I work on problems related to how we apply and use large data sets and new computational techniques to new and long-standing problems in geomorphology, sedimentology and, increasingly, eco-geomorphology.
A lot of this involves investigating how we re-purpose existing instrumentation capabilities to measure additional properties of interest in natural environments. This involves algorithm development to develop proxies between remotely sensed signals and physical (and biological) properties, as well as research into machine learning and inverse methods, data simulation, fusion, and assimilation.
I work mostly in aquatic (freshwater and marine) systems, combining field, theoretical, and experimental approaches to the field of sediment transport, hydrodynamics, and computational eco-hydro-sediment acoustics and optics. This understanding is applied to developing systems for monitoring river and coastal water and sediment, in order to address problems in river and coastal engineering and management, including hazards (floods, storms, tsunami, and landslides), river and sedimentation engineering, dam and reservoir management, water resources and wastewater management, treatment of contaminated sediments, engineering problems in ports and harbors, river restoration, beach nourishment, dredging activities, and renewable hydro-power projects.
Examples of this approach from my work include
1) using swath sonar designed for bathymetric mapping into quantitative estimates of the properties of benthic substrates and the life they support; and
2) using machine learning methods to semantically understand and extract features from large-scale remote sensing data sets, for application to automated environmental monitoring and data-model assimilation.
My current funded research projects (as of June 2018) are listed below
- Sediment and sandbar dynamics in the Colorado River, Grand Canyon National Park, AZ, including acoustic sedimentology of bed and suspended sediment, and bedload studies.
- Sediment dynamics in the Chippewa River, WI, including acoustic sedimentology of bed and suspended sediment, and bedload studies.
- Acoustic characterization of submerged vegetation and partially vegetated substrates in the Colorado River, Glen Canyon National Recreation Area, AZ, and Lake Ontario, NY
- Integrating machine and deep learning computational algorithms into large-scale Structure-from-Motion workflows, image and point cloud classification, various applications across the United States
- Application of deep learning to image classification problems in benthic habitats, Glen Canyon and Lake Ontario
- Sediment dynamics and fluvial geomorphology of the Colorado and Green Rivers, Canyonlands National Park, UT, including acoustic sedimentology of bed and suspended sediment, and bedload studies.
- Multispectral acoustics of invasive and native freshwater mussels, laboratory and field experimentation.
I develop instrumentation, computational and analytical tools which allow measurements of sediment at unprecedented scales and resolutions. In the coming years, methods for measuring the reflectance and scattering of sound (and light) to remotely characterize terrestrial and underwater surfaces and infer sediment properties will continue to mature. I have pioneered the field of acoustic bed-sediment characterization as applied to the management of large regulated rivers in the United States, including the Green and Yampa Rivers in Dinosaur National Monument, CO, the Green and Colorado Rivers in Canyon National Park, UT, and the Colorado River in Grand Canyon, AZ. This work is being applied to sandbar and native fish restoration efforts, as well as understand the controls on sediment transport and export from critical reaches of river.
Machine Learning & Geospatial Tools for Geomorphic Research
I develop open-source software for the wider aquatic geomorphology/sedimentology/ecology community, for applications such as grain size analysis (www.digitalgrainsize.org), sonar imaging of benthic habitats (http://dbuscombe-usgs.github.io/PyHum/) and analysis of massive topographic point clouds (http://dbuscombe-usgs.github.io/pysesa/). I apply modern geospatial science techniques (e.g., sonar, SAR, LiDAR, Structure from Motion and other photogrammetry, remote sensing, and advanced image processing) and Big Data analytics (e.g. machine learning, data-driven hypothesis testing) typically involving the analysis of very large and diverse data sets. My continuing interest in the development of scalable, massively parallelized or distributed research tools in computational geosciences, especially open source software for roughness, grain size, and automated land and seascape feature extraction, will continue to be applied to the quantitative characterization of Earth surface environments. Recently, I have been involved with initiating large-scale monitoring of the California coast, principally from aerial platforms, developing techniques for topographic analysis of massive (order billion point) photogrammetric point clouds.
Experimental Hydraulics, Hydrodynamics & Sediment Transport
For my doctoral thesis, I studied the mechanics of gravel transport under energetic water waves and gravel beach morphodynamics, which included working in the world's largest wave flume, investigating gravel barrier-lagoon dynamics at prototype scale. I continue to work on the complex inter-relations between fluid flows and sediment transport in coastal and fluvial environments. I have researched these processes by developing novel field-deployed optical and acoustic imaging systems designed for measuring aspects of flow or sediment at a very high rate, and computational algorithms for small-scale sediment hydroacoustics, in-situ particle and bed imaging, and flow-field/turbulence measurements. As part of this work, I have developed extensive experience with field data collection and laboratory experimentation, numerical methods and stochastic modeling techniques, as well as instrument design and fabrication.
'Hybrid' Modeling in Ecohydrology and Ecogeomorphology
Models of Earth surface processes should leverage all available data to build semi-empirical parameterizations of key processes, `beyond curve fitting', or calibration and constraint of tunable model parameters. This `hybrid' approach to modeling surface processes combines machine learning approaches with traditional, deterministic, physics-based models. Combining the strengths of inductive (data-driven) and deductive (physics-based) approaches in a single hybrid model has enormous potential in ecohydrology and ecogeomorphology where many of the key interactions do not follow laws of energy or mass conservation.