Sensitivity analysis of the aggregation weights
Methodology version: 2026.08.06. Scenario set: 24 golden scenarios
(276 scenario pairs). Each of the six final-aggregation weights was
varied by +/-25% with the remainder renormalised to a unit sum, and the
full scenario set re-scored under each of the 12 perturbations.
Generated by tools/sensitivity_analysis.py; regenerate whenever any
aggregation weight changes.
Per-perturbation results
| Weight |
Perturbation |
Rank stability |
Mean displacement |
Max displacement |
Category migrations |
| heat_priority_index |
-25% (0.75x) |
0.9529 |
0.65 |
2.00 |
1 |
| heat_priority_index |
+25% (1.25x) |
0.9891 |
0.57 |
1.77 |
1 |
| cooling_potential |
-25% (0.75x) |
0.9819 |
0.79 |
2.05 |
2 |
| cooling_potential |
+25% (1.25x) |
0.9601 |
0.69 |
1.79 |
1 |
| nbs_suitability |
-25% (0.75x) |
0.9891 |
0.54 |
1.49 |
0 |
| nbs_suitability |
+25% (1.25x) |
0.9891 |
0.49 |
1.38 |
1 |
| vulnerability |
-25% (0.75x) |
0.9891 |
0.50 |
1.75 |
0 |
| vulnerability |
+25% (1.25x) |
0.9928 |
0.45 |
1.62 |
1 |
| co_benefits |
-25% (0.75x) |
0.9964 |
0.34 |
1.01 |
1 |
| co_benefits |
+25% (1.25x) |
0.9891 |
0.32 |
0.96 |
0 |
| cost_feasibility |
-25% (0.75x) |
0.9928 |
0.14 |
1.04 |
0 |
| cost_feasibility |
+25% (1.25x) |
0.9928 |
0.13 |
1.00 |
0 |
Score displacement distribution (all 12 perturbations pooled)
- observations: 288
- mean 0.47 points; median 0.34; p75 0.67; max 2.05
Category migrations
| Weight |
Perturbation |
Scenario |
Score (category) change |
| heat_priority_index |
-25% |
s19_explicit_unknowns_everywhere |
59.58 (moderate) -> 60.3 (strong) |
| heat_priority_index |
+25% |
s24_productive_landscape_evidence_gap |
59.89 (moderate) -> 60.61 (strong) |
| cooling_potential |
-25% |
s03_arid_urban_forest_ceiling_clip |
80.92 (high_priority) -> 79.63 (strong) |
| cooling_potential |
-25% |
s24_productive_landscape_evidence_gap |
59.89 (moderate) -> 61.4 (strong) |
| cooling_potential |
+25% |
s19_explicit_unknowns_everywhere |
59.58 (moderate) -> 60.39 (strong) |
| nbs_suitability |
+25% |
s24_productive_landscape_evidence_gap |
59.89 (moderate) -> 60.59 (strong) |
| vulnerability |
+25% |
s24_productive_landscape_evidence_gap |
59.89 (moderate) -> 60.09 (strong) |
| co_benefits |
-25% |
s24_productive_landscape_evidence_gap |
59.89 (moderate) -> 60.21 (strong) |
Influence ranking (mean absolute displacement per weight, both directions)
| Rank |
Weight |
Mean displacement |
Max displacement |
| 1 |
cooling_potential |
0.74 |
2.05 |
| 2 |
heat_priority_index |
0.61 |
2.00 |
| 3 |
nbs_suitability |
0.51 |
1.49 |
| 4 |
vulnerability |
0.48 |
1.75 |
| 5 |
co_benefits |
0.33 |
1.01 |
| 6 |
cost_feasibility |
0.14 |
1.04 |
- Worst-case rank stability across all perturbations: 0.9529
- Pooled mean score displacement: 0.47 points
- Total category migrations: 8 of 288 scenario-perturbation combinations