News

A little bit more detail on individual research outputs, field campaigns, comings and goings, and goings on ...

 
 

 Example images of coastal environments (left column), with their sparse (center) and dense (right column) semantic segmentations

Example images of coastal environments (left column), with their sparse (center) and dense (right column) semantic segmentations

July 2018. New "deep learning" article published in Geosciences

The application of deep learning, specifically deep convolutional neural networks (DCNNs), to the classification of remotely-sensed imagery of natural landscapes has the potential to greatly assist in the analysis and interpretation of geomorphic processes. However, the general usefulness of deep learning applied to conventional photographic imagery at a landscape scale is, at yet, largely unproven. If DCNN-based image classification is to gain wider application and acceptance within the geoscience community, demonstrable successes need to be coupled with accessible tools to retrain deep neural networks to discriminate landforms and land uses in landscape imagery. Here, we present an efficient approach to train/apply DCNNs with/on sets of photographic images, using a powerful graphical method called a conditional random field (CRF), to generate DCNN training and testing data using minimal manual supervision. We apply the method to several sets of images of natural landscapes, acquired from satellites, aircraft, unmanned aerial vehicles, and fixed camera installations. We synthesize our findings to examine the general effectiveness of transfer learning to landscape-scale image classification. Finally, we show how DCNN predictions on small regions of images might be used in conjunction with a CRF for highly accurate pixel-level classification of images

Open access article accessed here


D800_20170308_214403-0_r.png

July 2018. Deep Learning workshops

MAPPING LAND-USE, HAZARD VULNERABILITY AND HABITAT SUITABILITY USING DEEP NEURAL NETWORKS

Over this summer and fall, I am teaming up with colleagues to

  • Curate data sets for training and testing of machine/deep learning methods on sets of images of natural environments and processes
  • Hold two training events for applying deep learning to image classification problems, using python and tensorflow
  • Develop online materials, examples, and software for applying deep learning to image classification

The project is funded by the USGS Community for Data Integration, For more details, go to: https://sites.google.com/view/usgsdeeplearning/home


 May 7 - 11, Santa Barbara California. https://www.geohab2018.org/

May 7 - 11, Santa Barbara California. https://www.geohab2018.org/

May 2018. GeoHab 2018 conference

  • Tues May 8, 10:30: R2Sonic Booth. Poster and software demo of PriSM (Probabilistic Acoustic Substrate Mapping toolbox)
  • Tues May 8, 15:15: talk entitled "Probabilistic Substrate Mapping in Rivers and Seas with Conditional Random Fields of Acoustic Backscatter"
  • Tues May 8, 16:00: poster entitled "Repeat-mapping bathymetric change and substrates in a large river: application to sediment budgeting and physical habitat mapping in the Colorado River in Grand Canyon"

 Co-conspirator David Dean of the USGS Grand Canyon Monitoring and Research Center

Co-conspirator David Dean of the USGS Grand Canyon Monitoring and Research Center

April2018. Fieldwork on the Chippewa River, WI.

We spent a week in southern Wisconsin repeat-mapping dune fields on the bed of the Chippewa River near Durand, WI, and downstream of the I35 bridge above Lake Pepin, where the Chippewa meets the Mississippi river. In collaboration with the USACE and Minnesota Water Science Center, we're working on quantifying total sand load in order to address sedimentation issues downstream.


 Hamill D, Buscombe D, Wheaton JM (2018) Alluvial substrate mapping by automated texture segmentation of recreational-grade side scan sonar imagery. PLoS ONE 13(3): e0194373. https://doi.org/10.1371/journal.pone.0194373

Hamill D, Buscombe D, Wheaton JM (2018) Alluvial substrate mapping by automated texture segmentation of recreational-grade side scan sonar imagery. PLoS ONE 13(3): e0194373. https://doi.org/10.1371/journal.pone.0194373

March 2018. Article published in PLOS-ONE

Side scan sonar in low-cost ‘fishfinder’ systems has become popular in aquatic ecology and sedimentology for imaging submerged riverbed sediment at coverages and resolutions sufficient to relate bed texture to grain-size. Traditional methods to map bed texture (i.e. physical samples) are relatively high-cost and low spatial coverage compared to sonar, which can continuously image several kilometers of channel in a few hours. Towards a goal of automating the classification of bed habitat features, we investigate relationships between substrates and statistical descriptors of bed textures in side scan sonar echograms of alluvial deposits. We develop a method for automated segmentation of bed textures into between two to five grain-size classes. Second-order texture statistics are used in conjunction with a Gaussian Mixture Model to classify the heterogeneous bed into small homogeneous patches of sand, gravel, and boulders with an average accuracy of 80%, 49%, and 61%, respectively. Reach-averaged proportions of these sediment types were within 3% compared to similar maps derived from multibeam sonar.


gsalogo.gif

River Science session at the 2018 Geological Society of America joint Cordilleran/Rocky Mountain section meeting. 

The meeting will be May 15-17, 2018 in Flagstaff, Arizona. Abstract submission deadline Feb 20

While our session description emphasizes the decades of river science resulting from river management in the Colorado River basin and Intermountain West, we encourage submissions on all aspects of river science in natural or human-modified river systems. Please contact us with any questions.  T3. Advances in River Science in the Intermountain West.

 
360_AGU-fm17-633.png

2017 AGU Fall Meeting, New Orleans, LA, 11-15 December

Go say hi!!

Grams, P.E., Buscombe, D. Topping, D.J., Mueller, E.R. (2017) Identification of discontinuous sand pulses on the bed of the Colorado River in Grand Canyon.

Leary, K., Buscombe, D., Schmeeckle, M., Kaplinski, M. (2017) Assessing the Importance of Cross-Stream Transport in Bedload Flux Estimates from Migrating Dunes: Colorado River, Grand Canyon National Park. 

Platt, A. S., Buscombe, D., Porter, R. C., & Grams, P. (2017), Estimating Sediment Thickness from Riverbed to Bedrock within the Colorado River in the Grand Canyon. 

 
2017-Journal_of_Geophysical_Research__Earth_Surface.png

October 2017. Figure from our paper selected for the front cover of the October 2017 edition of the Journal of Geophysical Research - Earth Surface

Figure shows example imagery for each of 10 unique substrate classes easily identifiable by eye from our underwater video system (LOBOS), arranged in two groups of five. The first group are found in sites where the riverbed is completely unvegetated (top four rows). The second group (bottom four rows) are found in partially vegetated riverbeds. The substrate codes shown in the first image in every group are those defined in Table 1 and colored the same as how they are represented in Figure 2 in the following paper: 

Buscombe, D., Grams, P. E., & Kaplinski, M. A. (2017). Compositional signatures in acoustic backscatter over vegetated and unvegetated mixed sand-gravel riverbeds. Journal of Geophysical Research: Earth Surface, 122, 1771–1793. https://doi.org/10.1002/2017JF004302

 
 Buscombe, D., Grams, P. E., & Kaplinski, M. A. (2017). Compositional signatures in acoustic backscatter over vegetated and unvegetated mixed sand-gravel riverbeds. Journal of Geophysical Research: Earth Surface, 122, 1771–1793.  https://doi.org/10.1002/2017JF004302

Buscombe, D., Grams, P. E., & Kaplinski, M. A. (2017). Compositional signatures in acoustic backscatter over vegetated and unvegetated mixed sand-gravel riverbeds. Journal of Geophysical Research: Earth Surface, 122, 1771–1793. https://doi.org/10.1002/2017JF004302

October 2017. Article published in Journal of Geophysical Research - Earth Surface

Multibeam acoustic backscatter has considerable utility for remote characterization and mapping of spatially heterogeneous bed sediment composition over vegetated and unvegetated mixed sand-gravel riverbeds of mixed sand and gravel from a moving vessel.

However, high-frequency, high-resolution acoustic backscatter collected in shallow water is hampered by significant topographic contamination of the signal. In this paper, a two-stage method is proposed to filter out the small-scale topographic influence on acoustic backscatter. This process strengthens relationships between backscatter and sediment composition.

A probabilistic framework is presented for classifying vegetated and unvegetated substrates based on acoustic backscatter at decimeter resolution. This capability is demonstrated using data collected from diverse settings within a 386 km reach of a canyon river whose bed varies among sand, gravel, cobbles, boulders, and submerged vegetation.

 
mb07970_hs.png

September 2017. Presented at the Rivers, Coastal and Estuarine Morphodynamics ("RCEM 2017") Symposium in Padova, Italy.

Many scaling relationships have been proposed to link subaqueous sand dune dimensions with aspects of flow. This is useful for estimating dune dimensions from flow depth and velocity, which can be used to estimate dissipation of flow energy, bottom turbulence generation and bedload sediment transport.

However, when data is compiled from rivers worldwide, one observes a more than order of magnitude variation in dune height or length for a given flow depth. Why? We investigate using a very large data set collected from the Colorado River in Grand Canyon. We find that dune dimensions are controlled by grain size and degree flow constriction.

 
bedoad_space_time.png

October 2017. New sensors installed on the Chippewa River, WI, to measure "total load" 

A new project has started in collaboration with the USGS Minnesota and Wisconsin Water Science Centers, and the Grand Canyon Monitoring and Research Centers, whose goal is to quantify 'total load' on the Chippewa River. Total load refers to the amount of sand moving both in suspension and as bedload. We have installed acoustic sensors to measure both components of load near Durand, WI. The aim is to continuously measure suspended sediment and track dunes moving along the bed to estimate bedload. The sensors will be in place for 3 years and will be used to estimate the amount of sediment that is entering Lake Pepin, hence the requirements for sediment management in the reservoir.

 
raw_dwnhi0.png

August 2017. Presented at the American Fisheries Society annual meeting in Tampa, Florida.

In the past few years, low-cost, consumer-grade (hereafter, ‘recreational-grade’, to distinguish from ‘survey’ or ‘scientific’ grade) sidescan sonar platforms, have been developed for leisure activities such as fishing and hobbyist archeology. Recreational-grade sonar lack standardization in (and description of) the acoustic signal processing used, often without high-quality (‘survey grade’) positioning and measured boat attitude (heave, pitch, yaw, etc), and reporting of those quantities. It is not usually possible to process data from such sonars using conventional commercial hydrographic surveying software, to post-process the positioning of the scans or carry out a calibration that corrects for radiative properties of individual transducers. However, these inexpensive, lightweight (portable) sidescan sonar units can be deployed on almost any waterborne craft without the requirement of specialist knowledge of sonar and geodetics, and with little to no experience with acoustic remote sensing. This accessibility is behind the rapid increase in popularity of these sonar systems, among the scientific research community for benthic imaging in a range of aquatic environments, both marine and freshwater, lotic and lentic

A special session was held on using low-cost sonar technology in rivers and streams to support. I presented on the development of PyHum program and the ongoing developments in automated bed sediment characterisation from recreational grade sidescan sonar systems.

 
lobos_boat1.jpg

April 2017. Channel Mapping in Western Grand Canyon

They said it couldn't be done. 60 miles of channel mapped in 2 weeks, with a crew of 25, 7 boats, 2 multibeam sonars, 1 boat-mounted lidar, 1 sub-bottom profiler, 1 sidescan sonar, 1 underwater video camera system, 4 robotic total stations, and lots and lots of gunners and nams!!

 
Buscombe_EMS_Fig2_v2.png

March 2017. Article published in Environmental Modelling and Software

In recent years, lightweight, inexpensive, vessel-mounted ‘recreational grade’ sonar systems have rapidly grown in popularity among aquatic scientists, for swath imaging of benthic substrates. To promote an ongoing ‘democratization’ of acoustical imaging of shallow water environments, methods to carry out geometric and radiometric correction and georectification of sonar echograms are presented, based on simplified models for sonar-target geometry and acoustic backscattering and attenuation in shallow water. Procedures are described for automated removal of the acoustic shadows, identification of bed-water interface for situations when the water is too turbid or turbulent for reliable depth echosounding, and for automated bed substrate classification based on singlebeam full-waveform analysis. These methods are encoded in an open-source and freely-available software package, which should further facilitate use of recreational-grade sidescan sonar, in a fully automated and objective manner. The sequential correction, mapping, and analysis steps are demonstrated using a data set from a shallow freshwater environment.