Drought forecasts, particularly at seasonal scales, offer great potential for managing climate risk in water resources and agricultural systems. In this context, the importance of assessing the economic value of such forecasts and determining whether a decision-maker should adopt them cannot be overstated. Value-assessment studies often, however, ignore the dynamic aspects of forecast adoption, despite evidence from field-based studies suggesting that farmers’ forecast-adoption behavior fits the general framework of innovation diffusion, i.e. that forecast adoption is a dynamic learning process that takes place over time. In this study, we develop an agent-based model of drought forecast adoption to study the role played by heterogeneous economic and behavioral factors (i.e. risk aversion, wealth, learning rates), forecast characteristics (i.e. accuracy), and the social network structure (i.e. inter- and intra-county ties, change agents, self-reliance) in the process of forecast adoption and diffusion. We consider two learning mechanisms: learning by doing, represented by a reinforcement-learning mechanism, and learning from others, represented by a DeGroot-style opinion-aggregation model. Results show that, when social interactions between agents occur, forecast adoption follows a typical S-shaped diffusion curve. By contrast, when agents rely only on their own experience, the adoption pattern is close to linear. Our numerical experiment shows additionally that forecasts are never adopted if forecast accuracy drops below 65 percent. Finally, the proposed model also provides a flexible tool with which to test the effectiveness of extension targeting strategies in facilitating the diffusion of forecasts.