River connectivity and climate behind the long‐term evolution of tropical American floodplain lakes

Abstract This study presents the long‐term evolution of two floodplains lakes (San Juana and Barbacoas) of the Magdalena River in Colombia with varying degree of connectivity to the River and with different responses to climate events (i.e., extreme floods and droughts). Historical limnological changes were identified through a multiproxy‐based reconstruction including diatoms, sedimentation, and sediment geochemistry, while historical climatic changes were derived from the application of the Standardised Precipitation‐Evapotranspiration Index. The main gradients in climatic and limnological change were assessed via multivariate analysis and generalized additive models. The reconstruction of the more isolated San Juana Lake spanned the last c. 500 years. Between c. 1,620 and 1,750 CE, riverine‐flooded conditions prevailed as indicated by high detrital input, reductive conditions, and dominance of planktonic diatoms. Since the early 1800s, the riverine meander became disconnected, conveying into a marsh‐like environment rich in aerophil diatoms and organic matter. The current lake was then formed around the mid‐1960s with a diverse lake diatom flora including benthic and planktonic diatoms, and more oxygenated waters under a gradual increase in sedimentation and nutrients. The reconstruction for Barbacoas Lake, a waterbody directly connected to the Magdalena River, spanned the last 60 years and showed alternating riverine–wetland–lake conditions in response to varying ENSO conditions. Wet periods were dominated by planktonic and benthic diatoms, while aerophil diatom species prevailed during dry periods; during the two intense ENSO periods of 1987 and 1992, the lake almost desiccated and sedimentation rates spiked. A gradual increase in sedimentation rates post‐2000 suggests that other factors rather than climate are also influencing sediment deposition in the lake. We propose that hydrological connectivity to the Magdalena River is a main factor controlling lake long‐term responses to human pressures, where highly connected lakes respond more acutely to ENSO events while isolated lakes are more sensitive to local land‐use changes.


| INTRODUC TI ON
Tropical floodplain lakes are subject to natural hydrological dynamics imposed by the main river channel and thus are exposed to extreme floods and droughts (Death, 2010;Poff & Ward, 1989;Resh et al., 1988). These hydrological events are known to influence primary productivity and community assembly (Junk et al., 1989), and increase or interrupt ecological connectivity (Amoros & Bornette, 2002), which in turn impacts habitat quality (Lake, 2000).
However, natural hydrological dynamics of tropical floodplains can be affected by long-term (decades-centuries) human-derived modifications such as river damming, deforestation, land-use change, and climate change (Angarita et al., 2018;Salgado et al., 2020;Van Looy et al., 2019).
The Magdalena River in Colombia is one of the largest rivers (1,540 km) of South America discharging over 7,100 m 3 /s into the Caribbean Sea and hosting over 70% of the nation's population and gross domestic product-GDP (Mojica et al., 2006). It dissects the country from south to north, running through the Central and Eastern Andean Cordilleras, producing around 320,000 Ha of floodplains ( Figure 1). These offer essential ecosystem services including flood regulation, support, and provision to the local human communities (Montoya & Aguirre, 2009). The river contains one of the largest fish provisions in the region, with key economic species such as the Magdalena catfish (Pseudoplatystoma magdaleniatum) and the Magdalena prochilodontid (Prochilodus magdalenae; Caballero et al., 2001). Its floodplains, lakes, wetlands, and primary riparian forests are within the Tumbes-Choco-Magdalena biodiversity hotspot (Myers et al., 2000) and host endemic birds such as the critically endangered, Blue-billed Curassow (Crax alberti), and other migratory bird species such as the Fishing Eagle (Pandion haliaetus), the Yellow-billed Cuckoo (Coccyzus americanus), and the Eastern Kingbird (Tyrannus tyrannus; Angel-Escobar et al., 2014). Other endangered and charismatic vertebrates found in the region include the Brown Spider Monkey (Ateles hybridus), the American Manatee (Trichechus manatus), the Lowland Tapir (Tapirus terrestris), and the River Otter (Lontra longicauda; Angel-Escobar et al., 2014).
Deforestation in the Magdalena River basin has been steadily increasing over the last six decades, with current rates being threefold higher than those from the 1950s (Ayram et al., 2020;Etter et al., 2006;Restrepo & Escobar, 2018). This profound transformation of the landscape has come with a great environmental burden as the river and associated lakes have experienced excess in sediment yields, water pollution, habitat fragmentation, and freshwater fish population declines (Best, 2019;Restrepo, 2015;Restrepo & Escobar, 2018). In addition, more than 20 large dam projects (>20 MW hydropower capacity) across the Magdalena F I G U R E 1 (a) Map of the Magdalena River catchment showing the associated floodplains and lakes (blue) and the location of the study area (green circle); (b) study area showing Barbacoas and San Juana Lakes (dark blue). Connecting rivers are indicated by straight blue lines; (c) aerial zoom into the study lakes showing the coring location for San Juana Lake (LSAN1) and Barbacoas Lake (LBARB1). An NDVI pixel plot is also presented to show the differences in vegetation cover associated with the two study lakes and how it changes according to ENSO. Photographs taken from Goggle Earth (Google Earth V9.132.0.6-WebAssembly with threads. March 19, 2021) River and tributaries have been constructed or are on their way of implementation resulting in higher fish extinction risks and severe river flow reduction (Angarita et al., 2018;Carvajal-Quintero et al., 2017).
In large riverine ecosystems, the way in which aquatic communities are organized and respond to climatic or human-derived stressors largely depends on the degree and magnitude of the disturbance, and on the spatial arrangement of lakes within the main hydrological network (Eros et al., 2012;González-Trujillo et al., 2020;Grant et al., 2007;). Connected lakes to the main river channel may be for instance, more dependent on the hydrological dynamics of the main river, and hence be prone to greater biological resilience and recovery through source-sink dynamics than in isolated lakes (Salgado et al., 2019). In turn, isolated lakes are likely to present lower dispersal, and thus, local environmental change such as nutrient inputs (eutrophication) is likely to exert a greater control over the community structure through temporal species turnover (Salgado et al., 2019). However, increased habitat connectivity may also disrupt ecosystem resilience by homogenizing lake communities (Strecker & Brittain, 2017). A long-term perspective is of particular value, as it captures environmental and hydrological change at centennial and millennial scales (Salgado et al., , 2019Salgado, Sayer, Brooks, Davidson, Goldsmith, et al., 2018) and supplies limnological information from a scarcely monitored region. This approach has shown to provide continuous data on sedimentological changes and aquatic communities over time allowing to track back in time the effects of land-use change and its hydrological and limnological effects (Liu et al., 2012;Salgado et al., 2020;Zeng et al., 2018).
By combining multiproxy (fossil diatoms and sediment geochemistry) paleoecological data with historical climatic records from two floodplain lakes (Barbacoas and San Juana) associated with the Magdalena River, this study aims to provide new information on the long-term limnological responses of these lake systems to both natural hydrological and human-induced stressors. We expect that the diatom communities will reflect whether the lakes are being affected by anthropogenic activity directly, reflected in a eutrophic diatom composition as a response to local land-use changes, or indirectly, reflected in diatom dilution due to increases in sediment yield and a sensitivity to hydroclimate reflected on the lake's species turnover regarding extreme dry and wet events. We further hypothesize that in a more connected lake ecosystem (Barbacoas Lake), climatic variation imposes greater controls over the limnology of the lake, while diatom communities are subject to dispersal assembly mechanisms, resulting in lower species turnover, greater variation in species abundance, and a dilution of the within lake signal (Leibold & Norberg, 2004). In less connected lakes (San Juana), the diatom communities are expected to experience marked species turnover, driven by niche assembly rather than dispersion (Leibold & Norberg, 2004). In this sense, the diatom species turnover in these floodplain lakes is expected to be controlled by local abiotic factors (Rodríguez-Alcalá et al., 2020).

| Study area
Barbacoas Lake (6°44′26″N 74°14′36″W) is located on the western margin of the Magdalena River and is directly connected to it via the Barbacoas creek, which has an approximate length of 6.3 km long ( Figure 1). Barbacoas is a shallow lake (average depth = 1.2 m) with a superficial area of 10 km 2 , brown-stained waters (mean secchi depth = 0.39 ± 0.47 cm), pH of 7.25 ± 0.26, and mean daily surface water temperature of 33.7 ± 0.35°C (Table 1). San Juana Lake is located (6°38′32″N 74°09′24″W) on the eastern margin of the Magdalena River (Figure 1). It has a superficial area of 1.05 km 2 and is characterized by average water depths of 2.2 m, brown-stained waters (mean secchi depth = 53 ± 0.55 cm), and a mean daily surface temperature of 30.15 ± 0.93°C (Table 1). The lake is fed by the San Juan River on the south that outflows on the west joining later the Carare River before spilling into the Magdalena River near the town of Bocas del Carare (6°46′48″N 74°06′14″W). The hydrological river distance between the Magdalena and the San Juana Lake is approximately of 18.5 km.
Both lake areas are comprised by wetlands with relatively minor transformations, and small isolated patches of tropical rainforests embedded in a matrix of pastures for extensive cattle ranching that have been pervasively transformed the broader middle Magdalena River basin during the last decades ( Figure 1c).

| Core sampling
A single, short sediment core from semi-littoral areas near the mouth of the outflow of each lake was collected using a wide-bore (diameter of 10 cm) corer (Aquatic Instruments Inc.). The core from the San Juana Lake (LSAN1) was collected at a water depth of 100 cm (6°38′32″N 74°09′24″W). The core from Barbacoas Lake (LBARB1) was retrieved at a water depth of 90 cm (6°44′26″N 74°14′36″W). Each core was subsampled in the field at 1-cm intervals. The sediments were then kept refrigerated at the Tropical Palynology and Paleoecology Laboratory, Universidad de Los Andes, for further analyses.

| Dating and age-depth model
The LSAN1 and LBARB1 cores were dated using radionucleotide measurements of 210 Pb, 226 Ra, 137 Cs, and 241 Am by direct gamma assay (Appleby et al., 2001) in the Environmental Radiometric Facility at University College London, UK. For LSAN1, the top 20 cm of the core was dated and an age model beyond the top 20 cm was fitted, by simulating new ages using the "scam" package in R (Pya & Pya, 2021). The age model followed a shape-constrained generalized additive model (GAM), with the age-model spline constrained to be monotonic decreasing (Simpson, 2018a). For LBARB1 core, sediment samples were dated every three centimeters, along the whole length of the core. Because of irregular changes in unsupported 210 Pb activities ( Figure S1), the 210 Pb chronology could not be resolved using the CIC (constant initial concentration) dating model. Thus, the chronology was corrected using the CRS (constant rate of 210 Pb supply) dating model (Appleby et al., 2001).

| Geochemical analysis
The organic matter (OM) content of each core was measured using the method of loss-on-ignition (LOI; Dean, 1974). Sampling resolution for LSAN1 core was at 1 cm for the top 20 cm samples and at every 2 cm for the remaining 30 cm of the core samples. For LBARB1 core, sampling resolution was at 1 cm throughout the core. Shifts in OM content were used as a proxy of flooding (river influence) following Schillereff et al. (2014). During high floods, OM is expected to decline through dilution from a greater deposition of terrigenous sediments associated with the river flow (Rapuc et al., 2019). In turn, OM is expected to increase during dry periods through increased in-lake primary production and decreased allochthonous input (Schillereff et al., 2014).
Sediment geochemistry was measured using X-ray fluorescence (XRF) with an Xmet 7500 portable analyzer spectrometer (Oxford Instruments Inc.). Three grams of dry sediment, 1-cm-thick sample was analyzed for XRF. A sampling resolution of every 1 cm was used for the top 18 cm sediment samples in both cores and of every 3 cm for the remaining bottom samples of both cores. We obtained two XRF readings (1 min of length) for each sediment sample, and the median value of both readings was used for statistical analysis. The XRF portable analyzer spectrometer was calibrated against certified material prior to analysis (Conrey et al., 2014), and all XRF measurements were run using the Mining method, which detects elements occurring in very low (<0.01 ppm) concentrations (Gasdia-Cocharne, 2017). The selected sediment samples from LSAN1 and LBARB1 cores were analyzed for iron (Fe), manganese (Mn), titanium (Ti), calcium (Ca), zirconium (Zr) phosphorus (P; only for San Juana Lake). The ratios of these elements (excluding P) were used as proxies for river-borne and in-lake processes as follows: grain size Zr/Fe (Davies et al., 2015;Schillereff et al., 2014), detrital input Ti/Ca (Davies et al., 2015;Salgado et al., 2020), and oxygenation of the water column Mn/Fe (Davies et al., 2015). Generally, during low flow periods, rivers deliver comparatively less sediments to lakes, and these are normally finegrained silts (Schillereff et al., 2014). During slightly elevated flows, clays and Fe commonly occur, whereas during peaks of high floods, coarse-grained (Zr) sediments are expected to increase (Schillereff et al., 2014). Ti is an unambiguous indicator of allochthonous coarser inputs from the catchment (Cohen, 2003), while Ca is often associated with in-lake production (Tjallingii et al., 2007). As such, higher values of Ti/Ca ratio may indicate greater detrital input (Salgado et al., 2020). Variation in Fe and Mn provides information about changing redox conditions (Davies et al., 2015). In a reducing (low oxygen) environment, the solubility of Fe and Mn increases, being Mn more readily affected (Boyle, 2002). An increase in Mn/Fe ratio can thus indicate the onset of aerobic conditions. As so, greater river influences into the lakes were inferred by increases in grain size, sedimentation rates, detrital inputs, and concomitant reductions in OM.

| Diatoms
Approximately 0.3 g of dry sediment per sample was used for diatom analyses following Battarbee (1986). Each sample was placed in a beaker with 30 ml of hydrogen peroxide (10%) for approximately 24 hr, or until the reaction stopped. After, 100 ml of distilled water was added to each of the samples and they were left until the water column was clear. Then, 0.6 ml of each sample was placed on a microscope slide and allowed to dry after which it was mounted using Naphrax, and then, 400 diatoms were counted and classified. For LSAN1 core, sampling resolution was every 1 cm in the top 20 cm, and every 4 cm for the remaining of the core, for a total of 27 samples.
For LBARB1 core, we used a sampling resolution of 2 cm throughout the core, for a total of 22 samples. The differential lake sampling resolution was due to the differences in sedimentation rates and the temporal resolution we wanted to achieve for the recent decades. The diatom species were identified using Lange-Bertalot and Metzeltin (2007), Krammer and Lange-Bertalot (1986, 1991a, 1991b, Lange-Bertalot and Metzeltin (1998), and the Diatoms of North America database (diatoms.org). Diatoms were then grouped into the following functional groups according to their ecological preference: Aerophil, Benthic, and Planktonic (Table 2). For the Benthic category, ecological preferences related to productive and acidic/dystrophic waters were also included (Viktória et al., 2017).

Species of the genus Eunotia, Pinnularia, Nitzschia, Encyonema, and
Gomphonema had very low counts and therefore they were aggregated into a single category according to their respective genus.

| Changes in diatom assemblages
To detect major zones of temporal diatom and geochemical change, The SPEI index data were downloaded from https://spei.csic.es/map/ maps.html#month s=1#month =3#year=2020 for the interval be- The total surface area of each lake during these events was calculated through polygons in a supervised image classification analysis in Qgis desktop 3.10.5.

| Gradients of ecological, geochemical and climatic change
The main temporal gradients of ecological, geochemical, and climatic change for each lake were assessed using multiple factor analysis-MFA (Pagès, 2002). MFA allows to cluster the different geochemical parameters (element ratios, LOI, and sedimentation rates), diatom functional groups (acidic/dystrophic, aerophil, planktonic, and benthic/productive), and the SPEI annual data (mean, maximum and minimum) into specific groups and assess simultaneously the amount of variation explained by each group. Trends in trajectory of change in the multidimensional space can be also visually assessed. The geochemical, diatom, and SPEI groups were standardized by applying a weight equal to the inverse of the first eigenvalue of the analysis of the group (Pagès, 2002). The MFAs were performed in R using the package FactoMineR (Pagès, 2002).

| Significant periods of change
Generalized additive models ("mgcv" package in R, Wood & Wood, 2016) were then used to estimate significant trends of temporal change using smooth functions following Simpson (2018a). The GAMs were fitted to the MFA DIM 1 scores of each lake against time (Beck et al. (2019) and the residual maximum-likelihood (REML) method was used to penalize overfitting trends. A Gaussian distribution with an identity link was used to model the time series data, and diagnostic Q-Q plots were performed to check for homogeneity of variances in the residuals. To account for uneven observations in the time series and for age uncertainty (heteroscedasticity) in the models, we used the amount of time per sediment sample/divided it by its means as a weight following Simpson (2018a). A base function (k) of 15 was used to achieve the best model fit (see Table S1 in Appendix for more details). The first derivative function of each GAM was identified and used to determine significant trends in the TA B L E 2 Diatom species recorded in the sediment record of the San Juana Lake and Barbacoas Lake time series data using the "gratia" package in R (Simpson, 2018b).
Here, trends that deviated from 0 (no trend) indicated periods of significance (Simpson, 2018). The strength of nonlinearity in the driver response relationships was also assessed using the effective degrees of freedom (edf) of the GAMs (Hunsicker et al., 2016). An edf equal to 1 is equivalent to a linear relationship, whereas an edf >2 implies a highly nonlinear relationship and thus most likely to exhibit ecosystem threshold responses (Hunsicker et al., 2016).

| Age model and sedimentation rates
The LSAN1

| San Juana Lake
A total of 19 diatom taxa were found in LSAN1 (Figures 3a and S4).
Zone 3 (22.5-0.5 cm;c. 1967 This zone was dominated by Pinnularia spp. and A. granulata Year Species

| Barbacoas Lake
A total of 21 diatom taxa were found in LBARB1 core (Figures 3b   and S5). Clustering analysis and RCA indicated three main temporal zones of diatom assemblage change: Zone 1 (38. 5-24.5 cm;c. 1959c. -1984 In this zone A. granulata, Aulacoseira sp., A. alpigena, and Actinella disjuncta dominated the assemblages (Figure 3b). In minor proportions and mainly restricted to this zone were A. distans and D. confervacea.

| SPEI analysis
Extreme dry and wet periods ( Figure

| GAMs and derivatives
The statistical results for each lake GAM fits can be found in the Supplementary Data (Table S1). For San Juana Lake, the edf value was >2 indicating a nonlinear behavior (Figure 7). The first derivate results detected a significant threshold after c. 1858 that gradually F I G U R E 5 Multiple factor analysis (MFA) plots for the core LSAN1, showing (a) the variation and contribution of diatoms functional groups (plankton in green, acidic/dystrophic in yellow and benthic/productive in gray), selected geochemical ratios and elements (purple) Fe/Mn, Ti/Ca, Zr/Fe, and P, organic matter content (LOI) and sedimentation rates and SPEI data (SPEI = annual average; < minimum annual value; > maximum annual value). Major zones of diatom community change detected by coniss analysis are indicated with brown (Zone 1), green (Zone 2), and blue (

| D ISCUSS I ON
Our results were in general agreement with previous riverine metacommunity studies (Eros et al., 2012;González-Trujillo et al., 2020;Grant et al., 2007) and support our initial hypothesize that in a more connected lake ecosystem, climate and dispersalrelated assembly mechanisms are more important than at the less connected lake, where niche assembly rather than dispersion and climate results in marked species turnover. For instance, since the establishment of modern conditions, the more isolated San Juana Lake has been less dependent on the hydroclimatic variations and thus less vulnerable to interannual climate variability. Moreover, the post-1980 extreme dry and wet events did not greatly affect the lake's surface area or its physical-chemical conditions.
However, the marked temporal successional trend in diatom species and the more recent (post-1960s) increases in diatom species associated with productive environments (Nitzschia spp., C. cuspidata and G. eriense; Table 2), coupled with gradual increases in P, OM, and sedimentation rates, all indicate greater importance F I G U R E 6 Multiple factor analysis (MFA) plots for the core LBARB1, showing (a) the variation and contribution of diatoms functional groups (plankton in green, acidic/dystrophic in yellow, and benthic/productive in gray), selected geochemical ratios and elements (purple) Fe/Mn, Ti/Ca, Zr/Fe, and P, organic matter content (LOI), and sedimentation rates and SPEI data (SPEI = annual average; < minimum annual value; > maximum annual value). Major zones of diatom community change detected by coniss analysis are indicated with brown (Zone 1), green (Zone 2), and blue ( and nutrient runoff from agriculture and husbandry (Bennion et al., 2004;Salgado et al., 2019).
In contrast, we found that Barbacoas Lake is prone to be highly sensitive to hydroclimatic events such as ENSO, where changes in diatom dominance are far more important than turnover. The geochemistry of the lake also strongly correlates with ENSO. We found greater OM production occurring during dry periods, while greater detrital inputs during wet periods. Spikes in sedimentation rates were also found to correlate with dry ENSO periods (Figure 2), likely associated with the first coming rains of the wet season. Nonetheless, the increasing tendency of sedimentation rates post-2000 suggests that other factors rather than climate are also influencing sediment deposition in the lake. A recent study by Restrepo and Escobar (2018) similarly found that after accounting for the effects of precipitation, sedimentation rates have severely augmented since 2000s in the Magdalena River in response to the increasing forest clearance in the river catchment.
Ecosystem regime shifts are usually characterized by a gradual environmental change which eventually pushes an ecosystem to a critical threshold (Dakos et al., 2015). Determining the existence and nature of such ecological changes is, therefore, key for floodplain conservation and for understanding ecosystem resilience (Dakos et al., 2015). As shown here, regime shifts are challenging to identify as they could be slow and derived from multiple natural and humanderived causes (Bunting et al., 2016). More importantly, the rate of ecosystem response to environmental change could also depend on the degree of hydrological connectivity (Salgado et al., 2019). We show that connected lake systems may respond slowly in a gradual lineal fashion but with high variance determined by ENSO. In turn, isolated lakes may be prone to respond faster but in a nonlineal way with low variance. These results open a window for tropical floodplain lake landscapes as model systems to test ecological theory of long-term ecosystem regime shifts controlled by multi-annual hydrological connectivity.

| San Juana Lake
The San Juana Lake transitioned over time from a river-governed system, to a wetland, and eventually to the lake it is today, probably as a consequence of a progressive disconnection from the Carare River, tributary of the Magdalena River (Figure 1). This is interpreted from the dominance in planktonic diatom species such as A. ambigua and A. granulata, and C. amphisbaena around the early c. 1600s and late 1700s (Zone 1) that reflect water mixing and river influence (Table 2). This is in accordance with the geochemical data that indicate low concentration of OM, high detrital inputs, and grain size. After this period, our analysis identified a significant ecosystem threshold at around the mid-1800s (Figure 7), linked to an isolated wetland-like system transition (Zone 2). This limnological change was indicated by decreases in the riverine-planktonic A. ambigua along with a marked increase in A. alpigena, a species associated with low waters levels in tropical freshwater lakes ( Table 2). The prevalence of S. pinnata and the presence of the aerophil D. confervacea further support a very shallow, wetland-like environment (Table 2). Accordingly, the declines in grain size and detrital inputs further suggest a disconnection to the river (Salgado et al., 2020;Schillereff et al., 2014). Such long-term transition from river, to a wetland-like environment, fits other ontological process of South American floodplain lakes (Fayó et al., 2018). These commonly originate from a natural cutting off the meandering neck of a river (Gaiser & Rühland, 2010), likely assisted by pronounced shifts in precipitation regimes and sediment load that alter connectivity (Amoros & Bornette, 2002). As shown here, ENSO can bring extreme dry periods in the study area; hence, the change from a river-dominated system to a wetlandlike environment would have been likely derived from a historical strong ENSO event (Li et al., 2011). Such dry climatic conditions would have promoted a disconnection of a meander from the main river channel (Fayó et al., 2018).
Permanent modern lake conditions were established around the late 1960s, with moderately water acidic conditions (Zone 3), as indicated by the dominance of benthic/tychoplanktonic rather than planktonic diatoms including Stauroneis neohyalina, Sellaphora alastos, Gomphoneis eriense, and Encyonema minutum var. pseudogracilis (Table 2). The transition to a lake system is also supported by further, and more pronounced, declines in detrital inputs and grain size along with an increase in autochthonous productivity (OM content) and water column oxygenation. Replacements of coarser sediment beds, to finer sands and clays in response to a river-lake transition, have been similarly recorded by Kuerten et al. (2013) in the Pantanal floodplains, in Brazil. Similarly, increases in OM and water column oxygenation after artificial damming, or as in our case, after the fluvial input declined, have been also described after the Panama Canal construction (Salgado et al., 2020) and in river-dammed floodplain lakes in the Yangtze River, China (Liu et al., 2012;Zeng et al., 2018).

| Barbacoas Lake
The sediment record of Barbacoas Lake showed that the system has been responding to a varying climatic influence and associated degree of connectivity to the main river but without reaching any ecosystem threshold (as indicated by the lineal response in GAM; promoting disconnection from the Magdalena River, but also must have increased evaporation rates resulting in the partial desiccation of littoral areas, a process that was observed during a subsequent dry spell in 2020 across the lakes in the study area (personal observation by Lopera, L & Salgado, J). In such smaller surface lake area with greater exposed littoral areas, aerophil species such as L. mutica and D. confervacea (Table 2) would have again found ideal opportunities to thrive ( Figure 6).
In a similar fashion, the direct connection to the Magdalena River was suggested to be re-established and enhanced during extreme and moderate wet periods, where the lake recovered its original surface area (Figure 4). In 2011, for instance the country experienced one of the greatest La Niña events of the last century, where most of the Magdalena River catchment experienced unforeseen floods (Euscategui & Hurtado, 2011). Our record shows that, at the time, diatom counts were barely legible in the fossil record while sedimentation rates increased significantly. Such large increases in river sediment inputs would have affected diatom preservation and/or dilute the diatom concentrations (Reed, 1998;Salgado et al., 2020). We also found that during moderate wet periods, grain size, water column oxygenation, and detrital inputs all increase in the lake. Diatoms species such as the planktonic A. herzogii, A. granulata var. angustissima, and A. granulata and the benthic Nitzschia spp. also increased in abundances during these wet periods.

| Paleoflood interpretation
Overall, we were able to discern the level of connectivity and river influence on the lakes through our multiproxy approach. For instance, we found a good concordance between dry years (SPEI values < 0), and high OM content and low grain size ratio (Schillereff et al., 2014). It is expected that with a lower river influence, the OM in the systems increases through in-lake primary productivity (Schillereff et al., 2014). As seen in Barbacoas Lake, the increasing river influence is thus reflected by a decrease in OM being deposited by the river and by an increase in detrital inputs (Rapuc et al., 2019), an opposite trend to what was observed in the San Juana Lake since its formation. The prevailing benthic/productive diatom functional group associated with these dry periods also concurred with previous floodplain lake literature (Fayó et al., 2018;Gell & Reid, 2014;Liu et al., 2017). Similarly, we found lower OM content and high grain size and detrital inputs to be associated with wet (SPEI < 0) periods (Rapuc et al., 2019;Schillereff et al., 2014). Planktonic diatom species also prevail during wet periods, in particular A. granulata var.
angustissima and A. herzogii (Fayó et al., 2018;Gell & Reid, 2014;Liu et al., 2017). These results provide therefore an exciting tool to assess with confidence lacking long-term monitoring data on hydrological and ecological change in the Magdalena River system.

| Limitations
Paleoecological data have limitations and can be biased. For instance, uncertainties in the age model of San Juana Lake and potential sediment reworking by waves and bioturbation may have introduced discrepancies in our age models. In particular, the simulated ages (pre-1900s) beyond the 210 Pb dated portion in the San Juana core must be understood as modeled and taken with caution. Similarly, as floodplain lakes combine riverine and lacustrine features, limnological processes may have large spatial variations across the lakes that may have been not fully accounted by our single-core approach (Maavara et al., 2015). Nonetheless, our multiple and independent lines of evidence of change in biotic and abiotic variables are all in general agreement. The magnitude and timing of changes observed in our record coincide with the known basin-wide natural climatic and anthropogenic history of the Magdalena River (e.g., Restrepo & Escobar, 2018) as well as with other tropical floodplain lake system dynamics in general (Kuerten et al., 2013;Liu et al., 2012;Zeng et al., 2018). The wide range of proxies used in our study also represents a different geographical extent. For example, geochemical elements integrate basin-wide information (Davies et al., 2015), while diatoms represent what is occurring in the lake scale (Pla-Rabés & Catalan, 2018); thus, the combination of several proxies would also imply the integration across spatial scales. Being so, the agreement among proxies implies not only a synchronic behavior but also a coincidence among scales.
Thus, we are confident that despite our single-core approach, our data are reflective of a general historical change in each lake.  2014;Etter et al., 2006;Restrepo & Escobar, 2018;Suescún et al., 2017). Forest clearance has greatly promoted erosion, increasing sediment loads and nutrients into the Magdalena River (Restrepo & Escobar, 2018). The signal of these marked increases in sedimentation in both lakes suggests that the current tendency of land-use change and higher erosion should be a factor to consider when assessing future management strategies, as the impacts of this are not yet clear in these lakes.

| CON CLUDING REMARK S
We also showed that the two studied lakes have had a very different ontological histories controlled by the degree of connectivity to the Magdalena River and climatic variation. The higher degree of connectivity of Barbacoas Lake makes it far more sensitive to ENSO events than the more isolated San Juana Lake, which in turn is suggested to be more sensitive to local changes associated with its watershed. Future climate change scenarios suggest that drier conditions will prevail in our study area (IDEAM, 2017). For the more isolated lakes, this could likely increase water retention times promoting in-lake productivity and generating cascade effects such as anoxia and eutrophication (Chislock et al., 2013). The degradation of these smaller and more isolated lakes is of great concern. These lakes play key roles in water regulation and in offering temporal ref- uge for the aquatic biota including endangered large mammals such as the river otter and the American manatee (Wild Conservation Society-WCS Colombia, 2015), and key economic species such as the Magdalena catfish or the Magdalena prochilodontid. As observed in Barbacoas, drier climates will inevitably lead to reductions in lake size, which will diminish habitats and the hydrologic regulatory capacity of these lakes (Amoros & Bornette, 2002). These processes will likely have positive feedbacks as more sediment will be accumulated allowing for emergent plants to colonize newly formed habitats, which in turn, will promote the lake disconnection (Amoros & Bornette, 2002). This, with the added concern of the increased sedimentation rates, will put the lake at risk of rapid clogging, and disappearance.

ACK N OWLED G M ENTS
We thank the Palynology and Paleoecology Research Laboratory Collected lake sediment material was transported to Los Andes University laboratories under the ANLA permit 2016004245-1-000.

CO N FLI C T O F I NTE R E S T
We declare no conflict of interest regarding patent or stock ownership, membership of a company board of directors, membership of an advisory board or committee for a company, and consultancy for or receipt of speaker's fees from a company. Writing-review & editing (equal).

DATA AVA I L A B I L I T Y S TAT E M E N T
The data supporting the results are archived in Pangaea Data