Doubt That Data Dissolves

A model trained to predict house prices shows high predictive variance in a neighborhood with very few sales, but that variance shrinks a lot once more sales data is collected. This is characteristic of which kind of uncertainty, and what practical tool is often used to estimate it at test time using a standard trained network?

Aleatoric uncertainty, which is irreducible no matter how much data is added

Epistemic uncertainty, estimated via MC dropout (stochastic forward passes at inference)

Aleatoric uncertainty, estimated via MC dropout (stochastic forward passes at inference)

Epistemic uncertainty, estimated by increasing the batch size at inference