Introduction

As mentioned above, the aim of SA is to determine the impact changes in model inputs, such as parameters, forcing functions, and boundary or initial conditions,

Table 2 Applicability of different outcomes of SA and UA to ecological models with different purposes

SA / UA outcome

Prediction

Forecasting

Understanding

Decision-making

Model validity

V

V

V

V

Model input sensitivity

(V)a

(V)a

V

(V)a

Confidence limits on model predictions

V

V

X

V

Risk-based performance measures

X

X

X

V

aPotentially indirectly applicable - see accompanying text.

aPotentially indirectly applicable - see accompanying text.

have on model outputs. In order for SA to be useful in an ecological modeling context, it should be able to assess the impact parameters have over their plausible case-study-specific range, cater for model nonlinearities, and take into account interactions between parameters. Other considerations include computational efficiency and the usefulness of the information provided.

It is particularly important to identify the purpose of the SA (e.g., why an SA is being conducted, what information is ultimately required). For example, an SA to help guide process-driven model formulation should be different from an SA utilized to guide what model inputs are most critical to accurately specify when a particular model is applied to a specific case study.

Traditional SA methods can be divided into two categories: derivative based and sampling based (Figure 2). Derivative-based SA methods are not well suited to ecological models, as they generally only explore the impact of model inputs on output(s) in the vicinity of model input base values, rather than over their entire plausible range, are linear and are problematic when used to identify higher-order interactions between parameters. For example, although numerically evaluating higher-order partial derivatives is at least theoretically possible across the entire plausible range of multidimensional parameter space to better evaluate parameter interactions, this approach quickly becomes computationally intractable due to the curse of dimensionality. Although sampling-based SA methods are more computationally expensive than common derivative-based methods, they cater for nonlinearities, higher-order interactions, and consider the sensitivity of model output(s) to inputs over their full range. In addition, the information provided by sampling-based methods is more comprehensive.

Solar Power

Solar Power

Start Saving On Your Electricity Bills Using The Power of the Sun And Other Natural Resources!

Get My Free Ebook


Post a comment