License Information

Use of this function requires a license for Whitebox Workflows for Python Professional (WbW-Pro). Please visit www.whiteboxgeo.com to purchase a license.

Description

This tool calculates several multiscale curvatures and curvature-based indices from an input DEM (--dem). There are 18 curvature types (--curv_type) available, including: accumulation curvature, curvedness, difference curvature, Gaussian curvature, generating function, horizontal excess curvature, maximal curvature, mean curvature, minimal curvature, plan curvature, profile curvature, ring curvature, rotor, shape index, tangential curvature, total curvature, unsphericity, and vertical excess curvature. Each of these curvatures can be measured in non-multiscale fashion using the corresponding tools available in either the WhiteboxTools open-core or the Whitebox extension.

Like many of the multi-scale land-surface parameter tools available in Whitebox, this tool can be run in two different modes: it can either be used to measure curvature at a single specific scale or to generate a curvature scale mosaic. To understand the difference between these two modes, we must first understand how curvatures are measured and how the non-multiscale curvature tools (e.g. ProfileCurvature) work. Curvatures are generally measured by fitting a mathematically defined surface to the elevation values within the local neighbourhood surrounding each grid cell in a DEM. The Whitebox curvature tools use the algorithms described Florinsky (2016), which use the 25 elevations within a 5 x 5 local neighbouhood for projected DEMs, and the nine elevations within a 3 x 3 neighbourhood for DEMs in geographic coordinate systems. This is what determines the scale at which these land-surface parameters are calculated. Because they are calculated using small local neighbourhoods (kernels), then these algorithms are heavily impacted by micro-topographic roughness and DEM noise. For example, in a fine-resolution DEM containing a great deal of micro-topographic roughness, the measured curvature value will be dominated by topographic variation at the scale of the roughness rather than the hillslopes on which that roughness is superimposed. This mis-matched scaling can be a problem in many applications, e.g. in landform classification and slope failure modelling applications.

Using the MultiscaleCurvatures tool, the user can specify a certain desired scale, larger than that defined by the grid resolution and kernel size, over which a curvature should be characterized. The tool will then use a fast Gaussian scale-space method to remove the topographic variation in the DEM at scales less than the desired scale, and will then characterize the curvature using the usual method based on this scaled DEM. To measure curvature at a single non-local scale, the user must specify a minimum search neighbourhood radius in grid cells (--min_scale) greater than 0.0. Note that a minimum search neighbourhood of 0.0 will replicate the non-multiscale equivalent curvature tool and any --min_scale value > 0.0 will apply the Gassian scale space method to eliminate topographic variation less than the scale of the minimum search neighbourhood. The base step size (--step), number of steps (--num_steps), and step nonlinearity (--step_nonlinearity) parameters should all be left to their default values of 1 in this case. The output curvature raster will be written to the output magnitude file (--out_mag). The following animation shows several multiscale curvature rasters (tangential curvature) measured from a DEM across a range of spatial scales.

Alternatively, one can use this tool to create a curvature scale mosaic. In this case, the user specifies a range of spatial scales (i.e., a scale space) over which to measure curvature. The curvature scale-space is densely sampled and each grid cell is assigned the maximum absolute curvature value (for the specified curvature type) across the scale space. In this scale-mosaic mode, the user must also specify the output scale file name (--out_scale), which is an output raster that, for each grid cell, specifies the scale at which the maximum absolute curvature was identified. The following is an example of a scale mosaic of unsphericity for an area in Pole Canyon, Utah (min_scale=1.0, step=1, num_steps=50, step_nonlinearity=1.0).

Scale mosaics are useful when modelling spatial distributions of land-surface parameters, like curvatures, in complex and heterogeneous landscapes that contain an abundance of topographic variation (micro-topography, landforms, etc.) at widely varying spatial scales, often associated with different geomorphic processes. Notice how in the image above, relatively strong curvature values are being characterized for both the landforms associated with the smaller-scale mass-movement processes as well as the broader-scale fluvial erosion (i.e. valley incision and hillslopes). It would be difficult, or impossible, to achieve this effect using a single, uniform scale. Each location in a land-surface parameter scale mosaic represents the parameter measured at a characteristic scale, given the unique topography of the site and surroundings.

The properties of the sampled scale space are determined using the --min_scale, --step, --num_steps (greater than 1), and --step_nonlinearity parameters. Experience with multiscale curvature scales spaces has shown that they are more highly variable at shorter spatial scales and change more gradually at broader scales. Therefore, a nonlinear scale sampling interval is used by this tool to ensure that the scale sampling density is higher for short scale ranges and coarser at longer tested scales, such that:

ri = rL + [step × (i - rL)]p

Where ri is the filter radius for step i and p is the nonlinear scaling factor (--step_nonlinearity) and a step size (--step) of step.

In scale-mosaic mode, the user must also decide whether or not to standardize the curvature values (--standardize). When this parameter is used, the algorithm will convert each curvature raster associated with each sampled region of scale-space to z-scores (i.e. differenced from the raster-wide mean and divided by the raster-wide standard deviation). It it usually the case that curvature values measured at broader spatial scales will on the whole become less strongly valued. Because the scale mosaic algorithm used in this tool assigns each grid cell the maximum absolute curvature observed within sampled scale-space, this implies that the curvature values associated with more local-scale ranges are more likely to be selected for the final scale-mosaic raster. By standardizing each scaled curvature raster, there is greater opportunity for the final scale-mosaic to represent broader scale topographic variation. Whether or not this is appropriate will depend on the application. However, it is important to stress that the sampled scale-space need not span the full range of possible scales, from the finest scale determined by the grid resolution up to the broadest scale possible, determined by the spatial extent of the input DEM. Often, a better approach is to use this tool to create multiple scale mosaics spanning this range, thereby capturing variation within broadly defined scale ranges. For example, one could create a local-scale, meso-scale, and broad-scale curvature scale mosaics, each of which would capture topographic variation and landforms that are present in the landscape and reflective of processing operating at vastly different spatial scales. When this approach is used, it may not be necessary to standardize each scaled curvature raster, since the gradual decline in curvature values as scales increase is less pronounced within each of these broad scale ranges than across the entirety of possible scale-space. Again, however, this will depend on the application and on the characteristics of the landscape at study.

Raw curvedness values are often challenging to visualize given their range and magnitude, and as such the user may opt to log-transform the output raster (--log). Transforming the values applies the equation by Shary et al. (2002):

Θ' = sign(Θ) ln(1 + 10n|Θ|)

where Θ is the parameter value and n is dependent on the grid cell size.

References

Florinsky, I. (2016). Digital terrain analysis in soil science and geology. Academic Press.

See Also

gaussian_scale_space, accumulation_curvature, curvedness, difference_curvature, gaussian_curvature, generating_function, horizontal_excess_curvature, maximal_curvature, mean_curvature, minimal_curvature, plan_curvature, profile_curvature, ring_curvature, rotor, shape_index, tangential_curvature, total_curvature, unsphericity, vertical_excess_curvature

Function Signature

def multiscale_curvatures(self, dem: Raster, curv_type: str = 'profile', min_scale: int = 4, step_size: int = 1, num_steps: int = 10, step_nonlinearity: float = 1.0, log_transform: bool = True, standardize: bool = False) -> Tuple[Raster, Raster]: ...

Project Links

WbW Homepage User Manual Support WbW