Data analysis
We evaluated the efficiency of our overall sampling effort for basimontane and premontane levels using rarefaction curves and the Chao1 species estimator for data pooled across sample years (Chao, 1984). Extrapolating a species accumulation curve to its asymptote and estimating species richness via rarefaction and the Chao 1, respectively, both provide an estimate of the performance of the proposed sampling method (Magurran, 2004; Chao et al. , 2009)
We evaluated diversity components using both univariate and multivariate analyses. We used generalized linear mixed effects models to evaluate the effect of altitude on species richness, total abundance, and effective diversity (i.e., eH’, where H’ is the Shannon diversity index; Jost 2006). Sampling year was a random effect because we used a repeated measures sampling design. Sample sites were also included as a random effect to address unmeasured differences in sites. As a result of this analytical model, results are general for altitudinal effects on fish diversity in the Caquetá River through years, including pre and post conflict time-periods, and among sites. Alternative models were compared using the corrected Akaike’s Information Criterion (AICc) with the bbmle package (Bolker, Team and Giné-Vázquez, 2022), where we emphasized AICc weights to identify the most plausible model after discounting for model complexity (Burnham and Anderson, 2002). Different alternative residual distributions (e.g., Gaussian, Gamma, negative binomial) in generalized linear models were iteratively evaluated and compared to log-transforms of response variables to best meet model assumptions (using check_model in the performance package; Lüdecke 2023). Given a chosen distribution, alternative models were then compared by AICc, where compared models included a null, random effects only, potential fixed effects (i.e., a dummy variable representing pre- and post-cease-fire conditions and altitude effects), and combinations of fixed and random effects. Finally, we used nonmetric multidimensional scaling (NMDS) with PERMANOVA to confirm differences in community structure between basimontane and premontane sites.
Data management and statistical analyses were performed in R (R Core Team, 2021) using functions from the glmmTMB (Magnusson et al. , 2020), vegan (Oksanen et al. , 2020), devtools (Hadley, Hester and Chang, 2022), iNEXT (Hsieh, Ma and Chao, 2016), ggplot2 (Kassambara, 2018), MASS (Ripley et al. , 2019), multcompView (Graves and Piepho, 2022), and performance (Lüdecke, 2023) packages.
Results
Samples included 4216 fish, belonging to 100 species, 58 genera, 24 families, and six orders (Supplementary file F1). Rarefaction curves reached asymptotes for both basimontane and premontane levels (Figure 3a). Premontane sites had >3 times more fish and >5 times more species than those recorded in basimontane sites (Figure 3b). In addition, basimontane sites had a different taxonomic composition than premontane sites. Basimontane sites had similar numbers of Characiformes and Siluriformes and some (<10%) of the Blenniiformes (mainly in the Cichlidae family; Figure 3c). In contrast, the premontane level was clearly dominated (>75%) by the Characiformes fishes which comprised a higher diversity at the order level (Figure 3c). Fish community structure disparities between altitudinal levels are confirmed by the PERMANOVA (F = 3.04, R = 0.23, P<0.05 ), and the NMDS in which two clear groups, corresponding to the basimontane and premontane sites, are depicted in the multidimensional space representing the analysis (Figure 3d).
Species richness was most efficiently modeled as a power law (i.e., log-log) function of altitude and the random effects of year and site (supplementary Table S1; Table 2). Although altitude alone has a good predictive power (R2 = 0.42) predictions improve with the random effects (R2 = 0.62), as confirmed by the model performance procedure (supplementary Figure S1). The model output suggests a significant mean decrease by 0.98 log(species richness) per unit of log-altitude (Table 2; see Figure 4a for regression line).
As for species richness, a power law model for total abundance using altitude as a fixed effect and the random effects was most predictive compared to other models (Supplementary Table S2; Table 3). Altitude alone represented 42% of variation in species richness, but model predictions improve with random effects (R2 = 0.62; supplementary Figure S2). The model output suggests a 0.72 significant decrease in log(fish abundance) per unit of log-altitude (Table 3; see Figure 4b for regression line).
Effective diversity was also most plausibly modeled as a power law (Supplementary Table S3; Table 4). Altitude alone has substantial predictive power (R-squared = 0.433), but adding year as a random effect slightly improved fit (R-squared = 0.456; Supplementary Figure S3). The model output suggests a significant 0.834 mean significant decrease in log(effective diversity) per unit of log-altitude (Table 4; see Figure 4c for regression line).
Overall, for all of the diversity components evaluated herein, using altitudinal level as fixed and site and year as mixed effects was rather predictive, and no pre- and post-conflict difference was apparent in results (i.e., results here represent conditions before further human land use effects). Fish diversity predictably decreased with altitude as a power law function.
Discussion
Our study is the first of its kind to evaluate fish community structure in the Colombian Andes-Amazon transition zone. Results here provide a baseline for conservation of regional streams across premontane and basimontane zones because data represent the period when armed conflict ended but before potential development of formerly avoided lands (Calle-Rendón, Moreno and Hilário, 2018).
Our results confirmed that altitudinal gradients in fish community assembly are important (van der Hammen and dos Santos 1995) and that fish diversity is greater at lower altitudes, no matter how we estimated diversity (e.g., Jaramillo-Villa et al. 2010, Lomolino et al. 2010, Lujan et al. 2013). Considering that effective mitigation of anthropogenic pressures on streams and fish assemblages should account for site-specific conditions and at different scales (Poiani et al. , 2000; Newbold et al. , 2015), altitudinal gradients in the study area may be predictive for conservation and management. We note that the basimontane streams are important to conserve despite their relatively low diversity because fishes there are notably different from those downstream.
Changes in community composition through time are typically complex and depend on spatial and land use contexts (Allen et al. , 2019; Hillet al. , 2021). We expect that fish diversity might be most heavily affected at the premontane level over time due to human activities that will stronger and earlier at lower altitudes. Conservation practices (e.g., riparian buffer zones, runoff settling ponds) to maintain stream conditions can be implemented early in regional land changes to best conserve diversity.
Given species extinction rates due to habitat degeneration in ecosystems similar to those sampled here (Manjarrés-hernández et al. , 2021), this study is valuable as a baseline of barely-impacted conditions soon after the conflict ended in Colombia and before encroaching development of formerly avoided lands (Calle-Rendón, Moreno and Hilário, 2018). We expect future sampling will compare results to those reported here to document changes in fish assemblages due to land use changes. We also expect land use to be most changed at lower altitudes, where soils and slopes are more amenable to agriculture, roads, and housing. If so, then it is possible that fish assemblages will be degraded by increased land use to become simpler and later affected by upstream effects (Figure 1). If ongoing deforestation and encroaching anthropogenic land use in the hyper-diverse Amazon basin is to be managed to minimize species losses, then some lands and streams must be preserved (Mckinney and Lockwood, 1999; Granado-Lorencio, Cerviá and Lima, 2007; Rull, 2007). Results here suggest that land surrounding premontane streams should take priority to conserve the most species in areas that are most vulnerable. Upstream catchments in basimontane areas can also be preserved to maintain water quality flowing into premontane streams and conserve unique fishes in those upstream reaches(Booth and Jackson 1997, Klauda et al. 1998, Araújo and Tejerina-Garro 2009, de Melo et al. 2009). Future work in the study area should evaluate relative effects of changing land use in upper and lower catchments on fish diversity in the lower reaches.
So far, most emphasis on ecological studies on understanding diversity has been on spatial patterns of biological diversity rather than analyzing temporal patterns. Understanding temporal changes is essential to predict possible scenarios of the most diverse world’s biota in natural ecosystems of the Amazon. Fish assemblages are affected by processes occurring at multiple scales, including those covered in our analyses (Livingstone, Rowland and Bailey, 1982; Jackson, Peres-Neto and Olden, 2001; Tedesco et al. , 2005; Oberdoff et al. , 2011). Future surveys could integrate data at scales obtained here with even greater spatial and temporal extents. We regard results here as a potential baseline for future work in the same drainages, where future work may reveal effects of coming anthropogenic land use in the Amazon basin.