We propose a hybrid physics-AI model incorporating urban surface energy processes and physical priors for improved local urban climate emulation.