At the moment much of my work is on model simulations to disentangle the effects of habitat heterogeneity (patchiness) in river networks and species-specific dispersal abilities and its implication on the colonization speed of fish species. Therefore I came across two interesting studies discussing the important role of sensitivity analysis in ecological/biological modelling:
Very interestingly, Marino et al. (2008) describe the Latin Hypercube Sampling (LHS) Approach, a stratified sampling approach (without replacement) with the great advantage of reduced sample sizes to yield similar high result accuracy as simple random sampling with large sample sizes. Given this convincing argument I am also trying this approach for fish dispersal modelling in river networks.
Furthermore, in his blog, Mathieu Fenniak provides a nice-to-read introduction to LHS, what is the sampling scheme behind and provide a python example to clearly show the advantage of reduced sample sizes. Moreover, he concludes that Latin hypercube sampling is capable of reducing the number of runs necessary to stablize a Monte Carlo simulation by a huge factor of up to 30% fewer calculations to create a smooth distribution of outputs.