Land use scenarios, seasonality, and stream identity determine the water physicochemistry of tropical cloud forest streams

Study area and sampling sites

Our study area is located in the headwaters of the La Antigua watershed, a major watershed that drains part of the states of Puebla and Veracruz before discharging into the Gulf of Mexico. We selected streams located within the Tropical Mountain Cloud Forest (TMCF), at elevations between 1,100 and 2,200 m a.s.l. The climate of this zone is temperate humid and 80% of annual precipitation falls as convective storms during the wet season (May/June–October). During the dry season (November–April) the weather is mostly stable and dry, but eventual cold fronts produce light rains and fog. Mean annual rainfall in this altitudinal strip ranges from 1,385 mm to 3,185 mm and mean daily temperature is 15 °C to 19 °C (Muñoz-Villers & McDonnell, 2013). The soils of the entire study area are Andosols formed on shallow volcanic ash deposits over semi-permeable fractured andesitic-basaltic bedrock (Rossignol et al., 1987).

TMCF is one of the most biodiverse forests in Mexico (Rzedowski, 1996; Ruiz-Jiménez, Téllez-Valdés & Luna-Vega, 2012). The area is under severe deforestation pressure and land use conversion from forest into different agricultural practices and urbanization is common (Espinoza-Guzmán, Borrego & Sahagún-Sánchez, 2023). Current land use in the area is a mosaic of forest (mostly restricted to steep slopes), agricultural fields, shade coffee plantations, pastures, and urban areas (Hernández-Pérez et al., 2022; Martínez et al., 2009). Agricultural land uses typically continue along the slopes down to the streambank without a differentiated riparian vegetation buffer, rather, sometimes only dispersed remnant TMCF vegetation occurs a long a narrow riparian strip (<5 m wide).

Nine perennial headwater streams were selected for the study (Fig. 1). In each stream, we identified the transition between two contrasting land uses (e.g., forest, pasture, coffee plantation) and selected a 100 m section in each land use (for a combined 200 m study reach per stream) to set up three scenarios (see also Table 1 and Fig. 1). Scenario F-P = streams with upstream forest followed by pasture; Scenario P-F = streams with upstream pasture followed by forest; and Scenario F-C: streams with upstream forest followed by shaded coffee plantation. The fourth possible scenario, streams that drain coffee plantation and move into forest, was not included due to lack of accessible sites under this scenario. Forest stream sections had dense riparian vegetation of either natural forest or secondary growth. The riparian zone of shaded coffee plantation sections included the same shade tree species as the plantations typically reach the stream margin. Pasture were for cattle grazing and some had a narrow strip of riparian vegetation present (i.e., scattered trees and shrubs remnant of the former forest).

Data collection

Mean canopy cover over the stream was estimated with the HabitApp mobile application by averaging leaf coverage of 10 photographs taken with a leveled cell phone at points separated ~10 m along each study section. The channel and riparian zone slopes were measured using a clinometer. Physicochemical characterization of stream water was conducted with samples collected at three points along each 200 m-long reach under study: at the upper most point (0 m), a middle point which is the boundary between the two land uses (100 m), and at the end of downstream section (200 m). Sampling was conducted during the dry (March–April), dry-to-wet transition (June), and wet (October) seasons of 2019.

We estimated discharge at each sampling point by measuring depth and instantaneous velocity at 10 cm intervals across a perpendicular transect. Flow was measured with a Flow Probe 101-FP201 by placing the sensor at 60% of depth at each sampling point. Discharge (m³/s) was calculated as Q = Av, where A is the transversal area and v is flow (Gore & Banning, 2017). Water temperature (°C), dissolved oxygen (mg/L), and conductivity (μS/cm) were measured with a portable Yellow Spring Instruments meter (YSI, Model 85), and pH with a Barnant potentiometer (Model 20). A water sample was collected in polyethylene bottles (1 L) and filtered using Whatman GF/C filters to determine physical and chemical variables: total suspended solids (TSS, dry weight), alkalinity (as CaCO₃, phenolphthalein), silica (SiO_2, molybdate), chloride (Cl⁻, titration with phenolphthalein) and sulfate (SO₄²⁻, turbidimetric technique). Calcium (Ca²⁺) and magnesium (Mg²⁺) were measured using an atomic absorption spectrophotometer (Shimadzu Mod. AA6501), and sodium (Na⁺) and potassium (K⁺) with a flame photometer (Corning Mod. 410). Nitrogen was measured as ammonium (N-NH₄⁺, Nessler), nitrate (N-NO₃⁻, brucine), we also report dissolved inorganic nitrogen (DIN) as the sum of those two components. A second filtered water sample was collected using 250-ml glass bottles to measured reactive phosphorus (PO₄, ascorbic acid) and another one unfiltered to determine total phosphorus (TP, persulfate digestion, and ascorbic acid). Determinations were conducted per APHA spectrophotometric techniques (APHA, 2005). All samples were kept refrigerated (4 °C) until analysis.

Benthic organic matter (BOM) was estimated by collecting three samples with a corer sampler (14 cm diameter) in areas of pools at each study section (six per stream) following methods in Lamberti et al. (2017). Coarse matter was manually collected using a 1 mm sieve. Then, sediments within the corer were disturbed and the water filtered through 250 and 65 µm sieves. Up to 5 L of water were filtered for fine particles. All samples were transported to the laboratory and dried for 24 h at 70 °C to obtain dry mass and ashed for 1 h at 500 °C to obtain ash free dry mass (AFDM). Both fractions were combined to obtain BOM. Suspended organic matter (SOM) was determined by filtering the water column using a drift net (mesh 250 μm) for 15–20 min. We measured the area of net submerged and water flow to obtain the volume sampled. For fine SOM, 20 L of water were filtered using a 65 μm sieve. Samples were processed as for BOM and both fractions were added to obtain SOM. Chlorophyll a was assessed by collecting three rocks (upper side of ~10 cm²) per sampling point and placing them in separate bottles with 90% methanol. Samples were kept cool and in the dark until analyzed within 24 h in the laboratory. Chlorophyll a was measured spectrophotometrically and concentration (mg/m²) was determined through Holden’s equations (Meeks, 1974). We obtained rock area using aluminum foil to cover the area exposed to solar radiation (i.e., upper side of rock) and applied a weight-area conversion.

Statistical analyses

Principal components analysis (PCA) was applied to the correlation matrix to identify patterns in water physicochemistry between the different scenarios and seasons. The variables were transformed with log₁₀ (X+1) before the analysis.

A two-way ANOVA was applied to evaluate how the scenario, season, and their interaction affected each physicochemical variable measured. We used a Shapiro test of normality and a Fligner-Killeen test of homogeneity of variances in each case. When normality or homogeneity of variance were not met, we used a Kruskal Wallis test. Streams were nested within scenarios. Tests were run using the statistical package R, version 4.1.2 (R Core Team, 2021).

We applied a PCA to each scenario independently to assess within scenario patterns during each season. A Procrustes analysis was performed to assess whether the three land use scenarios resulted in different stream physicochemistry during each season (dry, transition, wet). Procrustes analysis rotates and scales an ordination (a PCA in our case) to allow for the comparison of two ordinations and provides a correlation coefficient and P-value between the first axes of the two ordinations. We performed two comparisons per season: F-P against P-F and F-C. We considered F-P as the scenario with the potential lowest impact, as the upstream section drains forest while the downstream is used as pasture, thus we used this scenario as the reference. We generated a PCA for each scenario using three streams and three sampling points per stream: upstream (forest or pasture), middle point, and downstream (forest, pasture, or coffee). Analyses were run in R using the package Vegan and the functions Procrustes and Protest with 9,999 permutations.

To assess the effects of physicochemistry on biotic components (i.e., chlorophyll concentration), we used Pearson correlations to relate chlorophyll concentrations against major physicochemical variables at each scenario.

Stream physicochemistry

Our cloud forest streams presented strong seasonality, with the highest discharge values (up to 248.52 m³/s) during the wet season (Table 2). It is noteworthy that the discharge on the F-C on the wet season was almost as low as the other scenarios in the dry season. Water temperatures ranged from 14.63 to 19.66 °C and had moderately oxygenated water (range: 5.96 to 6.55 mg/L). Water pH was slightly acidic to neutral in most streams, range 5.52 to 6.96, with lowest values recorded during the wet season (Table 2). High TSS concentrations were observed in the wet season, with the highest values (14 mg/L) present in streams draining coffee plantation (Table 2).

Alkalinity was similar among streams, range 16.09 to 31.92 mg CaCO₃/L, but the highest values were recorded in scenarios that included coffee plantation. Alkalinity decreased during the wet season in all scenarios (Table 2). Most streams were nutrient poor, except for those draining coffee plantation (Table 2). For example, TP ranged from 0.19 to 0.47 µM in streams draining forest and pasture, but those that drained coffee plantation had over two times those concentrations (range: 1.02 to 1.33 µM, Table 2). Most ions showed a tendency to decrease in concentration during the wet season, except for Cl⁻ and SO₄²⁻ that increased (Table 2).

BOM ranged from 1,020 to 5,227 g/m², with both extreme values recorded in the forest-pasture scenario (Table 2). SOM ranged from 390 mg/L in F-C during the dry season to 45,389 mg/L in F-P during the wet season. Wet season values in the F-C scenario were highly variable due to a single sample that measured twice the amount of SOM than the rest (Table 2).

PCA analysis generated groups that corresponded to our scenarios and seasons (Fig. 2). Axis 1 accounted for 36% of the variance, with axis 2 accounting for an additional 28%. Positively associated with Axis 1 were alkalinity, SiO₂, and pH, while discharge and SOM were negatively associated. Axis 2 was positively associated with temperature, N-NO₃⁻, N-NH₄⁺, and oxygen. Streams within each scenario clustered in groups in ordination space. PCA analysis showed that all three streams within each scenario remained close together during each season (dry, transition, and wet) (Fig. 2). Even during the wet season, when hydrological connectivity between the stream and its watershed is expected to be highest, we found that scenarios remained separate in ordination space (Fig. 2).

Axis 1 ordered streams along a seasonal gradient from dry to wet season (Fig. 2). In the dry season, the three scenarios were grouped at the extreme right of axis 1 and the three sampling points (upstream, middle point, and downstream) of each scenario also grouped together. Dry season scenarios were associated with high alkalinity, pH, and BOM. The transition to the wet season clustered in the center of the graph and the wet season on the left. The wet season corresponded to the highest discharge and concentrations of SOM. Axis 2 aligned our different scenarios along a physicochemical gradient, with the P-F scenario always located in the lower quadrants of the graph, followed by the F-P scenario in the middle and the F-C scenario in the upper quadrants (Fig. 2). This gradient was consistent during the three seasons and the highest physicochemical values were those in streams draining the F-C scenario.

Effects of scenarios and seasonality

Of all variables measured (Table 2), only four presented significant differences among scenarios and seasons. Water temperature, TP, SiO₂, and N-NO₃⁻ were significantly different among scenarios (Fig. 3). Water temperature and TP were significantly lower in F-P and P-F than in F-C (F = 18.3, P < 0.001; F = 11.5, P < 0.001, respectively) (Figs. 3A and 3B). SiO₂ was also highest in F-C, followed by F-P, and it was lowest in P-F (F = 134.8, P < 0.001) (Fig. 3C). N-NO₃⁻ was similar between F-P and F-C, and significantly lower in P-F (H = 16.5, P < 0.001) (Fig. 3D).

We found a significant scenario-season interaction for discharge (F = 56.7, P < 0.001), conductivity (F = 9.5, P < 0.001), pH (F = 4.5, P = 0.011), N-NH₄⁺ (F = 5.4, P = 0.005), Cl⁻ (F = 12.2, P < 0.001), and SO₄²⁻ (F = 8.8, P < 0.001). Dry season discharge was not significantly different among scenarios (Fig. 4A), but during the transition season it was significantly higher in F-P and P-F than in F-C. During the wet season, discharge was significantly higher in F-P than in P-F and F-C (Table 2). Conductivity was significantly different among scenarios and seasons (Fig. 4B), but it was consistently highest in F-C, followed by F-P, and lowest in P-F. Dry season pH was similar among scenarios (Fig. 4C), but in the transition season, it was significantly higher in F-P than in F-C; and in the wet season pH was lowest in P-F (Fig. 4C). N-NH₄⁺ concentrations were lowest in P-F during dry and wet seasons, but were similar among scenarios in the transition season (Fig. 4D). During the dry season, Cl⁻ was lowest in F-P and highest in F-C; this pattern changed during the other seasons, with lowest values in P-F (Fig. 4E). During the transition, Cl⁻ was highest in F-P, but in the wet season, it was highest in F-C (Fig. 4E). SO₄²⁻ had similar concentrations among scenarios in the dry and transition seasons (Fig. 4F). In the wet season, SO₄²⁻ was highest in F-C and lowest in P-F.

Patterns within scenarios

There were longitudinal changes in water physicochemistry, as streams flowed from one land use to another within each scenario. Water temperature showed a general increase in the downstream direction for each study stream (n = 9). In F-P, the pattern showed a positive linear trend, the lowest temperature was recorded in forest and the highest in pasture (Figs. 5A, 5D and 5G). In P-F, a similar pattern was observed, but temperature was lowest in the middle point, where the streams flowed from pasture into forest (Fig. 5B). The F-C scenario did not show strong trends, except during the transition season, when temperature increased downstream and was highest in coffee plantation (Fig. 5C). N-NO₃⁻ was generally lowest in forest than in any other land use (Figs. 6A–6I). In F-P, N-NO₃⁻ increased downstream, in pasture, except during the transition season when it was high in forest and in pasture (Figs. 6A, 6D and 6G). In P-F, N-NO₃⁻ tended to be higher in pasture than in forest, but it was always highest at the middle point, where streams flowed from pasture into forest (Figs. 6B, 6E and 6H). The F-C scenario showed a strong pattern of high N-NO₃⁻ concentrations in the coffee plantation section (Figs. 6C, 6F and 6I). TP was similar among all scenarios and seasons, with some tendency toward high concentrations in coffee plantation (Fig. 7). Chlorophyll a was consistently low in forest sections, and high values were found in pasture and coffee plantation during the dry season (Table 2).

Stream identity played a more important role than land use in determining water physicochemistry within individual scenarios (Fig. 8). Accordingly, PCA analysis formed groups by stream number (i.e., identity) and not by land use type. In the F-P scenario, during the dry season, stream 3 had high N-NH₄⁺, temperature, and SO₄²⁻ in both forest and pasture sections (Fig. 8A). During the transition season, stream 2 differed from the rest as it had higher discharge, SO₄²⁻, and temperature (Fig. 8D). During the wet season, the pattern was similar as the one observed in the dry season (Fig. 8G). In the P-F scenario, stream 5 showed a strong physicochemical differentiation from the other two streams in all seasons, having higher N-NO₃⁻, discharge, SiO₂, alkalinity, conductivity, P-PO₄, and pH than the other two (Fig. 8B). In the F-C scenario, stream 7 was also different from the other two during all seasons, having higher conductivity, alkalinity, SiO₂, pH, N-NO₃⁻ and N-NH₄⁺ than the other two streams (Figs. 8C, 8F and 8I).

Comparison of scenarios using Procrustes with F-P as reference indicated that both P-F and F-C scenarios had significantly different physicochemical patterns than F-P during all seasons (Table 3). Differences were of similar magnitude during the dry and transition seasons, but there was a tendency for smaller differences among scenarios during the wet season, relative to the dry and transition seasons (Table 3).

Biotic responses

Mean chlorophyll a values ranged from 0.25 to 1.20 mg/m² across scenarios (Table 2). ANOVA showed that the scenario (F = 5.36, P = 0.015) and season (F = 3.81, P = 0.42) had a significant interaction (F = 5.24, P = 0.006) over chlorophyll a concentration (Fig. 9). In the dry season, there were no significant differences in chlorophyll a concentration between scenarios (P > 0.05) (Fig. 9). However, in the transition season, concentrations in F-C scenario were significantly higher than in F-P and P-F scenarios (P < 0.05). In the wet season, P-F scenarios presented higher concentrations than F-P and F-C scenarios (Fig. 9).

Few physicochemical variables were related to chlorophyll a concentrations during the dry (only Cl⁻) and wet seasons (pH and Ca²⁺) (Table 4). However, half of the physicochemical variables measured during the transition season strongly correlated with chlorophyll a (Table 4); discharge, BOM, and SOM had a significant negative correlation with chlorophyll, while temperature, alkalinity, conductivity, TP, SiO₂, Na⁺, and K⁺ had a significant positive correlations (Table 4).