Accurate forecasts of ecological dynamics are critical for ecosystem management and conservation, yet the drivers of forecastability are poorly understood. Environmental change and diversity are considered major challenges to ecological forecasting. This assumption, however, has never been tested experimentally because forecasts have high data requirements. In a long-term microcosm experiment, we manipulated species richness of 30 experimental protist communities and exposed them to constant or gradually decreasing light levels. We collected finely-resolved time series (123 sampling dates over 41 weeks) of species abundances, community biomass, and oxygen concentrations. We then employed data-driven forecasting methods to forecast these variables. We found that species richness and light had a weak interactive effect on forecasts of species abundances: richness tended to reduce forecast accuracy in constant light but tended to increase forecast accuracy in declining light. These effects could partially be explained by differences among time series in variability and autocorrelation. Forecasts of aggregate properties (community biomass, oxygen), however, were unaffected by richness and light, and were not more accurate than those of species abundances. Our forecasts were based on time series that were detrended and standardized. Since real-world forecasting applications require predictions at the original scale of the forecasted variable, it is important to note that the results were qualitatively identical when back-transforming the forecasts to the original scale. Taken together, we found no strong evidence that higher diversity results in lower forecastability. Rather, our results imply that promoting diversity could make populations more predictable when environmental conditions change. From a conservation and management perspective, our findings suggest that diversity conservation might have beneficial effects on decision-taking by increasing the forecastability of species abundances in changing environments.