Point bars (red) and counter point bars (blue) mapped from time-lapse Landsat imagery along the Mamoré River, Bolivia, colored by the ‘bar type index’
Most of my work is about channels — rivers, and the submarine channels that are their deep-water counterparts — and about how the landscapes they shape turn into stratigraphy. I am interested in how much of what happens at the surface survives into the rock record, how to read it, and how to make that reading quantitative and reproducible. The data come from three-dimensional seismic surveys, satellite imagery, outcrops, cores, and well logs; the tools are mostly simple models and open-source Python code. Much of this work is done through the Quantitative Clastics Laboratory research consortium at the Bureau of Economic Geology, with Jake Covault and a group of students, postdocs, and collaborators.
Rivers: meandering, migration, and the point-bar record
Meandering rivers are the simplest yet - in my opinion - the most interesting and beautiful of channel systems. We have shown that, contrary to a widely held view, the curvature of a meandering river is a good predictor of its migration rate, and that rivers not strongly affected by variations in bank erodibility are well described by simple curvature-based kinematic models. The same model, implemented in the open-source package ‘meanderpy’, explains how counter point bars develop through autogenic downstream translation, and how bend cutoffs reset the kinematics of neighboring bends.
More recently, we have been using time-lapse Landsat and Planet imagery to track river banks and centerlines automatically, with ‘rivabar’, a graph-based Python tool; the goal is to understand what controls channel migration and the complexity of the resulting point-bar stratigraphy. In the ancient record, we used digital outcrop models to document the three-dimensional anatomy of a Cretaceous river avulsion in the Cedar Mountain Formation of Utah — an analog for intersecting fluvial ridges on Mars. And in the Arctic, we have used a convolutional neural network to map ice-wedge polygons on point bars, which turn out to record the growth history of the bars.
High curvatures drive river meandering (Geology, 2019) • Autogenic translation and counter point bar deposition (GSA Bulletin, 2021) • Meander-bend curvature, sediment supply, and migration (JGR Earth Surface, 2021) • Post-cutoff bend curvature on the Trinity River (GSL Special Publications, 2024) • Meandering streamflows across landscapes and scales (GSL Special Publications, 2024) • Three-dimensional anatomy of a Cretaceous river avulsion (Geology, 2024)
Talks: Accreting, fast and slow (2026) • A frozen history of meandering (2023) • The three-dimensional geometry and stratigraphic record of meandering (2021) • meanderpy (2019)
Posts: Does river migration slow down in high-curvature bends? • Point bars and counter point bars • Rivers through time, as seen in Landsat images
Aggradational meandering river model
Submarine channels and deep-water systems
I spent the first part of my career working on turbidites and submarine channels — in outcrops in the Carpathians, in cores, and especially in three-dimensional seismic reflection data, which remains unparalleled when it comes to understanding the true three-dimensionality and spatial context of depositional systems. Much of that work continues. A recurring theme is that submarine channels evolve in a surprisingly river-like way: they migrate through bend expansion and downstream translation, they develop cutoffs, and the cutoffs create knickpoints that propagate upstream and sculpt terraces and remnant channel deposits — all without any external forcing. The vertical trajectory of a channel through time, combined with this lateral migration, is a first-order control on the stratigraphic architecture of channel-levee systems, an idea we have developed with kinematic models and tested against seismic data from the Gulf of Mexico, offshore Angola, the Niger Delta, the Campos Basin, and the Taranaki Basin.
Other threads in this work include turbidite bed-thickness statistics, three-dimensional numerical simulations of turbidity currents, cyclic steps and other supercritical bedforms, the (lack of) evidence for a Coriolis effect on submarine channel sinuosity, and the stratigraphic evolution of shelf-margin deltas and the submarine channels and landslides that they feed.
Seismic stratigraphy of a shelf-edge delta and linked submarine channels (SEPM Special Publication, 2012) • A model of channel-levee evolution based on channel trajectories (Marine and Petroleum Geology, 2011) • Cutoff-related knickpoints in submarine channels (Geology, 2016) • The stratigraphic record of submarine-channel evolution (The Sedimentary Record, 2016) • Channel trajectories control deep-water stratigraphic architecture (The Depositional Record, 2022) • Fluvial-style migration in the Joshua Channel (GSL Special Publications, 2024) • Giant meandering channels in the Campos salt basin (Geosphere, 2021) • Submarine channels ‘swept’ downstream after bend cutoff (The Depositional Record, 2019) • Submarine-channel meandering reset by landslide filling (The Depositional Record, 2024) • Landslide influence on delta stratigraphy (Geosphere, 2026) • Rapid adjustment of channel architecture to changes in sediment supply (JSR, 2015) • Linking seafloor dynamics to a submarine channel fill in outcrop (JSR, 2020) • Cyclic steps and related supercritical bedforms (Marine Geology, 2017) • Submarine fans and their channels, levees, and lobes (Springer, 2018) • Turbidite bed thickness distributions (Sedimentology, 2007)
Seismic image from two channel-levee systems (Fuji and Einstein) in the Eastern Gulf of Mexico
Time in the stratigraphic record
How much time is actually recorded in a sedimentary succession, and how is it distributed? Chronostratigraphic (Wheeler) diagrams are the natural way to think about this, and with the Python package ‘stratigraph’ we have made them easy to build and analyze in three dimensions, from forward models, flume experiments, and seismic data: maps of stratigraphic completeness, the nature and age structure of erosional surfaces, and the way deposition and erosion are partitioned in space and time.
The same way of thinking underlies ‘ChronoLog’, a proprietary package for correlating a large number of geophysical well logs. Pairwise correlations are computed with the dynamic time warping (DTW) algorithm; a least-squares optimization then stretches and squeezes all the logs into a relative-geologic-time framework that is consistent in three dimensions — correlation as a transform into the time domain, rather than as a set of lines drawn on a cross section. With USGS colleagues, we applied DTW to the question of whether turbidites along the Cascadia subduction zone — potential records of megathrust earthquakes — can be confidently correlated over large distances.
I have also become interested in the parts of the record that are not fragmentary. The stratigraphic record is famously incomplete, and signals are often shredded; but laminated siltstones of the Permian Spraberry Formation can be correlated lamina by lamina over 15 km, cross sets of the Jurassic Navajo Sandstone record seasonal wind reversals year by year, and point bars of rapidly migrating rivers are stacks of flood deposits. At these scales, seasonality is the dominant pacemaker of the sedimentary record, and these basic architectural elements preserve signals that larger-scale and more generic analyses suggest should be lost.
Stratigraphy in space and time (Earth-Science Reviews, 2024) • Automated multi-well correlation using relative geologic time (Basin Research, 2023) • Turbidite correlation for paleoseismology (GSA Bulletin, 2024)
Talks: The strange ordinariness of high-frequency signal preservation (2026) • Stratigraphic correlation: Old concepts inform new approaches (2022) • Chronostratigraphic diagrams revisited (2021)
Wheeler’s (1964) chronostratigraphic diagram, and its equivalent computed from a simple forward model with ‘stratigraph’
Stratigraphic forward modeling
Simple forward models are, I think, the best way to build intuition about how sedimentary systems respond to their boundary conditions — and to find out which of our intuitions are wrong. With diffusion-based models (Dionisos, Badlands) we have looked at how eustasy, grain size, and discharge control sand delivery to deep-water fans and slope basins, and how tectonic and climatic changes in a source area are transformed — and sometimes garbled — by the sediment-routing system before they are recorded in basin fill. Margin progradation turns out to be a good proxy for sediment supply, while provenance and sediment supply can respond quite differently to the same forcing. A surface-based mass-balance model of salt-withdrawal minibasins shows how onlap and offlap patterns emerge from the interplay of subsidence and sediment supply. At the reservoir scale, rule-based and process-based models capture architecture that geostatistical methods cannot. The question that remains is: how can we condition these more realistic models to data. Deep neural networks, especially diffusion models, seem quite promising in this regard.
How do basin margins record long-term tectonic and climatic changes? (Geology, 2020) • Conversion of tectonic and climatic forcings into records of sediment supply and provenance (Scientific Reports, 2019) • Eustatic control on continental-margin sand distribution (JSR, 2016) • Grain-size and discharge controls on submarine-fan deposition (Frontiers in Earth Science, 2019) • Slope-fan architecture from forward stratigraphic models (Marine and Petroleum Geology, 2018) • Stratigraphic evolution of intraslope minibasins (AAPG Bulletin, 2015) • Stratigraphic rule-based reservoir modeling (Bulletin of Canadian Petroleum Geology, 2015) • Conditioning process-based models to well data (Mathematical Geosciences, 2016)
Posts: Stratal patterns in salt-withdrawal minibasins • Salt and sediment: a brief history of ideas
Forward models of continental-scale sediment-routing systems, with different subsidence and precipitation histories (Zhang et al., 2020)
Applications to the subsurface
A lot of this work has direct applications to characterization of oil and gas reservoirs and, increasingly, to aquifers and carbon storage. Deep-water channel reservoirs are often modeled as randomly stacked channel fills; building reservoir models instead from the systematic migration that we see in seismic data — bend expansion, downstream translation, cutoff — and running flow diagnostics on them shows that cutoff-related architectures control sweep efficiency during production. Automated well-log correlation with ChronoLog makes it possible to build regional stratigraphic frameworks quickly, for example for CO2 storage assessment. Much of this is done with, and for, the sponsors of the Quantitative Clastics Laboratory.
Reservoir architecture and dynamic connectivity in deep-water channel systems (Interpretation, 2025) • Channel-base drapes in turbidite reservoirs (AAPG Memoir, 2010) • Detecting baffle mudstones using microfossils (Palaeo3, 2014) • Automated multi-well correlation (Basin Research, 2023) • Accelerated stratigraphic framework building for CO2 storage assessment (First Break, 2024)
Flow diagnostics in reservoir models of sinuous deep-water channels with bend cutoffs (Morris et al., 2025)
Machine learning, computer vision, and remote sensing
Image segmentation turns out to solve a surprising number of geoscience problems. ‘segmenteverygrain’ is an open-source Python package for segmenting and measuring grains in images of all kinds — thin sections photographed in transmitted light, sand grains in reflected light, gravel and boulders on beaches, river bars, and outcrops. A U-Net model trained on a variety of grain images produces an initial segmentation, which is then refined with Meta’s Segment Anything model, a more general-purpose segmentation algorithm. We have used the same U-Net approach to map ice-wedge polygons on Arctic point bars, and graph-based methods (‘rivabar’) to extract river centerlines and banklines from time-lapse satellite imagery. I helped develop the course ‘Machine Learning Applications in the Geosciences’ at UT Austin with Sergey Fomel, and several of these projects started out as course projects.
segmenteverygrain (Journal of Open Source Software, 2025) • GitHub
Talks: Accreting, fast and slow (2026) • A frozen history of meandering (2023)
Beach gravel with segmented grains and their orientations
Open-source software
Most of the code behind the work above is on GitHub:
- meanderpy — a simple kinematic model of meandering, based on Howard and Knutson (1984); it is also the source of the vector art behind the prints on the Art page.
- rivabar — graph-based extraction of river centerlines and banklines from satellite imagery.
- meandergraph — a graph-based approach to describing meanders and point bars.
- curvaturepy — code and data for the curvature–migration rate analysis.
- stratigraph — reproducible analysis and visualization of stratigraphy in space and time.
- segmenteverygrain — grain segmentation with U-Net and Segment Anything.
- chronolog_data — example data and supplementary material for ChronoLog.
- detritalPy — a toolset for detrital geo- and thermochronology, led by Glenn Sharman.
- notebooks — Jupyter notebooks on decompaction, grain settling, the diffusion equation, channel sinuosity, and other topics covered on the blog.