Calibration

guide

Calibration means using observations to learn which parameter values make a model reproduce the patterns you care about. In plant, this usually involves three separate choices:

1. What to compare: for example, individual growth, stand biomass, size distributions, or recruitment.
2. Which parameters to estimate: restrict this set to parameters that the data can plausibly inform.
3. How to measure agreement: define an observation model or loss function that accounts for the scale and uncertainty of the data.

These choices come before the optimisation or sampling algorithm. A parameter set that matches one summary, such as total biomass, may still give the wrong size distribution or temporal dynamics.

ImportantCurrent package boundary

plant runs individuals and patches for specified parameter values. Fitness, demographic-equilibrium, and evolutionary-assembly tools have moved to the regnans package. Use regnans when your calibration target depends on invasion fitness or resident equilibrium; those functions are no longer part of plant.

A practical calibration workflow

Before fitting parameters, build a forward simulation that is easy to rerun:

1. create one function that receives a named parameter vector;
2. construct the plant strategy and environment inside that function;
3. run the model and return only the summaries observed in your data;
4. test the function at several plausible parameter values;
5. only then connect it to an optimiser or Bayesian inference method.

Keep the calibration data, units, model version, environmental drivers, and random seeds together. These details determine what the fitted values mean and whether the analysis can be repeated.

The Example analysis shows how to turn patch output into quantities that could be compared with observations.