Urban heat island (UHI), a phenomenon involving increased air temperature of a city compared to the surrounding rural area, results in increased energy use and escalated health problems. To understand the magnitude and characteristics of UHI in Seoul and to accommodate for the high temporal variability and spatial heterogeneity of the UHI which make it inherently challenging to analyze using conventional statistical methods, we developed two deep learning models, a temporal UHI-model and a spatial UHI model, using a feed-forward deep neural network (DNN) architecture. Data related to meteorological elements (e.g. air temperature) and urban texture (e.g. surface albedo) were used to train and test the temporal UHI-model and the Spatial UHI-model respectively. Also, we develop and propose a new metric, UHI-hours, that quantifies the total number of hours that UHI exists in a given area. Our results show that UHI-hours is a better indicator of seasonal UHI than the commonly used index, UHI-intensity. Consequently, UHI-hours is likely to provide a better measure of the cumulative effects of UHI over time than UHI-intensity. UHI-hours will help us to better quantify the effect of UHI on, for example, the overall daily productivity of outdoor workers or heat-related mortality rates.