sensitivity_analysis: Parameter sensitivity analysis

In ashiklom/hector_permafrost_emit: Sensitivity of Hector model to permafrost emissions

Description

Perform a PEcAn-like parameter sensitivity analysis to analyze the contribution of parameters to predictive uncertainty.

Usage

sensitivity_analysis(dat, xcols, ycol, .type = "additive")

Arguments

xcols	Character vector of parameter column names
ycol	Name of column containing response variable
.type	Whether the multivariate fit is "additive" (Y ~ a + b + c, default) or "interactive" (Y ~ a * b * c). This argument supports partial matching via base::match.arg().
df	data.frame of Hector results.

Details

This analysis produces three key metrics:

• "Coefficient of variation (CV)" describes the relative uncertainty of the input parameter. It is calculated as the ratio between the input parameter variance and its median value.

• "Elasticity" is the normalized sensitivity of the model to a change in one parameter.

• "Partial variance" is the fraction of variance in the model output that is explained by the given parameter. In essence, it integrates the information provided by the CV and elasticity.

The theory and implementation are based on the sensitivity analysis described by LeBauer et al. (2013), but with several key differences:

• LeBauer et al. use a cubic spline interpolation through each point in the model output. This function uses a generalized additive model regression (mgcv::gam()).

• LeBauer et al. fit individual splines to each parameter-output combination where other parameters are held constant at their median. This function fits a multivariate generalized additive regression model, and then uses that fit to calculate the partial derivatives.

Value

data.frame of sensitivity analysis results. See Details.

Author(s)

Alexey Shiklomanov

ashiklom/hector_permafrost_emit documentation built on March 26, 2020, 12:15 a.m.