Efficient sampling of parameter space to maximise the number of minima we find while minimising the total number of simulation we need to run is a book in itself. There are many algorithms we can use to sample from the prior that are guided towards good solutions.

The outcome of the classification process is to allow us to sample from both the parameter space of the models (in this case affinity and rotation) to vary the model morphology, while also allowing us to switch interpretation.

Sampling from the prior
The prior is our existing set of beliefs about the reservoir based on the data and knowledge available to us. As we discussed above the prior can be a set of distributions that describe uncertain model parameters and our “true reservoir” exists somewhere in this space – what we call the global minima.

The only way to guarantee hitting all your targets is to fire off a lot of shots. If the cost of calculating the misfit value is low computationally we need not worry too much as we can easily use a Monte Carlo approach to extensively sample the misfit response in parameter space, therefore sampling from the likelihood surface to calculate the posterior probability.

X - величина/параметр, отсчеты/значения которой/которого выбираются
Y - пространстве/множестве, в котором производится взятие отсчетов / выборка значений