How can I identify outliers using a Kernel Density Estimation for multivariate data??

I have this time-series dataset of freshwater chemicals. I would like to perform KDE on the PCA score matrix (using fex. the first 2 scores) and then find a "cutoff density threshold" in order to detect outliers as samples that lay in the space with a certain (low) density. I am using the kde2d function of Dr Z. Botev.