When you have lots of various explanations variables in your analysis that would corroborate one with another , the method would allow you to identify the components of dataset that maximizes variance , making standard lineal models that are potentially more powerful.
For an analysis of water quality then, you might have some overall measure of ecological health and various environmental variables that you’d expect to be correlated.
Variables such as urbanization, salinity, levels of contaminants in the water, etc, might be correlated. instead of doing a multiple regression model with all these potentially interconnected variables, you could do a PCA of your explanatory variables.