During deep-well casting, thermal deformation of the crystallizer induces molten aluminum leakage, potentially causing steam explosions from contact with cooling water. To investigate this process, a multi-physics numerical model coupling radiative and convective heat transfer is developed for the analysis of temperature anomalies in the crystallizer region. The temperature field distribution in the surrounding air domain is characterized, representative leakage scenarios are simulated, and the radial, axial, and circumferential attenuation behaviors related to leakage morphology are quantified. A mapping relationship is established to determine sensitive monitoring zones and gradient thresholds, providing a basis for early warning and structural optimization. Results demonstrate that the radial temperature field follows an exponential decay pattern, exhibiting a “rapid de-cay-stable transition” mode. In sensitive regions, the boundary temperature is observed to decrease axially at a rate of 5.39 K/mm. The temper-ature distribution is found to be jointly influenced by leakage morphology, thermal gradient, and leakage area, and the radial average cooling rate of linear heat sources is calculated to be 16.93-19.35 K/mm. Although gradient walls effectively mitigate localized heat accumulation, expansion of the leakage flow area in-creases the tempera-ture on the 10 mm monitoring surface by 30.16 K and extend the high-temperature impact zone. These quantified thermal response characteristics provide theoretical and numerical foundations for designing real-time monitoring systems.