Property:CSDMS meeting abstract
From CSDMS
This is a property of type Text.
2
Our understanding of temporal changes in long term regional erosion rates in hundreds of years resolution is currently limited. The existing published research is either restricted in spatial coverage, the time frame is in millions of years, or the erosion rate through time is considered constant. We strive to fill this gap by mapping the regional erosion at a resolution of 3.70° latitude by 3.75° longitude (approximately 412 km x 416 km at the equator) at 500-year resolution for the past 21 ka for all land areas. We employ a transfer function that relates mean annual air temperature (MAAT) with erosion that is derived from published field observations recorded during recent times and transferred through the past 21 ka by using space-for-time substitution with paleo-temperature data from TraCE-21ka, a global climate model. Averaging erosion rates and MAAT by latitudes of the Kӧppen climate zones, a non-linear relationship was established. Among the many findings, the largest variation in erosion rates through time is found in cold regions, such as higher elevations and the Arctic. Furthermore, tropical or sub-tropical regions show minimal variation in erosion through time. +
Outburst floods and debris flows often incorporate large volumes of erodible bed sediment along their runout path. Although this phenomenon is widely recognized and often implicated for volumetric growth of debris flows, the effect of this process on the dynamics and runout extent of large flows has not been directly modeled extensively or systematically. Rather, models that account for this process traditionally utilize simple static volumetric and/or rheological adjustments. However, this process dynamically influences flood and debris-flow evolution in a complex spatiotemporal fashion.
We used D-Claw, a depth-averaged granular-fluid model that accommodates the incorporation of bed material into overlying flow and resultant changes in flow rheology across a wide range of solid concentrations, from dilute suspensions to dense-granular debris flows. We modeled hypothetical lake outburst floods from Spirit Lake, WA into the erodible sediment rich Toutle River Valley. Downstream flood dynamics of clear-water flows were compared to floods that entrain material and transform into down-valley debris flows. We found that while the entrainment of bed material may significantly increase total flow volume (>150%), downstream discharge and runout extent are more similar to clear-water floods than might be expected by volumetric considerations alone. We postulate that the relationship between entrained volume and flow extent depends on complicated site-specific factors such as location of erodible sediment and evolving rheological factors. +
Outlet glaciers convey large quantities of ice, sediment, and water from the interior of ice sheets to the coastal ocean. Beneath ice sheets, sediment is transported by melt water, entrainment in basal ice layers, and deformation of the till layer. Till deformation occurs when the ice sliding velocity exceeds a certain threshold, causing buried clasts to plough the sediment layer (Zoet and Iverson, 2020). Because ice velocity tends to decrease below the ice equilibrium line, but the threshold velocity to induce ploughing stays constant, the glacier will deposit till around the “till equilibrium line,” where the ice sliding velocity drops below the threshold velocity (Alley et al., 1989). Investigating the controls on till equilibrium lines will improve our understanding of erosion and sediment transport beneath glaciers and ice sheets. Here, we implement a numerical model of steady-state till equilibrium line position under a synthetic outlet glacier. We explore the influence of ice sliding velocity, clast sizes and distribution, and effective pressure at the bed. Additionally, we consider the case where the threshold velocity to induce ploughing is not constant, but instead depends on ice and sediment properties.
Zoet, L. K., & Iverson, N. R. (2020). A slip law for glaciers on deformable beds. Science, 368(6486), 76-78.
Alley, R. B., Blankenship, D. D., Rooney, S. T., & Bentley, C. R. (1989). Sedimentation beneath ice shelves—the view from ice stream B. Marine Geology, 85(2-4), 101-120. +
Over 10 percent of the worlds’ population lives less than 10 meters above sea level(McGranahan et al,. 2007), putting them at risk for rising seas and sinking coasts. Additionally, coastal inhabitants preferentially live in locations that are subsiding (Nicholls et al,. 2012), representing a flooding threat to people and infrastructure in coastal cities. Findings from the Intergovernmental Panel on Climate Change (IPCC 2018) outline the risks and impacts of sea level rise on flooding, and go on to identify a knowledge gap regarding the combined effects with coastal subsidence. When drivers of subsidence combine, they can generate sinking rates of 6-100mm/yr, significantly more than the 3-10mm/yr for sea level rise alone (Erkens et al,. 2015), making subsidence an order of magnitude more threatening to coastal cities.
The recent growth in access to C-band Synthetic Aperture Radar (SAR) data through the European Space Agency (ESA) Sentinel-1A/B satellites and the upcoming NASA-ISRO Synthetic Aperture Radar (NISAR) mission provide increased opportunities for differential interferometric synthetic aperture radar (DInSAR) monitoring. Here we developed a dockerized supercomputer workflow that allows us to rapidly generate InSAR pairs from Sentinel 1 imagery using ISCE processing software(Rosen et al., 2012) at ~10 meter resolution.
Results from this workflow are used to create a timeseries of subsidence for Lagos, Nigeria, where rapid urban growth has led to accelerated subsidence throughout the city. This growth has resulted in various flash floods due to infiltration and drainage issues in the last fifteen years(Atufu 2018), and the city is also vulnerable to coastal flooding. Next steps will include inputting our time series to determine how future flood events may impact specific areas of Lagos. Understanding where floods are a higher risk can allow for better distribution of rescue resources, and allow for targeted remediation and recovery efforts.
Over the recent geological past, alternations in the sediment budgets of low-order dryland catchments have left behind pronounced topographic imprints and are observed worldwide. The modern consequences of similar shifts extend from damage to ecosystems and crop yields to worldwide geopolitics. Quantifying erosion trends and understanding potential drivers for changes in the sediment budget is, however, a long-lasting challenge; ephemeral catchments are commonly situated in hydrological and vegetation boundary zones, greatly impacted by fine-scale and episodic erosion-triggered climatic events that are hard to observe and computationally complex over landscape evolution time scales.
This poster presents a set of catchment-scale landscape evolution numerical experiments designed according to field observations and projected climate records for the end of the 21st century derived from a convection-permitting climate model. The results demonstrate how changes in minute-scale rainfall bursts, as measured under modern global warming, could significantly enhance erosion rates and impact multi-century landscape evolution, even without a change in the general wetness level. Further, the presented modeling approach translates the future climate projection over the High Plains of Colorado into changes in vegetation cover along with an increase in sediment yield, which is contrary to the predicted general reduction in precipitation over the region. Modeling results, in conjunction with hints from a study area, reveal key relations between catchment morphology, soil/lithological properties, and vegetation cover that contribute to the understanding of landscape evolution under climatic changes. +
Particle settling velocity and bed erodibility impact the transport of suspended sediment to the first order, but are especially difficult to parameterize for the muds that often dominate estuarine sediments. For example, fine grained silts and clays typically form loosely bound aggregates (flocs) whose settling velocity can vary widely. Properties of flocculated sediment such as settling velocity and particle density are difficult to prescribe because they change in response to several factors, including salinity, suspended sediment concentration, turbulent mixing, organic content, and mineral composition. Additionally, mud consolidates after deposition, so that its erodibility changes over timescales of days to weeks in response to erosion, deposition, dewatering, and bioturbation. As understanding of flocculation and consolidation grows in response to recent technical advances in field sampling, numerical models describing cohesive behavior have been developed.
For this study, we implement an idealized two-dimensional model that represents a longitudinal section of a partially mixed estuary that mimics the primary features of the York River estuary, VA; and accounts for freshwater input, tides, and estuarine circulation. Suspended transport, erosion, and deposition are calculated using routines from the COAWST (Coupled Ocean-Atmosphere-Wave-and-Sediment Transport) modeling system. Here we evaluate the impact that bed consolidation and flocculation have on suspended sediment dispersal in the idealized model using a series of model runs. The simplest, standard model run neglects flocculation dynamics and consolidation. Next, a size-class-based flocculation model (FLOCMOD) is implemented. The third model run includes bed consolidation processes, but neglects flocculation; while the last model run includes both processes. Differences in tidal and daily averages of suspended load, bulk settling velocity and bed deposition are compared between the four model runs, to evaluate the relative roles of the different cohesive processes in limiting suspension in this partially mixed estuary. With an eye toward implementing these formulations in a realistic-grid model, we also consider the computational cost of including flocculation and consolidation.
Particle-based methods in computational fluid dynamics are capable of characterizing the propagation of the inertial terms and complex behavior of a fluid in low-viscosity systems onto an interface with highly viscous or solid materials, providing a high resolution window into fluid dynamics within environments that are fundamentally defined by fluid-solid interaction. The rate limiting feedbacks of wave action, erosion and sediment transport are a multiscale problem, involving kilometer-scale climate forcing and local, submeter Earth responses. We use real-world inputs of dynamic water levels at the mesoscale to drive local particle-based wave solutions towards natural coastal landforms effectively coupling the multiscale transfer of forces produced by topographic relief, wave action and ice collisions. 3D temporal solutions of nearshore currents at multiple scales make it possible to handle the specifics of fluid-solid interactions as the basis for training algorithms and subsequent expansion to larger regions. +
Passive margin stratigraphy contains time-integrated records of landscapes that have long since vanished. Quantitatively reading the stratigraphic record using coupled landscape evolution and stratigraphic forward models (SFMs) is a promising approach to extracting information about landscape history. However, the most commonly used SFM, which is based on an approximation of local, linear slope-dependent sediment transport, fails to produce diagnostic features of passive margin stratigraphy such as long-distance transport from the continental shelf and slope onto the abyssal plain. There is no consensus about the optimal form of simple SFMs because there has been a lack of direct tests against observed stratigraphy in well constrained test cases. Here we develop a nonlocal, nonlinear one-dimensional SFM that incorporates slope bypass and long-distance sediment transport, both of which have been previously identified as important model components but not thoroughly tested. Our model collapses to the local, linear model under certain parameterizations such that best-fit parameter values can be indicative of optimal model structure. Using seven detailed seismic sections from the South African Margin, we invert the stratigraphic data for best-fit model parameter values and demonstrate that best-fit parameterizations are not compatible with the local, linear diffusion model. Fitting the observed stratigraphy requires parameter values consistent with important contributions from slope bypass and long-distance transport processes. The nonlocal, nonlinear model yields improved fits to the data regardless of whether the model is compared against only the modern bathymetric surface or the full set of seismic reflectors identified in the data. Results suggest that processes of sediment bypass and long-distance transport are required to model realistic passive margin stratigraphy, and are therefore important to consider when inverting the stratigraphic record to infer past perturbations to source regions.
Past decades have seen rapid advancements in the field of soil erosion modelling, with a shift away from lumped empirical models and towards fully-distributed physically-based erosion models. The benefits of this shift is that distributed erosion models facilitate the spatial predictions of erosion and deposition across the landscapes by computing runoff and modelling the subsequent detachment, transport, and deposition of sediments. Despite the ability to represent the physical process of erosion spatially, distributed erosion models are validated to discharge and sediment yield at catchment outlets. Spatial information on erosion and deposition rates are seldom used to validate distributed models; this is because both plot and field-scale data on erosion rates are rare. Structure-from-motion (SfM) and multi-view stereo (MVS) algorithms coupled with the use of unmanned aerial vehicles (UAVs) have become a popular tool in geomorphology for modelling topographic change-detection on complex landscapes. We demonstrate the viability of using these techniques to generate spatial validation data; repeat UAV surveys of an agricultural field are used to identify dominant sediment flow paths, depositional zones, and rill/gully erosion processes. This unique spatial dataset allows us to tackle issues of spatial equifinality, model parameterization, and the accurate discretization of the landscape. +
A
Patterns of sediment transport and particle residence times influence the morphology and ecology of shallow coastal bays in important ways. The Virginia Coast Reserve (VCR), a barrier island-lagoon-marsh system on the Eastern Shore of Virginia, is typical of many shallow coastal bay complexes that lack a significant fluvial source of freshwater and sediment. Sediment redistribution within the bays in response to storms and sea-level rise, together with the dynamics of marsh and lagoon-bottom plants, largely governs the morphological evolution of this system. There are also important feedbacks between sediment and ecosystem dynamics. This is particularly true in the VCR, which is relatively unaffected by human activities. As a step towards evaluating the impact of hydrodynamics on sediment and ecological processes in the VCR, we employ a single unified model that accounts for circulation, surface waves, wave-current interaction, and sediment processes. This three-dimensional unstructured grid finite-volume coastal ocean model (FVCOM) is validated with field observations of wind- and tide-induced water flow (water level and current velocities) in Hog Island Bay, centrally located within the VCR. We present here the resulting patterns of sediment transport and particle residence times over event and seasonal time scales. Water and particle exchange within the VCR and between the VCR and the ocean is examined with the Lagrangian particle-tracking module in FVCOM. We focus on 3 bays with strongly varying bathymetry and coastline geometry, which are also located along a gradient of nitrogen input to the system. The results indicate that residence time of particles within the system vary greatly depending on the location of particle release, bay morphology, and wind conditions. The implications for morphologic evolution and ecosystem response to climate and land-use changes are evaluated. +
2
Permafrost thaw is of growing concern due to its potential to weaken slope stability and influence the magnitude and frequency of rapid mass movements in an area. Therefore, modelling of permafrost distribution and dynamics is needed for mountain areas, where such events happened or might happen. For instance in Iceland, permafrost thaw has been recently recognized as a new factor influencing landslide triggering, whereas the instability and slow movement of the unstable rock slope Mannen in southwestern Norway might be connected to permafrost conditions.
CryoGrid is a suite of permafrost models for solving different problems related to permafrost processes. This presentation will show examples of application of the two model schemes, the one-dimensional model CryoGrid 2 (Westermann et al., 2013) and the two-dimensional model CryoGrid 2D (Myhra et al., 2017). Both models solve the transient heat diffusion equation that additionally accounts for the latent heat effects due to ground freezing or thawing. The CryoGrid 2 model is forced with time series of air/ground surface temperature and snow depth data, whereas the CryoGrid 2D model is forced with ground surface temperature.
We applied CryoGrid 2 to the regional modelling of permafrost dynamics in Iceland for the last six decades. To account for snow redistribution, we ran the model for three scenarios of the snow depth, having a large impact on the modelled permafrost distribution.
CryoGrid 2D was employed to model ground temperature in the Mannen rock slope since the 1860s. Our preliminary results show that permafrost can occur in the Mannen slope in areas with considerably reduced snow depth, i.e. under steep parts of the slope.
REFERENCES
Myhra, K. S., Westermann, S., & Etzelmüller, B. (2017). Modelled distribution and temporal evolution of permafrost in steep rock walls along a latitudinal transect in Norway by CryoGrid 2D. Permafrost and Periglacial Processes, 28(1), 172-182. DOI: 10.1002/ppp.1884
Westermann, S., Schuler, T., Gisnås, K., & Etzelmüller, B. (2013). Transient thermal modeling of permafrost conditions in Southern Norway. The Cryosphere, 7(2), 719-739. DOI: 10.5194/tc-7-719-2013
Planktic foraminifera abundances and distributions, i.e. faunal assemblages, have been used to reconstruct oceanographic conditions from Earth's ancient past. Here we examine the utility of Species Distribution Models (SDMs) in characterizing the ecology of modern foraminifera and how this can inform reconstructions of past oceanographic states from Earth’s climatic history. Standard faunal assemblage proxy reconstructions often reduce multidimensional environmental data into a single variable, typically temperature. However, when environmental covariates feature strong spatial autocorrelation, these traditional methods may incorrectly interpret information from other variables as a temperature signal. Using modern machine learning-based SDMs we show that while temperature is an unquestionably important control on foraminifera distribution, other environmental factors appear to play a non-trivial role. This has implications not only for temperature values derived from standard assemblage-based reconstructions, but also may help reconcile apparent mismatches between proxies and climate model simulations +
Post-fire debris flow is a common hazard in the western United States. However, after decades of efforts in the debris flow research community, universally applicable post-fire debris flow predict methods are still lacking. Large discrepancies in the post-fire debris flow initiation mechanism are the main source that limits the predictive accuracy of debris flow. Improve and understanding these discrepancies is significant to possibly improve the debris flow modeling. In this work, we propose a data-driven, physics-informed machine learning approach for reconstructing and predicting debris flows. By using a classic supervising modern learning technique based on logistics regression, the logistics regression functions are trained by existing direct field measurements and debris flow numerical simulations from Las Lomas after 2016 Fish fire and then used to predict debris flow in different drainage basin where data are not available. The proposed method is evaluated by two classes of simulations: sediment transport model and runoff model. In runoff simulations, five drainage basins are considered: Las Lomas, Arroyo Seco, Dunsmore 1, Dunsmore 2, Big Tujunga. In sediment transport model, Las Lomas and Arroyo Seco watersheds are applied. Excellent predictive performances were observed in both scenarios, demonstrating the capabilities of the proposed method. +
Post-wildfire debris flows are a major source of geomorphic change that by nature of the large amounts of mass they mobilize can be deadly and destructive. These landslides are triggered by the interaction of fire-induced changes to both hydrologic and geomorphic responses. Representing the cascading effects of fire on landslides requires linking information from hydrologic models and debris flow models and presents both technical and theoretical challenges. Statistical models of debris flows have been used successfully for decades to assist in disaster prevention and mitigation. However, physically-based models that may provide additional insight into underlying processes and behavior under extreme conditions are rarely used.
We present a case study to begin addressing these challenges, focusing on a basin burned by the Thomas Fire in southern California in 2017. Soil water content maps and sediment fluxes produced by the Distributed Hydrology Soil Vegetation Model (DHSVM) in areas at risk for debris flows are compared with times and locations of known landslides. The degree of correspondence between modelled debris flow risk factors is compared for different potential methods of representing fire in DHSVM, including: changes to soil depth, soil infiltration characteristics, vegetation cover, and vegetation properties. Finally, future challenges of linking information across hydrologic and landslide models are discussed, towards more accurately representation of the spectrum of debris flow processes. +
Predicting river hydrodynamics through computational models is critical for advancing science and engineering practices to manage rivers and floodplains. Traditional hydrodynamic models pose computational challenges, often demanding extensive processing time for large-scale 2D flood simulations. While data-driven algorithms have shown promise in improving simulation efficiency, existing efforts have primarily concentrated on generating inundation maps only at event peaks. In this research, we introduce a novel deep learning model designed to provide accurate and rapid simulation of the temporal evolution of floods, providing 2D prediction of both water depth and flood inundation maps across an entire event. We trained and evaluated this model based on a dataset that was developed using HEC-RAS, a physics-based model, for a segment of Ninnescah River, in Kansas. This was done using a deep learning model to integrate the spatial advantages of a convolutional neural network along with the temporal sequence capabilities of a long-short term memory network. The hybrid model demonstrates remarkable proficiency in capturing the dynamic nature of flood events. Evaluation of the inundation maps, at the highest testing peak, exhibited exceptional performance, with precision exceeding 0.99 and an F1-score approaching 0.98. Moreover, this hybrid model showed robust performance in predicting water depth maps, with RMSE values of 0.03 m on average during testing and 0.08 m at the highest peak time-step. This study represents a significant advancement in our ability to conduct long-term simulations of hydrodynamics and sediment transport. +
Predictive understanding of the variation and distribution of substrates at large spatial extents in aquatic systems is severely lacking. This hampers efforts to numerically predict the occurrence and distribution of specific benthic habitats, which must be observed in the field. Existing survey methods are limited in scale, require heavy and technically sophisticated survey equipment, or are prohibitively expensive for surveying and mapping. Recreation-grade side scan sonar (SSS) instruments, or fishfinders, have demonstrated their unparalleled value in a lightweight and easily-to-deploy system to image benthic habitats efficiently at the landscape-level. Existing methods for generating geospatial datasets from these sonar systems require a high-level of interaction from the user and are primarily closed-source, limiting opportunities for community-driven enhancements. We introduce PING-Mapper, an open-source and freely available Python-based software for generating geospatial benthic datasets from recreation-grade SSS systems. PING-Mapper is an end-to-end framework for surveying and mapping aquatic systems at large spatial extents reproducibly, with minimal intervention from the user. Version 1.0 of the software (Summer 2022) decodes sonar recordings from any existing Humminbird® side imaging system, export plots of sonar intensities and sensor-derived bedpicks and generates georeferenced mosaics of geometrically corrected sonar imagery. Version 2.0 of the software, to be released Summer 2023, extends PING-Mapper functionality by incorporating deep neural network models that automatically locate and mask sonar shadows, calculate independent bedpicks from both side scan channels, and classify substrates at the pixel level. The widespread availability of substrate information in aquatic systems will facilitate development of the next generation of in-stream models for routing flows of sediment and water, as well as more sophisticated simulations of specific habitats.
Preserved in the morphology of bedrock river valleys is a recorded history of geologic and climactic conditions experienced by that river over time. A deep, narrow valley suggests that conditions favored vertical incision over lateral erosion, and the presence of a wide bedrock valley indicates that lateral erosion and valley widening outpaced vertical incision. While vertical incision and the rates and mechanisms by which it operates are relatively well understood, the processes of lateral erosion and valley widening remain more enigmatic. Utilizing bedrock valley morphology as an interpretive tool is impossible without first improving our knowledge of the valley widening process. For a bedrock valley to widen the river must first laterally erode the bedrock wall until the overlying stresses cause the wall to collapse into a talus pile on the valley bottom. Once the material has collapsed, it then must be transported away from the bedrock wall so that the river can regain access to the wall and lateral erosion can continue. In this two-step conceptual model of valley widening, the size of the individual talus blocks and the volume of the pile itself plays a large role in the rate of valley widening over time. In this study, I use numerical modeling to estimate the long-term breakdown and removal of talus material in a river through chemical and physical weathering. Inputs for the model include measured talus pile characteristics from a bedrock river with wide and narrow bedrock valleys (Buffalo River) and long-term flood simulations generated by the LandLab tool, Random Precipitation Distribution Generator. Model results may offer some insight into the potential role of talus in the bedrock valley widening process and improve our understanding of the conditions favorable for the development of wide bedrock river valleys. +
A
Previous studies have found that the ratio between valley spacing and mountain range width is relatively constant across the globe, but the processes responsible for its uniformity are not well understood. To determine the reasons for this uniform ratio, we firstly need to explore why valleys are evenly distributed in a mountain range, and what factors can impact valley spacing. Recent research has found that the critical length between hillslope and fluvial processes is an important control on the valley spacing of first order fluvial channels. In this study, we use the CHILD landscape evolution model to explore how the critical length affects valley spacing in higher order fluvial channels, and we use these results to help explain the narrow range of observations in the valley spacing ratio. We find that valley spacing has a linear relationship with critical length in higher order channels and, for a given order channel, the ratio between valley spacing and critical length is relatively constant. This relationship demonstrates that the competition between hillslope and fluvial processes influences the distribution of higher order channels across the landscape. However, we also find that valley spacing is influenced by model initial conditions and variability across the landscape, such as orographic precipitation patterns. Moreover, for a fixed domain in our model, although the critical length may vary, the ratio between the valley spacing of trunk channels and mountain width remains in the range observed in real landscapes. The reason for this is that the order of trunk channels varies with the critical length. Therefore, for a given domain size (or mountain range width), a larger critical length can produce lower order trunk channels but with the same spacing value as higher order trunk channels with a smaller critical length. This may be one of the reasons why the spacing ratio is relatively constant across diverse natural settings. +
2
Proper quantification of sediment flux has always been an area of interest both for scientist and engineers involved in hydraulic engineering and management of rivers, estuaries and coastal waters. In spite of the importance of bedload flux globally, either for monitoring water quality, maintaining coastal and marine ecology or during dam construction or even for food security, bedload data, especially for large rivers, extremely scarce. This is due to the fact that bedload flux measurements are relatively expensive and time consuming and introduce large spatial and temporal uncertainties. Lack of adequate and continuous field observation is a hindrance to developing a globally accepted numerical model. We developed a new global riverine bedload flux model as an extension of the WBMsed framework. Here we present an evaluation of the model predictions using over eighty field observations for large rivers (over 1000 km2), collected from different sources. This model will be used to study various aspects of fluvial geomorphology globally, which is most common interest area for the researcher to see the impacts of different issues at global scale. Also, considering the contribution of bedload as sediment in the global level, it will elucidate the relationship between suspended sediment and bedload. The observational dataset we compiled is in itself a unique product that can be instrumental for future studies. +
Quantifying the spatial distribution of vegetation on coastal dunes is critical for understanding their morphological evolution and ecological functioning. However, effective large-scale mapping of dune vegetation presents a challenge for coastal managers at the resolution suitable for detailed analysis of these berm-dune systems. This study presents an integrated machine learning and geographic information system (GIS) methodology to accurately classify vegetation, sand, and shadows from high-resolution unmanned aerial vehicle (UAV) imagery of an undeveloped dune system in Long Branch, New Jersey. Using a Zenmuse X5S 3-band multispectral sensor and a Zenmuse X5 RGB sensor, we collected data from a single survey in October, 2023. We then combined the band data from this flight in order to develop additional model inputs of normalized difference vegetation index (NDVI) and green NDVI (gNDVI). Thirteen model runs were performed and compared using 3 to 8 inputs including spectral bands of Red, Green, Blue, and near-infrared (NIR). The spectral indices of NDVI and gNDVI, were developed using Red, Green, and NIR band data. In our research, the random forest machine learning algorithm was employed, with optimal hyperparameters determined through empirical testing. Classifiers with 350 decision trees were found to yield highest accuracy on the training data across different spectral indices combinations. The trained random forest models were then applied to map vegetation distribution on an unseen "training grid" area. Results show that model runs containing the NDVI and gNDVI yielded higher accuracy in classification than model runs only containing a combination of R, G, B, and NIR bands. Incorporating additional Red and Green bands from our multispectral sensor produced less accurate model classification as these additional inputs lead to overfitting and increased misclassification of vegetation and shadows. Incorporating NDVI and gNDVI indices significantly improved classification performance compared to model runs that only used spectral band information. The classified vegetation, sand, and shadow maps were integrated into ArcPro GIS software for visualization and validation against other established methods of manual class identification and map digitization. Rigorous accuracy assessments confirmed the robustness of the machine learning approach that can be used to quickly and accurately identify vegetation distribution in a dune complex.
