# Maine geography package — FLAGS (v1.2, 2026-10-14) ## 1. Area figures are not the Census Bureau's "land area" The GeoLibrary town polygons include inland lakes and rivers; only coastal (tidal) water is drawn as separate polygons, which this package strips off. So `area_sqmi_excl_coastal_water` is slightly larger than the Census Bureau's land area where a town has big lakes or a wide river (Lewiston 35.5 vs the Bureau's 34.2 sq mi because of the Androscoggin; Portland 21.4 vs 21.3). `pop_density_2020_per_sqmi` is population over that area. If a page needs the Bureau's land area exactly, it has to be pulled from TIGERweb (not done). ## 2. Places whose population differs from the Census 490 of the 493 organized places with a Census row match the 2020 Census exactly. The three that do not: * **Perkins Twp (Franklin), 37 people, vs the Census "Perkins UT"** — the Census's "Perkins UT" (2302358070) is **Perkins Twp Swan Island in Sagadahoc County** (0 people), not the Franklin township. The Franklin Perkins Twp (registration unit MPD-0139) is counted by the Census inside South Franklin UT, so it has no Census row of its own; here it is counted from the blocks inside its GeoLibrary polygon. (v1 described this as the Census assigning the Franklin residents to a "Perkins UT"; corrected in v1.1, and the three Swan Island blocks that had been attached to the Franklin township were moved back to Swan Island.) * **Drew Twp (Penobscot) 26, Codyville Twp (Washington) 13** — Codyville deorganized in 2019 and Drew in 2023 (State Auditor's list of deorganizations; v1.1 said 2019 for both). The 2020 Census still counted them as Drew and Codyville plantations; the ACS now reports them as Drew UT and Codyville UT. Since census data v1.2 / geography v1.1 every table uses the ACS codes (2301918582, 2302913612). * Not a mismatch, but note: **Indian Island** (Penobscot Nation reservation, 370) is its own polygon and unit; Greenbush 1,444 is the town without it. Every township population is a block count and depends on the Census internal point lying inside the right polygon; for tiny townships along a boundary a handful of people can land on the wrong side. Treat township counts under ~20 as approximate. ## 3. Split towns and small overlaps (2021 plan) 44 places have residents in more than one State House district, 4 in more than one Senate district (Yarmouth 74 % SD 25 / 26 % SD 23; also Bangor-area and Lewiston/Auburn precinct splits), 5 in more than one county commissioner district. The state's official block equivalency files also put very small pieces of some towns in a second district — Kennebunkport 13 people in HD 140, Cape Elizabeth 191 in HD 121, Brewer 292 in HD 24. These are the state's own block assignments, so they are kept; whether the Secretary of State actually reports separate returns for those pieces is a question for the election files: the 2024 general-election file does report Cape Elizabeth votes in both HD 121 (187) and HD 123, and Brewer in both HD 20 and HD 24 (141), but Kennebunkport only in HD 134 — so the 13-person HD 140 piece may have no voters. The 9 North Haven people that census data v1 placed in CD 2 / HD 40 / SD 11 are **not** in the state's files (all 417 are in CD 1 / HD 42 / SD 12); census data v1.1 adopts the state's assignment. ## 4. Older plans were drawn on older blocks The 2003 and 2013 plans were drawn on 2000 and 2010 Census blocks; this package measures them with 2020 blocks and 2020 population. That is exactly what you want for "who lived where under the old map" comparisons using today's counts, but the shares of a split town can differ slightly from what the plan's own documents show, and a sliver of a town that had a few 2010-era residents but no 2020 internal point disappears from the crosswalk. Known cases where the election files report a town in a district the crosswalk lacks: * 2003 plan (2004–2012): Milford in HD 11 (2004–06), Portage Lake HD 1, Winterville Plt HD 2, Blaine HD 7, Bridgewater HD 7, Castle Hill SD 34, Ashland SD 35 — all tiny Aroostook/Penobscot pieces; the GeoLibrary's 2003 layer does not cover the unorganized townships and coastal water at all (640 people had to be attached to the nearest polygon), so it is the least precise of the three plans. **Edinburg (Penobscot) is a real 2003-plan split that the crosswalk misses** (added v1.2, audit II GC-14): it reported in both HD 12 and HD 13 in every general election 2004–2008 (73 + 69, 55 + 51 and 72 + 89 votes, and each pair adds up to Edinburg's vote for President or Governor that year; in 2010 and 2012 only HD 12 has an Edinburg row), but all 18 of its 2020 blocks fall in HD 12 under the GeoLibrary's 2003 layer, so `town_district_crosswalk_all_plans.csv` has 100% of Edinburg in HD 12. The block table gives no way to divide it (the plan's own split used 2000 blocks), so the crosswalk is left as is; use the election files for Edinburg's 2004–2012 House district. * 2013 plan: Ellsworth HD 136 (44 votes in 2014 — a real sliver), Sherman HD 144, Milford HD 137, Old Town HD 121, Strong / Wilton HD 112, New Portland / Solon HD 118 — rows with 0–3 votes apart from Ellsworth. * 2021 plan: Millinocket SD 4, Eustis SD 5, Milbridge / Steuben / Franklin / Gouldsboro HD 18, Brownville HD 30, Jackman HD 72, Medway HD 8, Albion SD 3, Madrid Twp SD 3 — rows with 0–7 votes in the June primary files, which look like SOS spreadsheet artefacts rather than real districting. **Exception (v1.2 wording):** Millinocket's SD 4 row is not only a June-primary artefact — the November 2022 general-election file also has a Millinocket row in SD 4 (49 + 23 + 2 = 74 votes) although all of Millinocket is in SD 2 under the plan and the block files. Cause unexplained. * 2011 congressional plan: Washington (Knox County) 13 votes in CD 2 in 2018 — the crosswalk has all of Washington in CD 1; unexplained. * Legislative elections before 2004 used the 1993 plan, which the GeoLibrary does not publish; no block-based crosswalk for 1990–2002 legislative districts. Town-to-district lists for the 1983 and 1993 plans, read from the statute and court order, are in `district_plans_1983_1993/` (no populations). The 2003 congressional layer does not describe 1990–2002 either (14–23 towns differ). ## 5. County commissioner districts The 2021 plan layer has 58 districts; Franklin County went from 3 to 5 commissioners in 2024, so the 2024 update layer's five districts replace Franklin's three. The 2013 layer is as published (58). No 2003-era commissioner layer exists on the GeoLibrary. ## 6. Layers downloaded but not turned into tables `sources/voting_districts_2010.geojson.gz` (the 2010 Census voting districts with the SENATE / HOUSE / CNTYCOMM in force in 2010) and `sources/school_units.geojson.gz` (full school-unit polygons; only the town → school unit lookup was kept, in `town_attributes.csv`). Precinct boundaries for the cities that report by ward/precinct are not on the GeoLibrary; the SOS precinct results still have no map. ## 7. Simplification Web files are simplified (towns ~40 m, districts ~60 m, counties ~80 m), so adjacent polygons can show hairline gaps or overlaps at high zoom and areas computed from them will be off by a fraction of a percent. Areas in `town_attributes.csv` were computed from the full-resolution polygons (in `sources/`) before simplification. ## Regions: two 2020 population figures (noted 2026-10-03, audit II) `regions_summary.csv` sums `regions.csv` (GeoLibrary town and township polygons, 2020 block populations) — total 1,362,359. `census-data/census_profile_regions.csv` sums Census county subdivisions (`regions_census_cousub.csv`). The two bases differ for three regions by a few people to a few dozen (Western Maine 170,813 vs 170,807; Central Maine 196,290 vs 196,279; North Woods 48,612 vs 48,629) because a few unorganized-township boundaries differ between the GeoLibrary and Census shapes. Use `regions_summary.csv` for map and area figures and `census_profile_regions.csv` for Census indicators; neither is wrong. ## 8. GIS slivers in the older plans (v1.1, 2026-10-05) The 2003 and 2013 layers are measured with 2020 block internal points (§4), so a block whose point lies a few metres over a town line can land in the neighbouring town's district. Two such cases carried real population and were corrected by hand in `block_assignments_all_plans.csv` (plan documents are not available, so the judgement rests on the block geometry, the rest of the county and the election files): * 2003 congressional plan — Harrison block 230050150001025 (91 people), internal point 5 m from the Harrison/Oxford County line: was CD 2 and House 100, now CD 1 and House 98. Cumberland County is otherwise entirely in CD 1 (5,552 of 5,553 blocks) and Harrison was never reported in CD 2 or House 100. * 2013 county-commissioner plan — Westbrook block 230050029023008 (42 people), 15 m from the Gorham line: was Cumberland-1, Senate 30, House 27; now Cumberland-4, Senate 28, House 34 like its five nearest Westbrook blocks. Westbrook never reported in those districts in 2014–2020. Other single-block pieces of towns in the older plans were not reviewed one by one; the §4 list of election-file mismatches is still the best guide to them. ## 9. Region tables vs. Census codes (v1.1) `regions_census_cousub.csv` now uses the ACS 2020–2024 county-subdivision codes, the same as `census-data/census_profile.csv`. With census data v1.2 the two 2020 population figures in the note above are unchanged (they differ only where GeoLibrary and Census township lines differ), and every region's geoid list now joins to the profile: before, Drew, Codyville and Magalloway (26/13/45 people in 2020) were listed under codes the profile does not carry, so their ACS figures were missing from North Woods and Downeast and their 2020 counts did not match a profile row. ## 10. Historical units without a certain region (v1.2, 2026-10-14) `regions_reporting_units.csv` now has the 49 units that election results v3.11 added (MPD-1033 … MPD-1081: 1831–1839 Governor-return places, two 1872–73 Annual Register places, Summit Plt from 1892, and 32 tabulation-book lines that are not yet identified with a present-day place). A region was given only where it is certain: * successor town or present-day township (Danville → Auburn, Clinton Gore → Clinton, Summit Plt → Summit Twp, …); * a county that lies wholly in one region *at the time the place voted* (Kennebec → Central Maine; Washington after 1839 → Downeast). Before 1839 Washington County also held present-day southern Aroostook, so a bare "No. 25 (Washington)" from 1831 is not placed; * every candidate location in one region (e.g. Penobscot "No. 2", which is Milton/Orneville, Lagrange or the Medford area — all North Woods; Oxford numbered townships, because the only Oxford places outside Western Maine are Buckfield and Canton, both incorporated before 1831). The other **18 have a blank region** with the reason in `basis`, and are left out of region sums like the 1918 soldier votes: Gore between Poland & Raymond; Page's Mills and Township No 9 (Hancock); No. 3 8th Range, Sunusville and Township No 3 (Penobscot); the four Haskell's Plantation spellings (candidates Kingsbury and Wilson are North Woods, Mayfield is Central Maine); three Somerset bracketed township lines; No 25 / No. 17 / No. 25 (Washington, 1831–32); Wellington and Pl. No. 10 (returned through Houlton, 1831). All are 1831–1839 lines with a handful of votes each.