Introduction
Despite the importance of river systems for water supply and other
ecosystem services, such as regulation of nutrient cycles (e.g. nitrogen
and phosphorus) and supporting fisheries, they are highly degraded
ecosystems due to anthropogenic stressors such as modified flows,
urbanisation, agriculture and wastewater (Vörösmarty et al., 2010). By
altering physical, chemical and biological components of freshwaters,
anthropogenic interventions play an important role influencing climate
change through greenhouse gas (GHG) emissions. River systems globally
contribute estimated annual CO2 emissions equivalent to
20-24% of fossil fuel emissions, 35-65% of the CH4emissions from all sources, and 4-5% of N2O total
emissions (Rosentreter et al., 2021, Friedlingstein et al., 2022, Battin
et al., 2023). However, global estimates of river GHG emissions remain
highly uncertain, due to sparse data availability and inconsistent
monitoring practices, perpetuating large gaps in international emissions
inventories and preventing effective management (Rudee & Phillips,
2021). The Paris Agreement signed at the UN Climate Change Conference
(COP21) in 2015 recognized the crucial need to quantify GHG emissions
from sources and sinks that have not yet been adequately quantified.
More effective river catchment monitoring and management are needed
urgently for countries to respond effectively to global climate change
threats by better managing carbon emissions.
Quantifying aquatic carbon cycle processes is challenging. Processes
such as photo-oxidation, metabolism (production, respiration) and
methanogenesis can be estimated from dissolved gas measurements, organic
matter degradation assays, GHG emissions (e.g., floating chambers) or
from dissolved gas concentrations relative to atmospheric concentrations
(Duc et al., 2013, Appling et al., 2018, Aho et al., 2021). However,
most studies have collected short-duration datasets in-situ, at small
numbers of sites, with low temporal resolution. Estimates of
photo-oxidation and decomposition with experimental manipulations are
also typically resolved at weekly-to-monthly timescales. Even where
daily-to-weekly sampling takes place, it often occurs at selected
locations during daylight hours, or misses important events such as flow
peaks (Bieroza et al., 2023). Thus, we lack a clear understanding of how
river stressors and management activities influence emission ‘hotspots’
in space, and/or ‘hot moments’ in time (Zhang et al., 2021b), risking
either over- or under-estimation of emissions. Recent reviews and
opinion articles have broadly outlined a need for global river
observation systems for river carbon monitoring (Battin et al., 2023,
Dean & Battin, 2024) but lacked details on how these networks could be
implemented. Here we evaluate how recent advances in autonomous (field
deployable and wireless) sensor networks, and robotic mobile sensing
platforms, can be harnessed to meet this requirement by combining
high-frequency, continuous data at multiple locations, with machine
learning (ML) models to improve carbon emission estimates and overall
water management in river networks.
The emergence of sensor technologies for high-resolution space/time
monitoring offers the potential to evaluate fundamental linkages between
hydrological regime, physicochemical conditions and nutrient dynamics to
fill knowledge gaps in understanding processes related to carbon
emissions. Links between river physical properties, network structure
and ecosystem carbon cycle parameters, including metabolism, have
advanced notably with Cole et al.’s (2007) concept of “leaky pipes”
for carbon loss along the land-ocean aquatic continuum (LOAC), and the
Pulse-Shunt Concept, which added transport vs reaction timescales
related to flow (Raymond et al., 2016). Wollheim et al. (2018) proposed
a similar River Network Saturation (RNS) concept, describing how river
networks become saturated with carbon at high flows, particularly in
low-order streams, where terrestrial carbon is “pulsed” to river
networks and “shunted” downstream because high flows restrict time for
uptake reactions in quantifiable amounts. Thus, most annual downstream
carbon export occurs during a small number of high flow events (Raymond
et al., 2016). At low flows, particularly in high-order rivers, carbon
uptake fluxes and subsequent emissions are much higher as transport
timescales are long and reactions can occur by photomineralization and
co-metabolism on bio-aggregates (Battin et al., 2008). Continuous
measurements of dissolved oxygen have enabled many of these advances in
understanding river carbon cycling processes of primary production and
respiration, but the spatial distribution of monitoring systems remains
limited and globally unbalanced. For example, across North America, the
relatively widespread availability of sensor data (Figure 1), has
promoted an understanding of key drivers of river carbon cycling and
CO2 emissions, as demonstrated through the StreamPULSE
project (Appling et al., 2018). A range of datasets are also collected
in regional initiatives (e.g. Figure 1b, c) yet for large parts of the
world, including much of the global south, we still have only patchy
knowledge of the parameters needed to quantify carbon transformations
and emissions, or data collected are not open access (Dean & Battin,
2024). Even in countries with advanced sensor networks, there are still
large gaps spatially between sensor locations (Fig 1b), and high-order,
poorly mixed rivers, which present challenges to developing
representative datasets, unless multiple sensors are deployed across
river cross sections.
Sensor network developments can improve our understanding of spatial and
temporal carbon dynamics significantly (Segatto et al., 2023) but cost
prevents monitoring all rivers. Coupling sensor developments with
advances in fixed sensor technology and data analytics, as well as
mobile robotics and ML, will be vital to achieve spatially continuous
data and interpolate spatially explicit datasets to derive whole
catchment understanding (O’Grady et al., 2021, Khandelwal et al., 2023).
By automating sensors using computer science advances and telemetry
systems, it is becoming possible to monitor, in near real-time, how
aquatic ecosystems are functioning. Additionally, the Internet of Things
(IoT) offers significant potential in delivering up-to-date water
quality data with a high level of precision and accuracy, enabling the
detection of even minor fluctuations in water quality. IoT facilitates
the connection of various instruments, including electronic devices and
sensors, utilising the communication infrastructure and cloud computing
resources already in place (Amador-Castro et al., 2024). This offers the
potential to validate existing carbon dynamic scientific models and
develop the next generation of catchment-scale numerical predictive
models. There is now a potential for a step-change in adaptive
management, moving away from current low-resolution, relatively slow
turnover data collection, with delayed analytics that impede effective
decision-making, to faster and more accurate workflows, even at national
scales. This will subsequently enable scientists to advance emission
quantifications at national to global levels and develop intervention
plans.