Code associated with Brigham and Crider's (2022) metric for calculating along-landform morphologic variability.
The shape of a landform changes with time depending on the type and rate of the processes active upon it. Identifying and untangling the contribution of each factor to landform shape is difficult, but patterns in profile form along linear landforms can be examined to determine the dominant processes. We develop a methodology to quantify these variations in profile form and define a “morphologic variability metric.”
Using the example of fault scarps in jointed bedrock, we first manually classify fault-scarp profiles extracted from the SfM-derived point clouds into six morphologic categories defined through observations of common forms across scarps.
Our morphologic variability metric quantitatively describes this change in shape. Then, we use principal component analysis with singular value decomposition, a data-reduction technique, on the positional data of the scarp-normal profiles to quantitatively distinguish between these morphologic classes. We follow this by employing a supervised learning algorithm, the support vector machine (SVM) method, to build a classifier, using the principal-component coordinates of the classified profiles in principal component space as a training set. Classification performance was assessed using 5-fold cross validation (81% accuracy) and with independent test data (80% accuracy). Finally, we define a morphologic variability metric and calculate it by determining the number of classes represented and the standard deviation of their proportions in a moving window along the scarp. This metric is then normalized, so that a value of 0 represents a section of scarp where all profiles belong to the same class, and a value of 1 represents a section where all classes are represented in equal proportion.
The morphologic classes are user-defined based on the profile forms represented in the overall dataset and the approach does not rely on any landform-specific metrics, instead using data-reduction techniques on the positional data of the landform profile itself as the basis for classification. This allows for a widespread application of this approach to many different types of landforms, as it eliminates the need to find a sufficient number of landform-specific parameters to quantitatively characterize the shape of the profile. Our metric is a promising tool to understand how many types of landforms evolve.
In this notebook, we extract landform-normal profiles using three user-provided data: a DEM of the region of interest, a polygon shapefile that can be used to crop the DEM to the sepecific lanform if desired, and a line shapefile (contained within the cropping polygon) that traces the center of the landform.
The output of the notebook is a folder with text files for each landform-normal profile. Each text file has columns for the Easting, Northing, UTM Zone, and elevation of each point along the landform-normal profile. These text files are used as the data input for morp-var 2.
In this notebook, we use the output of the first notebook (morph-var 1), which is a folder containing text files with the coordinates and elevation data of points along each landform-normal profile. This notebook focuses on formatting and normalizing the profile data, so that they are ready for use in our classification algorithm (morph-var 3).
To normalize the profiles, we calculate the two-dimensional coordinates of each profile, automatically determine the position of the crest and toe of the scarp, crop the the profile between the crest and the toe and normalize the profile by the scarp height. The output of the notebook is a landform-specific folder with text files for the processed profiles. If there are multiple landforms to be analyzed, the notebook will have to be re-run for each landform.
Before running morph-var 3, the user will have to manually inspect all or a sub-selection the processed profiles obtained from morph-var 2 and classify them according to the desired morphologic attributes. The notebook is currently set up for six morphologic classes, but this can be adapted for any number of classes. Once the the selected profiles have been inspected and classified, all of the training profile text files must be put in the same folder, which will be used as the input folder for morph-var 3. A list of file names for each class will be needed as well.
The outputs of the notebook are the best parameters to use in the SVM model (which will be needed in morph-var 4) and the cross-validation statistics of the classification model.
In this notebook, we use the classifier built and trained in morph-var 3 to classify the profiles of a new landform. Extract and process the profiles using morph-var 1 and 2, storing the processed profiles in a single folder if there are multiple landforms. This folder will be one input folder for the notebook, with the other being the training data folder created for morph-var 3.
The output of this notebook is a text file with the class predictions for each profile.
In this notebook, we use the landform profile classification results obtained from morph-var 4 to calculate the number and proportion of classes represented in a moving window along a landform.
The output of this notebook is a text file with the varib
Brigham, C. A. P., and Crider, J., G.: A New Metric for Morphologic Variability Using Landform Shape Classification via Supervised Machine Learning, Geomorphology, 399, (2022). https://doi.org/10.1016/j.geomorph.2021.108065