Dengue is currently the fastest-spreading vector-borne disease, with an estimated 100-400 million infections per year and nearly 6 billion people at risk--more than half the world's population. However, there is no specific antiviral treatment, thus the primary method for preventing infection is through targeted control of the principal mosquito vector, Aedes aegypti. The objective of this retrospective study is to evaluate the efficacy of forecasting dengue risk using human cases and an adaptation of DYCAST: a biologically based, spatiotemporal clustering model that proved operationally successful as a West Nile virus early warning system. Results demonstrate that DYCAST provided accurate and early detection of high-risk areas for dengue in Ribeirao Preto, Brazil, including a cryptic interepidemic focus of transmission preceding a severe epidemic. Using suspected cases, DYCAST predicted 90.3% (4,237/4,690) of confirmed cases, with a mean lead time of 66.3 days prior to illness onset. High-risk cells had >10x higher relative risk of containing a confirmed case, and sensitivity and specificity were 83.8% and 72.3%. These findings suggest that DYCAST could be utilized in real time to prospectively identify areas at high risk of dengue, for rapid and precise targeting of mosquito mitigation and public education. Implementation using free, open-source software offers a scalable technological strategy for the prevention and control of dengue around the world.