Chwaka Bay (Fig. 1a) is located in the east coast of Zanzibar. It is surrounded by mangroves in the south and sandy beaches and coral rubble on the west and east sides. Several channels run from the mangrove area to the mouth of the bay. Extensive seagrass meadows mixed with macroalgae cover the bay, together with bare carbonate sand and coral rock and rubble. The seagrass meadows in the selected site (− 6.1489°, 39.4528°) were dominated by Thalassia hemprichii (Ehrenberg) Ascherson 1871, mixed with Halophila stipulacea (Forsskål) Ascherson in Anon. 1868. The macroalgae Valonia sp. C.Agardh 1823, Halimeda sp. J.V.Lamouroux 1812, nom. et typ. cons., and Caulerpa sp. J.V.Lamouroux 1809 were also present in the study site mixed with the seagrass.
Chwaka Bay is surrounded by several villages which depend on fishing as their main economic activity, with seaweed farming as a secondary source of income. Seaweed farms are present in the west coast of the bay, in shallow waters, and close to the coastal areas, mainly over seagrass meadows but also on sandy bottoms. During the experimental time, the temperature of the seawater was 1.9 ± 0.3 °C, the conductivity 52.9 ± 0.3 μS/m, and the pH 8.46 ± 0.1 (averages ± standard errors).
Seaweed farm experimental plots
Two local seaweed farmers provided the material and instructed on how to build the farms in the same method as they do in Chwaka Bay. We delimited three blocks (A, B, and C, separated by approximately 50 m) in an area dominated by T. hemprichii with presence of H. stipulacea and a mixed macroalgae community. We built three treatment plots (each 3 m wide and 4 m long, with an area of 12 m2) nested within each block: one seaweed farm plot, one trampling plot, and one control plot (each separated by 3–5 m within a block, Fig. 1b), resulting in a total of three replicate plots per treatment.
The seaweed farm consisted of four seaweed lines (ropes) tied to eight wooden pegs (two for each line), as is done in the off-bottom method. In each of the ropes, twelve E. denticulatum seedlings were attached by using the “tie-tie” method with a total of 48 seedlings per farm. The seedling size was standardized based on typical farming practices for E. denticulatum used at the beginning of the harvesting cycle. The trampling plot consisted of the same line structure as the seaweed farm plot, but without E. denticulatum attached to the ropes. The control plot consisted of four wooden pegs pressed into the sediment within the seagrass meadow, delimiting an area of the same size as the seaweed farm plot.
The experiment ran for 96 days (25.11.2015–29.02.2016) and was sampled approximately every 15 days for a total of seven sampling times (days 0, 15, 34, 49, 63, 81, and 96). The experiment was carried out during the Kaskazi season, which is generally considered unfavorable for seaweed growth (Hassan and Othman 2019). Nevertheless, the seaweeds grew uniformly and no die-off or problems with seaweed growth were detected (personal observation). The seaweed was not harvested until the end of the experiment. In every sampling, the seaweed and the seaweed farm ropes were cleaned from epiphytes, debris, and sediment by carefully shaking the algae to remove sediment and loose epiphytes and hand picking any epiphytic algae and anemones growing on the algae or on the ropes. Loose epiphytes were then flushed by the tidal current or directly thrown out of the plots.
Trampling pressure was exerted by one scientist walking between the lines in the seaweed farm and trampling plots during the sampling and cleaning time. The total time in which trampling was exerted in each seaweed farm and trampling plot was of about 40 min per sampling campaign. There was no trampling between samplings in order to reproduce the same effect that actual seaweed farmers would have on their farms. At the end of the experiment, a total of 4 h and 40 min of trampling was exerted in each seaweed farm and trampling plot by one person. With this design, shading and trampling were applied in the seaweed farm plots, trampling on the trampling plots, and no-disturbance on the control plots.
Seagrass and benthic macroalgae variables
Percentage cover and shoot density of seagrasses were measured taking three random points in each plot at each sampling time with 0.25 m2 and 0.01 m2 quadrats, respectively. Percentage cover was included as the only variable for the measurement of benthic macroalgae as macroalgae could not be assessed by shoot density. The Braun-Blanquet scale was used for the cover measurements (Mueller-Dombios and Ellenberg 1974), consisting of a scale of 8 numbers each one referring to an interval of percentage cover. The scale goes as follows: 0.1 = < 5% solitary, 0.5 = < 5% sparse, 1 = < 5% numerous, 2 = cover ≥ 5% - ≤ 25%, 3 = cover > 25% - ≤ 50%, 4 = cover > 50 - ≤ 75%, and 5 = cover > 75%. The cover categories were transformed into the midpoint cover range (Braun-Blanquet 1964). The percentage cover was then divided by 100, to obtain proportions between 0 and 1. Shoot density was measured counting the individual shoots of the two seagrass species within the 0.01 m2 quadrat.
A Li-Cor 1500 with a PAR (photosynthetically active radiation) sensor was used to measure the degree of shading by the seaweed growing over the seagrass meadow at each sampling time. Three random points were chosen within the seaweed farm plots and outside the farmed plots. The underwater PAR sensor (μmol photons m−2 s−1) was placed just below the water surface and at the seagrass canopy, while recording also the water depth at the canopy. Light reduction was calculated as the percent difference in surface and bottom PAR irradiance for the seaweed farm plots and the control plots. The worksheets with PAR data in three samplings were unfortunately lost (days 49, 81, and 96); therefore, we present only the data from the days 0, 15, 24, and 63 (up to 9 weeks of the experiment). Nevertheless, as the typical harvesting cycle in Zanzibar is between 35 and 45 days (Hurtado et al. 2017), these values are representative of the light reduction experienced by seagrass and its associated community due to seaweed farming.
Statistical software R (R Core Team 2016) was used for the statistical analysis. Graphs were plotted in R using the package “ggplot2” (Wickham 2016) and “ggpubr” (Kassambara 2020). Aesthetical editing of the figures was done using InkScape (Version 0.92.2).
To test whether there was a change in shoot density and cover in time due to shading and/or trampling, the slopes of regression models were compared. Time, treatment, and their interaction (slope) were tested as fixed effects. Due to the use of the experimental block design, block and its interaction with both sampling time and treatment were included as fixed effects. Due to the added complication of including extra interactions with block, analysis of deviance tests and likelihood ratio tests were used to simplify the models if the interactions with block did not add explanatory power. Both treatment and sampling and their interaction were never dropped from the model as they were part of the hypothesis testing. Block was never dropped from the model to avoid pseudoreplication (see Supplemental Material for further details).
The differences in light reduction inside and outside of the seaweed farms (introduced in the model as “area”) at each sampling time were analyzed using a linear model. Block and its interaction with the sampling area (outside or inside the seaweed farms) were added as a covariate to avoid pseudoreplication. Analysis of deviance test was used for model comparison and simplification. The significance in the differences in light reduction inside and outside the seaweed farms was obtained with analysis of variance (Type II test). Another linear model was used for the analysis of the differences in light reduction within an area (inside or outside seaweed farms) among the sampling times. Block and its interaction with sampling area were included in the model as fixed effects. The significance in the differences in light reduction within a sampling area in time was obtained with analysis of variance (Type II test). Post hoc pairwise comparisons were tested by using permutations in the models.