Swimming organisms can enhance mixing in their natural environments by creating eddies in their wake and by dragging water along. However, these mixing mechanisms are inefficient for microorganisms, because swimming‐induced variations in velocity, temperature, and dissolved substances are evened out before they can be advected. In bioconvection, however, microorganisms induce water movement not by propulsion directly but by locally changing the fluid density, which drives convection. Observations of bioconvection have so far mainly been limited to laboratory settings. We report the first observation and quantification of bioconvection within a stratified natural water body. Using in situ measurements, laboratory experiments, and numerical simulations, we demonstrate that the bacterium Chromatium okenii is capable of mixing 0.3 to 1.2 m thick water layers at around 12 m water depth in the Alpine Lake Cadagno (Switzerland). As many species are capable of driving bioconvection, this phenomenon potentially plays a role in species distributions and influences large‐scale phenomena like algal blooms.

Three-dimensional direct numerical simulations are performed that give us an in-depth account of the evolution and structure of the double-diffusive interface. We examine the diffusive convection regime, which, in the oceanographically relevant case, consists of relatively cold fresh water above warm salty water. A 'double-boundary-layer' structure is found in all of the simulations, in which the temperature (T) interface has a greater thickness than the salinity (S) interface. Therefore, thin gravitationally unstable boundary layers are maintained at the edges of the diffusive interface. The TS-interface thickness ratio is found to scale with the diffusivity ratio in a consistent manner once the shear across the boundary layers is accounted for. The turbulence present in the mixed layers is not able to penetrate the stable stratification of the interface core, and the TS-fluxes through the core are given by their molecular diffusion values. Interface growth in time is found to be determined by molecular diffusion of the S-interface, in agreement with a previous theory. The stability of the boundary layers is also considered, where we find boundary layer Rayleigh numbers that are an order of magnitude lower than previously assumed.

The diffusive regime of double‐diffusive convection generates staircases consisting of thin high‐gradient interfaces sandwiched between convectively mixed layers. Simultaneous microstructure measurements of both temperature and conductivity from the staircases in Lake Kivu are used to test flux laws and theoretical models for double diffusion. Density ratios in Lake Kivu are between one and ten and mixed layer thicknesses on average 0.7 m. The larger interface thickness of temperature (average 9 cm) compared to dissolved substances (6 cm) confirms the boundary‐layer structure of the interface. Our observations suggest that the boundary‐layer break‐off cannot be characterized by a single critical boundary‐layer Rayleigh number, but occurs within a range of O(102) to O(104). Heat flux parameterizations which assume that the Nusselt number follows a power law increase with the Rayleigh number Ra are tested for their exponent η. In contrast to the standard estimate η = 1/3, we found η = 0.20 ± 0.03 for density ratios between two and six. Therefore, we suggest a correction of heat flux estimates which are based on η = 1/3. The magnitude of the correction depends on Ra in the system of interest. For Lake Kivu (average heat flux 0.10 W m−2) with Ra = O(108), corrections are marginal. In the Arctic Ocean with Ra = O(108) to O(1012), however, heat fluxes can be overestimated by a factor of four.