This study provides some new insights into earthquake forecasting models that are applied to regions with subduction systems, including the depth component for forecasting grids and time-dependent factors. To demonstrate the importance of depth component, a forecasting approach, which incorporates three-dimensional grids, is compared with an approach with two-dimensional cells. Through application to the two subduction regions, Ryukyu and Kanto, it is shown that the approaches with three-dimensional grids always demonstrate a better forecasting ability. I thus confirm the importance of depth dependency for forecasting, especially for applications to a subduction environment or a region with non-vertical seismogenic structures. In addition, this study discusses the role of time-dependent factors for forecasting models and concludes that time dependency only becomes crucial during the period with significant seismicity rate change that follows a large earthquake.