Flood Susceptibility Map (M-AHP, 2015)

An interactive web map of flood susceptibility for the Ankara region. A susceptibility raster produced with the Modified Analytic Hierarchy Process (M-AHP) is converted into a single static PMTiles vector tile archive and served with MapLibre GL — no tile server or backend required.

Based on the academic study: Sozer, B., Kocaman, S., Nefeslioglu, H. A., Firat, O., and Gokceoglu, C.: PRELIMINARY INVESTIGATIONS ON FLOOD SUSCEPTIBILITY MAPPING IN ANKARA (TURKEY) USING MODIFIED ANALYTICAL HIERARCHY PROCESS (M-AHP), Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLII-5, 361–365, https://doi.org/10.5194/isprs-archives-XLII-5-361-2018, 2018.

How the susceptibility map was generated (M-AHP)

The susceptibility raster (data/mAHP_2015_son_dp5_clip.tif) was produced in the study above. Several flood-conditioning factors are scored and combined with M-AHP; see the paper for the full methodology, factor weights, and validation.

flowchart TD
    L8[Landsat-8] --> LC[Land cover]
    TV[TopoVT vector data] --> DTM[Digital terrain model]
    TV --> HY[Hydrology data]
    DTM --> TWI[TWI / Flow accumulation / Slope]
    HY --> DS[Distance to dry & permanent streams]
    LIT[Lithology data]

    LC --> SC[Parameter scoring]
    TWI --> SC
    DS --> SC
    LIT --> SC

    SC --> MAHP[M-AHP] --> FSM[Flood susceptibility map]

Loading

Web-map pipeline

This repository starts from the produced raster and turns it into a web map in three steps:

clip.tif  ──(1)──>  data/sel.geojson  ──(2)──>  docs/sel.pmtiles  ──(3)──>  GitHub Pages
 (UTM 36N)          (WGS84, class 1-5)          (vector tile)               (MapLibre)

Stage	Script	Output
1. Classify + polygonize + reproject	scripts/classify_polygonize.py	data/sel.geojson, data/breaks.json
2. Vector tiles	scripts/build_pmtiles.ps1 (Docker tippecanoe)	docs/sel.pmtiles, docs/breaks.json
3. Publish	GitHub Pages	live map

Requirements

• Python 3.10+ — runs the classification/polygonization step.
• Docker — runs tippecanoe (no native Windows build) to make the vector tiles.
• Node.js — only for local preview (npx serve).

You do not need GDAL on PATH, a GIS desktop app, or a tile server.

Setup (one-time)

python -m venv .venv
.\.venv\Scripts\pip install -r requirements.txt
# tippecanoe has no native Windows build -> build a local Docker image:
docker build -t tippecanoe-local -f scripts/tippecanoe.Dockerfile .

Build the map

1. Classify + polygonize + reproject

scripts/classify_polygonize.py:

• Assigns the UTM 36N (EPSG:32636) CRS to the raster (from the .tfw + boundary shapefile .prj).
• Splits the continuous mAHP values (~3.91–45.93) into 5 classes using Jenks natural breaks.
• Removes speckle with sieve (merges noise into the neighboring class).
• Polygonizes connected same-class cells and reprojects UTM 36N → WGS84.
• Produces data/sel.geojson (each feature has a class 1–5) and data/breaks.json.

.\.venv\Scripts\python scripts/classify_polygonize.py

2. GeoJSON → PMTiles

scripts/build_pmtiles.ps1 runs tippecanoe in Docker with -l susceptibility --coalesce --reorder --drop-densest-as-needed -Z6 -z14, producing docs/sel.pmtiles (~8.5 MB) and copying breaks.json into docs/.

pwsh scripts/build_pmtiles.ps1

3. Publish (GitHub Pages)

• Commit docs/index.html, docs/sel.pmtiles, and docs/breaks.json.
• GitHub repo → Settings → Pages → Deploy from a branch → branch main, folder /docs.
• GitHub Pages supports HTTP range requests, so PMTiles works without a server, and sel.pmtiles (~8.5 MB) is well below GitHub's 100 MB file limit.

Local preview

Python's http.server does not support range requests (which PMTiles needs); use a range-capable server:

npx serve docs -l 8765
# http://localhost:8765

The map

docs/index.html is a self-contained page (MapLibre GL + pmtiles from CDN, keyless CARTO basemap). It colors the polygons green→red by the class (1–5) attribute and builds the legend and initial view (fitBounds) from breaks.json.

Susceptibility classes (Jenks, 2015)

Class	mAHP range	Susceptibility	Color
1	3.91 – 10.75	Very low	#1a9850
2	10.75 – 16.99	Low	#a6d96a
3	16.99 – 22.02	Medium	#fee08b
4	22.02 – 28.00	High	#fdae61
5	28.00 – 45.93	Very high	#d73027

Project structure

data/                          Source raster (+ .tfw world file) and study-area boundary
docs/                          Published site: index.html, sel.pmtiles, breaks.json
scripts/
  classify_polygonize.py       Raster -> classified WGS84 polygons + breaks.json
  build_pmtiles.ps1            GeoJSON -> docs/sel.pmtiles (Docker tippecanoe)
  tippecanoe.Dockerfile        Builds the local tippecanoe-local image
requirements.txt               Python dependencies

Notes

• The source raster has no embedded CRS; positioning comes from the .tfw world file (UTM Zone 36N / EPSG:32636, 10 m pixels, Ankara region).
• data/sel.geojson (~80 MB intermediate output) and .venv/ are git-ignored.