TopoBathySim

Policy-driven topobathymetric fusion runtime for data-first digital worlds.

TopoBathySim fuses heterogeneous elevation sources (global bathymetry, global land DEMs, and optional high‑resolution regional datasets) into seamless, analysis- and simulation-ready grids. It produces data-first outputs (xarray.Dataset) with per-pixel provenance (source masks), and optionally serves the same products over HTTP for real-time clients.

• Primary output: arrays + provenance, not pretty pictures
• PNG tiles and the viewer are debug/QA tools (seams, datum issues, coverage), not the product
• Fusion behavior is defined by YAML policy presets you can edit and share

What you can do

• Generate a fused DEM anywhere from a global basemap policy (coarse but robust) with clear provenance.
• Overlay regional high-resolution datasets (e.g., GBR100) on top of the basemap and blend edges smoothly.
• In data-rich coastal regions, prioritize “survey truth” (e.g., BAG / topobathy lidar) over compiled products, while retaining global fallbacks.
• Run a local tile/data service for Unity/Unreal/Omniverse workflows:
  • request fused elevation grids for simulation
  • visualize seams and provenance in the viewer for QA

Core concepts

1) Providers

A provider fetches a dataset layer (elevation or covariate), normalizes CRS/metadata, caches locally, and returns an xarray.DataArray.

Providers can be global (e.g., GEBCO + global land DEMs), regional compiled products (e.g., BlueTopo), or local high-resolution surveys (e.g., BAG, topobathy lidar). Adding a provider is the main contribution pathway.

2) Policies (YAML)

A policy is a YAML file that defines per-variable fusion strategies:

• which providers participate
• how to overwrite in trusted zones
• how to blend in transition zones
• fallback order for gap filling
• required outputs (e.g., source masks)

Policies are selected by filename, so you can keep policy presets in the repo (or your own project) and share them easily.

3) Operators

An operator is a pure, testable transformation used by policies (e.g., seamline feathering in meters for smooth edges).

4) Provenance (first-class)

Every fused variable includes a source mask (e.g., source_elevation) identifying the contributing provider for each pixel. Provenance is how you debug seams and validate that “truth” data really won.

Data sources and built-in providers

TopoBathySim includes a growing catalog of dataset providers. The current set reflects our initial focus on the Northeast US (where we have dense public coastal survey coverage). We warmly welcome contributors adding providers and policy presets for the EU, Asia-Pacific, and other regions.

Global basemap (worldwide)

• GEBCO 2025 (global topo+bathy baseline)
• Global land DEMs via STAC (e.g., Copernicus DEM / NASADEM where available)

Regional compiled bathymetry (mostly US coastal)

• NOAA BlueTopo™ (compiled best-available bathymetry for US waters)
• NOAA CUDEM (Ninth Arc-Second Topobathy) (regional integrated tiles; useful as a coastal gap-filler)

Local high-resolution “truth” layers (survey / lidar)

These layers are typically prioritized where they exist and may overwrite compiled products.

• NOAA NCEI BAG (Bathymetric Attributed Grids) High-resolution multibeam/sonar survey grids. When present, these are treated as “survey truth” for bathymetry.

• NOAA Topobathymetric Lidar (green lidar) Nearshore shallow-water bottom mapping (intertidal/subtidal) that complements airborne topographic lidar.

• USGS 3DEP / coastal lidar (topography) High-resolution land elevation, often used to mask water noise and sharpen the land–water interface.

See docs/source/data_sources.rst for citations and acknowledgements.

Docker

The service can be built and run in Docker. The Dockerfile defaults to nvidia/cuda:12.6.3-devel-ubuntu24.04 (x86_64). Override BASE_IMAGE for other targets:

# x86_64 / WSL2 with NVIDIA GPU (default)
docker build -t topobathysim .
docker run --rm --gpus all -p 9595:9595 topobathysim

# NVIDIA Jetson AGX Orin (JetPack 6.x)
docker build --build-arg BASE_IMAGE=nvcr.io/nvidia/l4t-jetpack:r36.2.0 -t topobathysim .
docker run --rm --gpus all -p 9595:9595 topobathysim

# macOS (Apple Silicon or Intel) via OrbStack / Docker Desktop — no GPU
docker build --build-arg BASE_IMAGE=ubuntu:24.04 -t topobathysim .
docker run --rm -p 9595:9595 topobathysim

Validated platforms: x86_64 (RTX 5080, CUDA 12.6, Ubuntu 24.04), NVIDIA Jetson AGX Orin (JetPack 6.1 / L4T r36.2), and macOS Apple Silicon (OrbStack).

Quick start (library)

Install (editable is recommended for development):

git clone https://github.com/lhzn-io/topobathysim.git
cd topobathysim
pip install -e .

Fuse elevation for a bounding box using a policy preset:

from topobathysim.runtime import run

# bbox = (west, south, east, north)
ds = run(
    policy_path="policies/global_default.yaml",
    bbox=(-74.05, 40.65, -73.70, 40.95),
    variables=["elevation"],
)

print(ds)
# ds["elevation"] -> float grid
# ds["source_elevation"] -> per-pixel provider id

Write a GeoTIFF (elevation only):

import rioxarray  # noqa: F401  (ensures .rio accessor)
ds["elevation"].rio.to_raster("fused_elevation.tif")

Quick start (service)

Run the server:

python service/run_server.py --host 0.0.0.0 --port 9595 --workers 4

Open the viewer:

• http://localhost:9595/viewer

Use the viewer to:

• inspect seams and coverage
• switch to provenance/source view
• verify that the policy is doing what you expect

For cache sizing and practical runtime expectations, see the documentation page: docs/source/performance.rst.

Policy preset gallery

Policy presets live in policies/:

• policies/global_default.yaml Global basemap fusion (coarse but robust). Good for quick demos and worldwide coverage.

• policies/iconic/great_barrier_reef.yaml Demonstrates how to overlay a high-resolution regional dataset (GBR100) on top of the global basemap and feather edges smoothly.

• (Northeast US) presets Examples that prioritize BAG + topobathy lidar + BlueTopo/CUDEM where available.

See policies/README.md for details, expected coverage, and suggested bounding boxes.

Output contract (data-first)

TopoBathySim’s canonical output is an xarray.Dataset:

• elevation (y, x) — fused elevation/depth in meters
• source_elevation (y, x) — integer source mask (provider ID per pixel)

Metadata is attached at the dataset and variable level, including:

• policy_hash (stable hash of the YAML policy)
• provider legend (mapping source IDs to provider names)
• CRS and vertical reference assumptions

International-friendly by design

TopoBathySim is global-first: you can run it anywhere with the global basemap policy.

Some datasets and datum transformations are region-specific. When vertical datum or reference is ambiguous, TopoBathySim surfaces this explicitly in metadata and provenance rather than hiding assumptions.

Contributing

Add a provider (best first contribution)

• Implement the Provider interface (fetch → normalize → cache → return DataArray + metadata).
• Add a policy preset that uses your provider.
• Add/update docs and citations.

If you’re outside the US: we would especially value providers and policy presets for:

• national hydrographic office products
• EMODnet / Copernicus Marine / regional lidar catalogs
• local survey archives and higher-update-frequency coastal products

Add a policy preset

If you have domain expertise for a region, you can encode it as a policy preset:

• overwrite in trusted zones
• blend at edges
• fill gaps in a sensible order
• publish provenance expectations

Documentation

• docs/source/ — API, methodology, CLI tools
• docs/planning/roadmap.md — roadmap (including time-varying covariates and Sentinel plans)
• policies/README.md — policy presets and how to run iconic demos

Built with agentic engineering

TopoBathySim is developed using agentic engineering practices — professional software development amplified by AI coding agents. The tools we use include Antigravity and Claude Code.

Domain expertise, architectural decisions, and quality judgment remain human responsibilities. The agents accelerate implementation, catch bugs, and handle boilerplate — but every provider, policy, and fusion strategy reflects deliberate engineering choices informed by geospatial domain knowledge.

License

MIT License (code). Datasets are provided by third parties under their respective terms; see docs/source/data_sources.rst and provider metadata for citations and licensing notes.