The 18 stream sites are located within the humid tropics of Far North Queensland, Australia (Fig. 1; Table 1). The region spans a range of climate types, from Af (tropical rainforest) and Am (tropical monsoon) in the lowlands (0–500 m a.s.l.) to Cfa (humid subtropical) at higher elevations (500–1000 m a.s.l.) (Beck et al., 2023). Annual rainfall in the region ranges between approximately 1500 and 4000 mm and follows a strong seasonal pattern, with 65 %–75 % falling during the wetter months (November to April) and 25 %–35 % during the drier months (May to October) (Bureau of Meteorology, 2025). The geology of the region is diverse, with the coastal areas to the East dominated by Quaternary alluvial deposits, while the mountain range and Atherton Tableland to the West are characterised by Cenozoic and Palaeozoic basaltic lava flows and metamorphic rocks. The area does not feature any significant organic-rich or carbonate-rich sedimentary formations (Jell, 2013).

Samples were collected in six streams that drain relatively pristine rainforest remnants (R1 to R6) and 12 streams that drain some degree of agriculturally disturbed land, including six with land converted to pasture (P1 to P6) and six with cropland (mostly sugarcane and banana plantations; C1 to C6) (Fig. 1). Some of the agricultural sites drain a mixture of different land uses (i.e., forested area, agricultural area (pasture/crops) and urban area; Table 1). However, urban land use remains below 4 % at all sites. The six rainforest sites are generally located in more rugged terrain, resulting in higher mean catchment slopes (0.25 mm-1) compared to the 12 disturbed sites (0.11 mm-1) (p=0.005; Mann-Whitney U-test). Streams range from second to fourth order, and while the extent of groundwater connection is unknown, it is likely that all streams receive some level of shallow groundwater inputs.

Stream water samples were collected once during the dry season from 25 to 28 August 2020, and a second time at the end of the wet season from 21 to 26 March 2021 (Fig. S1). At each site, electrical conductivity and temperature were measured using a portable conductivity meter (WTW), while the partial pressure of CO2 (pCO2) was measured only during the wet season campaign using a pre-calibrated sensor (eosGP, Eosense) that was allowed to equilibrate for at least 30 min prior to measurement. At each site, samples were taken for later analysis of DOC concentration, δ13C-DOC, 14C-DOC, dissolved organic matter (DOM) composition, and water stable isotopes (deuterium, δ2H and oxygen-18, δ18O). In addition, at three sites (i.e. R1, P1 and C4; randomly selected from each land use type), samples were collected for 14C-DIC analyses. Two samples (P1 and C3 in the wet season) were damaged during analysis and therefore could not be analysed for 14C-DOC.

We collected samples in 1 L high density polyethylene (HDPE) bottles (14C-DIC), 2 L HDPE bottles (14C-DOC), 125 mL HDPE bottles (DOM composition), 40 mL pre-acidified borosilicate amber glass vials (DOC and δ13C-DOC), and 60 mL polyethylene centrifuge vials (δ2H and δ18O). All HDPE bottles were acid-leached with hydrochloric acid and Milli-Q rinsed prior to sampling. Samples for 14C-DIC, 14C-DOC, and DOM composition analyses were filtered in the field using a peristaltic pump (Geopump, GEOTECH) connected to Tygon tubing and high-capacity 0.45 µm polyethersulfone filters (FHT-45, Waterra). The Tygon tubing was pre-rinsed in 10 % hydrochloric acid and Milli-Q water between samples, and both the tubing and the high-capacity filters were then pre-conditioned with the water to be sampled for at least 30 s before sampling.

We also extracted two shallow soil cores in two rainforest sites (R3 and R6) using a hand auger. We sampled soil organic C (SOC) at three depths along these cores (0–5, 10–15, 20–30 cm) for later analysis of 14C-SOC. For each depth, approximately 100 g of soil were collected, double Ziploc bagged, and frozen upon return to the laboratory.

Both 14C-DOC and 14C-DIC analyses were conducted at the Australian Nuclear Science and Technology Organisation (ANSTO) in Sydney (NSW, Australia). 14C-DOC samples were acidified to pH 2, sparged with N₂ gas and neutralised to just below pH 7, rotary evaporated to a concentrate, and freeze-dried to a powder. 14C-DIC samples were acidified on a water line under vacuum, with CO2 released using phosphoric acid. The liberated CO2 was then captured in a vial using liquid nitrogen. Samples for 14C-DOC and 14C-DIC were then combusted using CuO, Ag and Cu wire. CO2 was cryogenically purified before being converted to graphite, as outlined in Hua et al. (2001). The graphite was analysed by accelerator mass spectrometry (Wilcken et al., 2015). We report all 14C data as normalised percent Modern Carbon (pMC), calculated according to Stuiver and Polach (1977). Mean standard errors were ±0.41 pMC for 14C-DOC and ±0.36 pMC for 14C-DIC.

All soil samples were sent to the Radiocarbon Laboratory at the Australian National University (Canberra, ACT, Australia) for 14C-SOC analysis. Prior to analysis, rootlet fragments were sieved out and samples were pretreated with hydrochloric acid to remove any carbonates. The analyses were conducted on a single stage accelerator mass spectrometer following the method described in Fallon et al. (2010). Results are reported as pMC, calculated according to Stuiver and Polach (1977). The mean standard error for 14C-SOC based on all six measurements was ±0.28 pMC.

DOM composition was analysed on an Aqualog optical spectrometer (HORIBA) at ANSTO, with excitation wavelengths of 240–600 nm and emission wavelengths of 250–800 nm. Data were blank corrected, interpolated, corrected for inner filter effects and for Raman scattering and first and second order Rayleigh scattering, and normalised using the Aqualog software (Hansen et al., 2018). We derived peak intensities following Coble (1996), and calculated the humification index (HIX) as per Ohno (2002) and the biological index (BIX) as per Huguet et al. (2009) and Fellman et al. (2010). Peak intensities, HIX and BIX were determined using the R package staRdom (Pucher et al., 2019).

DOC concentrations and δ13C-DOC were measured at the Environmental Analysis Laboratory in Southern Cross University (Lismore, NSW, Australia) using a Flash elemental analyser (Thermo Fisher) connected to a Delta V Plus isotope ratio mass spectrometer (Thermo Fisher), as per Carvalho (2023). δ2H and δ18O were analysed at Charles Darwin University (Darwin, NT, Australia) on a L2130-i cavity ring-down spectrometer (Picarro) connected to a diffusion sampler as described in Munksgaard et al. (2011).

We conducted all spatial analyses in QGIS 3.4 (QGIS Development Team, 2019). Catchment boundaries and average catchment slopes were estimated using LiDAR-based elevation data at 5 m resolution (Geoscience Australia, 2015). In areas where the LiDAR data were incomplete or unavailable, we used elevation information from the SRTM 30 (Shuttle Radar Topography Mission, NASA). Land use proportions were derived from Sentinel-2 satellite imagery captured in October 2020. The classification between rainforest, pasture, cropland, and infrastructure was conducted using the Semi-Automatic Classification Plugin (Congedo, 2021) in QGIS 3.4.

There were very few differences between agricultural sites dominated by pasture and those dominated by cropland, with the only significant differences observed in elevation and stable isotopic values during the dry season (Table S1). Therefore, we grouped these two land uses into a single “agricultural” category. To assess seasonal differences within the same category (rainforest or agricultural), we used the Wilcoxon signed-rank test (signrank function in MATLAB). To compare the two land use categories (rainforest and agricultural) within the same season (wet or dry), we used the Mann-Whitney U-test (ranksum function in MATLAB). Given the small sample sizes, p values lower than 0.1 were viewed as being indicative of differences.

To explore the variability in 14C-DOC and identify potential drivers of these variations, we ran generalised additive models (GAMs) using the fitrgam function in MATLAB, as most relationships between 14C-DOC and predictors were nonlinear (Table S2). The fitrgam function uses cubic B-spline functions by default for continuous predictors. We developed separate GAMs for the dry and wet seasons as we expected the drivers to differ between our two sampling campaigns. All predictor variables were standardised (centred to the mean and scaled to the standard deviation) prior to analysis, and predictors that were strongly collinear were excluded. To assess collinearity, we calculated Pearson correlation coefficients and removed variables with coefficients greater than 0.5 (Fig. S2). This led us to remove δ18O and δ2H (strongly correlated with each other and with temperature), elevation (strongly anticorrelated with temperature), δ13C-DOC (strongly anticorrelated with DOC), HIX (strongly correlated with DOC and anticorrelated with BIX), conductivity (strongly anticorrelated with slope), and the fraction of forest (strongly correlated with slope). Because slope significantly differed between the two land use categories (p=0.005; Mann-Whitney U-test), we added an interaction term (“land use ⋅ slope”) to account for the non-independence between these two predictors. We assessed model performance by comparing Akaike Information Criterion (AIC) values, testing models with and without this interaction term to determine the best fit. The resulting GAM structure was as follows: 114C-DOC∼intercept+s(area)+s(temperature)+s(DOC)+s(BIX)+s(land use⋅slope)+ε where s() indicates a smooth term and ε is the error term. We then used partial dependence plots to visualise the marginal effect of each predictor on 14C-DOC based on the fitted GAMs.

For the three sites where we collected 14C-DIC data, we used a two-endmember mixing approach to estimate the relative contributions of soil CO2 inputs and DOC mineralisation to stream DIC. The expected absence of any significant geogenic DIC source in the study catchments (Jell, 2013) allowed us to undertake this analysis. To represent the soil CO2 endmember (C_soil), we used the mean ± standard deviation of 14C-SOC from the four shallowest soil samples (0–15 cm), giving 104.3±4.3 pMC. To characterise the DOC endmember (C_DOC), we used the 14C-DOC value measured at the site on the same day. The fraction of DOC mineralisation (fDOC) to stream DIC (C_DIC) was then estimated as follows: 2fDOC=CDIC-CsoilCDOC-Csoil Uncertainty in fDOC was estimated for each site and season by propagating the uncertainty in C_soil, using its upper and lower bounds to generate corresponding bounds in fDOC. This simple approach provided only first-order estimates of the sources of stream DIC, because 14C-DOC likely represents an older, less labile fraction of DOC, which could result in underestimations of fDOC. However, the low HIX values (0.4–4.1) measured across all sites (see Results) indicate a predominance of highly biolabile DOM, suggesting that our mixing model provided coarse yet reliable results.

We observed high variability in DOC age across sites, with no consistent differences between rainforest and agricultural sites; a result that allows us to reject our first hypothesis that older DOC is exported after land conversion to agriculture. A range of factors may explain this cross-site variability. First, our two soil cores had very distinct age profiles (Table S3), with SOC at >20 cm depth being either modern or centuries old. Such discrepancy may be linked to geomorphic controls: the R6 core, which contained modern SOC at depth, was taken from a low point near the stream, where organic matter likely accumulates more rapidly. In contrast, the R3 core, which contained old SOC at depth, was taken from a gentle slope. Such spatial variability in SOC age is not uncommon in Australian landscapes (Hobley et al., 2017; Bowman et al., 2004) and might have contributed to some of the differences in DOC age observed across sites. Shallow (0–30 cm) soil C in tropical forests has been reported in the range 95–104 pMC (Shi et al., 2020), i.e. from modern to ∼400 years BP (5th–95th percentiles). This range broadly encompasses the integrated age of our two soil profiles, with R3 at the older end and R6 at the younger end of the range. Given its location on a gentle slope, we believe the R3 core may be more representative of broader landscape conditions where DOC is produced and exported.

Hydrological flow paths also appeared to play an important role. The negative effect of DOC concentration on DOC age (years BP) in both seasons (Fig. 4) (despite this trend being weak in the raw data during the dry season; Table S2) suggests that some streams may be fed by slightly deeper subsurface contributions, with older and more processed, lower concentration DOC, whereas others were fed by shallower flow paths, which tend to contain younger and more concentrated DOC (Barnes et al., 2018; Campeau et al., 2019; Sanderman et al., 2009) as natural soil formation processes allow for a decreasing age with depth. Various factors can influence flow path depth, including seasonal changes in flow regime and catchment morphology. Steep catchments tend to have shallower flow paths and shorter residence times (Remondi et al., 2019; Tetzlaff et al., 2009), leading to the export of younger DOC. Steep forested catchments are also likely to generate high runoff containing freshly leached organic material from both throughfall (Van Stan and Stubbins, 2018; Behnke et al., 2023) and leaf litter (Janeau et al., 2014; Miranda and Avelar, 2022). In our study, slope was an important predictor of DOC age, particularly in the wet season and for rainforest sites, where steeper catchments tended to export younger DOC. A similar influence of catchment morphology on DOC age has been observed in subtropical mountainous catchments (Chen et al., 2023). Interestingly, DOC ages (years BP) in the agricultural catchments had no significant relationship with slope in either season, contrasting with the negative relationship observed in rainforest catchments in the wet season (Fig. S3). This finding suggests that in agricultural catchments, the rate and magnitude of soil erosion, rather than flow path depth, may be a key control on DOC ages. In these systems, it is possible that steeper catchments might even increase the mobilisation of aged soil C, likely exposed by tillage and/or cattle trampling (Sickman et al., 2010; Drake et al., 2019).

The sources of DOC were relatively similar between rainforest and agricultural catchments, with δ13C-DOC compositions indicating similarly depleted, likely C3 plant-derived biomass as the dominant source. This is despite the agricultural catchments having available C4 pools (sugar cane and some pasture grass species), and suggests that most DOC in disturbed catchments originated from older, pre-clearing soil C pools – a result similar to that of Moyer et al. (2013). Lower contributions of C4-derived C sources than would be expected based on land use have been related to potentially large contributions from riparian forests and to differences in the biolability of different DOC sources (Bouillon et al., 2012, 2007). While we lack δ13C-SOC data to confirm the importance of older soil C sources to the stream DOC pool in disturbed catchments, the generally older wet-season DOC ages at these sites tend to support our conclusion (see following section). The relatively low HIX values across sites and seasons indicate that regardless of DOC age, highly biolabile compounds, which are typically low molecular weight and less aromatic in nature (Berggren et al., 2009; Li et al., 2023), may be dominating the DOC pool. This is not unexpected, as previous studies have shown that aged C can also be relatively enriched in energy-rich organic matter at the molecular level and be highly biolabile and readily oxidised by microbes (Mann et al., 2015; Drake et al., 2019; Spencer et al., 2015).

Despite differences across sites and no clear distinction between DOC ages in rainforest versus agricultural catchments, our data revealed some unexpected findings. In contrast to most previous tropical studies, we found that DOC in small, headwater streams of humid tropical regions can be centuries to millennia old (with a median of 82.5 pMC or 1553 years BP across sites and seasons), even in pristine rainforest areas. Recent studies have reported similar occurrences of old DOC at low flow in small, relatively undisturbed catchments in temperate regions (Tittel et al., 2022; Rhyner et al., 2023). Our findings suggest that this pattern may also apply to tropical regions, challenging the view of faster DOC turnover in the tropics (Spencer et al., 2012; Mayorga et al., 2005; Müller et al., 2015; Marwick et al., 2015). One possible explanation for this result is the higher elevations (hence lower temperatures) of our six pristine sites (mean elevations 345–948 m a.s.l.; Cfa climate class), which may have led to relatively slower decomposition rates and older SOC accumulating over time, compared to what would be expected in warmer, lowland tropical areas. Another potential factor is the contribution of old groundwater-derived DOC to streams (McDonough et al., 2020), a possibility we discuss further in Sect. 4.3.

Our results show that DOC age varied significantly between dry and wet seasons, with marked differences between rainforest catchments, where DOC tended to be younger at high flow, and agricultural catchments, where DOC was similar or older at high flow (Fig. 6). These findings support our second hypothesis that DOC is younger at high flow for rainforest sites, but contradict it for agricultural sites. We interpret these differences between land use categories as reflecting fundamental differences in system dynamics. Rainforest sites were in an “equilibrium” state, where DOC age was a direct function of flow path depth and reflected the hydrological shift between dry and wet seasons. This pattern is well documented in temperate and northern high-latitude catchments, where during high flow conditions the shallower riparian or organic-rich surface soil and litter layers become connected to streams and mobilise younger DOC (e.g. Tittel et al., 2015; Barnes et al., 2018; Evans et al., 2022; Campeau et al., 2019; Aiken et al., 2014). Additionally, the younger DOC observed in the wet season may also be attributed to increased leaching of fresh leaf litter (abundant in the rainforest catchments) through surface flow pathways, as reported in similar steep tropical rainforest regions (Miranda and Avelar, 2022). Leaf litter is a source of young, unprocessed, low molecular weight, biolabile C (e.g. Hudson et al., 2018; Harfmann et al., 2019), and the low HIX and BIX values observed at our rainforest sites during the wet season are a strong indication that leaf litter leaching contributed to DOC in these streams.

In contrast, agricultural sites operated under unsteady-state conditions, where old DOC was exported during both wet and dry periods, although flow conditions still influenced DOC age. This likely results from anthropogenic disturbance removing the bulk of modern C from organic-rich surface soil and litter layers (Drake et al., 2019; Moore et al., 2013; Lee et al., 2021). At some agricultural sites, higher flow conditions even led to the export of older DOC, a result that aligns with Moore et al. (2013) who observed the export of older DOC during the wet season from two disturbed peat catchments in Borneo. We propose that at these sites, processes such as tillage, cattle trampling and bank erosion may have exposed deeper, older C at the surface, increasing the leaching and transport of older DOC via surface rather than subsurface pathways during the wet season (Fig. 6). Although we do not present direct evidence linking older soil C to disturbance, all our agricultural sites were formerly rainforest (Kemp et al., 2007; Vanclay, 1996) and there is a long history of erosion in the Great Barrier Reef region (e.g. Kroon et al., 2016; McKergow et al., 2005), suggesting that substantial loss of young SOC due to anthropogenic disturbance is likely. While the role of surface erosion on DOC export is complex and largely unclear, studies have reported enhanced DOC mobilisation with increased erosion (Van Gaelen et al., 2014; Chen et al., 2023). Alternatively, older DOC at higher flow could also originate from 14C-dead organic chemicals transported via runoff (Sickman et al., 2010), such as synthetic fertilisers commonly used in sugarcane cultivation – however, we suggest this explanation is likely not a major control on DOC age across our sites, as it would not apply to pasture sites where fertilisers are not used.

In the wet season, the two land use categories had significantly different DOC sources (p=0.053), with more depleted values (median ∼-31 ‰) in rainforest sites and slightly more enriched values (median ∼-29 ‰) in agricultural sites. These differences may be related to increased contributions of young, C3 plant-derived organic C in the rainforest sites compared to older SOC in agricultural sites, as the latter is typically more ¹³C enriched than C3 plant-derived biomass (Moyer et al., 2013; Drake et al., 2019). These apparent differences in DOC sources were not reflected in the BIX and HIX indices, where seasonal variations were more pronounced than the differences between land use categories (Fig. 2). The decrease in BIX values from dry to wet seasons suggests a shift from a mixture of aquatic/microbial and terrestrial DOM sources to mostly terrestrial sources (Huguet et al., 2009; Fellman et al., 2010). This shift is consistent with increasing hydrological connectivity between soil C pools and streams, and in some cases, with higher surface erosion and runoff. The low HIX values across all sites indicate low humification, which is relatively common in tropical streams and rivers (Ji et al., 2024; Iles et al., 2022; Lambert et al., 2016). This suggests that even older SOC pools underwent limited humification before being mobilised into streams as DOC. Additionally, at sites with gentler slopes, slower streamflow and greater evaporation, older C might have been photodegraded over time, transforming aromatic-rich organic matter to lower molecular weight, more biolabile moeities (Stubbins et al., 2010). Alternatively, the dominance of low molecular weight DOM could suggest potential connection to regional groundwater systems at certain sites, as groundwater has been shown to contain old C with low molecular weight (McDonough et al., 2020). This explanation is explored further in the following section.

Unlike DOC, DIC was mostly modern (96.5 < pMC-DIC < 101.2) and always younger than DOC, consistent with our third hypothesis that DIC is younger than DOC in the study region. At the three sampled sites (R1, P1 and C4), DIC was likely to be primarily sourced from young, soil-derived CO2 inputs to streams (37 %–88 %) rather than from internal respiration of older DOC (12 %–63 %). While these estimates should be interpreted cautiously given the limited number of DIC measurements, our results suggest a decoupling between DIC and DOC cycling, regardless of land use. Consistent with our work, previous studies have also observed a disconnect between the sources and controls of DIC and DOC in streams, as reported in both boreal and tropical areas (Campeau et al., 2019; Moyer et al., 2013). Younger CO2/DIC has been associated with the lateral export of fresh soil CO2 (Campeau et al., 2019) or to direct inputs of atmospheric CO2 (Moyer et al., 2013). The dominance of soil-derived CO2 inputs to the stream DIC pool is well-documented in Australian tropical streams (Solano et al., 2024; Duvert et al., 2020) and, more generally, in low-order streams (Hotchkiss et al., 2015; Battin et al., 2023). However, in-stream DOC respiration can also be a substantial source of stream DIC (e.g. Solano et al., 2023), and it is interesting to note that even in these relatively steep headwater streams characterised by short residence times, stream DOC mineralisation contributed a non-negligible fraction of the stream DIC pool. This may have been facilitated by potentially high DOM biolability across sites, as indicated by the low HIX values, and relatively high BIX values, particularly in the dry season.

Groundwater is increasingly recognised as a potentially significant source of old, biolabile DOC to surface systems (McDonough et al., 2022, 2020). There is a possibility that at our sites, some old DOC could have entered streams via groundwater rather than solely through the mobilisation of old soil organic C. However, the DIC data suggest this is unlikely, as DIC remained modern or near-modern in both the wet and dry seasons. Any groundwater contributions to streams likely originated from shallow flow paths composed of recently recharged waters, as further corroborated by the low EC values across all sites and seasons and the significant seasonal changes in water stable isotopes, which suggest minimal regional groundwater influence and short residence times in the subsurface. Our DIC dataset includes only three of the 18 sites, so additional DIC age data would be necessary to definitively rule out groundwater as a source of old DOC. This would be particularly relevant in larger, flatter catchments where older, regional groundwater contributions may be more prevalent.