Property:CSDMS meeting abstract
From CSDMS
This is a property of type Text.
2
Flow routing calculations are routinely performed in geomorphic and hydrologic analyses. These require an appropriate flow-routing surface, which is generally a digital elevation model which has been pre-processed to remove all depressions from its surface. This allows the flow-routing surface to host a continuous, integrated drainage network. However, real landscapes contain natural depressions that can store water and break up the drainage network. These are an important part of the hydrologic system, and should be represented in flow-routing surfaces. The challenge is in removing from a DEM only those depressions which would be filled under reasonable hydrologic conditions at a given location, and not all depressions indiscriminately. To address this problem, we developed FlowFill, an algorithm that routes a prescribed amount of runoff across the surface in order to flood depressions, but only if enough water is available. This method conserves water volume and allows a user to select a runoff depth that is reasonable for the region of interest. Typically, smaller depressions or those in wet areas or with large catchments are flooded, while other depressions may not be completely filled, thus permitting internal drainage and disruptions to hydrologic connectivity. Results are shown at a sample location using a range of runoff depths, with the resulting flow-routing surfaces with filled and unfilled depressions and the drainage network structure associated with the result. +
Fluid-driven granular flows sculpt Earth's surface through processes such as soil creep, landslides and debris flows, and river bed-load and suspended sediment transport. In the case of river bed-load transport, grains may move by rolling, sliding, and jumping within a thin layer known as bed-load layer. In this layer, it is common for grains to segregate by size (a process that has been extensively studied) or shape, which has only recently been recognized as an important control. Here we perform numerical simulations to examine how shape-driven particle segregation is controlled by 1) purely granular interactions and 2) fluid-granular dynamics. To isolate granular dynamics, we construct a DEM model using LIGGGHTS to examine segregation of dry grains of different shapes in a rotating drum. To fold in the role of fluid drag, we use a CFD-DEM model (OpenFOAM + LIGGGHTS) to study particle segregation in open channel laminar flow. To efficiently simulate different shapes, we use bonded spherical particles to construct spheres, cubes, and cylinders. For the former, we use a horizontal cylinder filled with the same particles, and rotate it at low angular velocities. Meanwhile, for the latter, we set a periodic channel filled with spherical and non-spherical particles, of equal mean volume, sheared by a viscous Couette flow which imposes enough shear stress to move the particles by bed-load transport. For both, we investigate the statistical properties of the segregation by size and shape that non-spherical grains experience in the systems by tracking hundreds of individual trajectories throughout the entire bed, and the mechanisms involved that are mainly driven by particle collisions and fluid-grain interactions. These results illuminate the role of grain shape in controlling sediment transport, with implications for natural rivers, hillslopes, and aeolian systems. +
Fluvial bedload is a fundamental process by which coarse sediment is transferred through landscapes by fluvial action and is characterized by cyclic sequences of particle motion and rest. Bedload transport has many complex physical controls but may be well described stochastically by distributions of grain step length and rest time obtained through tracer studies. However, none of these tracer studies have investigated the influence of large wood on distributions of step length or rest time, limiting the applicability of stochastic sediment transport models in these settings. Large wood is a major component of many forested rivers and is increasing because of disturbances such as wildfire and insect infestations as well as the use of wood in rivers as part of ‘natural flood risk management’ practice in the UK. This study aims to investigate and model the influence of large wood on grain-scale bedload transport.
St Louis Creek, an alpine stream in the Fraser Experimental Forest, Colorado, is experiencing increased wood loading resulting from the infestation of the mountain pine beetle in the past decades. We inserted 957 Passive Integrative Transponders (PIT) tagged cobbles in 2016 upstream of a wood loaded reach and measured and tagged >20 pieces of large wood in the channel. We resurvey the cobbles and wood on an annual basis after snowmelt, building distributions of rock-step lengths and rest time distributions as well as observing any changes and transport of large wood. Additionally, we are developing novel active tracer tags, with integrated accelerometer technology, which will help to constrain these distributions and investigate the influence of woody debris.
We observed increased probabilities of grain deposition around large wood in the first 3 years of resurvey data, and preliminary statistical analysis suggests a significant influence of wood presence, and its relative stream position, on transport likelihood and distance, although additional annual data is required to verify its reproducibility. Over the next two snowmelt seasons, active tags will provide detail on the transport behaviour of cobbles at unprecedented levels, allowing us to refine stochastic bedload transport models in environments where biota is significantly interacting with earth surface processes.
Fluvial deltas have worldwide socio-economic importance as human development and infrastructure centers and provide several ecosystem services, including storm protection and nursery habitats. Their subsurface architecture also holds clues to past climate and sea-level change that can be reconstructed from stratigraphy. A significant challenge in inverting stratigraphy is separating the signals of external forcing, such as variations in sea level, and internal processes, such as the dynamics of the fluvial surface and channel network variations. In a previous work, we analyzed laboratory flume data from the Tulane Delta Basin using an experimental run with oscillating sea level conditions and constant sediment supply. We found that the dynamics of the fluvial surface play an important role in delaying the response of the upper portion of the subaerial topset. To further quantify this phenomenon, we couple this flume experiment with a numerical modeling framework that integrates the topset with a subaqueous offshore region or foreset. The numerical model can explain the topset slope, convexity dynamics, and sediment partitioning between the topset and the foreset under sea level variations. For example, it captures how during sea-level rise (SLR), low sedimentation near the topset's center reduces the subaerial slope and increases convexity, while during sea-level fall (SLF), high sedimentation increases the slope and concavity. Moreover, the model can explain the counterintuitive observation of higher sediment topset bypass to the foreset under SLR than SLF due to the reduction in subaerial slope, partially explained by a higher presence of active channels during SLR than SLF. These results underscore the importance of internal processes such as fluvial surface and channel dynamics, which can result in net erosion during SLR and net deposition during SLF, potentially complicating the reconstruction of paleo sea-level from deltaic deposits. +
Fluvial incision patterns help us understand the role of precipitation in river formation and evolution. The effects of drainage area, sediment supply and precipitation are closely linked and disentangling them is a challenging task.
In this study, we model different precipitation scenarios and use the stream power law to analyse river profiles. We focus on the analysis of the χ coordinate, a transformation of the stream-power law to capture changes in slope with distance downstream. The value of this coordinate is controlled by the concavity index, θ, which sets the steepness of the rivers downstream.
Similar χ profile shapes can be caused by different precipitation patterns, tectonic forcings or lithologies. However, choosing different θ coefficient values will lead to patterns similar to those arising from the natural forcings above, distorting the original physical signal.
In this study, we use the modelling framework Fastscape to generate landscapes that evolve to steady state under different precipitation scenarios. We test multiple precipitation models and calculate the χ profiles of the resulting simulated rivers using LSDTopoTools. We complete the analysis by comparing the model results to real topographic data from sites featuring a strong precipitation gradient, such as the Pyrenees, the Alburz mountains and the Andes.
This piece of research provides further insight on the importance of constraining the θ coefficient in χ profiles, in particular when disentangling the role of precipitation in river incision mechanisms. +
Fluvial incision since late Miocene time has shaped the modern transition between the Central Rocky Mountains and the adjacent High Plains of North America. Incision has formed a distinctive pattern of deep gouges at the mountain front centered around large drainages, most notably the Arkansas and South Platte Rivers. While there is a clear contrast in material strength and erodibility between the crystalline rocks that comprise the core of the mountains and the sedimentary packages that overlie the plains, researchers seldom account for this contrast when attempting to model the geomorphic evolution of the plains. In this study we set an explicit boundary across which erodibility changes from a value representative of granite, for the Central Rockies, to a value representative of coarse sandstone, for the High Plains. We then evolve the landscape with constant, uniform uplift and fluvial incision with sediment transport dependent upon a characteristic transport length. We find that with no external forcing beyond steady uplift and even on a landscape of modest gradient, it is possible to recreate deep incision at the mountain front simply by running water across substrates with highly contrasting erodibilities. This preliminary result has applications to future studies of the geomorphic evolution of the High Plains as it causes us to re-evaluate the sensitivity of this landscape to the material properties of the mountains and plains. In future work, this may guide us to look more closely at intrinsic properties of the landscape as an explanation for geomorphic expression before considering external forces. +
Fluvial sediment dynamics play an important role in the functioning and connectivity of the earth’s natural systems. It is not only one of the primary drivers of landscape development and channel morphodynamics, but also has important implications for water resources, ecology, geochemical cycling, and socio-economic aspects. Although anthropogenic influences are a major cause of changes in river sediment transport processes, it is widely accepted that these processes are also sensitive to climate change. Future climate changes particularly rises in temperature driven by increased greenhouse gas emissions, are projected to considerably impact 21st-century precipitation distribution which will alter fluvial processes, soil erosion and sediment loads worldwide. Predicting the responses of riverine fluxes to future climate is, therefore, vital for the management of fluvial systems. In this study, we conduct a global scale analysis of future suspended sediment and water discharge dynamics in response to the changing climate. We use a spatially and temporally explicit global scale hydrogeomorphic model, WBMsed. Changes in the earth’s climate system were obtained by forcing the model with downscaled precipitation and temperature projections generated by multiple General Circulation Models (GCMs), each driven by four Representative Concentration Pathways (RCPs). We investigate climate-induced spatial and temporal trends and variability in global suspended sediment loads and river discharge dynamics in the 21st century. +
Fluvial terraces are commonly interpreted as recorders of past environmental (e.g. tectonic or climatic) conditions. However, controls on terrace formation through river incision, and on the destruction of terraces through lateral erosion are poorly understood. Here, we present results from a physical experiment performed at the St. Anthony Falls Laboratory that provide insights into the formation and preservation potential of alluvial terraces, into dynamics of alluvial valley width, and the dependence of these parameters on external forcings: primarily on river response to base level fall. The model was performed in a wooden box with dimensions of ~4 meters by ~2.5 meters by ~0.5 meters, which was filled with silica sand with a unimodal grain size distribution (D50= 0.14 mm). Sediment and water were mixed and fed into the box via a gravel diffuser to inhibit scour. A single channel incised down to the base level, which was steadily lowered by a weir. Six experiments were performed, each with a constant water discharge of 0.1 L/s and a sediment flux of 0.022 L/s, and with a base-level fall rate of 0mm/hr, 25mm/hr, 50mm/hr, 200mm/hr, 300mm/hr, and 400mm/hr. We collected aerial photographs every 20 seconds and digital elevation models (DEMs) every 15 minutes throughout each experiment. Terraces formed in the experiments with base level fall due to incision and headwards knickpoint retreat. Major sidewall collapses and progressive valley widening were observed and controlled by the lateral migration of the channel. +
Fluvial valley width is determined by a combination of factors including regional lithology and drainage organization, as well as regional glacial and uplift history. In both topographic analysis and numerical modeling-based studies, valley width has been observed to follow a power law scaling relationship with drainage area. Local to regional scale studies have also demonstrated the influence of lithology, differential uplift, and drainage reorganization on this relationship. Yet, significant uncertainty remains regarding how these trends extend to the scale of large river networks and how they are influenced by transient forcing. The Upper Mississippi River Valley, initially incised during the early Pleistocene, presents a case study that encompasses a wide range of valley forms likely influenced by some combination of these factors, including by not limited to Spatially variable glacial history, bedrock lithology, and punctuated drainage reorganization events. This research aims to analyze the variable contribution of lithology, downstream changes in drainage area and history of reorganization, and regional variability in glacial isostatic adjustment in determining downstream trends in valley morphology. By isolating these effects, we aim to determine whether there is an extractable signal of the conditions during initial valley incision embedded in modern valley topography. Here we present a dataset of high-resolution valley aspect ratio and curvature, paired with longitudinal trends in drainage area and bedrock lithology. This is compared with empirical expectations for valley width scaling. Preliminary analysis found an overall downstream valley widening trend, however with multiple perturbations, including narrow gorges, and locally beveled valley walls caused by a combination of lithologic transitions and differing drainage integration histories. +
Following pioneering modeling work examining the evolution of wave-influenced deltas (Ashton et al., 2013; Nienhuis et al., 2013), we coupled the River and Floodplain Evolution Model (RAFEM) to the Coastline Evolution Model (CEM). Results of a recent suite of model experiments (conducted using the CSDMS software stack and Dakota) lead to new insights: 1) The preferred location of avulsions (a distance from the river mouth scaling with the backwater length), previously observed in laboratory models and in the field, can arise for geometric reasons that are independent of those recently suggested (Chatanantavet et al., 2012; Ganti et al., 2016). This alternative explanation applies when the river longitudinal profile tends to diffuse more rapidly than the floodplain longitudinal profile. 2) Although the timescale for avulsions is expected to increase with increasing wave influence (Swenson, 2005), we find that this depends on the angular wave distribution. When wave influence is strong and the angular mix of wave influences tends to smooth a nearly straight coastline (coastline diffusion), progradation is slowed and avulsions delayed. However if the angular wave distribution produces anti-diffusive coastline evolution, a strong wave influence still leads to cuspate delta shapes, but avulsions are barely delayed. 3) Although increasing sea-level-rise rate is expected to cause more rapid avulsions, and does in laboratory deltas, we unexpectedly find that this is not true for river-dominated deltas in our model (or for anti-diffusive wave climates). The explanation, involving the role of sea-level-rise related transgression (or decreased progradation), raises potentially important questions about geometrical differences between laboratory deltas and natural deltas. 4) The magnitude and timescale of autogenic variability in sediment delivery rates at the river mouth depends on wave climate, sea-level-rise rate (for some wave climates), and on the amount of super elevation of the river channel (relative to the surrounding floodplain) required to trigger avulsions.
* Ashton, A. D., Hutton, E. W., Kettner, A. J., Xing, F., Kallumadikal, J., Nienhuis, J., and Giosan, L. (2013), “Progress in coupling models of coastline and fluvial dynamics,” Computers & Geosciences, 53, 21–29.
* Chatanantavet, P., Lamb, M. P., and Nittrouer, J. A. (2012), “Backwater controls of avulsion location on deltas,” Geophysical Research Letters, 39.
* Ganti, V., Chadwick, A. J., Hassenruck-Gudipati, H. J., Fuller, B. M., and Lamb, M. P. (2016b), “Experimental river delta size set by multiple floods and backwater hydrodynamics,” Science advances, 2, e1501768.
* Nienhuis, J. H., Ashton, A. D., Roos, P. C., Hulscher, S. J., and Giosan, L. (2013), “Wave reworking of abandoned deltas,” Geophysical research letters, 40, 5899– 5903.
* Swenson, J. B. (2005), “Relative importance of fluvial input and wave energy in controlling the timescale for distributary-channel avulsion,” Geophysical Research Letters, 32.
For a subset of global deltas, morphological evolution is due to the competing actions of the river, which brings about the delivery of terrestrial sediment, and waves, which redistribute the input sediment across the coastline. Given that there are many such coastlines where waves exert considerable influence worldwide, an improved understanding of the effect of waves on the morphological evolution of coastal delta settings is imperative, especially in view of the perceived declining influence of the river input. Accordingly, this study presents a preliminary numerical model approach applied to investigate the planform evolution of deltaic coastlines due to the interplay between flow discharge and waves. Model simulations were undertaken with the coupled Delft3D and SWAN (Simulating Waves Nearshore) numerical models for fluvial and wave input, respectively. Additionally, the idealized numerical model represents a straight, sandy deltaic coastline interrupted by two fluvial discharge outlets, and, at the same time, affected by waves approaching from a dominant direction. We found that the modelled deltas evolved into diverse shoreline - and - river–mouth forms under varying combinations of wave and river inputs. The modelling approach also makes a preliminary distinction between the relative effects of waves’ significant height (Hs) and incidence angle (αo) on deltaic planform morphological evolution. Future development of the model will focus on critically exploring the interaction between these two key morphodynamic processes along similar natural coastline settings. +
For landscapes to achieve a topographic steady state, they require steady tectonic uplift and climate, and a bedrock that is uniformly erodible in the vertical direction. Basic landscape evolution models predict that incising drainage networks will eventually reach a static geometric equilibrium – that is, the map-view channel pattern will remain constant. In contrast, natural rivers typically incise through heterogeneous bedrock, which can force reorganization of the drainage structure. To investigate how lithological variability can force landscape reorganization, we draw inspiration from formerly glaciated portions of the upper Mississippi Valley. In this region, depth-to-bedrock maps reveal buried dendritic river networks dissecting paleozoic sedimentary rock. During the Pleistocene, ice advance buried the bedrock topography with glacial till, resurfacing the landscape and resetting the landscape evolution clock. As newly formed drainage networks develop and incise into the till-covered surface, they exhume the buried bedrock topography. This then leads to a geomorphic "decision point": Will the rivers follow the course of the bedrock paleodrainage network? Or will they maintain their new pattern? Using a numerical landscape evolution model, we find that two parameters determine this decision: (1) the contrast between the rock erodibility of the glacial till (more erodible) and of the buried sedimentary rock (less erodible) and (2) the orientation of the surface drainage network with respect to the buried network. We find that as the erodibility contrast increases, the drainage pattern is more likely to reorganize to follow the buried bedrock valleys. Additionally, as the alignment of the two networks increases, the surface drainage network also tends to restructure itself to follow the paleodrainage network. However, when there is less contrast and/or alignment, the surface drainage pattern becomes superimposed on the bedrock topography, with streams cutting across buried bedrock ridges. Our results agree with field studies demonstrating that variability in erodibility exerts a first-order control on landscape evolution and morphology. Our findings can provide insight into how lithologic variation affects surface processes, drives drainage reorganization, and creates geopatterns.
For many deltas, their morphology reflects the 100-1000 year balance of wave, tidal, and river-driven sediment fluxes. Human-induced changes to these fluxes can also act on 100-1000 years and therefore influence delta morphology.
Wave, tidal, and river fluxes also change on much shorter (day-to seasonal) timescales. These fluctuations do not work their way into delta morphology immediately, but many studies have indicated substantial relevance, nevertheless.
How to marry these two timescales? In this poster I will investigate the concept of river sediment retention, or trapping efficiency, and its potential to relate seasonality to long-term fluxes. For example, wave, river, and tidal fluxes might each dominate for a few months every year. If the order and respective magnitude of these fluxes throughout the year influence tidal sediment retention, it can affect long-term morphology and make it deviate from a balance based on simple annual averages. +
For paleo environmental studies, a key challenge is to partitioning physical signals operated under multiple spatio-temporal scales. For example, paleo relative sea-level (RSL) data record a combined signal from global ice-ocean mass exchange induced global mean sea-level change and gravitational, rotational and deformational effects, along with regional and local RSL change caused by changing ocean density, groundwater storage and sediment redistribution. Here we present an open-sourced spatio-temporal hierarchical model framework (PaleoSTeHM) that is conceptually suitable for investigating this problem by separating the underlying phenomenon of interest and its variability from the noisy mechanisms by which this underlying process is observed. PaleoSTeHM is built upon a modern, scalable machine-learning framework and offers flexible modelling and analytical choices. In this presentation, we will show some of the modelling choices in PaleoSTeHM along with an example application for Holocene sea level change. Also, we will seek inputs from potential users for this framework in order to make this co-develop framework more sustainable and allows a wide range of paleo-sea level and -climate researchers to easily and robustly incorporate spatio-temporal statistical modeling into their work. +
Fractal geometry is a branch of mathematics pioneered by Benoit Mandelbrot in the 1970's with the goal of finding a mathematically rigorous way to define the geometry found in nature, including what he saw in river networks. Since then, much work on the geometry and structure of river networks has involved fractal method, from passing mention to assumed fractal characteristic's to trying to tie older geomorphic parameters to Mandelbrot's fractal math. However results on the fractal dimensions of river networks have been contradictory and not always well matched to theoretical explanations of fractal geometry. For example, in a 1988 work, Tarboton et al. found that the measured fractal dimension of river networks transitioned from close to 1 at small scales to close to 2 at large scales. They attributed this to switching from a regime where fractal dimension was dominated by Sinuosity to one where it was dominated by the branching characteristics of rivers. Neither of these matches Mandelbrot's prediction of a fractal dimension of 1.2 for river networks, which he derived from a Hack exponent of 0.6, used in the relation between stream length and basin area, which would likely be influenced by river branching. More recent unpublished calculation of the fractal dimension of large North American river basins found a dimension close 1.1, which conveniently would correspond to a Hack exponent of 0.55 which matches more recent empirical work on Hack's law. To better understand the connection between fractal dimension and Hack's Law, in this poster I present work comparing the fractal dimension of modeled river networks to physical ones, and look at what theoretical parameters may explain them variability in measured fractal dimensions of river networks. +
Fresh impetus has been given to efforts for a unified bio+geo understanding of seafloor physical properties. In part the requirement comes from practical needs in: the dependability of automated modules (Autonomous Underwater Vehicles), for object detection (e.g. unexploded ordinance), and for more accurate Acoustic Seafloor Classification in habitat mapping. By the combination of various techniques, and especially new information resources, the opportunities for fresh advancement in the field have recently increased.
The new information resources include semantic structures such as Encyclopedia of Life, WoRMS, Traitbank and others where the characteristics of organisms are described, including their lifecycles, engineering activities, morphologies. They also include environmental databases of ever increasing resolution and scope, such as photosynthetically available radiation, sediment types, water flows, particulate matter and nutrients.
The challenge is a significant one, to combine these factors, but there are some approaches which have been tested and found very promising. Some are described in this poster. They include simulations (rather than analytical models) with data formats derived from the 3D printing industry, agent-based approaches, population models of various types (including cellular models), and more.
Global change, often human-induced, is causing a re-balancing between 'barren' sediment-dominated areas and those which are intensely colonized. Models such as these are required to see ahead to the consequences and management of the changes. +
Freshwater inflow plays a substantial role in the water quality of coastal and estuarine watersheds and ecosystems. The salinity of an estuary can vary depending on the amount of freshwater received. In highly managed systems, such as St. Lucie and Caloosahatchee estuaries in south Florida, USA, understanding total freshwater inflow and the sources of inflow is very important for management decision-making. There is very little information on the quantity of freshwater inflow to St. Lucie and Caloosahatchee estuaries from their ungauged tidal basins. This study examines a linked hydrologic, hydraulic, and watershed water quality model (WaSh) for simulating freshwater inflow to these two systems. The WaSh model is a time-dependent simulation model that represents basic surface hydrology, groundwater flow, surface water flow, and water quality fate and transport. The WaSh model consists of four basic components; a cell-based representation of the watershed basin land surface, a groundwater component, a surface-water drainage system, and a water management component that can consider the effects of reservoirs, stormwater treatment areas, irrigation supply and demand, and land-use changes. The model is capable of simulating hydrology in watersheds with high groundwater tables and dense drainage canal networks, which is typical in South Florida.
The model was developed using long-time series of rainfall, temperature, evapotranspiration, basin boundaries, hydrography including streams and canals features, soils, land use, and land surface elevations. The results indicate that the model accurately simulates the distribution of freshwater over the coastal watersheds and the transport of freshwater through the estuary and that it is a valuable tool for understanding the dynamics of freshwater inflow to estuaries in coastal watersheds.
Keywords: Coastal Hydrology, Ungauged basin, WaSh model, estuary, Salinity, hydrologic and hydraulic model, water quality
Freshwater resources in coastal Bangladesh fluctuate with extreme periods of shortage
and abundance. Bangladeshis have adapted to these alternating periods but are still
plagued with scarce drinking water resources due to pond water pathogens, salinity of
groundwater, and arsenic contamination. The success of attempts to correct the
problem of unsafe drinking water have varied across the southern Bangladesh as a
result of physical and social factors. We use a multicriteria decision analysis (MCDA) to
explore the various physical and social factors that influence decisions about freshwater
technologies and management schemes in southern Bangladesh.
MCDA is a holistic, analytical tool for evaluation of alternatives. MCDA is used to
support public participation and provide structured, rational, and transparent solutions to
complex management problems. To determine the best freshwater technologies and
management schemes, we examine four alternatives, including managed aquifer
recharge (MAR), pond sand filter (PSF), rain water harvesting (RWH), and tubewells
(TW). Criteria are grouped into four categories: environmental, technical, social, and
economic. Weighting of social factors will be determined by community surveys, nongovernmental
organizations (NGO) opinions, and academic interviews. Data
include regional water quality perceptions, perceptions of management/technology
success, MAR community surveys, and interviews with NGO partners. Environmental
and technical feasibility factors are determined from regional water quality data,
geospatial information, land use/land change, and regional stratigraphy.
Survey data suggest a wide range of criteria based on location and stakeholder
perception. MAR and PSF technologies likely have the greatest environmental and
technical potential for success but are highly influenced by community dynamics,
individual perspective, and NGO involvement. RWH solutions are used less frequently
due to quantity limitations but are most successful at reducing the water security threats
of contamination by pathogens, arsenic, and salts. This MCDA informs us of community
and stakeholder water resource decisions, specifically related to their objectives and
values.
From the classic U-shaped glacial valley to the convex soil-mantled hillslope, geomorphic processes leave clear signatures on the landscapes they create. However, it has been challenging to develop topographic metrics that can be used to extract process parameters. While researchers have gained significant insights into geomorphic processes through metrics like mean local relief, channel steepness, and ridgetop curvature, it is still difficult to make quantitative predictions about processes from quantitative topographic measurements.
Prior modeling work has found that in 2D models that combine stream incision with diffusive hillslope processes, valley spacing is strongly controlled by the relative rates of advective and diffusive processes (Perron et al. 2008). In this work we train a simple convolutional neural network to predict the ratio of the coefficient of stream erosion (K) and coefficient of diffusion (D) used to generate the model topography. Across a test set of 1800 model runs with different K and D values, the network had a normalized root mean square error of 0.03, showing that convolutional neural networks have significant promise for extracting complex and geomorphically meaningful topographic signatures.
In this work we focus on interpreting the neural network to try to help explain what it is calculating in a theoretically grounded way. The output of activation maximization, Fourier analysis, neuron ablation, and other interpretability techniques are complicated, but might imply that the network is detecting patterns that are geographically meaningful. This poster will present these interpretability results. +
A
GEOtop 1.145 is used to model the thermal and hydrological state of the subsurface in the Kuparuk basin, Alaska. GEOtop is a distributed hydrological model with coupled water and energy budgets. The surface energy balance scheme includes sensible, latent and radiative heat fluxes at the air-soil or air-snow interface. The subsurface represents heat fluxes in the vertical and water fluxes in the vertical and horizontal directions. The ERA-Interim atmospheric reanalysis product, which is used to force the model, is compared to meteorological and radiation data from the Kuparuk Basin and other stations on the North Slope of Alaska. The use of ERA-Interim reanalysis to force GEOtop enables large-scale simulations to be performed over areas where in situ meteorological data is sparse, such as the North Slope of Alaska. Model simulations forced by ERA-Interim reanalysis data are validated using borehole observations of soil temperature. Model results will be presented demonstrating the interactions between soil properties, snow cover, vegetation and climate. +
