Ecological communities often contain many types of organism and many types of interactions. This complexity is one reason why forecasting their dynamics with enough skill to be useful may be very difficult. Yet some communities are less complex than others, and so may be more forecastable than others. Despite the idea that the complexity of communities may be the primary reason they are difficult to forecast, we are unaware of any empirical study testing it. Furthermore, communities experience environmental change, and this may affect how forecastable are their dynamics. We performed an experiment in which we manipulated community complexity (species richness) and environmental change (light availability), using experimental ecosystems containing microorganisms in a laboratory setting. During nine-months we recorded population sizes in the communities on three days of every week, resulting in time series with about 120 data points. We then forecasted the dynamics of population sizes using a variety of forecasting methods and quantified forecast error. We found that forecast error was greater in more diverse communities in constant environmental conditions, but that forecast error was lower in more diverse communities in the changing environment. Although there were effects of the treatments on time series autocorrelation, variability, and permutation entropy, there were no interactive effects of the treatments. The results suggest that the complexity of ecological communities is not a fundamental barrier to forecasting their dynamics, at least when an environmental change with strong effects is occurring.