Catchment land use drivers are weak predictors of lakes’ phytoplankton assemblage structure at functional group level

A number of studies evidenced the impact of human activities on aquatic environments worldwide. The availability of nutrients in lakes is strongly influenced by watershed land use patterns depending on the share of forestry, agriculture and urbanization level. Nearly all the previous studies, which examined the relationships between the phytoplankton community and the land use pattern on the watershed of lakes or reservoirs were performed on a taxonomic basis. In this study, 78 lakes were sampled to analyse how the different land use types affect their phytoplankton functional group compositions. Our results concluded that land use is a complex driver, and it cannot provide an unequivocally straightforward forecast, which could result in the selection of a specific functional group or taxa. Thus, land use alone is a weak predictor, but the shared effects may structure the phytoplankton assemblage composition.


Introduction
Numerous environmental drivers affect the composition and abundance of phytoplankton assemblages in lakes. However, in most cases the mixing pattern, the availability of light, and a few resources (carbon forms, N, P, Si) are considered as the main drivers. Phytoplankton composition might be strongly influenced by other drivers such as pH/bedrock geology (Geelen and Leuven, 1986), extreme weather events (Kasprzak et al., 2017), environmental extremities (Padisák & Naselli-Flores, 2021), invasions (Padisák, 1997;Crossetti et al., 2019), biotic interactions (Wyngaert et. al., 2022) and dispersaldriven processes (Ptacnik et al., 2010). A number of experiments have shown the importance of the top-down regulation (Levine et al., 1999;Huber et al., 2008) but under natural conditions this effect appeared less evident (Frau et al., 2019). These events/conditions may effectively determine the phytoplankton assemblages on different spatial and temporal scales. Moreover, not only the change in a variable's value is important, but its frequency and amplitude as well (c.f. Intermediate Disturbance Hypothesis; Connell, 1978;Reynolds et al., 1993). Environmental stochasticity can also be influential in structuring phytoplankton community composition (Seltmann et al., 2019).
Some of the factors can be either the cause or the consequence of another, consequently, causal relationships are often difficult to reveal (Selmeczy et al., 2019). Catchment land use is a complex driver that acts indirectly on aquatic ecosystems by mediating variables that directly affect phytoplankton communities (e.g. Zhang et al., 2020). Due to the water use in agriculture the discharge of many rivers decreased leading to the drying up of lakes (Foley et al., 2005), as in case of the Aral Sea. Water level may not change so drastically, but the physical properties of the waterbody are modified, such as the water temperature (LeBlanc et al., 1997) affecting the whole ecosystem. Changes in land use could alter the availability of light in aquatic environments (Julian et al., 2008) because intensive agriculture increases soil erosion and sediment load (Foley et al., 2005). Transport or resuspension of fine sediments may increase turbidity, which limits phytoplankton growth (e.g. Declerck et al., 2006). Chemical properties of the water often change as a consequence of land use. The availability of nutrients in lakes is strongly influenced by land use patterns in the watershed (Stomp et al., 2011;Vanni et al., 2011). Forestry, agricultural activities or the urbanization level alter this component significantly (Paul et al., 2012;Peng et al., 2021). The amounts of nitrogen and phosphorus loading remarkably increased at global scale because of the intensification of the fertilizer use in agriculture (Mekonnen & Hoekstra, 2018). The ratio of N to P in the catchments changes as well, due to the nitrate-rich fertilizers (Vanni et al., 2011;Chorus & Spijkerman, 2021). However, the amounts of nutrients exported from watersheds to lakes strongly depend on the intensity and temporal pattern of precipitation (Hayes et al., 2015). Extreme weather events such as heavy rainfalls occur more frequently and are expected to further increase in the future (Beniston et al., 2007) playing a key role in the nutrient transport through runoff events (Carpenter et al., 2015). Apart from nutrients, agricultural chemicals (e.g. pesticides) also reach the waterbodies through this process (Feld et al., 2016).
The variables altered by land use could significantly affect the phytoplankton communities. Phytoplankton forms the basis of aquatic ecosystems and provide numerous ecosystem services, like oxygen, food or fuel production (Naselli-Flores & Padisák, 2022). Change in land use is a global threat (Rockström et al., 2009), thus understanding its effects on freshwater ecosystems including the phytoplankton community is essential (Hasan et al., 2020).
In our study, we used Reynolds's functional classification (Reynolds et al., 2002) on the phytoplankton data as it is the most widely accepted approach to evaluate and interpret the structure of phytoplankton assemblages in an ecological context . Reynolds's functional groups (RFG) are based on deep knowledge on taxonomy and autecology of species, namely their relative tolerance and sensitivity to different environmental conditions (Reynolds et al., 2002). This classification has been used, inter alia, in the Water Framework Directive (e.g. Padisák et al., 2006;Poikane et al., 2011).
In this study, we explored how different land use types affect the abundance and composition of phytoplankton functional groups in small lakes. Our main questions: (i) is there any association among RFGs and land use types? (ii) is there any association among other environmental variables and land use types? and (iii) what is the importance of watershed land use types shaping lacustrine phytoplankton community at taxonomic and functional levels?
We hypothesized that urban and agricultural land use types have a strong effect on the phytoplankton communities by altering its abundance and composition. High-nutrient concentrations and RFGs (e.g. M, L M and P) with high-nutrient demands and preferring eutrophic environments were expected in these areas. We hypothesized that forested areas would offer habitat for RFGs, such as coda E and F, which commonly occur in lakes with moderate nutrient concentrations. Additionally, we expected, that biomass of RFGs W1 and W2 will be higher in lakes surrounded by or connected to wetlands.

Study area and the measured variables
A total of 78 lakes were included in this study (Fig. 1). All the lakes are located in Hungary (Central-Europe) and their surface areas range between 0.1 and 10 ha. The depths at the sampling sites ranged from 0.8 to 14 m. The sampling sites were approached by a belly boat and the depth of the lakes was measured with Deeper Smart Sonar pro + . Sampling was carried out during the summer (July-August) of 2017. Integrated phytoplankton samples were taken from the upper one meter of the water column in the pelagic zone of the lakes. In case of two lakes with depth less than one meter, samples were taken from the surface to 10 cm above the sediment. Physical and chemical variables of the lakes (temperature (°C), oxygen concentration (mgL −1 ) and saturation (%), conductivity (µScm −1 ), pH and turbidity (FNU)) were measured with a Hach Lange HQ40d multimeter and a Hach Lange TSS portable turbidimeter. Phytoplankton samples were preserved in Lugol's solution. At least 400 settling units (cells, filaments, and colonies) were counted in each sample using a Zeiss Axiovert 100 (Oberkochen, Germany) inverted microscope according to the classical methods developed by Lund et al. (1958) and Utermöhl (1958). Phytoplankton biomass was calculated based on cell volumes from the most similar geometric forms according to Hillebrand et al. (1999) using the Opticount cell counting software (Hepperle, 2008) and supposing 1 g cm −3 density. Phytoplankton species were sorted into Reynolds's functional groups according to the classification of Reynolds et al. (2002); updated by Padisák et al. (2009). Water samples were analysed for total phosphorus (TP), soluble reactive phosphorus (SRP), total nitrogen (TN), nitrite (NO 2 − ), nitrate (NO 3 − ), ammonium (NH 4 + ), soluble reactive silica (SRSi), sulphate (SO 4 2− ), chloride (Cl − ), bicarbonate (HCO 3 − ) and chemical oxygen demand (COD) according to APHA (2012). Nutrient concentrations and COD were expressed in µgL −1 and mgL −1 O 2 , respectively.

Land use types
Five land use categories were employed based on Corine Land Cover (CLC) database (European Union, Copernicus Land Monitoring Service 2018, European Environment Agency): urban areas, agricultural areas, forest and seminatural areas, wetlands and water. The land use dataset was created in QGIS 3.10 software. To calculate the area of the different land use categories, first, the sampled lakes were digitalized in  Bernabei et al., 2010) was created around the lakes. Then, this buffer zone was fitted on the CLC layer then was cut. As a result of these steps surface elements were obtained with the needed CLC data. Finally, we saved the ratios of different land use types for each lake.

Statistical analyses
Data analyses were carried out in R environment 4.2.0 (R Core Team, 2022) with the following packages: vegan (Oksanen et al., 2022), PerformanceAnalytics (Peterson et al., 2020), rgdal (Bivand et al., 2022), maps (Deckmyn, 2021) and tidyverse (Wickham et al., 2016). Principal component analysis (PCA) was conducted with the prcomp function on the scaled and centered dataset, where the physical, chemical and land use variables were analysed. All the screened variables are plotted on the ordination. The correlation among the variables (physical, chemical and land use types) were explored with the cor.test function using the Spearman's correlation because the normal distribution of land use data was violated. Hellinger transformation was applied with decostand function on the phytoplankton dataset as suggested by Legendre and Gallagher (2001) on community composition data containing many zeros and used in the statistical analyses. Variation partitioning was used to quantify the pure and shared effects of the environmental and land use variables on the taxonomy-based and RFG-based community composition. First, the most important environmental and land use variables were selected with forward selection in an RDA model using the ordistep function. The VIFs of the explanatory variables were limited to ten in the RDA model following Borcard et al. (2018). Second, variation partitioning was implemented with the varpart function and, third, the significance of the different fractions (the variation explained by a given set of variables) were tested by ANOVA with 999 permutations. A non-metric multidimensional scaling (NMDS) was carried out with metaMDS function on the RFG-based and Hellinger transformed phytoplankton dataset. The relationships among land use types and the phytoplankton community were assessed with the envfit function by fitting the analysed land use categories on the NMDS ordination plot. R 2 and significance values of the correlations are also indicated.

Results
General description of the phytoplankton assemblages of the lakes Altogether, 190 taxa were found in the 78 lakes belonging to 8 main taxonomic groups and 26 RFGs ( Fig. 2 and Table S1). Chlorophyta and Cyanophyta were the richest taxonomic groups with 87 and 42 taxa, respectively. Regarding RFGs, coda J and F contained the most, 34 and 22 taxa, respectively. The most frequently occurring groups were the X2, J and X1. X2 was present in all the sampled lakes, while J occurred in 74 and X1 in 73 lakes. Rhodomonas spp. and Chrysocromulina parva Lackey were the typical representatives of X2 and Scenedesmus spp. were the most abundant taxa in codon J.
In terms of biomass, Cyanophyta was the most important taxonomic group (Fig. 3). Raphidiopsis raciborskii (Woloszynska) Aguilera, Berrendero Gómez, Kastovsky, Echenique & Salerno, the only species in the S N codon, reached high biomasses in 15 lakes, thus this RFG was dominant in terms of biomass. Additionally, H1 and W1 coda were abundant (Fig. 3).
Seven significant correlations were found among the land use and the investigated physical and chemical variables. Positive significant correlation was found only between the urban land use category and three variables: temperature, dissolved oxygen, and pH (Table 1). Nitrate, bicarbonate and conductivity correlated negatively with forest land use category.
Additionally, agriculture land use type was negatively correlated with dissolved oxygen (Table 1).

Land use types and Reynolds's functional groups
In general, only weak relationships were found among the land use categories and RFGs. The strongest correlation was found between the agriculture category and codon T (Spearman's r = 0.28, P = 0.01; Fig. 5). Only two species were included in this group: Mougeotia spp. and Planctonema gelatinosum X.Liu, H.Zhu, B.Liu, Z.Hu & G.Liu. Additionally, a significant positive correlation was found between urban land use type and the total biomass (Spearman's r = 0.27, P = 0.016), as well as between urban land use type and cyanobacteria density (Spearman's r = 0.31, P = 0.006).

Land use types, environmental variables and RFGs
According to variation partitioning, five environmental variables (dissolved oxygen, chemical oxygen demand, ammonium, nitrite and SRSi) were important and constrained the phytoplankton functional group composition. However, none of the land use types were significant. Variation partitioning revealed that the phytoplankton assemblage categorised according to Reynolds's functional groups was determined purely by the selected environmental variables in 12.8% and the shared effect was only 0.6% (Fig. 6A). The pure environmental fraction had a slightly stronger effect on the community structure considering the main taxonomic groups (14.4%), whereas the shared effect was 3.4% (Fig. 6B).
Non-metric multidimensional scaling (NMDS) with fitting land use categories on the ordination, was conducted to explore and visualize specific relationships among the phytoplankton functional groups and land use types (Fig. 7). Forest land use showed weak (R 2 = 0.09) but significant (P < 0.05) correlation with the phytoplankton community and urban land   use correlated marginally significantly with the community structure (R 2 = 0.06, P = 0.083). Lakes with dominating forested surrounding area are grouped in the upper left corner on the ordination. Lakes in urban areas are placed mostly to the opposite direction, to the right from 0 on the x-axis. Furthermore, there is no recognisable pattern for lakes which are surrounded by agricultural areas. The physical and chemical variables, additionally the land use and phytoplankton dataset are given in the supplementary material (Table S2).

Discussion
Human activities have a significant impact on aquatic environments as evidenced worldwide (Vörösmarty et al., 2010) and in Hungary (Borics et al., 2016). However, we found much weaker relationships between the land use types and the phytoplankton composition or other environmental variables than expected. The correlation of urban land-use category with the NMDS ordination derived from the entire phytoplankton community of the lakes was only marginally significant, it showed significant positive  NMDS ordination based on the phytoplankton community with fitted environmental variables. Squared correlation coefficients are given for each environmental factor reflecting the strength of this factor as a predictor of the assemblage structure. Symbols * and ` indicate significant (P < 0.05) and marginally significant (0.05 < P < 0.10) relationships, respectively. The colour of the pie charts indicates the land use ratios of the lakes relationship with three abiotic and two biotic variables: temperature, pH, dissolved oxygen, total phytoplankton biomass and biomass of cyanobacteria. The positive correlation between urban areas and lake temperatures highlights the heat island effect of urban areas and its significant consequences. The heat islands in urban areas are a special case how land use alteration modifies local climate and influences terrestrial and aquatic ecosystems through the elevated temperature (Foley, 2005;Cosgrove and Berkelhammer, 2018). Heat islands may provide stepping stones for invasive heat tolerant species, which increases the sensitivity of both types of habitats (Teurlincx et al., 2019). Temperature has a fundamental effect on metabolism and growth rate of organisms (Brown et al., 2004). This could be one of the reasons for the increased total biomass in urban lakes compared to other types of waters as elevated concentration of nutrients were not detected in urban environments. Furthermore, temperature is a key factor in the development of the cyanobacteria populations in lakes, because many species belonging to this group grow faster at high temperatures (Carey et al., 2012). This statement is consistent with our results that high cyanobacterial biomass was detected in urban areas. Additionally, urban lakes are often well sheltered from wind action which also favours disturbance sensitive cyanobacteria (Padisák et al., 1988). The elevated values of DO and pH could be explained by the active photosynthesis of high algal biomass in these lakes. During photosynthesis CO 2 , or in some cases HCO 3 − , is removed from the water by photosynthetic organisms resulting in pH increase (Reynolds, 2006). Although higher total-and cyanobacteria biomass was registered in lakes with urban environment than in others, there were no phytoplankton functional groups which correlated significantly with urban areas.
The proportion of the forested land use area correlated significantly negatively with three environmental variables: conductivity, nitrate and bicarbonate. Catchment of lakes covered by forested areas or natural vegetation in high proportion retains substantial amounts of dissolved and particulate allochthonous materials (Nobre et al., 2020). Lakes are strongly linked to their watershed (Bucak et al., 2018), thus the buffer provided by the shoreline vegetation could effectively reduce the transport of nutrients or other anthropogenic pollutants to waterbodies (Mayer et al., 2007;Mellander et al., 2018). The reduced surface runoff, soil erosion and nutrient uptake by plants are the most significant mechanisms which are probably responsible for the observed reduced values of nitrate and conductivity. Nevertheless, we could not identify any RFGs correlateing significantly with the proportion of forested area. NMDS analyses with fitted variables found significant but weak relationship between phytoplankton community and forest land use type, however, variation partitioning found that all the land use variables alone are negligible. This seems contradictory at first glance; however, the observed discrepancy is attributed to the distinct levels of robustness between the two statistical techniques. Additionally, the outcome is supported by the variation partitioning analysis, which demonstrates that land use impacts the phytoplankton community through environmental variables as evidenced by the shared variance.
Only DO correlated significantly and negatively with the agricultural land use type. It is possible that the lower amount of DO were caused by the decomposition of organic material, as suggested by the associations between TP, ammonium, and agricultural land use type (Fig. 4). Although agriculture is considered the main source of nitrogen and phosphorus to aquatic environments, significant relationships were not detected among these variables. Water quality is often deteriorated in areas with intensive agriculture because it leaches nutrients and other chemicals into the water bodies (Foley, 2005). This does not contradict our general findings because the standing crop of phytoplankton is driven mainly by nutrient loading and not by the ambient nutrient concentrations (e.g. Padisák & Istvánovics, 1997). Unfortunately, we did not have nutrient load data for the lakes. In our study, only codon T correlated positively with agriculture cover. There were only two species in this group: Mougeotia sp. (planktonic) and Planctonema gelatinosum. The members of this codon are mostly found in the well-mixed epilimnia of deep lakes (Reynolds et al., 2002). These species are not frequent in Hungarian lakes (Borics, 2015) as lakes with these specific features are not common in Hungary. Typical representatives of lakes which can provide such a habitat are meso-eutrophic, deep, from medium to large sized lakes, like Lake Geneva or Lake Garda. Increasing biomass of Mougeotia sp. was detected during/after the nutrient load reduction phase in the epilimnion of these lakes (Tapolczai et al., 2015) and in others as well, for example in Lake Kinneret (Zohary et al., 2019). Typical waterbodies in Hungary, which can provide adequate habitat for these species are gravel pit lakes and deep oxbows. This is confirmed by the fact that relatively high biomass of this group was found in such lakes. Lakes with significant biomass contribution of codon T are located in agricultural areas, however deep oxbows and mine lakes can be found in areas with different land use type as well. This observation suggests the importance of the dispersal of species from the species pool though this process has been rarely analysed by phytoplankton ecologists (Cartuche et al., 2021). It is not clear, how agricultural activities increase the biomass of codon T; however, this codon, although does not prefer waters rich in nutrients, is sensitive to nutrient depletion (Reynolds et al., 2002), which may explain the growth of this group in lakes surrounded by agricultural areas.
Most studies focusing on the land use effect on phytoplankton communities were performed on taxonomic basis, except for a few studies (e.g. Zorzal-Almeida et al., 2017) which included functional groups or traits. Katsiapi et al. (2012) found higher cyanobacteria biomass in urban and agricultural areas, while chrysophytes with some other groups were abundant in forested ones. We also found higher total biomass and higher amount of cyanobacteria in urban areas than in other land use types. The high contribution of cyanobacteria in eutrophic lakes is a commonly observed phenomenon (Reynolds, 1998;Ferber et al. 2004). In a study in New Zealand, Paul et al. (2012) found that cyanobacteria correlated positively with the percentage of pasture and negatively with native forest in the catchment. However, in Ohio ecoregions, intensive agricultural activity did not always result in high-cyanobacterial biomass (Beaver et al., 2012). Microcystin concentration in the examined lakes and reservoirs was higher in agricultural areas than in forested ones in the continental US (Beaver et al., 2014), however, in a number of cases high microcystin concentrations were measured also in forested areas. Thus, it is possible to conclude from these studies that in different cases significant relationships were found between the biomass of cyanobacteria or the amount of their toxins and the land use type (mostly agriculture) in the catchment. In addition, though the land use type may affect the phytoplankton community, other variables play role as well and could be more important to structure the phytoplankton community than the land use type.
The results of variation partitioning, in addition to the explored relationships among land use types and environmental variables, highlighted the indirect way in which the complexity of the land use affect lentic phytoplankton communities. Variation partitioning shows that the measured physical and chemical variables significantly influence the phytoplankton assemblages. The shared effect of physical and chemical variables plus the land use type is low, additionally the effect of land use type alone is negligible. However, different environmental variables were significantly affected by land use types. Land use type around the lake is not a single variable, like turbidity or salinity. Latters may ultimately determine the phytoplankton assemblages as shown for example by Dokulil and Padisák (1994) in Lake Fertő: only those species can survive, which are able to tolerate the quasi-continuous light limitation and high salinity values in this lake. These environmental variables and other "environmental extremities" are very selective, thus clearly restrict the number of occurring species almost independently from their RFG affiliation since in such cases species-specific ecophysiological adaptations are of prior importance (Padisák & Naselli-Flores, 2021). The land use surrounding a lake is a very complex feature influencing different environmental variables and the sum of these impacts will finally affect the phytoplankton community. This result is in accordance with the findings of Feld et al. (2016) who examined more than 800 lakes across Europe disentangled the effect of land use and geo-climatic factors. In case of phytoplankton, variation partitioning showed that variance remained mostly unexplained or explained by geoclimatic factors. Similar results were found by Özkan et al. (2013) sampling nearly 200 lakes in Denmark where water chemistry and lake morphology were the strongest predictors of the phytoplankton community, while land use and climate contributed only little to the explained variation. Landscape and local environmental variables were analysed in floodplain lakes by Machado et al. (2016) and the authors concluded that land use did not affect the phytoplankton assemblages. These results are in line with a recent review on human impacts on lakes which only marginally mentions land use as a potential driver of lakes' phytoplankton (Salmaso & Tolotti, 2021).
These statements are in accordance with the study by Schneider et al. (2020) who showed littoral indicators (e.g. periphyton) can more successfully track nearshore land use than pelagic nutrient concentration. According to these studies, land use has a less obvious effect on pelagic ecosystems, which can be explained by its indirect effect. For example, the littoral region can act as an effective buffer zone, which can substantially mitigate the effect of land use. This is confirmed by studies on the relationship of small lotic systems and land use, where the buffer zone is not extensive, thus land use has a significant effect on these ecosystems (e.g. Walsh and Wepener, 2009;Stenger-Kovács et al., 2020;Tapolczai et al., 2021).
One can think that the effects of land use were masked by the morphological variation of lakes and data sub-setting would enhance the land use effect. Therefore, in an exploratory analysis, we split our data and analysed lakes only in given categories (e.g. oxbows). The variation explained by the land use variables remained low, thus our results cannot confirm this hypothesis, which is in accordance with the results of Feld et al. (2016).
Though most of the studies found correlation between higher taxonomic levels, mostly cyanobacteria, and land use type on the lake's catchment there could be another reason, why only one significant relationship (codon T and agriculture) was detected between functional groups and land use types. Namely, while the RFGs (Reynolds et al., 2002) describe well the phytoplankton based on their habitat preferences and can reflect seasonal shifts, they may not be suitable to distinguish the functional differences caused by land use activities. Different studies show that the use of RFGs or other functional approaches as MFG (Salmaso & Padisák 2007) and MFBG (Kruk et al, 2010) contributed significantly to our understanding of drivers of phytoplankton dynamics. However, there is hardly any method that is good for everything. For example, species-based analyses provided better results than RFGs in a metacommunity study with data from three ecoregions in China (Xiao et al., 2018) and in lakes with one or more extreme environmental variables where species are selected according to their ecophysiological tolerance limits and not by other adaptive features considered in the RFG concept (Padisák & Naselli-Flores, 2021). However, the reason for detecting only one significant relationship between land use types and RFGs is probably because the land use activity acts indirectly on the lentic phytoplankton.
Our conclusion is that land use activity affects the lentic phytoplankton communities indirectly, thus land use alone is a weak predictor, but the shared effects could structure the phytoplankton communities significantly.
Acknowledgements We thankful to the reviewers and the editor for their thorough reviews our manuscript. We are grateful to the current and former colleagues and students of the Center for Natural Science (former Department of Limnology) for their help on the field and in the laboratory. Additionally, we thank Franciska M. Tóth for her advice in relation to QGIS tasks and Edit Király for laboratory and microscopy assistance. This study was financed by the National Research Development and Innovation Office (NKFIH 120595 and FK137979).
Code availability Not applicable.

Conflict of interest
The authors declare that they have no conflict of interest apart that Judit Padisák is an associate editor of the Hydrobiologia therefore must be blinded during processing this manuscript.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/.