Spatial Analysis

At a glance

• Outcome: Three spatial questions taken far enough to act on, each reporting what the data supported, including two where the finding contradicted the hypothesis I started with.
• Status: Case studies, complete.
• Role: Solo: framing, data gathering and scraping, analysis, and cartography.
• Stack & libraries: ArcGIS Pro, Python (arcpy, pandas), and ESA SNAP for the radar processing.
• Scale: ~80,000 geocoded building permits, 548 scored electric service areas, and multi-scene Sentinel-1 SAR mosaics.
• Validation: Findings reported against the null result where there was one; four of five final candidate feeders were independently already flagged as worst-performing circuits.
• Limitations: Pacific Power publishes no feeder boundaries, so those service areas are Thiessen approximations from circuit linework, not ground truth.

Overview

Three spatial questions I wanted answered well enough to act on. For each I framed the question, gathered or scraped the data, and ran the analysis. I report what the data supported, including the two times my hunch was wrong. ArcGIS Pro and Python (arcpy, pandas), with ESA SNAP for the radar work.

SAR flood mapping: how close are the flood maps?

Question. During a real high-water event in the Coquille River valley (Coos County, OR), how well do FEMA’s flood-hazard maps, the ones flood insurance is priced against, match where the water actually went?

Method. Processed Sentinel-1 SAR in ESA SNAP (orbit correction, radiometric calibration, terrain flattening and correction for the valley’s hills), classified open water with ESRI’s pretrained Water Body Extraction model, and compared the detected extent on the flood weekend against the mapped zones. ALOS-2 PALSAR and Landsat 9 for reference.

Finding. The FEMA zones tracked the observed flooding closely. The valley-floor floodplain is a long-recognized, well-mapped feature, so here the maps and the radar largely agreed.

Water quality and school performance: does one track the other?

Question. Across Oregon, is there a spatial relationship between drinking-water quality and standardized test scores, or does any apparent link stand in for something else?

Method. Test scores from the Oregon Department of Education; water-quality demerit scores scraped and hand-curated from an Oregon water-system source; Census density and income. Aggregated to regions and symbolized as a bivariate choropleth.

Finding. The direct relationship was weak. The pattern lined up far better with density and income. Schools in eastern Oregon could post similar scores while differing enormously in resources. The interesting part was assembling the question from three separate public sources, including water data that existed publicly but had to be scraped to be usable at all.

Electrification and grid reliability: do electrifying neighborhoods lose power more?

Question. Do Portland’s distribution areas under the most electrification pressure, meaning EV adoption, heat-pump and electrical permits and densification, also have worse grid reliability? The question came from noticing frequent short outages in my North Portland neighborhood as multi-unit housing and chargers went in, against the rare outages in the deeper Southeast where I’d lived before.

Method. An eight-step ArcGIS Pro and pandas pipeline over ~80,000 geocoded building permits, tract-level EV registrations, PGE’s SAIDI/SAIFI filings to the Oregon PUC, feeder polygons, Pacific Power circuit linework, and ACS estimates. Pacific Power doesn’t publish feeder boundaries, so I built approximate service areas from circuit linework with Thiessen polygons, area-weighted the tract data onto feeders, scored every area on two independent axes, transition pressure and grid stress, then combined them so only areas elevated on both ranked highly.

Finding. Across 548 service areas the two axes were essentially uncorrelated (Spearman ρ ≈ 0.05). Electrification pressure and poor reliability don’t track each other system-wide, the opposite of my starting hunch. But the framework still isolated ten overlap hotspots worth an engineering look, the Sylvan-Barnes feeder and the Vernon substation area among them, and four of the five PGE picks were already flagged as worst-performing circuits. The useful output wasn’t a correlation. It was a repeatable way to target where the two problems actually coincide.