We conducted a Web of Science search in January 2024 using the following Boolean query within the “Topic” field: Australia AND (“dissolved organic carbon” OR “dissolved inorganic carbon” OR DIC OR DOC OR CO₂ OR CH₄ OR methane OR “carbon dioxide”) AND (river OR stream OR creek OR freshwater). We excluded review articles and papers outside the scope of this study, such as papers based on data from brackish environments (e.g. estuaries). For studies involving transect sampling from river to estuary, we only included data from the freshwater endmember. We also included data from theses that did not result in scientific publications. In total, we extracted data from 64 papers and 2 theses.

For each paper, we mined all available data, including geographic coordinates, catchment area, discharge, stream order, water physico-chemical parameters (temperature, pH, conductivity, dissolved oxygen), sampling date and time, CO₂ concentration, CO₂ measurement method (whether direct or indirect), CH₄ concentration, CO₂ flux, CH₄ flux, DOC concentration, DIC concentration, POC concentration, alkalinity, and the isotopic ratio of carbon-13 to carbon-12 (δ13C) of different C species. We included CO₂ values derived indirectly from pH and alkalinity, as we assumed the authors had ensured their accuracy and reliability. When the required data were not available in the supporting information, we contacted the corresponding author(s) to request digital databases. As a last resort, we digitised data from plots using the WebPlotDigitizer tool (Rohatgi, 2024). Lastly, we standardised all C concentrations to µmol L⁻¹ and all CO₂ and CH₄ emission fluxes to mmol m⁻² d⁻¹.

A total of 35 776 coupled alkalinity, pH and temperature measurements and 16 853 DOC measurements from 2287 sites were obtained from databases maintained by state departments. Data were obtained from five agencies: Department of Lands, Planning and Environment, Northern Territory (NT); Department of Environment, Science and Innovation, Queensland (QLD); WaterNSW, New South Wales (NSW); Department of Energy, Environment and Climate Action, Victoria (VIC); and Department of Water and Environmental Regulation, Western Australia (WA). These data are either in the public domain or published under the Creative Commons (CC-BY) license. We used the total alkalinity, water temperature and pH data to calculate DIC concentrations based on standard carbonate equilibrium equations (Kalka, 2021; Millero, 1995; Plummer and Busenberg, 1982). Indirect DIC calculations from alkalinity and pH are unlikely to be a source of significant uncertainty given the typical pH range of rivers (i.e. from 6 to 8). Unlike our approach for the literature-based dataset where we included indirect CO₂ concentrations (Sect. 2.1), we did not estimate CO₂ concentrations indirectly from alkalinity, pH and temperature for the state data. This decision was made to avoid potential errors, particularly in systems with low buffering capacity (e.g. headwater streams) and high DOC concentrations (e.g. forested streams and wetlands) (Abril et al., 2015; Liu et al., 2020), and because of the potentially lower reliability of pH data given the diversity of sources and methods across states.

In addition to collecting and aggregating already published data, we collected new riverine C data from tropical and temperate streams across Australia. In February–April 2023 we sampled 97 sites during the wet season in the Australian tropics (NT and northern QLD). In September 2023, these same 97 sites were revisited and sampled in the dry season for streams that had maintained flow. Sampling these sites twice was done to capture seasonal variability, with the first campaign occurring during the peak of the wet season (JFM) and the latter at the end of the dry season (JJAS). A third campaign was conducted in the temperate areas of VIC and Tasmania between November and December 2023, corresponding to the austral late spring and early summer. This third campaign increased the number of sampled sites by 68. All the sites were selected based on local knowledge, satellite imagery, and accessibility, with an emphasis on small headwater streams given the underrepresentation of low-order streams in national and global databases (Lauerwald et al., 2023; Liu et al., 2022c; Rocher-Ros et al., 2019).

At each site, samples for DIC, DOC, POC, dissolved CO₂ and CH₄ concentrations were collected and measured, although there were some exceptions to this due to resource constraints. In addition to these core C parameters, supplementary parameters were measured when appropriate and feasible. These included specific conductivity, pH, water temperature, dissolved oxygen, flow rate, flow velocity, fluxes of CO₂ and CH₄ to the atmosphere, gas transfer velocity, stream depth and width, and streambed slope (see Appendix A1 for a detailed description of field measurements).

The primary method employed for measuring DIC, CO₂, CH₄ concentrations, and their isotopic compositions involved collecting duplicate water samples in 12 mL Exetainer vials, preserved with 10 µL of 6 % mercuric chloride (HgCl₂) and sealed with septa-caps, containing no headspace. Samples were analysed at the Water Studies Analytical and Stable Isotope Facility, Monash University (VIC). CO₂ and CH₄ concentrations were measured on a trace gas analyser (VICI TGA 6k; VICI Valco Instruments) interfaced with a pulse-discharge helium ioniser. Just prior to analysis, 4 mL of water was replaced with helium. Samples were shaken to allow gas exchange between the water and headspace, and the equilibrated headspace was then analysed for CO₂ and CH₄ on the trace gas analyser. DIC was converted to CO₂ by adding phosphoric acid (12.5 mM) to the samples and leaving them on a shaker table overnight. CO₂ was then analysed using the trace gas analyser. The isotopic signatures of DIC, CO₂ and CH₄ were determined using a Gas Bench II interfaced with a continuous-flow isotope ratio mass spectrometer (Thermo Scientific Delta V Advantage). For δ13C-CH₄ analysis, dissolved CH₄ in water samples was completely purged, extracted, and trapped using an automated trace gas cryo-prep system (Thermo Scientific PreCon) before being analysed on the isotope ratio mass spectrometer. DOC samples were collected in the field by filtering 40 mL of stream water through a 0.7 µm G/GF glass microfiber syringe filter (Whatman) into a pre-acidified (200 µL of 98 % H₂SO₄) borosilicate amber vial. Samples were refrigerated and sent to the Environmental Analysis Laboratory at Southern Cross University (NSW) for analysis. DOC was measured using a total organic C analyser (TOC-VCPH; Shimadzu) and δ13C-DOC was measured using an O.I. Analytical 1010 TOC analyser interfaced to a Europa 20-20 isotope ratio mass spectrometer (PDZ Sercon Ltd.). To determine POC concentrations and isotopes, 1 L water samples were filtered through pre-weighed 0.7 µm G/GF glass microfiber filters that had been pre-combusted at 450 °C for 2 h. After filtration, the filters were oven-dried at 60 °C for 48 h and reweighed to determine the particulate mass retained on the filter. The filters were then analysed on an ANCA GSL2 elemental analyser interfaced to a Hydra 20-22 continuous-flow isotope ratio mass spectrometer (Sercon Ltd.) at Monash University.

Due to equipment availability at the sites in the NT and QLD, CO₂ and CH₄ concentrations were also measured using a headspace equilibration technique where 45 mL of water and 15 mL of air were equilibrated in a sealed 60 mL syringe by gently shaking for two minutes. The equilibrated gas was then transferred into a sealed 12 mL syringe via a three-way-stopcock and kept cool until analysed using a LI-7810 trace gas analyser (Li-Cor Inc.) within 12 h. The original concentration of gas in the water was calculated using a mass balance of gas in the equilibrated gas sample and the atmospheric gas used for mixing (Cawley et al., 2020), and corrected for carbonate equilibria when DIC and pH were available (Koschorreck et al., 2021). Despite sometimes measuring gas concentrations using both the headspace and Exetainer sampling methods, only one value is reported in the dataset. The selected method is specified in the database and details on how the chosen method was selected are in Appendix A2. This CO₂ concentration, in combination with water temperature and pH, was then used to calculate any missing DIC using carbonate equilibrium equations (Kalka, 2021; Millero, 1995; Plummer and Busenberg, 1982).

Fluxes of CO₂ and CH₄ to the atmosphere were measured using a tethered floating chamber connected to the LI-7810 gas analyser. Flux was calculated using the slope of the linear increase of CO₂ / CH₄ within the chamber. Linearity of concentration increase was assessed both visually and using simple linear regression to ensure the observed gas flux was solely diffusive flux and not ebullitive flux. Samples were removed from the dataset when the R2 value was lower than 0.85. Two to three replicate chamber measurements, each lasting three to five minutes were undertaken at each reach to capture different stream morphologies (e.g. pools and riffles). The reported flux was the weighted average of all replicates for the site, with weights assigned subjectively based on the distribution of the different hydro-morphological features that each replicate captured to limit the influence of extreme (high or low) fluxes when reporting only a mean value for the site. For sites where the stream was too turbulent to use a floating chamber, gas transfer velocity was estimated using propane gas injections (further details in Appendix A1).

OzRiCa comprises both existing published data (Sect. 2.1 and 2.2), and newly collected, unpublished data from our sampling campaigns as described above (Sect. 2.3). The database consists of three tables (Tables A, B and C, https://doi.org/10.4211/hs.9aa735254e7e424ca18603c047d02f50, Ulloa-Cedamanos et al., 2026). Table A contains the concentrations, fluxes and isotopic compositions of DIC, DOC, POC, CO₂ and CH₄ at all sites and for all sampling dates. POC is only briefly covered in the following sections due to the limited number of measurements across Australia (n=235) compared to DIC and DOC. In Table A, sites with at least two records in different months within the same calendar year were categorised as “seasonal”, whereas sites with at least two records across different calendar years were categorised as “interannual”. The rest of sites were categorised as a “spot” measurement. Table B contains a range of catchment characteristics, including the climatic region, stream order and catchment area of sampled sites, which were obtained using the Köppen-Geiger classification (Fig. C1, Beck et al., 2018) and the National Environmental Stream Attributes database (Stein et al., 2012) on ArcGIS Desktop (version 10.8; Environmental Systems Research Institute (ESRI)). Table B also contains water quality parameters when available, including dissolved oxygen, specific conductivity, water temperature, and pH (Fig. C2); geomorphological features of the site such as stream or river depth and width, flow rate and flow velocity. Table B is more comprehensive for sites derived from our sampling campaigns than for those based on published data. Lastly, Table C lists the publications and datasets from which we extracted the data. A more detailed description of each table can be found in the corresponding metadata tables provided in Appendix B.

OzRiCa contains 54 843 observations from 2879 unique sites across Australia (Fig. 1). The distribution of these observations, however, varies significantly across hydrological and climatic regions. Most sites are concentrated in areas where streams and rivers are denser and less intermittent. Non-gas C observations (i.e. DIC and DOC; Fig. 1a) are predominantly found in eastern (810 sites, 140–155° E) and western Australia (1669 sites, 110–125° E), while gas C observations (i.e. CO₂ and CH₄; Fig. 1b) are mostly concentrated along the eastern coast and the Murray River in the southeast (285 sites).

Although the arid climate zone covers over half of the Australian continent (Table 1), only 3 % of the data included in OzRiCa are from sites located there (Fig. 1C and D). Instead, 97 % of the data come from the other climate regions, which together contain 86 % of the river network surface area (Table 1). When expressed relative to river surface area, site density is strongly biased towards the Mediterranean region, with over 10 sites km⁻² of river surface area (Table 1). This high density reflects the long-term DIC and DOC monitoring program in southwestern WA that has been running since 1969. Site density is also somewhat high in the tropics (0.62 sites km⁻² of river surface area), largely because strong hydrological seasonality reduces the median river network surface area there, as rivers contract during the dry season (Rexroade et al., 2026). At the other end of the spectrum, arid regions have the lowest site density, with 0.12 sites km⁻² of river surface area (Table 1). Although river surface area is very limited in arid Australia, the number of sites is even smaller, resulting in a particularly low ratio. This reflects the extremely sparse monitoring effort in arid regions, where river systems are remote and often highly intermittent.

The distribution of sites in the database was also evaluated in terms of system size, using stream order as a proxy for size (Fig. 1E and F). Here the distribution was skewed towards sites draining small catchments (stream orders 1–3), accounting for 1344 sites out of 2744 sites for non-gas C data and 249 sites out of 420 for gas C data. Within these low-order streams, 11 % (non-gas C) and 61 % (gas C) of the data are unpublished as they originate from our field campaigns.

DIC and DOC measurements were the most prevalent C measurements across all climatic regions (Fig. 1C and D) and stream orders (Fig. 1E and F). DIC was the most frequently observed C species (although much of the DIC database was indirectly derived from alkalinity data), with two thirds of DIC sites concentrated in western Australia (Fig. 1A). The distribution of DOC was skewed towards eastern Australia (62 %), where most concurrent measurements of DIC and DOC were recorded. Sites with concurrent CO₂ and CH₄ observations mostly occurred in northern (45 %, 10–20° S) and southern Australia (32 %, 35–45° S), with 96 % of these sites stemming from our field campaigns. Most POC observations were from eastern Australia, particularly in tropical, subtropical and temperate regions and low-order (1–3) streams (Fig. C3).

The temporal coverage of C data at each site varied considerably, ranging from a single observation at 1368 sites (i.e. 47 %) to a maximum of 1307 observations at one site (Fig. 2A and B). CO₂ and POC were more commonly found at sites with a single observation, whereas CH₄ showed similar prevalence at sites with both single and multiple observations (Figs. 2A and B, C3). The longest DIC record included 1111 observations over 20 years (ID: 129001A, WA) and the longest DOC record included 1173 observations over 35 years (ID: 401201, VIC). The longest CO₂ record had 401 observations over 26 years (Murray River at Euston Weir, Bargrizan et al., 2022) and the longest CH₄ record had 20 observations in one year (Rocky Mouth Creek, Gatland et al., 2014).

The data in OzRiCa covered the period from 1966 to 2024 (Figs. 2C and D, C3). DIC observations spanned the entire period, while DOC, CO₂, POC, and CH₄ measurements began in 1978, 1980, 1989 and 2013, respectively. Concurrent measurements of non-gas (Fig. 2C) or gas C species (Fig. 2D) started after 1978 and 2013, respectively, with CH₄ mostly measured together with CO₂, while DOC was frequently measured independently of DIC.

Monthly observations revealed that sampling was more consistent and regular across all climatic regions for non-gas C sites (Fig. 3A). Mediterranean and semi-arid regions had the highest number of non-gas C observations, driven by extensive state agency monitoring in WA, which contributed 79 % of Mediterranean observations, while 89 % of semi-arid observations were sourced from the VIC and NSW state databases. In contrast, for gas C sites, consistent and regular sampling was limited to arid and semi-arid regions, with relatively frequent sampling in tropical and subtropical sites, and very irregular and less frequent sampling in temperate and Mediterranean sites (Fig. 3B). The strong temporal coverage of gas C in arid and semi-arid regions was in large part due to a long-term database from the Murray River that provided monthly observations over 26 years (Murray River at Cobram; Bargrizan et al., 2022) and 32 years (Murray River at Tailem Bend; Bargrizan et al., 2022). In contrast to other climatic regions which had relatively even numbers of CO₂ and CH₄ observations, at least 99 % of observations in arid and semi-arid regions were CO₂ rather than CH₄ concentrations. The subtropical and tropical regions had observations in nearly every month (Fig. 3B), reflecting the high number of sites in these regions (Fig. 1D; Table 1). However, the data were unevenly distributed, because most observations were from different studies. Peaks in gas C observations occurred in February, March, and September, coinciding with our first two sampling campaigns. In Mediterranean regions, where the number of sampled sites was high (Table 1), more than half of the months lacked gas C concentration data (Fig. 3B). Although the river network density was one of the highest for temperate regions (Table 1), the number and frequency of observations remained limited, with only nine sites having more than one gas C observation. November stood out as the month with the highest number of observations and the most concurrent observations, due to our last sampling campaign that captured both C species in temperate regions.

Although spot observations were the most common observation across all climatic regions, repeated sampling was frequent across sites (Fig. 4), with nearly half of the sites having more than two measurements (Table 1). Among seasonal sites, only 21 sites had 12 or more observations. In contrast, 193 interannual sites had records extending over 10 or more years, with 94 % sourced from state agencies in NSW, VIC, and WA. Seasonal and interannual sites primarily covered Mediterranean and semi-arid regions, with the latter corresponding, for instance, to the long-term study of the Murray River (Bargrizan et al., 2022).

The C data in OzRiCa, including different species, isotopes, and fluxes, spanned several orders of magnitude (Table 2, Fig. 5). Among C concentrations, DIC had the highest median values, followed by DOC, CO₂, POC and CH₄. DIC and CO₂ observations obtained directly versus indirectly were significantly different (p=0.000105 for DIC and p<10-15 for CO₂; Wilcoxon rank-sum test), with median direct CO₂ concentrations 125 % higher than median indirect CO₂ concentrations, while median direct DIC concentrations were 9 % higher than median indirect DIC concentrations (Table 2). The large difference for CO₂ estimates may arise because 92 % of indirect CO₂ data in the database are derived from large rivers (stream orders 7–9) (e.g. Bargrizan et al., 2022), whereas the direct CO₂ measurements include 78 % of observations from small streams and rivers (stream orders 1–3), which typically have higher gas concentrations. Isotopic compositions showed a large range, with CH₄ showing the most depleted values (median -50.5 ‰) and DIC the most enriched (median -9.8 ‰). The isotopic values of CO₂ (median -21.1 ‰) was slightly higher than those of DOC (median -25.1 ‰). The median atmospheric flux of CO₂ was in molar terms 272 times higher than for CH₄ (Table 2). Among C species, CH₄ concentrations and to a lesser extent indirect CO₂ and POC concentrations, exhibited the highest coefficients of variation, while DIC showed the greatest CV among stable C isotopes (Table 2). Most distributions were unimodal, but gas C fluxes and CH₄ concentrations (i.e. the species with fewer observations) had bimodal distributions (Figs. 5, C3).

No consistent increase or decrease in the magnitude of C species across stream orders was observed, likely due to the high variability between sites within the same stream order (Fig. 6). Low-order streams (1–3) had significantly lower median DIC concentrations than high-order streams (7–9) (p<0.01, Wilcoxon rank-sum test). A similar trend was observed for CH₄, which showed significantly higher median values with increasing stream order (1–5, p<0.01, Kruskal-Wallis test); however, the absence of CH₄ data from high-order streams limited this analysis. DOC and CO₂ concentrations had two distinct peaks, one for low to medium-order streams (2–4) and the other one for high-order streams (7–9), with 99.7 % of CO₂ concentrations in the latter derived through indirect estimation methods (mostly from various sites along the Murray River). Overall, median values of DIC and DOC isotopes, along with C gas fluxes, appeared to increase with stream order (1–9 for dissolved and 1–5 for gas C, p<0.01, Kruskal-Wallis test).

The growth of Australian riverine C studies over the past decade (Fig. 2) has improved the nationwide data coverage, yet the distribution of sites remains uneven across the country (Figure 1). The recent geographic expansion mainly involves DIC observations in eastern and western Australia, largely driven by indirect records from different state agencies, along with our previously unpublished observations of non-gas and gas C in northern and southern Australia. Western Australia also has DOC observations from 151 sites recorded between 1991 and 2019, but gas C remains largely unmeasured in this region. OzRiCa further highlights the uneven distribution of observations, with areas far away from the coasts particularly underrepresented. A key challenge in these mainly remote areas is their predominantly arid and semi-arid environments (Fig. 4), which together represent 52 % of Australia's river network surface area (Table 1), and are characterised by sparse and intermittent streams and rivers. These systems play a significant role in C and nutrient cycling, ecosystem function, food security (Shanafield et al., 2024) and C budgets (Stanley et al., 2023). For instance, recent studies have shown that including seasonal drying and rewetting of riverbed sediments increase global inland water C emissions (e.g. López-Rojo et al., 2024; Marcé et al., 2019; Qin et al., 2024). Adding data from arid and semi-arid regions will help improve our understanding of C cycling in intermittent systems, which drain over 70 % of Australia's river length (Sheldon et al., 2010) and have become more prevalent over recent decades (Sauquet et al., 2021). While tropical regions are also underrepresented in OzRiCa, this study includes previously unpublished data that improves their geographic representation. Tropical inland waters are recognised as significant hotspots for gas C emissions, and the lack of data in this region remains a global issue (Drake et al., 2018; Lauerwald et al., 2023). These climatic regions, which are still poorly studied, face pronounced wet-dry cycles (CV = 1.3, Table 1), growing pressure from rising human land and water demands (e.g. Duvert et al., 2022; Stringer et al., 2021) and climate change (e.g. Lian et al., 2021), underscoring both the need and the opportunity for targeted future monitoring.

The aggregation of data from numerous studies into OzRiCa has provided coverage across most stream orders, with low-order streams (1–3) being the most prevalent in the database (Fig. 1). This focus on low-order streams is particularly important, given their abundance across river networks (Downing et al., 2012; Marx et al., 2017), their sensitivity to anthropogenic and environmental changes (Ulloa-Cedamanos et al., 2024), their underrepresentation in global databases (Drake et al., 2018; Liu et al., 2024), and their relevance for total riverine C emissions, despite covering only a small percentage of the catchment surface area (e.g. Ågren et al., 2007; Marx et al., 2017; Natchimuthu et al., 2017; Wallin et al., 2018). High-order streams (7–9) play a crucial role in the downstream export of non-gas C species to the ocean. Although OzRiCa includes DIC and DOC observations from major rivers (Fig. 6), our analysis underscores the need for additional POC and CH₄ concentration observations, as well as stable C isotopes and atmospheric C emission in large river systems, where significant gaps still exist.

Unlike the spatial coverage of Australian riverine C data, which has improved over the last decade, most sites still suffer from limited temporal coverage. The predominant sampling strategy involves collecting one or a few samples from individual sites, resulting in short and sparse time series (Liu et al., 2024). This limited temporal coverage is specifically reflected in OzRiCa for C gas records, where only 32 sites had more than 10 observations, with temperate regions particularly low in time-series data, as most sites had just one or two observations (Fig. 3, Table 1). For non-gas C records, the situation is better with 476 out of 2744 sites having more than 10 observations (Fig. 2), largely due to monitoring by state agencies. Long-term data (> 5 years) are extremely infrequent for gas C and infrequent for non-gas C. These data come from a large-scale study in the arid and semi-arid region of the Murray-Darling Basin (Bargrizan et al., 2022) and from state agencies. The Murray-Darling Basin study focused largely on DOC and CO₂ observations in the Murray River, providing river-specific insights that might not be applicable to smaller streams and rivers of the arid and semi-arid regions. No such long-term data exist for CH₄, with the longest time series being up to two years (Atkins et al., 2017). Despite increasing efforts to address these temporal gaps, most studies are limited to a few years, as very few funding programs support long-term monitoring. The limited long-term observations are further challenged by the current downward trend in the number of long-term discharge monitoring stations (Shanafield et al., 2024), which are crucial for estimating C fluxes and understanding C dynamics.

The lack of data at finer temporal scales can lead to inaccurate C flux estimations, as critical C processes may be overlooked. For example, predominant sampling during daytime, excluding diel variability, may result in a consistent underestimation of global riverine GHG emissions, since nighttime CO₂ emissions are on average 30 % higher than daytime emissions (Gómez-Gener et al., 2021). Integrating hydrological events is pivotal to better constrain C export (Lauerwald et al., 2023), as more than half of the annual DOC export can occur during the early flood stages (e.g. Birkel et al., 2020; Ulloa-Cedamanos et al., 2021a). Australia's extreme hydroclimate variability, which has intensified over the past decades (Ayat et al., 2022), fosters these hot moments of C export. The previously identified data limitations underscore the need for high-resolution and long-term data to capture these critical temporal variations and improve the accuracy of C flux estimates across different climatic regions.

As discussed above, the increase in the availability of both gas and non-gas riverine C data has been remarkable, creating new opportunities to examine large-scale spatial and temporal patterns. However, specific challenges in C observations become evident when aggregating different studies together in a single database. For instance, gas flux estimates require the inclusion of different emission pathways, particularly for the CH₄ flux. Currently, OzRiCa only includes observations from the diffusive CH₄ pathway, although the ebullitive CH₄ flux can be substantial in riverine systems (Stanley et al., 2023). The analysis of C species also involves multiple steps and calculations, with field and laboratory protocols varying widely within the same C species. Although direct measurements of DIC and CO₂ should be prioritised, indirect estimations are still observed in environmental agency databases (Hartmann et al., 2014) and from the literature, particularly in studies using long-term data (e.g. Binet et al., 2020; Rosentreter and Eyre, 2020; Stets et al., 2014; Ulloa-Cedamanos et al., 2020). OzRiCa is not an exception, with indirect DIC data representing a large part of the DIC data, nearly all of which was sourced from state agencies. While indirect DIC calculations are generally reliable within the typical pH range of rivers, indirect CO₂ estimates can introduce significant errors in systems with low buffering capacity and high DOC concentrations (Abril et al., 2015; Liu et al., 2020). In OzRiCa, nearly all the indirect CO₂ concentrations come from eight peer-reviewed publications (with 90 % of these data from Bargrizan et al., 2022), where we assumed the authors had ensured the accuracy and reliability of their indirect estimates. A few other indirect CO₂ estimates come from our sampling campaigns and were included only when direct measurements were unavailable or deemed unreliable (see further details in Appendix A2). Although we observed a statistically significant difference between the median direct and indirect DIC concentrations, the magnitude of this difference was relatively small (9 %). The median direct and indirect CO₂ observations differed more substantially (125 %), but this was related to a bias in stream order between the two groups, with small streams – typically exhibiting higher CO₂ concentrations (Marx et al., 2017; Wallin et al., 2018) – overrepresented in the direct dataset. Based on these considerations, we have high confidence in the robustness of both direct and indirect CO₂ and DIC estimates included in OzRiCa.

Significant gaps in C databases include the limited number of isotopic C data and the lack of concurrent measurements of multiple C species (DIC, DOC, POC, CO₂, and CH₄) at the same sites. Isotopic data are key for identifying biogenic and geogenic sources of C (e.g. Campeau et al., 2017), in-stream processes and residence times of C. For instance, δ13C isotopes can help pinpoint CO₂ sources (Duvert et al., 2020b; Telmer and Veizer, 1999), which are often assumed to originate primarily from respiration in wetlands and upland soils (Liu et al., 2022c). Concurrent measurements of multiple C species are also necessary to better understand the mechanisms driving temporal and spatial variations in C sources and sinks. While our three sampling campaigns address this issue in OzRiCa, most other studies in the literature focus on either DOC or CO₂, with DIC and CH₄ often observed sporadically. In contrast, state agencies typically focus on either DOC or indirect DIC records. Including isotopic C and multiple species of C in future studies could offer deeper insights into the sources and processes of C in large-scale assessments.

OzRiCa represents a powerful tool for exploring the role of rivers in the global C cycle, functional ecology and ecosystem health of inland waters. This database provides a better understanding of spatial and temporal dynamics of C species in Australian rivers and streams, with particular strengths in addressing areas that were historically underrepresented, such as tropical and semi-arid areas, and low-order streams. These areas, critical in C cycling due to their sensitivity to anthropogenic and environmental changes, have been prioritised in this effort. We hope OzRiCa will help improve our comprehension of freshwater C processes and better prepare us to predict their responses to global environmental changes. OzRiCa also offers unique opportunities through the integrated analysis of C data (Table A) and basin-specific data (Table B), enabling the parameterisation of freshwater C models and the development of scalable products for regional and continental assessments. This approach allows researchers to benchmark model outputs against real observations from Australia, providing a robust framework for advancing our understanding of C dynamics in freshwater systems.