Land acquisition intel

Land Acquisition Intelligence Skill

Fetches real parcel data + infrastructure, scores every parcel against your use case criteria, outputs ranked CSVs of top sites. Zero paid APIs. Runs fully local.

Critical: Always Follow This Order

1. Understand intent — what land use, what region, any custom requirements
2. Resolve region — get center lat/lon + bbox
3. Run fetch script — parcels + infrastructure
4. Run score script — vectorized scoring with use-case weights
5. Report top candidates — table + Google Maps links + file paths

Never skip steps. Never invent scores without running the scripts.

---

Step 1: Understand Intent

Ask (or infer from context) before running:

Required:

• What is the land for? (datacenter / warehouse / solar / industrial / agricultural / custom)
• What region? (city, county, state, or coordinates)

Optional (infer defaults if not given):

• Minimum acreage? (default: none)
• Maximum distance from highway? (default: none)
• Any specific infrastructure required? (power/rail/fiber/water)
• Budget tier? (high-value = prefer smaller parcels near infra; land-banking = larger, more remote OK)

If use case is custom or unclear: ask "What's the primary value driver for this site? Power access, logistics, labor, cost, or something else?"

---

Step 2: Resolve Region

Extract city/county/state from user input. Map to coordinates:

Built-in coords (Western US defaults):

• Carson City NV: 39.1638, -119.7674
• Reno NV: 39.5296, -119.8138
• Las Vegas NV: 36.1699, -115.1398
• Sparks NV: 39.5349, -119.7527
• Fernley NV: 39.6077, -119.2125
• Phoenix AZ: 33.4484, -112.0740
• Salt Lake City UT: 40.7608, -111.8910
• Boise ID: 43.6150, -116.2023
• Denver CO: 39.7392, -104.9903
• Portland OR: 45.5051, -122.6750

For any other city: use Nominatim geocoder (free, no key):

https://nominatim.openstreetmap.org/search?q=CITY+STATE&format=json&limit=1

Default bbox: ~60-mile radius from center point.

---

Step 3: Fetch Data

Run: python3 scripts/fetch_data.py --lat LAT --lon LON --state STATE --output /tmp/land-intel/raw/

Fetches (all free, no API keys):

Layer	Source	Notes
Parcels	State/county ArcGIS REST APIs	APN, acres, owner, address
Parcels fallback	OSM building + landuse footprints	Always works
Highways	OSM Overpass	motorway, trunk, primary
Interchanges	OSM motorway junctions	exact ramp locations
Rail lines	OSM	freight rail, usage tag
Substations + power lines	OSM power layer	voltage included
Water infrastructure	OSM	rivers, reservoirs, water towers
Farmland / vacant land	OSM landuse	agricultural, brownfield, greenfield
Industrial zones	OSM landuse=industrial	existing industrial parks

Expected raw files:

/tmp/land-intel/raw/
├── parcels_raw.csv         # APN, acres, lat/lon, county
├── osm_buildings.csv       # OSM land use proxy
├── highways.csv
├── interchanges.csv
├── rail.csv
├── substations.csv
├── transmission_lines.csv
├── water_infra.csv

If a fetch fails: log it, skip that layer, continue. Missing infra = 999km default = lower score for that factor only.

---

Step 4: Score Parcels

Run: python3 scripts/score_parcels.py --raw /tmp/land-intel/raw/ --use-case USE_CASE --output /tmp/land-intel/output/

USE_CASE options: datacenter, warehouse, solar, agricultural, industrial, custom

Scoring Factors (all vectorized, no loops)

Universal factors (all use cases):

• Acreage fit: 0-20pts — calibrated to use-case ideal range
• Highway access: 0-20pts — distance to nearest major road
• Zoning compatibility: 0-15pts — industrial > commercial > agricultural > unknown > residential
• Water access: 0-5pts — nearest water body or tower

Use-case specific weights:

Factor	Datacenter	Warehouse	Solar	Agricultural
Power substation	35pts	5pts	15pts	5pts
Transmission line	5pts	3pts	10pts	2pts
Highway access	15pts	25pts	5pts	10pts
Interchange proximity	5pts	15pts	0pts	0pts
Rail access	0pts	15pts	0pts	5pts
Acreage	20pts	20pts	25pts	30pts
Zoning	15pts	15pts	10pts	15pts
Water	5pts	5pts	5pts	20pts
Fiber/telecom	5pts	0pts	0pts	0pts
Climate bonus	5pts	0pts	5pts	5pts
Labor market	0pts	10pts	0pts	0pts

Custom use case: prompt user for 3-5 scoring priorities, assign 100pts across them.

Scoring outputs:

• [use_case]_candidates.csv — top 5,000 sites, all score components
• all_candidates_ranked.csv — full dataset with composite score

Motivated seller signals (added to every parcel where data available):

• Out-of-state owner: +15pts signal
• Estate/trust in owner name: +20pts signal
• Tax delinquent: +30pts signal
• Long hold (pre-2010 acquisition): +15pts signal

---

Step 5: Report Results

After scripts complete, present findings in this format:

## Land Scouting Results: [USE CASE] in [REGION]

### Data Summary
- Parcels scored: X across N counties
- Infrastructure layers: highways ✓, rail ✓, power ✓, water ✓

### Top 10 Sites

| # | Score | Location | County | Acres | Key Factor | APN | Map |
|---|-------|----------|--------|-------|------------|-----|-----|
| 1 | 94 | 39.1638,-119.7674 | Carson City | 42ac | sub=0.3km | 1004169 | [link] |
...

### Score Distribution
Min: X | Median: X | Max: X | Sites >80: X

### Files Saved
- /tmp/land-intel/output/[use_case]_candidates.csv
- /tmp/land-intel/output/all_candidates_ranked.csv

Google Maps link format: https://maps.google.com/?q=LAT,LON

Always include top 10 in the response. Offer to show more or filter by criteria.

---

Error Handling

No parcel data:

• Use OSM buildings + 500m synthetic grid as fallback
• Note: "County parcel data unavailable, using OSM proxy — APN/owner data will be missing"

OSM rate limited (HTTP 429):

• Sleep 3s, retry once
• Split into smaller bbox quadrants
• Skip layer if still failing

Dependencies missing:

pip install pandas numpy requests tqdm shapely pyproj geopandas --break-system-packages

Region not found:

• Use Nominatim geocoder on city name
• If still fails: ask user for coordinates directly

---

Optional Enrichment

After base scoring, offer these if user wants to go deeper:

• Zoning details: actual zoning codes from county assessor (Municode or county GIS)
• Ownership lookup: cross-reference APN with county assessor for owner name/address
• Active listings overlay: Redfin CSV endpoint (recursive bbox subdivision, no API key)
• BLM/federal ownership: PADUS or BLM NILS layer — critical for Western US
• Power capacity: substation IDs from OSM → call utility for available MW
• Environmental flags: FEMA flood zones, EPA brownfields, wetlands

Prompt: "Base scoring complete. Want me to enrich the top 20 with zoning codes, ownership data, or active listing prices?"

---

Scope

Works well for:

• Any US county with public ArcGIS parcel data (most do)
• Western US: Nevada, Utah, Arizona, Idaho, Oregon, Colorado, California
• Eastern US: NC OneMap, county GIS portals (see references/state-endpoints.md)
• Any use case expressible as infrastructure proximity + acreage

Doesn't cover:

• International parcels
• Permitting or entitlement research
• Environmental impact analysis
• Residential property or single-family land
• Title/lien searches

Data freshness:

• OSM: real-time
• County parcel layers: typically updated annually (check SourceDate field)
• Infrastructure: OSM community-maintained, generally reliable