Evaluation of an urban tree population for future scenarios
Section 1: City Climate Analysis
RQ1: How does Malmö’s climate change under future scenarios, and what are the primary dimensions of that change?
Malmö’s Köppen-Geiger Reclassification
Malmö’s observed climate has crossed a major classification boundary. Using the high-resolution (1 km) Köppen-Geiger maps of Beck et al. (2023), which estimate that approximately 5% of the global land surface shifted to a different major class between 1901–1930 and 1991–2020, and the CitiesGOER database (Kindt, 2023), which assigns Köppen-Geiger zones to >52,000 cities across four 30-year observed periods and 14 future scenarios, Malmö was classified as Dfb (continental, warm summer, zone 26) throughout the 20th century (1901–1990) before shifting to Cfb (temperate oceanic, zone 15) in the most recent period (1991–2020).
The critical threshold is the coldest month mean temperature crossing 0 °C — the boundary between the continental (D) and temperate (C) macroclimate groups. The C/D boundary was designed by Köppen to correspond to a biome transition (Köppen, 1936; Essenwanger, 2001). The distinction carries direct biological consequences:
- Soil freezing regime. When coldest-month means are below 0 °C, soils regularly freeze, preventing root water uptake. Sempervirent (evergreen broadleaf) species cannot compensate for winter desiccation under these conditions, restricting them to the C (temperate) domain (Charrier et al., 2015). The loss of regular soil freezing alters the set of species that can persist at a site.
- Root frost damage. Fine roots are the most frost-sensitive tree organ (lethal at approximately −5 °C) and rely on snow cover and frozen-soil insulation for protection. As Dfb sites transition to Cfb, the shift from stable frozen ground to freeze-thaw cycling exposes roots to new damage patterns (Charrier et al., 2015).
- Tree growth responses. Across 35,217 trees in Canada, annual radial growth responses to freeze-day frequency varied by species, with gymnosperms (the dominant Dfb canopy trees) showing negative responses to high freeze-day frequency (D’Orangeville et al., 2022).
Malmö is not an isolated case. Across the CitiesGOER database, 1,742 cities made the same Dfb → Cfb transition between the 1961–1990 and 1991–2020 periods, concentrated along the continental/temperate frontier in northern Europe, the North American Great Lakes region, and East Asia (Figure 1). Under all SSP scenarios through both the 2050s and 2090s, Malmö remains classified as Cfb — the shift is irreversible within the projection horizon.
Figure 1. Global distribution of cities by Köppen zone under the 1991–2020 observed period, drawn from CitiesGOER (Kindt, 2023; n > 52,000 cities). Green points: cities classified as Cfb (zone 15, temperate oceanic) in 1991–2020. Blue points: cities classified as Dfb (zone 26, continental warm-summer) in 1991–2020. Orange points: the 1,742 cities that were Dfb in 1961–1990 and shifted to Cfb in 1991–2020. Black ring: Malmö. Layer controls allow each group to be toggled independently; hover labels show city name and country. Included to show the spatial pattern of the Dfb → Cfb transition and Malmö’s position within it; classification boundary follows Beck et al. (2023).
Variable Selection
From 48 to 11
Of the 48 bioclimatic variables in CitiesGOER (Kindt, 2023), 38 are forecastable. Many are algebraically related. Following Booth et al. (2014) and Dormann et al. (2013), 11 were retained based on ecological independence and semantic separation (see collapsible table below).
From 11 to 2
For species suitability screening (Section 2), two axes are used: bio01 (annual mean temperature, °C) and CMI (climatic moisture index, P/PET − 1). These represent thermal stress and water stress — the two independent modes of failure for established urban trees.
Resolution sensitivity
A PCA (principal component analysis — a method that compresses many correlated variables into a smaller number of summary axes) of the 11 variables was computed at two resolutions (Table 3, Figures 2–3). At the global scale (52,223 cities), bio01 aligns with PC1 (Pearson r — a linear correlation coefficient, where |r| = 1 is a perfect relationship and 0 is none — r = +0.957) and CMI with PC2 (r = +0.870). At the Cfb scale (8,079 cities), the variance structure changes: PC1 is dominated by bio04/continentality (|r| = 0.881) rather than bio01 (|r| = 0.735). For Malmö alone (n = 4 scenarios, p = 11 variables), PCA is not computable; Table 2 uses standardised change (SD — standard deviation, a measure of spread — of the Cfb peer-city distribution) instead.
NoteVariable selection: included and excluded
Included (11 variables):
| Variable | Description | Ecological role |
|---|---|---|
| bio01 | Annual Mean Temperature (°C) | Overall thermal regime; primary determinant of species range limits |
| bio05 | Max Temperature of Warmest Month (°C) | Heat stress ceiling; lethal thresholds for foliage and cambium |
| bio06 | Min Temperature of Coldest Month (°C) | Cold hardiness floor; frost-kill and overwinter survival |
| bio04 | Temperature Seasonality (SD × 100) | Continental vs oceanic regime; phenological cue reliability |
| bio12 | Annual Precipitation (mm) | Total water input to the system |
| bio15 | Precipitation Seasonality (CV) | Temporal distribution of rainfall; drought timing and duration |
| bio14 | Precipitation of Driest Month (mm) | Minimum water availability; drought floor |
| CMI | Climatic Moisture Index (P/PET − 1) | Net water balance integrating supply and atmospheric demand |
| MCWD | Max Cumulative Water Deficit (mm) | Drought severity; empirically linked to tree mortality thresholds |
| annualPET | Annual Potential Evapotranspiration (mm) | Evaporative demand; energy-driven water loss |
| continentality | Continentality (°C) | Maritime vs continental buffering; amplitude of seasonal extremes |
Excluded (27 variables) — grouped by reason for exclusion:
Temperature derivatives redundant with bio01, bio05, bio06 (15 variables): bio02 (Mean Diurnal Range = bio05 − bio06 within months), bio03 (Isothermality = bio02/bio07), bio07 (Temperature Annual Range = bio05 − bio06), bio08–bio11 (quarterly temperature means recoverable from monthly data underlying bio01/bio05/bio06), thermicityIndex (linear combination of bio01, bio06, and min temp of coldest month), growingDegDays0 and growingDegDays5 (cumulative temperature sums directly derivable from the monthly temperature distribution captured by bio01/bio04/bio06), maxTempColdest and meanTempColdest (collinear with bio06, r > 0.90 in Köppen 15), meanTempWarmest and minTempWarmest (collinear with bio05, r > 0.85), monthCountByTemp10 (step function of the temperature distribution), PETColdestQuarter (dominated by temperature; collinear with bio06 and bio11).
Precipitation derivatives redundant with bio12, bio14, bio15 (6 variables): bio13 (Precipitation of Wettest Month), bio16–bio19 (quarterly precipitation totals recoverable from bio12/bio14/bio15 and monthly distribution).
PET derivatives redundant with annualPET and continentality (3 variables): PETDriestQuarter, PETWarmestQuarter, PETWettestQuarter (quarterly decompositions of annualPET; seasonal partitioning captured by continentality and bio04). PETseasonality (correlated with continentality, r > 0.5 in Köppen 15).
Composite indices redundant with CMI (3 variables): aridityIndexThornthwaite (alternative formulation of moisture balance using the same P and PET inputs as CMI), embergerQ (precipitation-temperature ratio; partially collinear with CMI and bio01).
Table 1: Malmö’s Climate Envelope
Table 2: Magnitude of Climatic Change (11 Selected Variables)
Table 3: PCA Variance Explained at Two Resolutions
Figure 2: PCA Loadings in 3D Space — Global vs Cfb
Figure 2. PCA loadings (the coefficients that define how each variable contributes to each principal component) for the 11 curated bioclimatic variables on the first three PCs, computed at two resolutions: all CitiesGOER cities (n = 52,223; Global tab) and validated Cfb cities only (n = 8,079; Cfb tab). Each point is one variable; its position on the three axes shows its loading on PC1, PC2, and PC3 respectively. Colour indicates ecological domain (red = temperature, blue = precipitation, green = moisture and energy). Interactive plots can be rotated. Included to show how the variable clustering changes between resolutions: at the global scale, bio01 loads on PC1 and CMI on PC2 with minimal overlap; at the Cfb scale, bio01’s loading shifts toward PC2 and bio04/continentality dominate PC1 (see Table 3 and Variable Selection narrative).