Search

Many studies of Earth surface processes and landscape evolution rely on having accurate and extensive data sets of surficial geologic units and landforms. Automated extraction of geomorphic features using deep learning provides an objective way to consistently map landforms over large spatial extents. However, there is no consensus on the optimal input feature space for such analyses. We explore the impact of input feature space for extracting geomorphic features from land surface parameters (LSPs) derived from digital terrain models (DTMs) using convolutional neural network (CNN)‐based semantic segmentation deep learning. We compare four input feature space configurations: (a) a three‐layer composite consisting of a topographic position index (TPI) calculated using a 50 m radius circular window, square root of topographic slope, and TPI calculated using an annulus with a 2 m inner radius and 10 m outer radius, (b) a single illuminating position hillshade, (c) a multidirectional hillshade, and (d) a slopeshade. We test each feature space input using three deep learning algorithms and four use cases: two with natural features and two with anthropogenic features. The three‐layer composite generally provided lower overall losses for the training samples, a higher F1‐score for the withheld validation data, and better performance for generalizing to withheld testing data from a new geographic extent. Results suggest that CNN‐based deep learning for mapping geomorphic features or landforms from LSPs is sensitive to input feature space. Given the large number of LSPs that can be derived from DTM data and the variety of geomorphic mapping tasks that can be undertaken using CNN‐based methods, we argue that additional research focused on feature space considerations is needed and suggest future research directions. We also suggest that the three‐layer composite implemented here can offer better performance in comparison to using hillshades or other common terrain visualization surfaces and is, thus, worth considering for different mapping and feature extraction tasks. Plain Language Summary: Characteristics of the land surface, such as steepness, relative slope position (e.g., ridge vs. valley), and roughness, can be digitally represented using a variety of methods. These digital representations, along with human‐annotated labels, can be provided to artificial intelligence algorithms to generate maps of landforms, such as those associated with river systems, glaciers, or human‐induced changes to the landscape. Given the large number of terrain derivatives that can be generated from digital elevation data, it is unclear how the chosen inputs impact the utility of the resulting maps. This study suggests that artificial intelligence mapping algorithms are sensitive to representations provided to them since different inputs resulted in varying map accuracies. We suggest a combination of features, which collectively describe relative slope position, steepness, and local terrain texture, as a means to represent the landscape and to serve as input to artificial intelligence algorithms. Our results can make it easier to train artificial intelligence algorithms to consistently and objectively find surficial features of interest across large swaths of Earth's surface. Key Points: We test the sensitivity of convolutional neural network geomorphic feature extraction approaches to the input feature spaceA three‐layer composite incorporating slope and topographic position index outperformed common inputs like hillshades and slopeshadesResults held true across four use cases including both natural and anthropogenic landforms [ABSTRACT FROM AUTHOR]

Landslides are the main source of sediment in most mountain ranges. Rivers then act as conveyor belts, evacuating landslide-derived sediment. Sediment dynamics are known to influence landscape evolution through interactions among landslide sediment delivery, fluvial transport, and river incision into bedrock. Sediment delivery and its interaction with river incision therefore control the pace of landscape evolution and mediate relationships among tectonics, climate, and erosion. Numerical landscape evolution models (LEMs) are well suited to study the interaction among these earth surface processes. They enable evaluation of a range of hypotheses at varying temporal and spatial scales. While many models have been used to study the dynamic interplay between tectonics, erosion and climate, the role of interactions between landslide-derived sediment and river incision has received much less attention. Here, we present HyLands, a hybrid landscape evolution model integrated within the Topo Toolbox Landscape Evolution Model (TTLEM) framework. The hybrid nature of the model lies in its capacity to simulate both erosion and deposition at any place in the landscape due to fluvial bedrock incision, sediment transport and rapid, stochastic mass wasting through landsliding. Fluvial sediment transport and bedrock incision are calculated using the recently developed Stream Power with Alluvium Conservation and Entrainment (SPACE) model. Therefore, rivers in HyLands can dynamically transition from detachment-limited to transport-limited, and from bedrock to bedrock-alluvial to fully alluviated states. Erosion and sediment production by landsliding is calculated using a Mohr-Coulomb stability analysis while landslide-derived sediment is routed and deposited using a multiple flow direction, non-linear deposition method. We describe and evaluate the HyLands 1.0 model using analytical solutions and observations. We first illustrate the functionality of HyLands to capture river dynamics ranging from detachment-limited to transport-limited configurations. Second, we apply the model to a portion of the Namche-Barwa massif in Eastern Tibet and compare simulated and observed landslide magnitude-frequency and area-volume scaling relationships. Finally, we illustrate the relevance of explicitly simulating landsliding and sediment dynamics over longer timescales for landscape evolution in general and river dynamics in particular. With HyLands we provide a new tool to understand both the long and short-term coupling between stochastic hillslope processes, river incision, and source-to-sink sediment dynamics. [ABSTRACT FROM AUTHOR]

Surface mining may be humanity's most tangible impact on Earth's surface and will become more prevalent as the energy transition progresses. Prediction of post-mining landscape change can help mitigate environmental damage, but requires understanding how mining changes geomorphic processes and variables. Here we investigate surface mining's complex influence on surface processes in a case study of mountaintop removal/valley fill (MTR/VF) coal mining in the Appalachian Coalfields, USA. The future of MTR/VF landscapes is unclear because mining's effects on geomorphic processes are poorly understood. We use geospatial analysis—leveraging the existence of pre- and post-MTR/VF elevation models—and synthesis of literature to ask how MTR/VF alters topography, hydrology, and land-surface erodibility and how these changes could be incorporated into numerical models of post-MTR/VF landscape evolution. MTR/VF reduces slope and area–slope product, and rearranges drainage divides. Creation of closed depressions alters flow routing and casts doubt on the utility of models that assume steady flow. MTR/VF creates two contrasting hydrologic domains, one in which overland flow is generated efficiently due to a lack of infiltration capacity, and one in which waste rock deposits act as extensive subsurface reservoirs. This dichotomy creates localized hotspots of overland flow and erosion. Loss of forest cover probably reduces cohesion in near-surface soils for at least the timescale of vegetation recovery, while waste rock fills and minesoils also likely experience reduced erosion resistance. Our analysis suggests three necessary ingredients for numerical modeling of post-MTR/VF landscape change: 1) accurate routing and accumulation of unsteady overland flow and accompanying sediment across low-gradient, depression-rich, engineered landscapes, 2) separation of the landscape into cut, filled, and unmined regions, and 3) incorporation of vegetation recovery trajectories. Improved modeling of post-mining landscape evolution will mitigate environmental degradation from past mining and reduce the impacts of future mining that supports the energy transition. [Display omitted] • Mountaintop removal mining flattens topography and moves drainage divides. • Cut and filled domains generate different quantities of erosive runoff. • Creation of many closed depressions reduces applicability of common models. • Vegetation loss and material property changes increase erodibility. • Our work reveals the necessary elements for models of post-mining landscape change. [ABSTRACT FROM AUTHOR]

Models of landscape evolution provide insight into the geomorphic history of specific field areas, create testable predictions of landform development, demonstrate the consequences of current geomorphic process theory, and spark imagination through hypothetical scenarios. While the last 4 decades have brought the proliferation of many alternative formulations for the redistribution of mass by Earth surface processes, relatively few studies have systematically compared and tested these alternative equations. We present a new Python package, terrainbento 1.0, that enables multi-model comparison, sensitivity analysis, and calibration of Earth surface process models. Terrainbento provides a set of 28 model programs that implement alternative transport laws related to four process elements: hillslope processes, surface-water hydrology, erosion by flowing water, and material properties. The 28 model programs are a systematic subset of the 2048 possible numerical models associated with 11 binary choices. Each binary choice is related to one of these four elements – for example, the use of linear or nonlinear hillslope diffusion. Terrainbento is an extensible framework: base classes that treat the elements common to all numerical models (such as input/output and boundary conditions) make it possible to create a new numerical model without reinventing these common methods. Terrainbento is built on top of the Landlab framework such that new Landlab components directly support the creation of new terrainbento model programs. Terrainbento is fully documented, has 100 % unit test coverage including numerical comparison with analytical solutions for process models, and continuous integration testing. We support future users and developers with introductory Jupyter notebooks and a template for creating new terrainbento model programs. In this paper, we describe the package structure, process theory, and software implementation of terrainbento. Finally, we illustrate the utility of terrainbento with a benchmark example highlighting the differences in steady-state topography between five different numerical models. [ABSTRACT FROM AUTHOR]