1 What Is Geodesign?

Geodesign is a six-phase iterative framework for landscape-scale spatial decisions, developed by Carl Steinitz at the Harvard Graduate School of Design. Each phase asks a different question. The cycle repeats with increasing depth.

The framework arranges six "models" into two triads. The Assessment Triad (left) asks how the landscape works. The Intervention Triad (right) asks how it might be changed. Three feedback loops connect them — each loop represents a full pass through the analysis at increasing depth.

This study ran all three loops. What follows is the story of that process.

The Steinitz Framework for Geodesign Six models in two triads, connected by three iterative feedback loops ASSESSMENT TRIAD INTERVENTION TRIAD How does the landscape work? How might it be changed? 1. Representation Model How should the landscape be described? 44 GIS layers, 134,436 parcels, SSURGO soils, satellite imagery, Hansen forest change, FEMA flood, species inventories 2. Process Model How does the landscape operate? Stormwater (SCS-CN), canopy function, habitat connectivity, financial flows, property value dynamics, carbon cycling 3. Evaluation Model Is the current landscape working well? 20 MCDA objectives across 4 domains (environmental, economic, social, governance) with stakeholder weighting 4. Change Model How might the landscape be altered? 12 scenarios (A-J + hybrids), land-swap alternatives, alternative site inventory (221 public parcels scored) 5. Impact Model What differences might changes cause? 30-year projections, 10-order cascade analysis, Monte Carlo sensitivity, stakeholder trade-offs, regret 6. Decision Model Should the landscape be changed? COA comparison table, Pareto dominance, minimax regret, recommendation with caveats inform scenarios LOOP 1 Scaffold Which are viable? LOOP 2 Stress Do findings hold? LOOP 3 Design STAKEHOLDERS IN EVERY PHASE UNCA • City of Asheville • Buncombe County • Five Points • Save the Woods • USFS SRS • EBCI • Developer MOD-SIM-VIS at each phase: MODEL (spatial + process) • SIMULATE (run scenarios) • VISUALIZE (maps + charts for decisions) After Steinitz (2012), A Framework for Geodesign. Adapted with nested loops, stakeholder participation, and MOD-SIM-VIS cycle. Applied at Harvard GSD, MIT, ETH Zurich, and in national-scale planning across 20+ countries.

The key insight of the framework is that the six phases are not a linear pipeline. They form feedback loops. When a scenario fails stress-testing (Loop 2), you go back and redesign. When a design reveals new data needs (Loop 3), you go back and collect. The three loops in this study each represented a full pass of increasing depth — from screening to stress-testing to detailed design.

2 The Data Foundation

Before building any model, we assembled the most comprehensive public-data picture of this site ever constructed. All data is from authoritative public sources. Nothing was paywalled. Everything is reproducible.

GIS Layers
44
Vector and raster layers from 7 agencies: parcels, soils, flood zones, canopy, zoning, infrastructure, species
Buncombe County, City of Asheville, FEMA, USDA, NOAA, Esri, Hansen/UMD
Total Data
4.8 GB
Raw downloads, derived models, computed outputs, and scenario geometries
130+ files across 17 model categories
Parcels Classified
134,436
Full Buncombe County parcel inventory loaded, classified by ownership, tax status, and development suitability
Buncombe County GIS (nc_buncombe_parcels_poly.shp)
Aerial Images
16
Google Earth time-lapse from 1993 to post-Helene 2024, showing forest continuity and storm damage
Google Earth Pro historical imagery
Species Documented
1,768
Research-grade observations within 2 km of the site (incl. 9 Trillium species, 8 native orchids, 100+ birds, fungi, plants, mammals)
iNaturalist + eBird + NC Natural Heritage Program (retrieved 2026-04-28)
Research Documents
18
Peer-reviewed papers, technical reports, and regulatory references informing the models
Sun et al. (2026), Coweeta Hydrologic Lab, Hansen et al. (2013), NOAA Atlas 14, etc.

Data Acquisition Timeline

The data collection itself followed the geodesign phases — each loop required deeper data.

Day 1
Parcel & Ownership Data — Full Buncombe County parcel shapefile (134,436 records). Classified all public-owned parcels, identified 221 potential alternative sites for scoring.
Day 1
Flood & Soil Data — FEMA National Flood Hazard Layer, USDA SSURGO soil survey (17 map units, hydrologic soil groups). Foundation for the stormwater model.
Day 1-2
City & County GIS Layers — 27 vector layers from City of Asheville ArcGIS: zoning, infrastructure, canopy cover, historic districts, transit routes.
Day 2
Forest Change Rasters — Hansen/UMD Global Forest Change v1.12 (2001-2024). Esri 10m Land Use/Land Cover (2017-2023). Basis for canopy loss and land-use trajectory analysis.
Day 2
Ecological Inventories — iNaturalist research-grade observations (12,545 within 2 km), eBird hotspot data (220+ species at Beaver Lake), NC Natural Heritage Program species records.
Day 2-3
Aerial Imagery & Precipitation Data — 16 Google Earth time-lapse images (1993-2024). NOAA Atlas 14 precipitation frequency estimates for 5 storm recurrence intervals.
Day 3-4
Literature & Precedent Research — 18 research documents including Sun et al. (2026) on forested watershed hydrology, Coweeta LTER 90-year dataset, comparable project case studies.
Data Inventory by Source FEDERAL FEMA Flood Hazard SSURGO Soils NOAA Atlas 14 Hansen Forest v1.12 Esri 10m LULC NC Heritage Spp. COUNTY / CITY 134,436 Parcels Zoning Districts Canopy Cover Infrastructure Transit & Roads Historic Districts COMMUNITY SCI. iNaturalist 12,545 observations eBird 220+ species Google Earth Pro 16 aerial images Coweeta LTER 90 yr watershed data LITERATURE Sun et al. (2026) Hansen et al. (2013) 14 more papers DERIVED SCS-CN Model Financial Model MCDA Scores + 14 more All data publicly available. No paywalled sources. Full reproduction package: 4.8 GB.

3 Loop 1: The Scaffold

The question we asked
"Which scenarios are viable and which can be eliminated?"

The first pass through the geodesign framework is deliberately coarse. The goal is not precision but triage: build the simplest credible models, generate a rough ranking, and identify which scenarios deserve deeper analysis.

What we built

  • Stormwater model (SCS-CN method with SSURGO soils) — 17 soil map units, 5 storm recurrence intervals, curve numbers for forest (CN=58.7), open space (64.2), and impervious (98)
  • Financial model (30-year cash flow) — Three-perspective split: public cost, university revenue, developer return. Property tax, ecosystem services, and subsidy requirements for each scenario
  • MCDA framework — 20 objectives across 4 domains (environmental, economic, social, governance). Equal-weight scoring for initial screening. 12 scenarios (A through J plus variants)
  • Alternative site inventory — 134,436 parcels classified, 221 public parcels scored for stadium and housing suitability
Loop 1: Scaffold — From Data to First Ranking INPUTS 44 GIS layers 134,436 parcels SSURGO soil survey FEMA flood zones Hansen forest rasters NOAA precipitation Property tax records Comparable projects Enrollment data Species inventories 12 scenario definitions 20 evaluation objectives 4 weighting schemes MODELS SCS-CN Stormwater 5 storm events x 12 scenarios 30-yr Financial NPV at 3%, 5%, 7% 20-Objective MCDA 12 scenarios scored 0-100 Site Scorer 221 alt. sites ranked COA TABLE 12 scenarios ranked Pareto dominance 4 weighting schemes Strengths / weaknesses PROCEED / CONSIDER / ELIMINATE verdicts FINDINGS H ranks #1 (92.4/100, all schemes) A ranks last (stadium, every scheme) 10 sites at 90/100 (Brevard Rd, Fairway Dr) $0 tax from stadium (UNC exempt, G.S. 105) 5 scenarios PROCEED Loop 1 result: Stadium scenario (A) is dominated by every alternative. H leads all rankings. Five scenarios advance to stress-testing.

Loop 1 Headline Finding: The stadium scenario (A) ranks last of 12 under every weighting scheme. It is Pareto-dominated — there exists at least one alternative that is better on every dimension simultaneously. Meanwhile, 9 public parcels score 90/100 as alternative stadium sites. The top candidates — 1568 Brevard Rd (123 ac, County) and 226 Fairway Dr (111 ac, City) — are cleared, outside the floodplain, with no forest impact. (53 Birch St scored 100/100 algorithmically but was disqualified due to probable cemetery adjacency.)

Five scenarios advanced to Loop 2: H (forest preserved, housing elsewhere), E (land swap, minimal clearing), F (scaled-down stadium, land swap), I (conservation + research emphasis), and A (stadium, kept as a baseline for comparison).

4 Loop 2: The Stress Test

The question we asked
"Do the conclusions hold under different assumptions, or did we bake in our preferences?"

Loop 1 produced a clear ranking. But rankings are only as good as the weights behind them. If someone who wants the stadium assigns weights that favor economic development, does H still win? We needed to stress-test the entire framework.

What we did

  • 1,000 Monte Carlo draws — Random weights sampled uniformly across all 20 objectives. No human judgment in the weighting. Pure exploration of the objective space.
  • Adversarial scoring — Gave the stadium the most generous possible scores on every objective. What if every assumption breaks in the developer's favor?
  • Minimax regret analysis — For each scenario, what is the worst-case regret across all 1,000 draws? Which scenario minimizes maximum regret?
  • 10-order cascade analysis — Traced impacts through 10 cascading orders: direct, transportation, stormwater, property values, tax revenue, tourism, ecology, equity, governance, resilience.
  • 4-scale spatial analysis — Evaluated each scenario at site (45 acres), neighborhood (Five Points), watershed (Reed Creek), and city scale.
Monte Carlo Weight Sensitivity: 1,000 Random Draws How often does each scenario win under random objective weights? 1,000 750 500 250 Wins out of 1,000 1,000 H Forest Preserved 0 E 0 F 0 I 0 G 0 J 0 B 0 C 0 D 0 A Stadium Proposal H wins 100% of random draws No combination of objective weights produces a scenario where the stadium beats preservation. Method: 1,000 independent draws of 20 weights from Uniform(0,1), normalized. Random seed: 42. Even under adversarial scoring (developer's best case for every objective), H wins.

Loop 2 Headline Finding: Scenario H wins 1,000 out of 1,000 Monte Carlo draws. There is no weighting of the 20 objectives — no matter how much you value economic development over ecology — that makes the stadium rank first. Even under adversarial conditions designed to maximize the stadium's score, H dominates.

The 10-order cascade analysis confirmed the result from a different angle: the stadium scenario produces net harm at all four spatial scales (site, neighborhood, watershed, city). The minimax regret winner is H1, a hybrid variant that adds a community sports field to the forest-preserved scenario.

5 Loop 3: The Design

The question we asked
"What do the best alternatives actually look like in detail — with real costs, real timelines, and real trade-offs?"

Loops 1 and 2 established that preservation beats stadium development under every analysis. But "preserve the forest" is not a plan. Loop 3 takes the top two scenarios and designs them in detail: spatial layout, phasing, 30-year financial pro formas, and stakeholder trade-off analysis.

What we designed

  • H1: Community Field + Housing + Forest Preserved — The minimax-regret winner. 45 acres fully preserved. 2,000-seat community field on already-cleared Zillicoa MC land. 402 housing units (80% market-rate, 20% affordable) on MC parcels. $0 public subsidy. Breaks even by Year 3.
  • E2: Research Reserve + Conservation Housing — UNCA-centered variant. Forest becomes a formal research reserve with conservation easement. Housing emphasizes faculty/student/senior mix. Partnership with USFS Southern Research Station formalized.
  • 30-year pro forma for each — Year-by-year cash flows, NPV at 3 discount rates, cumulative tax generation, and comparison to Scenario A (stadium).
Loop 3: 30-Year Financial Comparison NPV at 5% discount rate, including ecosystem services H1: Community Plan Field + Housing + Forest Preserved NPV (5%) +$68.2M positive returns from Year 3 Public Subsidy $0 30-yr Tax to City $120M Housing Units 402 Forest Preserved 100% Break-even Year Year 3 RECOMMENDED vs A: Stadium Proposal 5,000-seat stadium on forest site NPV (5%) -$132.5M negative for entire 30-year horizon Public Subsidy $29M 30-yr Tax to City $0 Housing Units 0 Forest Preserved 0% Break-even Year Never ELIMINATED NPV Gap (at 5%) $200.7M H1 generates $120M in city tax over 30 years. The stadium generates $0 (UNC property is tax-exempt under G.S. 105-278.1). H1 requires $0 public subsidy. The stadium requires a $29M subsidy plus ~$15M in infrastructure.

Loop 3 Headline Finding: The NPV gap between H1 (community plan) and A (stadium) is $200.7 million at a 5% discount rate over 30 years. H1 breaks even by Year 3, requires zero public subsidy, preserves 100% of the forest, and generates $120M in property tax for the city. The stadium never breaks even, requires $29M+ in public support, clears the entire forest, and generates zero tax revenue because UNC property is exempt.

The H1 design includes a 2,000-seat community field on the already-cleared Zillicoa MC parcel — giving UNCA a sports venue without touching the forest. 402 housing units across two phases (80% market-rate, 20% affordable) address the city's housing shortage while generating permanent tax revenue.

6 What We Got Wrong

Every analysis has errors. The question is whether you find them or your opponents do. We found four claims that required correction during the analysis and corrected them before publication.

Self-Correction Log Four claims tested, four claims revised. The analysis is stronger for it. CLAIMED TESTED REVISED STRONGER ANALYSIS Opponents cannot attack what you already fixed FALSIFIED "Last significant urban forest in Asheville" "Increasingly rare and demonstrably valuable" (65 patches exist) CORRECTED 5.44 million gallon stormwater (point estimate) 3.7 - 7.0 million gallon range (uncertainty shown) DOWNGRADED "USFS conducted research on this site" "Hypothesized" (zero records found via formal search) REFRAMED "Land-use gerrymandering" (loaded framing) Neutral "allocation equity test" framework

Why Self-Correction Matters

An advocacy document hides its weaknesses. An analytical document exposes them. By catching these four errors before publication, the analysis gained something more valuable than the claims it lost: credibility.

Consider the first correction. The original claim — "this is the last significant urban forest in Asheville" — sounded powerful. But when we checked the Esri 10m land cover data, we found 65 forest patches of 45+ acres within the Asheville metro boundary. The claim was false. If an opponent had found it first, every number in the analysis would have been suspect.

The revised claim — "increasingly rare and demonstrably valuable" — is weaker as rhetoric but stronger as evidence. It is also true.

Quality Review Summary

The full analysis passed through five review passes (v1.0 through v6.0). Each pass:

The integrity principle: An analysis that self-corrects is more trustworthy than one that appears perfect. We are publishing the correction log because it is evidence that the remaining claims survived testing.

7 What Remains Unknown

Honest analysis requires honest boundaries. These are the things this study cannot answer and the caveats readers must carry forward.

7 Uncertainties

5 Mandatory Caveats

The bottom line: Even with all seven uncertainties resolved in the stadium's favor, the structural conclusion holds. The $200M NPV gap, the zero tax revenue, the floodway overlap, and the Pareto dominance of preservation scenarios are robust findings. What remains unknown can change the details but not the direction.