A probabilistic context-free grammar is a generative model, which can be used to generate random words in the language of the underlying context-free grammar. In order to do this, for each nonterminal $A$ we give a probability distribution on the set of rules of the form $A\to \alpha$. To generate a sentence, you start with the starting symbol $S$. As long as the current sentential form contains some nonterminal, say $A$, you apply a random derivation rule $A\to \alpha$ according to the probability distribution. If the probabilities are chosen correctly, then eventually a terminal word will be generated (almost surely).
On the left of your image, you can see the probability distributions. You can estimate the probabilities from a collection of parse trees, but you can also come up with them in any other way. In particular, it doesn't make too much sense to determine them from a single parse tree, like your description seems to imply.
On the right of your image, you can see the calculation that the given parse tree be generated by the PCFG on the left. This probability is the product of all derivation rules that need to be applied to obtain the parse tree (the order of application is immaterial since it doesn't affect the product).
Your underlying difficulty seems to be that you don't understand what PCFGs are for and how they are used. Hopefully, my description above helps to clarify this somewhat.
