GitHub

Transiograms

Empirical transiograms

Transiograms of categorical variables are matrix-valued functions $t_{ab}(h)$ that measure the transition probability from categorical value $a$ to categorical value $b$ at a given lag $h \in \R$. They are often used for the simulation of Markov processes in more than one dimension.

The Carle's estimator of the empirical transiogram is given by

\[\widehat{t_{ab}}(h) = \frac{\sum_{(i,j) \in N(h)} {I_a}_i \cdot {I_b}_j}{\sum_{(i,j) \in N(h)} {I_a}_i}\]

where $N(h) = \left\{(i,j) \mid ||\p_i - \p_j|| = h\right\}$ is the set of pairs of locations at a distance $h$, and where $I_a$ and $I_b$ are the indicator variables for categorical values (or levels) $a$ and $b$, respectively.

(Omini)directional transiograms

GeoStatsFunctions.EmpiricalTransiogram — Type
EmpiricalTransiogram(data, var; [options])

Computes the empirical (a.k.a. experimental) omnidirectional transiogram for categorical variable var stored in geospatial data.

Options

  • nlags - number of lags (default to 20)
  • maxlag - maximum lag in length units (default to 1/2 of minimum side of bounding box)
  • distance - custom distance function (default to Euclidean distance)
  • algorithm - accumulation algorithm (default to :ball)

Available algorithms:

  • :full - loop over all pairs of points in the data
  • :ball - loop over all points inside maximum lag ball

All implemented algorithms produce the exact same result. The :ball algorithm is considerably faster when the maximum lag is much smaller than the bounding box of the domain of the data.

See also: DirectionalTransiogram, PlanarTransiogram, EmpiricalTransiogramSurface.

References

source
GeoStatsFunctions.DirectionalTransiogram — Function
DirectionalTransiogram(direction, data, var; dtol=1e-6u"m", [options])

Computes the empirical transiogram for the categorical variable var stored in geospatial data along a given direction with band tolerance dtol in length units.

Forwards options to the underlying EmpiricalTransiogram.

source
GeoStatsFunctions.PlanarTransiogram — Function
PlanarTransiogram(normal, data, var; ntol=1e-6u"m", [options])

Computes the empirical transiogram for the categorical variable var stored in geospatial data along a plane perpendicular to a normal direction with plane tolerance ntol in length units.

Forwards options to the underlying EmpiricalTransiogram.

source

Consider the following categorical image:

using GeoStatsImages

img = geostatsimage("Gaussian30x10")

Z = [z < 0 ? 1 : 2 for z in img.Z]

cat = georef((; Z=Z), img.geometry)

cat |> viewer
Example block output

We can estimate the ominidirectional transiogram with

t = EmpiricalTransiogram(cat, :Z, maxlag = 50.)

funplot(t)
Example block output

Empirical surfaces

GeoStatsFunctions.EmpiricalTransiogramSurface — Type
EmpiricalTransiogramSurface(data, var;
                            normal=Vec(0,0,1), nangs=50,
                            ptol=0.5u"m", dtol=0.5u"m",
                            [options])

Given a normal direction, estimate the transiogram of variable var along all directions in the corresponding plane of variation.

Optionally, specify the tolerance ptol in length units for the plane partition, the tolerance dtol in length units for the direction partition, the number of angles nangs in the plane, and forward the options to the underlying EmpiricalTransiogram.

source
t = EmpiricalTransiogramSurface(cat, :Z, maxlag = 50.)

surfplot(t)
Example block output

Theoretical transiograms

We provide various theoretical transiogram models from the literature, which can can be combined with ellipsoid distances to model geometric anisotropy.

Models

Linear

GeoStatsFunctions.LinearTransiogram — Type
LinearTransiogram(; range, proportions)

A linear transiogram with range in length units, and categorical proportions.

LinearTransiogram(; ranges, rotation, proportions)

Alternatively, use multiple ranges and rotation matrix to construct an anisotropic model.

LinearTransiogram(ball; proportions)

Alternatively, use a custom metric ball.

source
funplot(LinearTransiogram())
Example block output

Gaussian

GeoStatsFunctions.GaussianTransiogram — Type
GaussianTransiogram(; range, proportions)

A Gaussian transiogram with range in length units, and categorical proportions.

GaussianTransiogram(; ranges, rotation, proportions)

Alternatively, use multiple ranges and rotation matrix to construct an anisotropic model.

GaussianTransiogram(ball; proportions)

Alternatively, use a custom metric ball.

source
funplot(GaussianTransiogram())
Example block output

Spherical

GeoStatsFunctions.SphericalTransiogram — Type
SphericalTransiogram(; range, proportions)

A spherical transiogram with range in length units, and categorical proportions.

SphericalTransiogram(; ranges, rotation, proportions)

Alternatively, use multiple ranges and rotation matrix to construct an anisotropic model.

SphericalTransiogram(ball; proportions)

Alternatively, use a custom metric ball.

source
funplot(SphericalTransiogram())
Example block output

Exponential

GeoStatsFunctions.ExponentialTransiogram — Type
ExponentialTransiogram(; range, proportions)

An exponential transiogram with range in length units, and categorical proportions.

ExponentialTransiogram(; ranges, rotation, proportions)

Alternatively, use multiple ranges and rotation matrix to construct an anisotropic model.

ExponentialTransiogram(ball; proportions)

Alternatively, use a custom metric ball.

source
funplot(ExponentialTransiogram())
Example block output

MatrixExponential

GeoStatsFunctions.MatrixExponentialTransiogram — Type
MatrixExponentialTransiogram(rate)

A matrix-exponential transiogram with transition rate matrix.

MatrixExponentialTransiogram(; range, lengths, proportions)

Alternatively, build Euclidean ball with given range and base transition rate matrix from mean lengths and relative proportions.

MatrixExponentialTransiogram(; ranges, rotation, lengths, proportions)

Alternatively, use multiple ranges and rotation matrix to construct an anisotropic model.

MatrixExponentialTransiogram(ball, rate)

Alternatively, use a custom metric ball.

Examples

MatrixExponentialTransiogram(lengths=(3.0, 2.0, 1.0))
MatrixExponentialTransiogram(proportions=(0.5, 0.5))

References

Notes

The final proportions of the matrix-exponential model can be different from the relative proportions specified during construction. They are also a function of the mean lengths.

source
funplot(MatrixExponentialTransiogram())
Example block output

PiecewiseLinear

GeoStatsFunctions.PiecewiseLinearTransiogram — Type
PiecewiseLinearTransiogram(abscissas, ordinates)

A piecewise-linear transiogram model with abscissas and matrix ordinates obtained from an EmpiricalTransiogram.

PiecewiseLinearTransiogram(ball, abscissas, ordinates)

Alternatively, use a custom metric ball.

References

source