Centered Log-Ratio Transformation¶
The Centered Log-Ratio (CLR) Transformation is a normalization technique used for compositional data, where values represent relative proportions of a whole rather than absolute quantities. It helps remove the closure effect, a mathematical constraint where components of a dataset sum to a fixed total (e.g., 100%), which can distort statistical analysis. The transformation converts compositional data into an unconstrained space, making it suitable for methods that assume independent variables.
As an example, in geochemistry, elemental concentrations in rocks are often expressed as proportions (e.g., major oxides summing to 100%). Directly applying statistical models to such data can lead to misleading results due to spurious correlations. The CLR transformation ensures that geochemical variables are properly normalized, allowing for more accurate pattern recognition, clustering, and geostatistical modeling. It is particularly useful for identifying geochemical trends in exploration, such as distinguishing between hydrothermal alteration zones and background compositions.
Warning
The input data must be in the same unit (e.g., percent) to ensure meaningful results. Additionally, if all points contain one no-data value, CLR cannot be performed. The method requires at least 2 properties.
Interface¶
The general parameters controlling the Centered Log-Ratio (CLR) Transformation are shown in the figure below.
The options are described as follows:
Object Selection¶
Client: The object containing the dataset to be transformed using the CLR method.
Data: The geochemical or compositional data to be transformed. Users can select multiple Float data properties to apply the transformation.
Property Group (optional): A selection of grouped compositional data properties for batch transformation.
Inverse Transformation¶
Invert: Enables the inverse CLR transformation, converting CLR-transformed data back into its original compositional space. This is useful when reverting processed data to its initial scale while maintaining consistency with the CLR-transformed dataset. However, the inverted values is rescaled so that their sum equals 1, preserving the compositional nature of the data but losing the original scale.
Output Names¶
Output Name (optional): The name of the output property group containing the CLR-transformed data. By default, it is set to Client’s name + “_CLR”. Users can modify this to specify a custom output name.
Once the parameters are selected, press OK to run the transformation.
Tutorial¶
The following video presents a tutorial on how to use the CLR application.
Open the application.
Select the object containing the data.
Select the data to be analyzed or a property group containing the data.
(optional) Choose to do the inverse transformation.
(optional) Choose the name of the output results.
Run the application.
Inspect the transformed data.
