SAR Flood Isolation Platform

SAR Flood Isolation Platform

Sentinel-1 thresholding comparison and settlement isolation · 2024 spring floods, Kazakhstan · research prototype

Synthetic benchmark · seed 42 grid —

Defence scenarios

PRESETS

Pre-processing

OBJ 1

Thresholding

OBJ 1–3

−1 dB is stricter (less water), +1 dB looser. Local tiles use Otsu only where Ashman's D > 2.

Road network

OBJ 4

Kulsary package from the Colab notebook (real Sentinel-1, HAND, OSM roads and places), or an OSM roads GeoJSON.

How to make the Kulsary package
  1. Open kulsary_gee_package.ipynb in Google Colab.
  2. Run all cells and sign in with the Google account that has access to the Earth Engine project.
  3. The last cell downloads kulsary_package.json (about 5–15 MB).
  4. Load that file here. Everything on this page then runs on the real scene.
Roads only: export from Colab
import osmnx as ox
G = ox.graph_from_place("Kulsary, Kazakhstan", network_type="drive")
edges = ox.graph_to_gdfs(G, nodes=False).reset_index()
edges["highway"] = edges["highway"].astype(str)
edges[["u", "v", "length", "highway", "geometry"]].to_file(
    "kulsary_roads.geojson", driver="GeoJSON")
from google.colab import files
files.download("kulsary_roads.geojson")
Building scene
Simulating Sentinel-1 backscatter, speckle and terrain…
→Choose a scenario on the left, or change any setting. Every number is recomputed in your browser.

Result

LIVE

Accuracy on test points

OBJ 2

Settlements

OBJ 4
Select a settlement to see why it is open, delayed or cut off.

Method comparison at the current settings

OBJ 2 · OBJ 4 · SUBQ 2

36-condition sensitivity matrix

OBJ 3 · SUBQ 1

Each cell is one full run: threshold → mask → road overlay → routing. Click a cell to open it.

Run log

P1 → P4