport. Ambulance (Kiirabi), car rental (Autorent) and fuel (Kütus) are not in it — they enter neither the biodiversity term nor the main formula.
Substitution inside a term (the children question). Each term is an additive sum across destination types, so it is fully compensatory — many childcares can offset the absence of a school; no “at least one of each type” rule is applied. If diversity should be required, the term can instead count distinct types present, cap each subtype, or use a geometric mean.
Where the weights come from. The 35 / 25 / 20 / 20 split is an expert judgement by the research team, informed by the literature rather than computed from it: the evidence supports the direction and ranking (population is the strongest factor in green-space siting; equity and child-health justify the heaviest weight on the deficit), but the exact numbers are calibratable. A sensitivity analysis or AHP is recommended.
Workshop process. Teams 1 and 2 defined their priority categories during the session; the index scores were computed afterwards by mapping those categories onto the available attributes (equal weights, presence-based). The teams did not see or tune numeric scores in the workshop.

What the method produces (the result)

For each of the 5,966 parcels the method outputs a 0–100 priority score, a 1–5 tier, and a per-term contribution breakdown. The actionable outputs are: (1) a prioritised map; (2) the top quintile — about 1,063 parcels — under the Fair-Share method; and (3) across all three narratives, a consensus shortlist of parcels every method ranks top-priority, recommended as the first wave of action.

Classification and diagnostics

For mapping, scores are reclassified into five quantile tiers (each ≈ 20% of parcels; tier 1 = top priority), in preference to fixed cut-offs because unpopulated cells carry a non-trivial deficit baseline (~35 points). Because the score is additive it is also decomposed into the points contributed by each term (c_demand 0–25, c_children 0–20, c_greendef 0–35, c_context 0–20, summing exactly to the score), and each parcel is tagged with its dominant driver — the recommended tool for auditing unexpected values.

Implementation in QGIS

An initial single Field Calculator expression returned NULL for every feature. Three compounding causes were identified: the source layer is read-only (delimited-text / WKT); numeric fields stored as text made arithmetic evaluate to NULL; and field references with Estonian diacritics did not resolve at evaluation time inside a nested-aggregate expression. The model was re-implemented as PyQGIS scripts that match field names via Unicode NFC normalisation, cast values tolerantly (treating “-”, blanks and text as 0), compute the layer maxima in Python, and write a new GeoPackage. The pipeline is reproducible and self-checking — each script prints its maxima, feature counts, score range and any unmatched field names.

Output field	Type	Meaning
regreen_score	Double	Composite priority index, 0–100
regreen_class	Integer	Quantile tier 1–5 (1 = top priority)
regreen_tier	Text	Tier label for the legend
c_demand / c_children / c_greendef / c_context	Double	Points contributed by each term (sum = score)
top_driver / top_share	Text / Double	Largest-contributing term and its share of the score

Assumptions and limitations

The following are offered for critical review; several bear directly on interpretation.

Issue	Effect on results	Recommended mitigation
Green access is a proxy	No direct vegetation / NDVI or park-area input; outdoor-recreation POI access stands in for “biodiversity”. A biologically green but POI-poor cell reads as deficit, and vice-versa.	Cross-check the deficit term against NDVI / land-cover; consider adding a vegetation index.
Deficit saturates in empty land	Where no outdoor-recreation POI is reachable, the full 0.35 term is awarded regardless of population, inflating remote / unpopulated cells.	Gate the score on dwellings > 0, or multiply the deficit term by a demand factor so it only counts where people are present.
Extent-relative normalisation	Min-max scaling is relative to the dataset maximum; scores compare only within this extent and are sensitive to outliers.	Use robust scaling (percentile capping) or fixed reference maxima for cross-study comparability.
Fully compensatory model	Linear addition lets a strong term mask a weak one (e.g. high reach hiding that a cell is already green).	Test a multiplicative / geometric-mean variant; always report term values beside the score.
Expert weighting	The 0.25 / 0.20 / 0.35 / 0.20 weights are subjective.	Sensitivity analysis (± weight perturbation) and/or AHP or stakeholder calibration.
Accessibility metric semantics	The exact definition of the W15 values (gravity / decay vs count) is unconfirmed, affecting interpretation.	Confirm the metric and its units with the data provider.
MAUP / grid effects	Results depend on the 125 m cell size and grid origin; sub-cell variation is hidden.	Test sensitivity to resolution; validate against parcel-level data where available.
Playground = demand	Playgrounds are treated as a children-presence signal, not green supply — a defensible but consequential choice.	Confirm with stakeholders; optionally test as partial green supply.

Quality assurance

Additivity check: the four contribution fields sum to regreen_score (verified by construction); spot-check a sample of cells.
Field-resolution log: each script prints any field names it could not match, catching schema or encoding drift.
Distribution review: inspect the score histogram and the printed quantile breaks and tier counts for a sensible spread.
Ground-truthing: overlay top-tier parcels on recent aerial imagery to confirm they are plausibly empty / greenable and not already vegetated.
Sensitivity: re-run with perturbed weights and a demand-gated deficit term; compare the rank stability of the top quintile.

References

Roslund MI, et al. Biodiversity intervention enhances immune regulation and health-associated commensal microbiota among daycare children. Science Advances. 2020;6(42):eaba2578.
Hanski I, et al. Environmental biodiversity, human microbiota, and allergy are interrelated. PNAS. 2012;109(21):8334–8339.
South EC, Hohl BC, Kondo MC, MacDonald JM, Branas CC. Effect of greening vacant land on mental health of community-dwelling adults: a cluster randomized trial. JAMA Network Open. 2018;1(3):e180298.
Konijnendijk CC. Evidence-based guidelines for greener, healthier, more resilient neighbourhoods: introducing the 3-30-300 rule. Journal of Forestry Research. 2023;34:821–830.
Assessing European cities with the 3-30-300 rule underscores the need for enhanced urban greening efforts. Nature Communications. 2026.
Optimal site selection strategies for urban parks green spaces under the joint perspective of spatial equity and social equity. Frontiers in Public Health. 2024;12:1310340.
Identification of optimal locations for green space initiatives through GIS-based multi-criteria analysis and the analytical hierarchy process. Environmental Systems Research. 2024.