น2.4. Parametric forecast error
Following Hartig and Dormann (2013), point estimates of the
various parameters (the medians of the parameters sampled during
the MCMC) were used to compute predicted future population sizes
for up to the next 50 time steps. Concretely, for each parameter point
estimates, 50 time series were computed starting at time step t (with t
in [50; 99]), and lasting during 100t steps. These time series were
using as initial X(t) values, a maximum-likelihood estimator based on
the last five observed population sizes N(t4)…N(t), and using the
model parameter point estimates. The mean square error between
the simulated X(t) values and the reference X(t) values was then
computed, and divided by the standard deviation of the reference
time series to obtain standardized root mean square errors (SRMSE,
see Hartig and Dormann (2013) for details, and Appendix for the R
scripts used). Constant forecasts equal to the population temporal
mean would lead to a SRMSE equal to one. Therefore, forecasts with a
SRMSE larger than one can be considered to have failed. Note that,