FathomWave Geospatial · Fairbanks, Alaska

Toward complete bathymetric coverage of Alaska's waters

Alaska's nearshore waters and rivers are critically undermapped, limiting planning and operations. We use machine learning and free satellite imagery to produce depth estimates with per-pixel uncertainty, delivered through an agentic framework that folds user knowledge and local data into the estimate. Our models support safer navigation, infrastructure planning, and emergency response.

See the products How it works

Physics-constrained Interpretable Uncertainty on every pixel Sub‑meter RMSE across 19 Alaskan sites*

19Alaskan coastal sites validated against LiDAR

sub-meter RMSE at every site*

2016 →any requested date since, on any coast with Sentinel-2 coverage

Sentinel-2 archive

1ststudent-led National I-Corps team in Alaska's history

NSF I-Corps

Winner2026 UAF Best Graduate / Post‑Graduate Innovation

University of Alaska Fairbanks

Approach

What We Do

We turn publicly accessible, regularly repeating satellite imagery into satellite-derived bathymetry (SDB) using physics-informed, probabilistic models that provide interpretable, defensible depth in the turbid, sediment-heavy water where older methods failed. We are also building agentic frameworks that fold community knowledge into the map itself, so what residents know about their coast becomes part of the estimate rather than a note beside it.

Products

One Model Family, Two Water Bodies

For nearshore coastal and riverine areas alike we deliver the same two rasters for any requested date since 2016: a best-estimate depth grid and a per-pixel uncertainty map. The two share a model family, but each is tuned to the physics of its own environment.

Nome, Alaska: airborne LiDAR depth (top) beside the PINN prediction (bottom) over the same nearshore area

Flagship

Nearshore coastal

Most SDB methods were built for clear tropical water. Ours was trained on the sediment-heavy, ice-affected Alaskan coast, where turbidity is the design case rather than the failure mode.

SWOT surface elevation, GRADES discharge, and USGS gauge records feeding a physics-informed neural network that maps channel depth along a whole river reach, with a cross-section showing depth, barge clearance, and the uncertainty band along the bed

In development

Rivers

A survey boat can cover only a localized section of river in a day. We can automatically map an entire reach from free open-source riverine data by embedding the physics of open-channel flow directly into the model. The model yields channel depth along with explicit uncertainty estimates.

Who it's for

Built With the People Who Use It

Currently we are conducting customer discovery with National NSF I-Corps. We work directly with Alaskan coastal communities: residents and regional experts flag what the imagery misses — a shifted bar, a new channel, ice — and our models are built to take that knowledge in and revise the map, not just be checked against it afterward.

Communities

Coastal villages & tribal organizations

Relocation planning, erosion monitoring, and safe barge and skiff landings.

Agencies

State & federal

Charting gaps, flood and storm-surge modeling, infrastructure permitting.

Engineering

Coastal & marine consultants

Pre-survey reconnaissance and change detection between vessel campaigns.

Research

Universities & observing systems

Repeatable, uncertainty-aware depth for habitat, sediment, and climate work.

One depth map, four users: barge approach depth for communities, storm-surge reach for agencies, bar migration for engineers, and year-over-year change for researchers

Translating geospatial data into decisions is no small feat. At FathomWave Geospatial, we integrate the strength of geospatial AI agents wrapped in a controllable agent harness to push data through the processing pipeline. Geospatial AI agents allow more data to be translated into decisions, ending in tangible products that can be used and enhanced by local knowledge holders.

If you're one of these and have a site in mind, tell us about it. Early partners shape what we build.

How it works

Probability and Physics Keep the Deep‑Learning Networks Honest

A depth estimate without an error bar is a guess. It might be right, but the number itself gives you no way to tell, and no way to defend the decision you make on it. Our networks must satisfy the physics of how light attenuates through water, and they return a probability distribution for every pixel, not a single number. That uncertainty map is the product: it tells a planner which depths to build on, which to verify, and where to send a survey vessel.

Satellite imagery

Free, global, revisited every few days. Sentinel-2 for optical depth and SWOT/GRADES for river data.

Sentinel-2 optical, Sentinel-1 SAR, and SWOT radar observing water extent, quality, and surface elevation

Physics-informed, probabilistic model

PINN and ProbAI architectures trained against LiDAR ground truth, with the physics residual as a loss term. The network predicts a distribution for every pixel, not a point, so the answer carries its own confidence. Interpretable by construction.

Physics-informed neural network training loop: inputs, depth and attenuation predictor, physics constraints, total loss, optimization

What you receive

Two GeoTIFF rasters for the date you ask for: the most probable depth and the uncertainty around it, pixel by pixel. Any coast with Sentinel-2 coverage, any date since 2016. Re-run after every storm season.

Nome: predicted depth (left) and total uncertainty (right) from the Bayesian KAN model

Validation

19 Sites, Sub‑Meter Accuracy at Every Site

From Akutan to Kaktovik, each site is validated against airborne LiDAR. These are the coasts where the problem is hardest: turbid, ice-affected waters that are largely uncharted.

Map of Alaska showing 19 validated coastal sites from Akutan to Kaktovik — Basemap: USGS 3DEP · Esri · GEBCO

Arctic coast 5

Utqiaġvik · Wainwright · Point Lay · Point Hope · Kaktovik

Kotzebue & Norton Sound 7

Kivalina · Cape Blossom · Shishmaref · Teller · Nome · Golovnin Bay · Unalakleet

Bering Sea 1

Hooper Bay

Cook Inlet 3

Ninilchik · Seldovia · Homer

Alaska Peninsula & Aleutians 3

Nelson Lagoon · Cold Bay · Akutan

RMSE< 1 m at all 19 sites*

Ground truthairborne topobathymetric LiDAR

ImagerySentinel-2, 2016 → present

Research

Peer-Reviewed Foundations

Spun out of the UAF Coastal Mapping Lab. The methods are peer-reviewed and open-sourced. * Accuracy figures on this page are from work currently under peer review.

In publication

A Checklist for Ethically Applied GeoAI with End-Users Using Earth Observation and Geospatial Data

Crowley, M.A., Trochim, E., Gongora-Svartzman, G., Harvie, J., Zammit, K., Engle, B., Cantin, A.S., Jain, P. & McFayden, C.B. In: The Applied Artificial Intelligence and Deep Learning Book. National Aeronautics and Space Administration (NASA).

In review

Physics- and Kolmogorov-Informed Neural Networks with Soft-Attenuation Constraints for Satellite-Derived Bathymetry

Flynn, B., Trochim, E.D. & Borger, L.

In preparation

Ontology-Driven Agentic Orchestration of Geospatial Workflows for Satellite-Derived Shoreline Extraction in Alaska

Trochim, E.D., Hutchinson, M., Helder, N. & McArthur, M.

In preparation

Examining Turbidity in the Beaufort Sea, Alaska Using Remote Sensing

Trochim, E.D., Flynn, B.A., Dykstra, S., Undiz, B. & Moriarty, J.

Insights

Notes From the Coast

Work published by the FathomWave team that shows how we think about the coast, including tutorials, explainers, courses, and field notes. We also point to the reports and observing-system briefs that frame the problem we work on.

Tutorial · 2025Medium

Satellite-derived bathymetry: a tutorial in Google Earth Engine

Rinzler, McArthur, Helder, Sierra & Trochim

Explainer · 2024Medium

Topobathy LiDAR 101

McArthur, Lomelino & Trochim

CourseedX · AlaskaX

Introduction to Topobathymetric Lidar Data

University of Alaska Fairbanks

Explainer · 2025Medium

LiDAR data security

Lomelino, McArthur & Trochim

Report · 2026US AON

Arctic Observation Storylines: Coastal Flooding

US AON technical brief on risk management and hazard mitigation

About us

FathomWave Geospatial is a student-led team of coastal scientists and machine-learning researchers building depth data for Alaskan communities. Our products are built for a range of end users, supporting community and relocation planning, navigation and safe landings, infrastructure and permitting decisions, and flood and storm-surge models. For most of Alaska's nearshore coast and rivers, the depth data we're creating is out of date or has never existed at all.

FathomWave spun out of the UAF Coastal Mapping Program. The methods our business is founded on are scientifically rigorous, peer-reviewed, built to scale, and open source wherever possible.

2026 UAF Best Graduate / Post‑Graduate Innovation — Winner