It can be challenging to differentiate community changes caused by human activities from the influence of natural background variability. Using gradient forest analysis, we explored the relative importance of environmental factors, operating across multiple spatio-temporal scales, in influencing patterns of compositional turnover in estuarine benthic macroinvertebrate communities across New Zealand. Both land-derived stressors (represented by sediment mud content and total sediment nitrogen and phosphorus content) and natural environmental variables (represented by sea surface temperature, Southern Oscillation Index, and wind-wave exposure) were important predictors of compositional turnover, reflecting a matrix of processes interacting across space and time. Generalized linear models were used to determine whether measures of benthic macroinvertebrate diversity, which are commonly used as indicators of ecological health on a local scale, changed in a way that was consistent with the compositional turnover along the environmental gradient. As expected, compositional turnover along land-derived stressor gradients was negatively associated with diversity indices, suggesting a decline in ecological health as land-derived stressors increase. This study moves towards an ecosystem-based management approach that focusses on cumulative effects rather than single stressors by considering how multiple land-derived stressors influence indicators of estuarine health, against a background of natural variability across several spatio-temporal scales.