Abstract Tropical lakes harbour high levels of biodiversity, but the temporal and spatial variability of biological communities are still inadequately characterised, making it difficult to predict the impact of accelerated rates of environmental change in these regions. Our goal was to identify the spatiotemporal dynamics of the planktic diatom community in the Cajas Massif in the tropical Andes. We analysed seasonal diatom and environmental data over a period of 1 year from 10 lakes located in geologically distinct basins and modelled community–environment relationships using multivariate ordination and variation partitioning techniques. Generalised additive models with a full‐subset information theoretic approach also were used to determine which environmental variables explain single‐species abundance. Although the lakes are monomictic and thus have variable thermal structure across the year, seasonal variability of water chemistry conditions was negligible, and seasonal differences in diatom community composition were small. Across space, diatom community composition was correlated primarily with ionic content (divalent cations and alkalinity), related to bedrock composition, and secondly with lake thermal structure and productivity. The ionic gradient overrode the effect of the thermal structure–productivity gradient at the diatom community level, whereas individual diatom species responded more sensitively to variables related to in‐lake and catchment productivity, including chlorophyll‐a and iron, and the proportion of wetlands in the catchment. Our results indicate that the spatiotemporal variability of Cajas lakes and their diatom communities is the result of multiple intertwined environmental factors. The emergence of the ionic and thermal structure–productivity gradients in a rather small tropical lake district suggests segregation of ecological niches for diatoms that also may be important in other high‐elevation lake regions. Future studies that track tropical Andean lakes under natural and anthropogenically mediated change, both in contemporary times and in palaeoenvironmental reconstructions, would benefit from the modelling approach (community and species levels) developed here.