A stock of phytoplankton (approximately 90% of the phytoplankton was Microcystis aeruginosa by microscopic observation) was collected from the Moselle River in southwest Germany on 9 August 2022 during a heavy Microcystis bloom. Colonies were collected from the water surface using a silk plankton net with a 40 μm mesh size. To select predominantly Microcystis colonies, the samples were first filtered through a 500 μm sieve to remove large particles and then through a 40 μm sieve. The filtered Microcystis colonies with sizes between 40 and 500 μm were stored and cultured at 20 ± 1°C and photosynthetically active radiation (PAR) of 15 μmol photons s⁻¹ m⁻².

Time-resolved observations of Microcystis colonies were obtained from video observations of the water surface using a down-looking digital camera (for the observation of surface scum and velocities of Microcystis) and of the bulk water (at depths of 0 – 4 cm, 12 – 16 cm and 26 – 30 cm) using side-looking cameras (for the observation of colony size and velocities of Microcystis). Videos of 90 s duration were hourly recorded. During the recordings at the water surface, illumination was provided by an LED spotlight, while during the recordings of the side-looking cameras, illumination in the bulk water was provided by a vertically oriented laser light sheet, which illuminated a central plane of the flow channel using a continuous-wave line laser (InLine HP, MediaLas, Germany; output power: 500 mW, green, 532 nm). Colonies within the approximately 3 mm thick light sheet were observed using three digital cameras in a perpendicular arrangement ( Supplementary Figure S1 ). The resolution and field of view of the videos recorded by the cameras used in the experiments are shown in Supplementary Table S2 .

Individual Microcystis colonies were detected in each image based on intensity thresholds during automated image processing. The analysis provided surface areal coverage of Microcystis (SAC, calculated as the percentage of water surface that was covered by colonies) for down-looking cameras, the equivalent spherical diameter of colonies for side-looking cameras, as well as colony velocities estimated using particle tracking velocimetry (Kelley and Ouellette, 2011). We estimated and summed up the area covered by the colonies for each frame, which was then divided by the camera’s field of view to calculate the SAC. The change rate of surface areal coverage (dSAC/dt) during low wind periods, representing the reformation rate of surface scum, was estimated by the difference of mean SAC between subsequent recordings. The volumetric median colony diameter (D_{v 50}, i.e., the colony size corresponding to the 50^th percentile of observed colony volumes) was used to characterize the average colony size as described in Wu et al. (2019) and Wu et al. (2020). More than 300 colonies on each of the videos were used to estimate D_{v 50}. The rate of change of colony size (dD_{v 50}/dt) was used to characterize the aggregation or disaggregation of colonies. Since surface scum formed during low wind periods and was dispersed during wind disturbances, the D_{v 50} near the water surface at the beginning of a wind disturbance, when the surface scum layer was dispersed, was considered as a representative colony size at the water surface. To compare the colony size dynamics at the water surface and in the bulk water, dD_{v 50}/dt in the bulk water was estimated as the difference of mean D_{v 50} near the water surface during wind disturbances between subsequent recordings. The dD_{v 50}/dt at the water surface was estimated by the ratio of the difference of mean D_{v 50} near the water surface between the beginning of the wind disturbance and the end of the preceding wind disturbance to the duration of the wind disturbance.

The mean velocities of colonies at the water surface and in the bulk water were estimated by observing the displacement of identified colonies via particle tracking velocimetry, using the predictive tracking algorithm described by Kelley and Ouellette (2011).

We did not observe a consistent size-dependence of colony velocities ( Supplementary Figure S2 ) and therefore considered the observed colony velocities as a proxy for flow velocity.

We use root-mean-square velocity fluctuations (U_rms ) as the measure of the intensity of turbulence, which scales with the square root of turbulence kinetic energy, and is calculated as:

where the fluctuating velocity components of the horizontal (u’) and vertical (v’) flow velocities are the difference between the instantaneous and the temporarily averaged velocities (Reynolds decomposition: u ' = u − u ¯ and v ' = v − v ¯ ). The temporarily averaged velocity was calculated by averaging the velocities of all colonies in each video.

Shapiro-Wilk test was used to assess the normality of data. A post hoc LSD (least significant difference) test with one-way ANOVA was used to compare the surface areal coverage (SAC) and the flow velocity during low wind periods at different depths at both stages among the different flumes if the data were normally distributed with homogeneous variance, otherwise, Kruskal-Wallis tests were used. The differences in D_v50 of colonies among different flumes as well as the difference between the rate of colony size change (dD_v50 /dt) at the water surface and in the bulk water were compared using the student t-test. The relationships between mean SAC and D_v50 , mean SAC and dD_v50 /dt, normalized flow velocity and SAC as well as between rate of change in SAC and time were fitted by linear regressions. All statistical analyses were performed using the software package SPSS 27.0 (IBM Corp, USA). Data are presented as mean ± standard deviation and were tested for statistical significance at a significance level (p) of 0.05 unless stated otherwise.

Surface scum layers developed during low wind periods and disappeared during wind disturbance over the seven-day experimental period ( Figure 2 ). During wind disturbances (1.5 – 4.5 m s^-1), the areal coverage of the water surface with Microcystis colonies was generally < 1%, whereas it increased during low wind period (0.5 m s^-1) gradually from 0.1 to 25% from Day 1 to Day 7. The observed rates of change in SAC (dSAC/dt) and rates of change in SAC per unit biomass (dSAC/(OD₆₈₀ dt)) during low wind speed linearly increased over time (p < 0.05, Figure 3 ), suggesting that the surface scum recovered from wind disturbances at increasing rates. Similar to surface coverage, the biomass of Microcystis colonies in the thin surface layer showed a 2 - 6 fold increase from Day 1 to Day 7 ( Supplementary Figure S3A ).

The formation of surface scum followed a diurnal pattern with the average surface coverage during low wind periods in the afternoon (stage II) being reduced by 87.7%, 36.6%, 31.3%, 44.5% and 51.1% (for F1 – F5) in comparison to that in the morning (stage I) ( Figure 2 ).

The dynamics of surface scum formation varied in dependence on wind conditions ( Figure 2 ). The prolonged duration of low wind periods promoted scum formation: the surface coverage in flumes F1 and F4 was significantly higher than that in F2 and F5 in most cases (p < 0.05). Wind magnitudes of 4.5 and 1.5 m s^-1 promoted the subsequent surface scum formation, in comparison to the intermediate magnitude of 3.8 m s^-1: In most cases, the surface areal coverage in F1 and F4 was higher than that in F3 (p < 0.05).

The colony size showed temporal trends during the 7-day experiment ( Figure 4 ). In the upper layer near the water surface, Dv ₅₀ increased from the beginning to Day 7 by a factor of 2.4, 3.1, 2.4, 4.4 and 2.7 in F1- F5, respectively. During the increase of Dv ₅₀, we observed the peaks in Dv ₅₀ of Microcystis colonies near the water surface at times when the wind speed changed from low to high from Day 3 – Day 7 ( Figure 4 ). Through visual observations, we found that such larger colonies suddenly appearing in the bulk water resulted from wind-induced entrainment of surface scum patches forming during preceding low wind periods.

The mean rate of change in colony size (dDv ₅₀/dt) at the water surface and in the bulk water is -0.12 and 0.14 mm h^-1, respectively. In each flume, the dDv ₅₀/dt at the water surface is significantly higher than that in the bulk water (p <0.05, Figure 5 ). At the beginning of wind disturbances, both Dv ₅₀ and dDv ₅₀/dt increased linearly with increasing SAC during the preceding low wind period ( Figure 6 ).

The median diameter of Microcystis colonies (Dv ₅₀) was affected by the different wind conditions ( Figure 4 ). Dv ₅₀ increased with decreasing magnitude of wind disturbances regardless of their duration ( Figure 4 ). The response of Dv ₅₀ to the duration of the wind disturbance was dependent on its magnitude, with longer duration of high-magnitude wind disturbances leading to decreased Dv ₅₀ (F1 < F2, Figure 4 ), whereas Dv ₅₀ increased for longer durations of low-magnitude wind disturbances (F4 > F5, Figure 4 ). The order of mean Dv ₅₀ of Microcystis colonies near the water surface was F4>F5>F3>F2>F1 when combining all measurements (p < 0.05).

At the beginning of the experiment, the surface flow velocities increased nearly linearly with the applied wind speed resulting in equivalent wind speeds in the atmospheric boundary layer (U ₁₀) of 2.0 m s^-1 under low wind speed conditions and 2.8 to 5.3 m s^-1 during the simulated disturbances ( Table 1 ).

The surface flow velocity and underwater flow velocities (at three depths) observed at a particular wind speed generally decreased from Day 1 to Day 7 ( Supplementary Figure S4 ). The normalized flow velocity (defined as the ratio of flow velocity to wind speed) at the water surface and in the bulk water generally decreased linearly over time (p < 0.05, Supplementary Figure S5 , Supplementary Table S3 ). In all flumes, the surface flow velocity and underwater flow velocities during low wind periods decreased to approximately 1 mm s^-1 on Day 7 ( Figure 7 ).

At the applied wind speed of 0.5 m s^-1 (low wind periods), significant differences in surface flow velocity among different flumes were observed (p <0.05, Figure 7 ). No consistent dependence of surface flow velocity on the duration of the intermittent wind disturbance was observed, while surface flow velocity during low wind periods generally increased with increasing preceding magnitudes of wind disturbance: The order of surface flow velocity during low wind periods was F1>F3>F4, except for Day 6 stage I and Day 7. Unlike the surface flow velocity, the underwater flow velocities during low wind periods did not differ significantly between different flumes in most cases ( Supplementary Figure S6 ).

Higher SAC was primarily observed at low surface flow velocities (< 5 mm s^-1). However, it is noteworthy that SAC at moderate surface flow velocities (5-12 mm s^-1) on Day 3 – Day 6 could also surpass some SAC at low surface flow velocities ( Supplementary Figure S7 ). Neither Dv ₅₀ of colonies nor (dDv ₅/dt) were linearly related to the root-mean-square velocities of colonies in the bulk water (U_rms , calculated using Equation 1) when combining all measurements ( Supplementary Figure S8 ). However, the proportion of dDv ₅₀/dt being smaller than -1 mm h^-1 (indicating colony disaggregation) increased with increasing U_rms . Conversely, the proportion of dDv ₅₀/dt larger than 0.5 mm h^-1 (colony aggregation) decreased with increasing U_rms when U_rms exceeded 3 mm s^-1and increased with increasing U_rms when U_rms was smaller than 3 mm s^-1 ( Supplementary Figure S9 ).

The presence of Microcystis affected the hydrodynamic parameters: In the flumes experiment, from Day 1 to Day 4, no linear relationships between the normalized flow velocity and SAC were found, but from Day 5 to Day 7, the normalized flow velocity decreased with increasing SAC (p < 0.05, Figure 8 ). The results of our supplementary dilution experiment showed that the surface tension decreased with the increasing optical density of Microcystis ( Figure 9 ).

Our experiments successfully reproduced the periodical formation and dispersion of Microcystis surface scum under intermittent wind disturbances ( Figure 2 ). During low wind periods (0.5 m s^-1), we observed gradually increased surface scum in all the flumes. Surprisingly, we found that scum was forming more rapidly following prolonged duration of high-magnitude wind disturbances compared to that following moderate magnitudes (F1, Figure 2 ). The differences in scum coverage in varying flumes can be attributed to the flow conditions, as they govern the migration of Microcystis. However, we found significant differences in flow velocities among flumes were only evident for the surface ( Supplementary Figure S6 ; Figure 7 ), suggesting that the surface flow field likely played a pivotal role in modulating the response of scum formation to preceding wind disturbances.

Higher scum coverage can be observed at moderate surface flow velocities (5 - 12 mm s^-1) compared to lower surface flow velocities in some cases ( Supplementary Figure S7 ). This result implies potential mechanisms for surface scum formation under moderate flow velocities, considering that surface scum is typically associated with lower flow conditions (Wu et al., 2015). Webster and Hutchinson (1994) suggested the trapping mechanisms of colonies by viscous sublayer. A possible explanation we proposed is the strong vertical velocity gradient within the viscous sublayer, where the higher flow velocities near the water surface result in lower static pressure and a net upward lift force on the colonies, of which the magnitudes depend on the colony size and morphology. Although this process was not resolved in our study due to the lack of small tracer particles, we speculate that these complex interactions may account for the higher scum coverage observed at moderate flow velocities following a prolonged period (6 h) of high-magnitude wind disturbance. Future studies should include a more detailed characterization of the hydrodynamic conditions at the water surface (e.g., adding additional small seeding particles to perform particle image velocimetry) to better understand this mechanism.

The observed increase in surface scum also led to more frequent aggregations among colonies at the air-water interface. The developing scum significantly increased the colony size ( Figure 6 ), thereby explaining the higher rates of change in colony size at the surface compared to in the bulk water ( Figure 5 ). Given the limited attention to these aggregations, the underlying mechanism is unclear. Here we proposed capillary forces could be a driver for the formation of Microcystis aggregations at the water surface. In the vicinity of immersed buoyant particles, attractive forces arise due to micro-deformation of the air-water interface (capillary immersion force, Kralchevsky and Nagayama, 2000; Paunov et al., 1993). The common phenomenon that floating objects such as bubbles tend to clump together or cling to the sides of the container is driven by capillary forces (Vella and Mahadevan, 2005). We speculate that capillary forces partly increase the encounter rate of Microcystis at the water surface during prolonged periods of low wind speed, which promotes their accumulation and aggregation.

During wind disturbance, surface scums were dispersed throughout the bulk water, in which the hydrodynamics was governed by turbulent flow. Our study showed that moderate wind-generated turbulence promoted the aggregation of Microcystis into larger colonies, whereas high wind-generated turbulence favored their disaggregation ( Supplementary Figure S9 ). This dual effect can be explained by the opposing effects of turbulence in increasing the collision frequency of cells and colonies and by increasing shear forces (Liu et al., 2019; O’Brien et al., 2004; Njobuenwu and Fairweather, 2018; Yao and Capecelatro, 2021). The latter likely limited the colony size in our experiments, which consistently decreased for increasing magnitudes of wind disturbances, regardless of disturbance duration ( Figure 4 ). The duration of wind disturbances thus affects colony size by controlling the duration of interactions of colonies with turbulence, which may explain the magnitude-dependent effect of wind disturbance duration on colony size.

Periodic wind disturbances shifted the habitat of Microcystis between the air-water interface and the bulk water, during which the surface aggregation of Microcystis and the combination of their collision and disaggregation in the bulk water were alternating. In our experiments, such cycles constituted a positive feedback regulation between colony size and formation of surface scum, by which the size of suspended colonies and re-formation rate of scum continuously increased ( Figure 3 ). It should be noted that colony size can also be affected by Microcystis growth (Xiao et al., 2018). However, the growth of Microcystis did not show notable differences among different flumes. In addition, the D_v50 changed in accordance with the wind conditions, with the peak in D_v50 occurring after the start of the disturbances ( Figure 4 ), suggesting that the colony size dynamics were likely governed by the hydrodynamic conditions rather than by growth.

We found a pronounced reduction in water surface tension with increasing Microcystis biomass with our supplementary experiment ( Figure 9 ), which can have important consequences for the hydrodynamic processes at the air-water interface. We found that the normalized flow velocity decreased with increasing scum coverage at the later phase of our experiment ( Figure 8 ), suggesting that the dense scum layer suppressed the momentum transfer from wind to water. At low wind speed, the reduction in momentum transfer can be explained by low water surface roughness, which is a consequence of low water surface tension (suppression of capillary waves, Wüest and Lorke, 2003). The increasing scum layer explained the decrease in water flow velocity generated by a given wind speed over time in our experiments ( Supplementary Figure S5 ) and in a previous study (Wu et al., 2019). These findings implied that floating Microcystis can promote the formation and persistence of surface scum by altering surface tension and counteracting wind-driven mixing. Such reduced flow velocity caused by the presence of scum layers additionally contributed to the increasing re-formation rate of surface scum ( Figure 3 ). The lack of direct measurements of surface tension in the flumes does not negate the surface tension-mediated hydrodynamic interactions between Microcystis and the air-water interface during the flume experiments, as indicated by the observed reduction in flow velocity.

The reduced surface tension caused by Microcystis colonies can also directly generate surface flows being directed towards regions of high surface tension (Marangoni effect (Roché et al., 2014; Vinnichenko et al., 2018), i.e., a spreading towards Microcystis-free regions. Although we could not directly observe this flow in our experiment as the Marangoni flow was masked by the wind-driven flow, it is ubiquitously present in environmental flows (Scriven and Sternling, 1960). Similar to the ‘soap boat’ (a well-known visualization of Marangoni convection), we propose that the lateral gradient of surface tension in the vicinity of initial scum patches can drive their horizontal surface drift and reshape their distributions in the absence of external force (e.g., weak wind forcing).

We scaled the wind speeds measured in the flumes to representative wind velocities at a standard measurement height of 10 m (U₁₀ , see Supplementary Table S3 ). In our experiment, the low wind speed (U₁₀ of 2 m s^-1) corresponds to the wind magnitude during weak wind periods, while the wind disturbances (U₁₀ of 2.8 – 5.3 m s^-1) correspond to amplitudes of periodic wind forcing frequently observed over lakes (Saber et al., 2018; Fernández Castro et al., 2021). The chosen 6 h and 12 h periods of wind disturbances represent the lower and the upper range of field observations over lakes (Fernández Castro et al., 2021).

Surface scum formed during low wind periods and was dispersed during wind disturbances, which is in line with field investigation that showed an upper threshold of U₁₀ for scum persistence is 2.6 - 3 m s^-1 (George and Edwards, 1976; Cao et al., 2006; Qi et al., 2018). Moreover, light-induced changes in the buoyancy of Microcystis, commonly observed in the field (Ibelings et al., 1991), were successfully reproduced ( Figure 2 ). These observations further demonstrate the validity of our experimental design.

The water depth in our study was limited by the dimensions of the flume to 30 cm, which is smaller than the amplitude of vertical migration of Microcystis in natural lakes. Consequently, the formation rates of surface scum that we observed in the flumes may be higher than those found in the field. In addition, the low water depth restricted vertical mixing at high wind speeds, exposing Microcystis in the bulk water to higher irradiance than in deeper water. However, the focus of this study was on the hydrodynamic interactions of surface scum with the air-water interface, which is not expected to be linked to the scale of the experimental setup or to water depth. As the light intensity affects both the growth and the buoyancy production of Microcystis, we utilized relatively high irradiance levels, as they often persist in surface waters in the field. Initially, our experiments also included the configuration of the low duration (3 h) and median-magnitude (3.8 m s^-1) wind disturbances. That data, however, had to be excluded due to technical issues with the flume operation. Future experiments should be conducted at higher resolutions (e.g., analyzing finer gradients of disturbance magnitude and duration) in order to provide a more comprehensive understanding of the partly non-linear interactions of wind with surface scum. Furthermore, buoyancy-driven migration could not be resolved in our measurements, as the flow velocity (generally exceeding 1 mm s^-1) surpasses the floatation velocity of most colonies (< 1 mm s^-1, Wallace et al., 2000).

Our study reveals the close connection between Microcystis dynamics and the air-water interface. The air-water interface emerges as a crucial habitat increasing colony size during scum formation (typically under low wind speeds), surpassing the impact of bulk water ( Figure 5 ). Recent studies have shown that the wind has declined in some regions over the past few decades (Vautard et al., 2010; Ranjbar et al., 2022). Consequently, this study indicates that the projected decrease in wind speed under future climate conditions will result in larger colonies and facilitate the recovery of surface scum upon wind disturbance. Additionally, our finding of denser scum after prolonged periods of high wind speeds may have implications for early warning of blooms during windy conditions.

Our study also suggests that Microcystis reduces surface tension, likely through the excretion of surface-active substances. This process constitutes a dispersion-avoidance mechanism under wind conditions, allowing Microcystis to persist at well-lit depths. Furthermore, reduced surface tension potentially drives lateral convection, facilitating the expansion of Microcystis and mitigating competition for resources during weak wind periods. These processes may represent adaptive responses of Microcystis to wind stressors, suggesting that water surface tension, affected by Microcystis, may be an important mediator by which scum-forming cyanobacteria shape their habitat in favorable directions. Given the oversight of the aggregation of Microcystis at the air-water interface and their hydrodynamic interactions in current studies, it is important to consider these processes appropriately for a more accurate prediction of Microcystis scum dynamics in future climatic conditions.