Infectious disease modeling has emerged as a powerful data driven tool for monitoring outbreaks, assessing intervention strategies, and allocating public health resources. This thesis addresses a variety of challenges faced in real-world infectious disease forecasting. We first present methods for aggregating forecasts made at different spatial scales, where explicitly modeling the spatial dependency would be computationally prohibitive. We then extend the mechanistic model framework to create an operational forecasting model capable of handling real-world COVID-19 surveillance system issues. Finally, we propose a new framework for merging mechanistic and statistical approaches to infectious disease forecasting. This framework allows modelers to construct “semi-mechanistic” models that draw from the strengths of both mechanistic and statistical paradigms. In an application setting of forecasting COVID-19 cases and deaths, we demonstrate that a semi-mechanistic model outperforms both a fully mechanistic model and a fully statistical model.