Capturing trends - rules of thumb
When you are modeling (continuous) properties you might want to capture a known trend or correlate your property with a secondary property. Within the Rock Properties workflow, there are two workflow steps where you can do so. Each option has its strengths and weaknesses. The first option is the Transform and Trends from, the second option is the Conditioning form. It depends on your input data and the type of trend you want to adhere to, which modeling step you have to use. Depending on the selected modeling method, the output distribution and mean is significantly impacted.
How do you capture a trend, and how do you correlate your data with a secondary property? These are the rules of thumb when working with JewelSuite:
- When there is a trend in your input data, e.g. you can see a decrease in porosity with depth in your well log data, you use the Transform & Trends form. The purpose of the Transform and Trends form is to apply data transformation(s) to an upscaled grid property. When you want to capture a trend that is already present in your data, you can use the 1D, 2D or 3D trend transformations. The selected trend is used to de-trend the upscaled property to make the data stationary. You can at this moment adjust the selected trend line. After the simulation is run (in the Run Model step), the trend is added back.
- When you know there is a trend based on additional information, you can use a trend map or trend property in combination with Conditioning to incorporate this trend into your property model. With the Conditioning form you can enforce the modeling algorithm to use a secondary property to influence the values simulated during the modeling run, while at the same time honoring the (measured) input data. Applying a trend using the Conditioning form requires the input data and trend to have the same data range and unit. When both for input and trend data a normal score is executed, the secondary data can be of a different unit and/or data range.
- A combination of the two methods.
When you choose to use a secondary property to steer your primary property, on the Conditioning form you have to select the '3D' option, and after selecting a property select an integration method: Locally Varying Mean (LVM) or CoKriging. In case you want to specifically use a 1D or 2D trend, select the Vertical & lateral option and then either the vertical or lateral option.
Locally Varying Mean - You use the selected trend as the Locally Varying Mean to guide your primary property when there is no correlation between the primary and the secondary data. There is no manual interference.
CoKriging - If you want to have some (manual) control over the secondary property, use CoKriging to as a means to use secondary data to guide your primary data. This method requires a correlation factor which will be involved in the kriging estimation.
Whatever method you choose to use, it is essential that you know how to handle your primary data and secondary data properly. When you have unexpected results, it is best to check your settings and parameters. When you use the Transform and Trends form to steer your data, you can also manually interact with the data. When you use the Conditioning form, the modeling algorithm determines most of the settings. There is (little) room for manual adjustments when you select CoKriging as an integration method.