In light of global environmental change, forecasting the future state of populations, communities, and ecosystems is critical. Human influences often change multiple stressors simultaneously, which could make forecasting particularly difficult. Currently, however, we lack empirical studies that test if stressors differ and interact in their effects on forecastability. We conducted a long-term experiment with aquatic protists to test if the forecastability of ecological dynamics depends on the number and identity of stressors. Over nine months, we exposed our experimental communities to a factorial manipulation of three stressors (temperature, salt, resources). The stressors increased gradually and then remained constant at their final levels in the last phase of the experiment. We collected data on species abundances three times per week, resulting in 120 timepoints. We then used data-driven forecasting methods (e.g. empirical dynamic modelling) to forecast species abundances and total community biomass. We forecasted different periods of the time series, i.e. time periods sooner or later after the final environmental conditions had been reached. We found that the stressors had interactive effects on forecastability and that the direction of effects depended on the forecasted time period. For example, shortly after reaching the final stressor levels, high salt increased the forecastability of species abundances, in particular when resources were high. In contrast, high salt reduced forecastability in the final phase of the experiment. Likewise, forecastability of community biomass was interactively affected by multiple stressors soon after the final stressor levels had been reached. Specifically, in low-resource environments, high salt increased forecastability when temperature was low but reduced forecastability when temperature was high. Towards the end of the experiment, however, forecastability of community biomass depended primarily on temperature which reduced forecastability. Taken together, our experiment revealed that environmental drivers can differ and interact in their effects on forecastability. Moreover, our results suggest that stressor impacts on forecasting can vary in time, which highlights the need for long-term and fine-grained time series in ecology.