Sensitivity Analysis

Sensitivity analysis involves varying input values within their valid ranges to see how the model's output changes. It can be regarded of as a black box that processes a set of inputs and calculates one or more quantities of interest (outputs) (Borgonovo et al., 2022). 

One of the major difficulties in agent-based modeling is determining the values of system parameters, which controls the agent behaviors and interactions. Estimated parameter values are often used for simulation. In such cases, sensitivity analysis is mandatory; namely, we need to perform simulations with various parameter settings to confirm the robustness of the conclusion that was obtained based on the estimated parameter values. Moreover, sensitivity analysis could provide insights into the modeled system as well as identify parameters that are critical for the system dynamics. (Niida et al., 2019)

• Local Sensitivity Analysis: One input at a time. Test how sensitive the model is to the value of each individual parameter. It helps understand the individual effect of each parameter on the model's output.
• Global Sensitivity Analysis: Several or all inputs are varied at the same time. Test how sensitive the model is when varying all parameters at the same time. This provides a more comprehensive view of how the model responds to input variations.
• OAT : One at a time. Input one factor and experiment output respon to that. Examole : vary maximum value at spesific ticks.
• OFAT: Use broader of possible setting in input. Keep all other input constant and record model output.
• Uncertainty Analysis: Looks at how uncertainty in parameter values affects the reliability of model results.
• Robustness Analysis: Explores the robustness of results and conclusions of a model to changes in its structure.

Studying an epidemic model: Varying the initial number of infected individuals to see how it affects the final outbreak size.

change number of agent surviving from sugarcane

Parameterisation and Calibration before simulation

Parametirasion : setting in model rules and equations. Quantitative and qualitattive remains constant through model simulation. Define it before simulation.

Direct parameterisation: literature, fieldwork,experiments, experts

Indirect parameterisation/calibration: 2nd base choise if you dont know your parameter value at all, narrow down to certain range, and find best value. define at least one model output as target run simulation varying unknown parameter. Analyze tun to identify parameter setting with desired output.

Callibration : to know which setting to run your model. Worfkflow: define at least one model output as target, run simulations varying unknown parameter, analyse runs to identify parameter setting with desired output. Example : number of plots within on district, costs enlarging house, desired output:  population growth in outer districts but still plots available at end.