Case File · 004 · FloodGuard-BD · In Development · GCP Credits Required

Near-real-time
flood maps.
Every satellite pass.
No manual steps.

Bangladesh floods every monsoon. Existing systems rely on manual satellite interpretation, cloud-obscured optical imagery, and coarse delayed outputs. FloodGuard-BD closes the gap — Sentinel-1 SAR to pixel-level flood map, with calibrated confidence at every pixel.

10m
Resolution per pixel
<30min
SAR overpass to map
ECE<0.03
Calibrated uncertainty
0.830
Dice on validation
Open FloodGuard Live Demo
Prithvi-100M foundation model Sentinel-1 SAR · all-weather Evidential deep learning Pixel-level uncertainty Active learning loop SPARRSO partnership Apache 2.0 open source Prithvi-100M foundation model Sentinel-1 SAR · all-weather Evidential deep learning Pixel-level uncertainty Active learning loop
§ 01 — The Exposure

80% of Bangladesh
is flood-prone.
The tools are
not keeping pace.

The 2024 monsoon inundated more than 40% of the country. Millions displaced annually. Agriculture and infrastructure losses in the billions. These are not projections — they are last year's record.

FFWC and UNOSAT rely on optical imagery that clouds make unreliable precisely when floods are forming. Manual tracing misses the window for pre-emptive resource deployment.

The data exists. Sentinel-1 SAR passes over Bangladesh regardless of cloud cover. The gap is the intelligence layer that converts those passes into actionable maps within minutes — with a confidence score per pixel so disaster managers know where to trust the output.

>40%
Bangladesh inundated 2024
336×
Faster than manual GIS
$4
Pilot training compute cost
$0
Licensing cost to agencies

"Every dollar of infrastructure translates directly into flood maps that reach disaster managers before the water does."

§ 02 — Architecture

Fully automated.
From satellite overpass
to map delivery.

Sentinel-1 new scene ↓ Pub/Sub trigger ↓ Cloud Function → GEE tile export → Cloud Storage (COG tiles) ↓ Vertex AI Endpoint (T4) ├── Preprocess (normalize, batch, HAND derivatives) ├── ONNX Runtime inference (Prithvi-100M + Evidential head) └── Post-process → GeoTIFF (probability + uncertainty map) ↓ Cloud Storage (public output) ↓ Firestore (scene ID, timestamp, district area stats) ↓ Next.js Dashboard (Cloud Run) ├── Leaflet map + temporal slider ├── Uncertainty layer toggle ├── District-level flood statistics └── SMS/Telegram alert when area > threshold
§ 03 — The Model

Prithvi-100M.
10× less labeled data.
Foundation model
advantage.

NASA and IBM's geospatial ViT masked autoencoder, pre-trained on global HLS Sentinel-2 data. We extend patch projection layers to ingest Sentinel-1 SAR alongside Sentinel-2 bands and HAND terrain derivatives — a multimodal stack that sees through monsoon cloud cover.

Why Evidential Deep Learning
over standard softmax.

Standard softmax gives overconfident predictions in ambiguous regions — urban water versus shadow, submerged vegetation. A Dirichlet distribution head jointly predicts flood probability and calibrated uncertainty: both aleatoric (data noise) and epistemic (model ignorance).

Final calibration with isotonic regression on held-out UNOSAT maps achieves ECE <0.03. Uncertainty maps correctly flag regions where annotator agreement is also lower — confirming the uncertainty head is meaningful, not decorative.

Disaster managers get a trust score per pixel. That is the difference between deploying resources and waiting.

§ 04 — Validated Performance

Real Bangladesh
flood events.

ModelDatasetDice
Baseline U-Net (threshold)Gaibandha 20200.781
Prithvi-100M fine-tuned (pilot)Gaibandha 2020 + synthetic weak labels0.830
Target — full pipeline2019–2024 monsoon seasons>0.850
Sylhet 2022

Haor region inundation — validated against SPARRSO official records. Worst-in-decades event.

Gaibandha 2020

Brahmaputra floodplain — primary training and validation set. Char territory inundation.

Feni 2024

Most recent monsoon event — current benchmark against DDM field records.

§ 05 — GCP Services

Fully managed.
Zero manual ops.

Google Earth Engine

Sentinel-1 GRD + Sentinel-2 L2A acquisition, Lee speckle filtering, radiometric terrain flattening, dynamic 512×512px tiling with 256px overlap, COG-format export. Non-commercial tier, $0.

Cloud Storage

Raw COG tile archive with Standard→Coldline→Archive lifecycle policies, public flood map output bucket, model artifact registry. Cost-optimized for 3 TB scale.

Vertex AI · Training

Single NVIDIA T4 spot instance, ~2 hours per training run, TFRecord streaming from Cloud Storage. Approximately $12 per run at spot pricing.

Vertex AI · Prediction

T4 prediction endpoint, ONNX Runtime inference, autoscaling. Real-time inference triggered per new Sentinel-1 scene via Pub/Sub.

Pub/Sub + Cloud Functions

New Sentinel-1 scene notifications trigger automated pipeline. Cloud Function calls GEE Python API, exports tiles, fires Vertex inference, writes output GeoTIFF and Firestore metadata.

Cloud Run + BigQuery

Next.js dashboard with Leaflet map, temporal slider, uncertainty toggle, district statistics. BigQuery stores active learning corrections for weekly retraining cycle.

Cloud Scheduler

Weekly Cloud Build trigger for active learning retraining cycle. Corrections from human validation feed back into the model — adapting to evolving flood patterns without manual relabeling.

§ 06 — Open Source + Handover

Built to be
handed over.
Not licensed.

All code is released under Apache 2.0. A Colab demo notebook ships with the repository. The system is designed for handover to FFWC, the Department of Disaster Management, SPARRSO, or UN OCHA at zero licensing cost.

Partnership discussions with SPARRSO — Bangladesh's national space research agency — are already underway. The validation methodology is explicitly calibrated against their official flood event records.

Apache 2.0 license — free for government and humanitarian use
Colab demo notebook for independent validation
SPARRSO partnership discussions underway
Post-pilot operational cost under $500/month — fundable via humanitarian grants
Zero vendor lock-in by design