South Portland Elementary School Redistricting Model

A reproducible geospatial model for analyzing school boundary scenarios in South Portland, Maine.

What This Does

Models the Kaler closure scenario and optimizes the remaining school zone boundaries to:

1. Respect hard capacity constraints — no school exceeds its enrollment cap
2. Maximize community continuity — contiguous zones; bussed neighbors go together
3. Minimize travel distance — walkability and drive time factored into assignments

The interactive webapp (see webapp/) shows community and grade-center boundary options for the Kaler-closed scenario.

How It Works

Three-stage assignment algorithm:

• Stage 1 — Capacity-bounded Voronoi flood-fill from guaranteed seed blocks
• Stage 2 — Hard capacity enforcement with tiered contiguity-preference moves
• Stage 3 — Bussed community cohesion smoothing (keeps non-walkable neighbors together)

Uses OSM road networks (via OSMnx) for real walk/drive distance matrices.

Schools

School	Capacity	Coordinates
Brown	260	43.6347, -70.2489
Dyer	240	43.6219, -70.2749
Small	240	43.6413, -70.2339
Skillin	380	43.6260, -70.3054
Kaler	240	43.6287, -70.2688

Note: Closing Skillin is excluded — remaining four schools (capacity 1,020) cannot absorb all 1,013 K-4 students.

Quick Start

pip install geopandas osmnx folium networkx shapely pandas numpy
python3 run.py

Outputs saved to outputs/:

• *_map.html — Interactive folium maps (open in browser)
• *_boundaries.geojson — Zone boundary polygons
• *_block_assignments.csv — Per-block assignments with walk/drive distances
• *_metrics.json — School utilization, walkability %, avg drive times
• scenario_summary.csv — Ranked comparison of all scenarios

Grade-Center Mode

The webapp supports a grade-center view where PreK–1 students attend one set of schools and Grades 2–4 students attend a different set.

Zone pairing (Kaler closed)

Zone	PreK–1 school	Cap	Grades 2–4 school	Cap
West	Dyer	240	Skillin	380
East	Small	240	Brown	260 ← binding

The Brown G2-4 cap (260) is the binding constraint — it limits how many students can be assigned to the East zone regardless of the PreK–1 capacity available at Small.

Shared boundaries

Rather than optimising the PreK–1 and Grades 2–4 zones independently (which produces different boundary lines that could confuse families), the base case uses a single geographic partition for both grade bands. Every block's West/East assignment is identical whether you are looking at the PreK–1 or the 2–4 map.

prek1_full (grade-center with full PreK expansion) uses the same zone geometry as prek1_current. The extra PreK seats affect only the capacity-utilisation display, not the boundary itself.

Algorithm (compute_shared_boundaries.py)

1. Starting partition — Load the independently-optimised Grades 2–4 assignments. By construction these satisfy the Brown 260 cap; any K-1 imbalance is then fixed in step 2.

2. Capacity rebalancing — Iteratively move border blocks from over-capacity zones to under-capacity zones using three escalating move types:

   • Single-block moves (most targeted)
   • Peninsula moves (bridge block + all trailing components, guarantees contiguity)
   • Border swaps (exchange adjacent blocks between zones for a net capacity gain)

3. Distance optimisation — Move border blocks to the zone with the lower student-weighted drive cost:

   zone_cost = K-1 students × drive_dist_to_prek1_school
             + G2-4 students × drive_dist_to_g24_school
   

   A block is moved only when the new zone has a lower cost and all four capacity constraints remain satisfied after the move.

Re-running the boundary computation

If you need to regenerate the shared boundaries (e.g. after changing school capacities or block-level student counts):

/opt/anaconda3/bin/python3 compute_shared_boundaries.py

This patches webapp/public/data/kaler_closed.json in-place. Requires geopandas and shapely (available in the Anaconda environment).

Data

Polygons.geojson — 316 census blocks (2020 Census, South Portland, ME) with population counts used to estimate student distribution. Block-level race/ethnicity data (2020 Census PL 94-171) is included to display estimated percent minority students per school zone.