Siting the Grid: A Geospatial Case for Smarter Data Center Location in Washington State

Composite Siting Suitability Explorer

Each cell is a 0.15-degree grid square (~14 km) scored on five normalized indicators. Adjust the weights below to see how priorities change the siting landscape. Existing and proposed data center locations are shown as markers.

Key Findings

Scores by cluster

Scores reflect the ~14 km grid cell containing each cluster. Composite updates live with the weight sliders above.

Note: Tukwila's community burden score (0.109) reflects Census ACS high poverty/minority rates in the SR-99 industrial corridor. Grid access and water are strong, but EJ burden is the highest of any active cluster in this analysis.

Top 10 Unclaimed Site Scores

Grid cells with no existing or proposed data center, ranked by composite score under the current weight settings. Locations are cell centroids (~14 km resolution). Updates live with the sliders above.

Individual Indicator Maps

Each indicator normalized 0–1 (green = favorable for siting). White diamonds mark existing and proposed data center locations.

Left to right: transmission proximity, water availability (annual precip), community burden (Census ACS Demographic Index, inverted).

Risk modifiers: seismic hazard and flood zones

Adding USGS National Seismic Hazard Model PGA values and FEMA flood zone designations as risk disqualifiers shifts the picture: the Puget Sound basin drops sharply on seismic score, and river corridor cells flag on flood risk.

Left to right: seismic safety (inverse PGAM), flood safety (outside FEMA SFHA), risk-adjusted composite (60% opportunity + 20% seismic + 20% flood).

Environmental Risk: Contamination and Waterway Sensitivity

Two additional risk modifiers quantify environmental liability independent of water scarcity. Contamination proximity scores each cell by distance to the nearest active EPA Superfund National Priorities List site in Washington State. Amazon's Wallula Gap campus scores 0.014 — effectively at the perimeter of the Hanford Site, the largest contaminated site in the Western hemisphere. Waterway sensitivity scores by distance to the nearest major regulated river (Columbia, Snake, Spokane, Skagit, Yakima) — a proxy for ESA Section 7 thermal discharge constraints and water withdrawal scrutiny. The same Wallula Gap site scores 0.015, placing it directly on the Columbia's bank.

Left: contamination proximity (1 = far from EPA NPL Superfund site). Right: waterway sensitivity (1 = far from major regulated river). White filled = existing DC; outline = proposed.

Geothermal Resource Opportunity

Surface heat flow (mW/m²) from the IHFC Global Heat Flow Database 2024 (664 measured borehole sites within the Washington State bounding box) is IDW-interpolated to each grid cell and normalized 0–1. Higher scores indicate elevated geothermal energy potential, primarily following the Cascade volcanic arc. Values are capped at the 95th percentile (162 mW/m²) to prevent hydrothermal anomalies near Mount Baker (up to 5,146 mW/m²) from dominating the spatial interpolation.

In the composite scoring, geothermal opportunity functions as an opportunity indicator: cells adjacent to elevated heat flow can in principle tap enhanced geothermal or direct-use systems to offset grid power demand. The Cascade foothills score highest; the eastern Columbia Basin, where most existing and proposed campuses are located, scores 0.30–0.35.

Surface heat flow interpolated from 664 IHFC 2024 borehole measurements (small dots). Inferno scale: 0 = low heat flow / 1 = high geothermal potential. White filled = existing DC; outline = proposed.

Terrain Flatness: A Physical Buildability Gate

Large data center campuses require 20–30 acres of essentially flat ground for foundations, cooling infrastructure, and security setbacks. This indicator asks not whether a grid cell is uniformly flat, but whether enough flat land exists within it to accommodate a campus pad.

Slope is computed at ~90m resolution from SRTM1 tiles (AWS Open Data terrain tiles bucket) using numpy.gradient with a per-latitude correction for east–west pixel width. For each 5 km grid cell, the score is the fraction of pixels with slope < 5° — the conventional threshold for large-pad civil grading.

Cells where fewer than 3% of pixels meet this threshold (< 185 acres of sub-5° terrain) are hard-gated as unbuildable and rendered as dark overlays on the map regardless of slider weights. 61 cells (6.3% of the analysis grid) are excluded by this gate — primarily the Cascade Range, the Olympics, and the Okanogan Highlands. Passing cells are scored 0–1 proportional to their flat fraction.

Flatness explains the desert clustering

The Columbia Plateau — where Quincy, the Tri-Cities, and all four proposed campuses sit — is a basalt plateau formed by successive lava flows. Wallula Gap and Richland score 1.000 flatness (fully flat cells). Quincy scores 0.769. In contrast, the Malaga / East Wenatchee cluster scores only 0.479 — it sits at the Columbia River canyon where the Cascades descend to the plateau, introducing enough relief to reduce the buildable flat fraction meaningfully. Developers who cite “proximity to power” as the reason for eastern WA concentration are correct, but terrain is an equally silent co-factor: the basalt plateau requires minimal site preparation on campuses measuring hundreds of acres.

The Tukwila corridor (HorizonIQ, ColoCrossing) scores 0.828 on flatness — the Duwamish industrial valley is engineered flat — but carries the lowest community burden score (EJ = 0.109) of any active cluster. The flatness gate removes the Cascade foothills cells that had been artificially inflated by strong transmission and water scores.

Fraction of each 5 km cell with slope < 5° (SRTM1 ~90m). Dark cells are hard-gated as unbuildable (< 3% flat area). YlGn scale: 0 = scarce flat land / 1 = highly flat. White filled = existing DC; outline = proposed.

Protected and Off-Limits Land: A Regulatory Buildability Gate

Siting a data center on land that is legally off-limits is not a planning error — it is a category error. National Parks, Wilderness Areas, Wildlife Refuges, military installations, and American Indian sovereign territories are not difficult to permit; they are impossible to permit. This indicator identifies and hard-gates those cells from the analysis.

Protection layer sources: Esri USA Federal Lands (ArcGIS Living Atlas), filtered to National Park Service, Fish and Wildlife Service, Department of Defense, and Forest Service units. Bureau of Land Management is excluded because BLM land is leasable for commercial use. Tribal sovereign territory is sourced from Census TIGER AIANNH (federally recognized American Indian and Alaska Native homelands). Both layers are merged and dissolved before computing overlap.

The gate threshold is 25%: a grid cell where more than one quarter of its 5 km extent falls within protected boundaries is excluded. The reasoning is practical — a cell that is majority-adjacent to a National Park boundary faces the same permitting exposure, utility corridor constraints, and community opposition as one that sits directly inside it. 82 cells (8.4% of the analysis grid) are excluded by this gate.

What the gate removes

The gate removes the entire Olympic Peninsula, the North Cascades, the southern Cascades south of Rainier, much of the Okanogan Highlands (Colville National Forest), and cells intersecting the Yakama Nation and Colville Reservation boundaries in central WA. Eastern WA agricultural land, the Puget Sound lowlands, and the Columbia Basin — where existing data center clusters operate — are largely unaffected. All nine existing and proposed cluster cells pass with protected fractions well below the 25% threshold.

Combined with the terrain flatness gate (61 cells excluded), the two buildability filters together remove 124 cells (12.7%) from the viable candidate pool, leaving 850 cells (87.3%) open to composite scoring.

Cells gated by protected land coverage (dark, 82 cells). Green boundary = merged protection layer (Esri Federal Lands: NPS, USFWS, DoD, FS; plus Census TIGER tribal areas). White filled = existing DC; outline = proposed.

Methodology

Analysis grid: 0.15-degree fishnet (~14 km cells) clipped to Washington State boundary. 974 cells total. Two hard buildability gates (terrain flatness, protected land) remove 124 cells, leaving 850 viable candidates for composite scoring. All scores normalized 0–1 before weighting.

Transmission proximity

Distance from each cell centroid to nearest high-voltage (≥100 kV) transmission line. Lines sourced from OpenStreetMap Overpass API (power=line, voltage filter). Distances computed in UTM Zone 10N. Score = 1 − (dist / max_dist).

Water availability

30-year mean annual precipitation (1991–2020) from Open-Meteo Historical Weather API (ERA5 reanalysis). Sampled at 50 interior points, IDW-interpolated to analysis grid. Score = normalized percentile rank.

Community burden

Demographic Index (average of poverty rate and minority rate) computed from US Census ACS 5-Year Estimates 2022 (variables B17001, B02001), replicating the EPA EJScreen Demographic Index formula. Tract boundaries from Census TIGER/Line 2022. Score = 1 − normalized index (inverted: lower burden = higher score).

Seismic safety

Peak Ground Acceleration (PGA_M) at 2% probability of exceedance in 50 years, from the USGS ASCE 7-22 design values API (2018 National Seismic Hazard Model, site class C). Sampled at 50 points, IDW-interpolated. Score = 1 − normalized PGA_M.

Flood safety

Special Flood Hazard Area (SFHA) designation from the FEMA National Flood Hazard Layer REST API (zones A, AE, AH, AO, AR, A99, VE, V). Queried in 40 tiles covering WA; some coastal tiles unavailable due to service limits. Score = 1 if cell centroid outside SFHA, 0 if within.

Contamination proximity

Distance from each cell centroid to the nearest active EPA Superfund National Priorities List (NPL) site in Washington State. 26 active WA sites hardcoded from the EPA state list; coordinates are approximate site centroids. Score = normalized distance (0 = adjacent to NPL site, 1 = farthest cell in state). Notable: the Hanford Site (46.55°N, 119.53°W) — the largest contaminated site in the Western hemisphere — sits immediately adjacent to the Tri-Cities proposed campus cluster.

Waterway sensitivity

Distance from each cell centroid to the nearest sample point along a major regulated Washington waterway: Columbia River (20 pts), Snake River (5 pts), Spokane River (4 pts), Skagit River (6 pts), Yakima River (4 pts). Proximity is a proxy for ESA Section 7 thermal discharge constraints and water withdrawal regulatory scrutiny under Endangered Species Act salmon protections. Score = normalized distance (0 = adjacent to major river, 1 = farthest cell in state).

Geothermal opportunity

Surface heat flow (mW/m²) from the IHFC Global Heat Flow Database 2024 (Lucazeau et al., DOI 10.5880/fidgeo.2024.014). 664 borehole measurements within the Washington State bounding box are IDW-interpolated (k=8 neighbors, power=2) to each grid cell centroid using EPSG:32610 metric coordinates. Raw values are capped at the 95th percentile (162 mW/m²) before interpolation to prevent hydrothermal anomalies near Mount Baker (up to 5,146 mW/m²) from dominating. Score = interpolated heat flow / maximum interpolated value (0 = low geothermal potential, 1 = high).

Terrain flatness

Slope computed at ~90m resolution from SRTM1 tiles (AWS Open Data terrain tiles bucket; 36 tiles covering WA bounding box). Raw HGT binary files parsed with NumPy; slope derived via numpy.gradient with per-latitude correction for east–west pixel width (dx = RES × cos(lat) × 111,320 m/deg). For each grid cell, flat_frac = fraction of ~90m pixels with slope < 5°. Hard gate: cells with flat_frac < 0.03 (< ~185 acres of sub-5° terrain) receive flatness_score = 0.0 and are excluded from composite scoring entirely. Passing cells: score = flat_frac / p95(flat_frac of buildable cells), clipped to [0, 1].

Protected land

Protected area boundaries from two sources: (1) Esri USA Federal Lands (ArcGIS Living Atlas public service), filtered to four agencies: National Park Service, Fish and Wildlife Service, Department of Defense, and Forest Service. BLM excluded (leasable for commercial use). (2) Census TIGER AIANNH (federally recognized American Indian and Alaska Native homelands). All sources reprojected to UTM Zone 10N (EPSG:32610), dissolved into a single multipolygon, and intersected with each grid cell using geopandas.overlay(how='intersection'). protected_frac = intersected area / cell area. Hard gate: cells with protected_frac > 0.25 receive protected_score = 0.0 and are excluded from composite scoring entirely.

Full methodology and reproducible notebooks available at github.com/simonhansedasi/wa-datacenter-suitability. Analysis uses Python / GeoPandas / Matplotlib. No proprietary data.